RADA PROGRAMOWA / PROGRAM COUNCIL PRZEWODNICZĄCY / CHAIRMAN: prof. dr hab. dr h.c. mult. Czesáaw CEMPEL Politechnika PoznaĔska REDAKTOR NACZELNY / CHIEF EDITOR: prof. dr hab. inĪ. Ryszard MICHALSKI UWM w Olsztynie CZàONKOWIE / MEMBERS: prof. dr hab. inĪ. Jan ADAMCZYK prof. dr hab. Wojciech MOCZULSKI AGH w Krakowie Politechnika ĝląska prof. dr. Ioannis ANTONIADIS prof. dr hab. inĪ. Stanisáaw NIZIēSKI National Technical University Of Athens – Greece UWM w Olsztynie dr inĪ. Roman BARCZEWSKI prof. Vasyl OSADCHUK Politechnika PoznaĔska Politechnika Lwowska – Ukraine prof. dr hab. inĪ. Walter BARTELMUS prof. dr hab. inĪ. Stanisáaw RADKOWSKI Politechnika Wrocáawska Politechnika Warszawska prof. dr hab. inĪ. Wojciech BATKO prof. Bob RANDALL AGH w Krakowie University of South Wales – Australia prof. dr hab. inĪ. Lesáaw BĉDKOWSKI prof. dr Raj B. K. N. RAO WAT Warszawa President COMADEM International – England prof. dr hab. inĪ. Adam CHARCHALIS prof. Vasily S. SHEVCHENKO Akademia Morska w Gdyni BSSR Academy of Sciences MiĔsk – Belarus prof. dr hab. inĪ. Wojciech CHOLEWA prof. Menad SIDAHMED Politechnika ĝląska University of Technology Compiegne – France prof. dr hab. inĪ. Zbigniew DĄBROWSKI prof. dr hab. inĪ. Tadeusz UHL Politechnika Warszawska AGH w Krakowie prof. dr hab. inĪ. Marian DOBRY prof. Vitalijus VOLKOVAS Politechnika PoznaĔska Kaunas University of Technology – Lithuania prof. Wiktor FRID prof. dr hab. inĪ. Andrzej WILK Royal Institute of Technology in Stockholm – Sweden Politechnika ĝląska dr inĪ. Tomasz GAàKA dr Gajraj Singh YADAVA Instytut Energetyki w Warszawie Indian Institute of Technology – India prof. dr hab. inĪ. Jan KICIēSKI prof. Alexandr YAVLENSKY IMP w GdaĔsku J/S Company "Vologda Bearing Factory" – Russia prof. dr hab. inĪ. Jerzy KISILOWSKI prof. dr hab. inĪ. Bogdan ĩÓàTOWSKI Politechnika Warszawska UTP w Bydgoszczy prof. dr hab. inĪ. Daniel KUJAWSKI Western Michigan University – USA

Wszystkie opublikowane prace uzyskaáy pozytywne recenzje wykonane przez dwóch niezaleĪnych recenzentów. Obszar zainteresowania czasopisma to problemy diagnostyki, identyfikacji stanu technicznego i bezpieczeĔstwa maszyn, urządzeĔ, systemów i procesów w nich zachodzących. Drukujemy oryginalne prace teoretyczne, aplikacyjne, przeglądowe z badaĔ, innowacji i ksztaácenia w tych zagadnieniach.

WYDAWCA: ADRES REDAKCJI: Polskie Towarzystwo Diagnostyki Technicznej Uniwersytet WarmiĔsko – Mazurski w Olsztynie 02-981 Warszawa Katedra Budowy, Eksploatacji Pojazdów i Maszyn ul. Augustówka 5 10-736 Olsztyn, ul. Oczapowskiego 11 tel.: 089-523-48-11, fax: 089-523-34-63 REDAKTOR NACZELNY: www.uwm.edu.pl/wnt/diagnostyka prof. dr hab. inĪ. Ryszard MICHALSKI e-mail: [email protected] SEKRETARZ REDAKCJI: KONTO PTDT: dr inĪ. Sáawomir WIERZBICKI Bank PEKAO SA O/Warszawa CZàONKOWIE KOMITETU REDAKCYJNEGO: nr konta: 33 1240 5963 1111 0000 4796 8376 dr in . Krzysztof LIGIER Ī NAKàAD: 350 egzemplarzy dr inĪ. Paweá MIKOàAJCZAK Wydanie dofinansowane przez Ministra Nauki i Szkolnictwa WyĪszego DIAGNOSTYKA’3(47)/2008

SPIS TREĝCI / CONTENTS

Sáowo wstĊpne...... 3

Bogdan ĩÓàTOWSKI – UTP Bydgoszcz ...... 5 Ku WyĪynom Nauki - Portret prof. zw. dr hab. Czesáaw CEMPEL, dr h. c. mult.

Zbigniew Witold ENGEL – AGH Kraków...... 13 Wkáad profesora Czesáawa CEMPELA w rozwój wibroakustyki

Marian W. DOBRY – Politechnika PoznaĔska ...... 19 Rozwój diagnostyki i kongresy diagnostyki technicznej w Polsce

Czesáaw CEMPEL – Politechnika PoznaĔska...... 23 Generalized Singular Value Decoposition in Multideimensional Condition Monitoring of Systems Uogólniony rozkáad wartoĞci szczególnych w wielowymiarowej diagnostyce maszyn –propozycja diagnostyki porównawczej

Jan M. KOĝCIELNY, Michaá SYFERT, Bolesáaw DZIEMBOWSKI – Politechnika Warszawska ...... 31 The Issue Of Symptoms Arising Delays During Diagnostic Reasoning Problem uwzglĊdnienia dynamiki powstawania symptomów we wnioskowaniu diagnostycznym

Roald TAYMANOV, Ksenia SAPOZHNIKOVA – D. I. Mendeleyev Institute for Metrology (VNIIM) St. Petersburg, Russia ...... 37 Automatic Metrological Diagnostics Of Sensors Automatyczna diagnostyka metrologiczna czujników

Valentyn SKALSKY, Oleg SERHIYENKO, Yevhen POCHAPSKY, Roman PLAKHTIY and Roman SULYM – National Academy of Sciences of Ukraine, Lviv...... 43 Portable Multi-Channel Device For Acoustic Emission Monitoring Of Structures And Products

V. DIAZ, D. GARCÍA-POZUELO, M. J. L. BOADA, A. GAUCHIA – Universidad Carlos III de Madrid...... 47 Steering Inspection By Means Of Tyre Force Measure

Walter BARTELMUS, Radosáaw ZIMROZ – Politechnika Wrocáawska...... 55 Problems And Solutions In Condition Monitoring And Diagnostic Of Open Cast Monster Machinery Driving Systems Problemy i rozwiązania w diagnostyce ukáadów napĊdowych maszynowych gigantów podstawowych górnictwa odkrywkowego

Elias A. YFANTIS, Panaghiotis KAPASAKIS – Hellenic Naval Academy, Piraeus – Greece...... 61 Marine Gas Turbine Performance Diagnostics: A Case Study

Zbigniew KORCZEWSKI – Akademia Marynarki Wojennej w Gdyni...... 65 Endoscopic Diagnostics Of Marine Engines Diagnostyka endoskopowa silników okrĊtowych

Nadiia BOURAOU, Olexandr ZAZHITSKIY – National Technical University of Ukraine "KPI"...... 71 Decision Making Of Aircraft Engine Blades Condition By Using Of Neural Network At The Steady- State And Non-Steady-State Modes

Alexander YAVLENSKY, Alexander BELOUSOV, Gleb ROGOZINSKY, Alexander VOLKOV – Saint-Petersburg State University...... 75 Digital Cinema Diagnostic System Based On Spectral Analysis And Artificial Intelligence Methods

Vasily SHEVCHENKO, Michael ZHILEVICH, Alexander KOROLKEVICH – National Academy Of Sciences, of Belarus ...... 79 Features Of Algorithms Of Diagnosing And Maintenance Of Reliability Of Hydraulic Drives Of Machines 2 DIAGNOSTYKA’3(47)/2008 SPIS TREĝCI / CONTENTS

Vitalijus VOLKOVAS, Ramunas GULBINAS, Edmundas Saulius SLAVICKAS – Kaunas University of Technology, Lithuania ...... 83 Evaluation Of Quality Of Heterogeneous Mechanical Systems Using Impedance Method

Igor DRUZHININ, Ksenia SAPOZHNIKOVA, Roald TAYMANOV – D. I. Mendeleyev Institute for Metrology St. Petersburg...... 89 Diagnostics Of Control Rod Drive. Possibilities To Extend Its Lifetime At Npp’s Diagnostyka napĊdu prĊta sterującego. MoĪliwoĞci zwiĊkszenia czasu eksploatacji napĊdów `w elektrowniach atomowych

Ryszard MICHALSKI – Uniwersytet WarmiĔsko-Mazurski w Olsztynie ...... 95 Diagnostics In Car Maintenance Diagnostyka w utrzymaniu pojazdów samochodowych

Bogdan ĩÓàTOWSKI – UTP Bydgoszcz ...... 101 Technical Diagnostics Of Folded Objects. Directions Of Development Diagnostyka záoĪonych obiektów technicznych. Kierunki rozwoju

Andrzej WILK, Bogusáaw àAZARZ, Henryk MADEJ – Politechnika ĝląska ...... 111 Gear Fault Detection Using Vibration Analysis Wykrywanie uszkodzeĔ przekáadni na podstawie analizy drgaĔ

Štefan LIŠýÁK, Pavol NAMEŠANSKÝ – University of Žilina, Slovakia...... 117 Human And Computer Recognition Of Nighttime Pedestrians

Asja SOKOLOVA, Felix BALITSKY – Russian Academy of Science ...... 121 On Some Critical Machinery Vibration Monitoring Algorithm And Its Application For Incipient Fault Detection And Localization

Stanisáaw NIZIēSKI – Uniwersytet WarmiĔsko-Mazurski w Olsztynie, Wáodzimierz KUPICZ – Wojskowy Instytut Techniki Pancernej i Samochodowej w Sulejówku...... 125 Method For Diagnosing Internal Combustion Engines With Automatic Ignition On The Basis Of Torque Measurement In Traction Conditions Metoda diagnozowania silników spalinowych o zapáonie samoczynnym na podstawie pomiaru momentu obrotowego, w warunkach trakcyjnych

Wojciech BATKO – AGH University of Science and Technology in Kraków ...... 137 Weight And Significance Of Object Dynamic Models In Constructing Monitoring Systems Rola i znaczenie modeli dynamicznych obiektu w budowie systemów monitorujących

Tadeusz UHL – AGH University of Science and Technology in Kraków...... 143 Mechatronics In Diagnostics Mechatronika w diagnostyce

Zbigniew DĄBROWSKI – Politechnika Warszawska ...... 153 About Need For Using Nonlinear Models In Vibroacoustic Diagnostics O koniecznoĞci stosowania w diagnostyce wibroakustycznej modeli nieliniowych

Stanisáaw RADKOWSKI – Politechnika Warszawska...... 157 Vibroacoustic Monitoring Of Mechanical Systems For Proactive Maintenance Monitoring wibroakustyczny systemów mechanicznych w proaktywnej strategii eksploatacji

I. NIKOLENKO – National Academy of Environmental Protection and Resort Development Krimea, A. OLEJNICHENKO – Joint-Stock Company “Strojhydravlika” Odessa, Ukraine...... 165 Substantiation Of Structure And Parameters Of Hydraulic Stands With Recuperation Of Capacities For Diagnostics Of Adjustable Hydromachines

Warto przeczytaü / Worth to read ...... 170 DIAGNOSTYKA’ 3(47)/2008 3 Słowo wstępne

Słowo wstępne Tak wówczas pisał Jubilat na łamach DIAGNOSTY jako informatora liczącego 8 stron W niniejszym wydaniu Diagnostyki 3(47)/2008 maszynopisu formatu B5, którego głównym mottem prezentowane są prace, na których podstawie było „diagnoza−geneza−prognoza – podstawą zostaną wygłoszone referaty w 11 sesjach każdej decyzji”. plenarnych IV Międzynarodowego Kongresu Następnie z inicjatywy Jubilata począwszy od nr Diagnostyki Technicznej w tym referaty wygłoszone 23(2000r.) Informator DIAGNOSTA został z okazji 70 rocznicy urodzin prof. Czesława Cempla. przekształcony w czasopismo naukowo-techniczne W kongresie udział wezmą przedstawiciele DIAGNOSTYKA. z następujących krajów: Anglii, Australii, Białorusi, Zasługi Jubilata, jako przewodniczącego rady Grecji, Hiszpanii, Kolumbii, Litwy, Nepalu, programowej czasopisma są nieocenione, w sensie Niemiec, Rosji, Słowacji, Ukrainy, USA oraz merytorycznym i koncepcyjnym. Do redakcji z Polski. Diagnostyki Jubilat stale zgłaszał nowe propozycje Przedstawione publikacje w tym wydaniu w zakresie treści i formy czasopisma. Pozwoliło to Diagnostyki obejmują podstawowe i kluczowe osiągnąć wysoki poziom i uznanie w środowisku zagadnienia na obecnym etapie rozwoju diagnostyki naukowym między innymi wyrażone 4 punktową technicznej w zakresie: budowy modeli, nowych ocena Diagnostyki na liście MN i SW. metod i środków diagnostycznych. Począwszy od To wyjątkowy Jubilat, określenia kierunków rozwoju diagnostyki, roli Rozpatrując dynamikę i akustykę, i znaczenia modeli dynamicznych w budowie Z zapałem wtargnął w diagnostykę, systemów monitorowania maszyn, mechatroniki By symptomowe toczyć boje, w diagnostyce, dynamikę powstawania symptomów Osiągnął wyżyny energetycznej ewolucji maszyn. we wnioskowaniu, metod i algorytmów Co zaś się tyczy Jubilata, diagnozowania wybranych układów i zespołów To tu wyznać muszę szczerze, maszyn oraz pojazdów. Że to jest wyjątkowy naukowiec, Pierwsze cztery prace poświęcone są Bo w to co robi – wierzy szczerze. Jubilatowi – prof. Czesławowi Cempelowi. A że w nauce zrobił wiele, Zestawiony dorobek naukowy Jubilata jest To nikt tutaj nie zaprzeczy, imponujący, zarówno w liczbach publikacji jak Że to wszystko ma sens i jest do rzeczy. i osiągnięciach naukowych w zakresie: Czesławie – Tyś wielki w Wibroakustyce − dynamiki i wibroakustyki maszyn i systemów, i Diagnostyce. − diagnostyki wibroakustycznej maszyn, Boś wiele zrobił na tym polu, − energetycznej teorii ewolucji maszyn, Brawo Ci za to – Jubilacie, w imieniu własnym − koncepcji wielowymiarowej informacji i przyjaciół wielu, diagnostycznej w ujęciu symptomowej macierzy Abyś nadal trzymał szyk i styl przez następnych obserwacji, wiele, wiele lat. − zastosowaniu teorii szarych systemów do prognozowania. Prof. Czesław Cempel posiada wyjątkową Redaktor naczelny osobowość i talent organizowania licznej rzeszy Ryszard MICHALSKI uczonych do rozwiązywania istotnych problemów badawczych. Tak zrodziła się polska szkoła diagnostyki technicznej w latach siedemdziesiątych XX wieku i Informator Polskiego Towarzystwa Diagnostyki Technicznej pt. „DIAGNOSTA”. W pierwszym numerze z 1990 roku ówczesny Prezes Czesław Cempel tak pisał (...) zainteresowanie diagnostyką techniczną w kraju sięga 1977 roku, kiedy to liczna grupa entuzjastów nowo powstałej dziedziny zebrała się w Białym Borze (...). Zatem celem działania PTDT, skupiającego naukowców, praktyków i wytwórców przyrządów diagnostycznych, jest rozpoznawanie znanych i sprawdzonych metod i środków diagnostycznych, a także stymulacja opracowań nowych. W zadaniach tych niebagatelną rolę będzie odgrywał Informator, jako płaszczyzna wymiany myśli między twórcami metod diagnostycznych, środków diagnostycznych a użytkownikami (...).

DIAGNOSTYKA’ 3(47)/2008 5 Ku WyĪynom Nauki – Portret Profesora ...

KU WYĩYNOM NAUKI – PORTRET PROFESORA prof. zw. dr hab. Czesáaw CEMPEL, dr h. c. mult. Politechnika PoznaĔska, Czáonek Korespondent POLSKIEJ AKADEMII NAUK

Bogdan ĩÓàTOWSKI

Uniwersytet Technologiczno-Przyrodniczy 85-796 Bydgoszcz, ul. Prof. S. Kaliskiego 7, e-mail: [email protected]

Są takie chwile w Īyciu czáowieka, kiedy w Katedrze Akustyki i Teorii DrgaĔ Uniwersytetu. zupeánie nieoczekiwanie pojawia siĊ okazja, by Tu zdobywa pierwsze doĞwiadczenia dydaktyczne dokonaü okresowego podsumowania dokonaĔ kogoĞ i badawcze, co zwielokrotnione dalej wytĊĪoną bliskiego, zaufanego i prawego w kaĪdej sytuacji. pracą są moĪliwe do obserwacji na drodze rozwoju Mam taką przyjemnoĞü i zaszczyt jako uczeĔ naukowego. Profesora sprawozdaü pokrótce o dokonaniach Mistrza, co wiąĪe siĊ z Jego 70 – rocznicą urodzin.

1. Dane biograficzne Profesor Czesáaw CEMPEL urodziá siĊ dnia 22 lipca 1938 roku w Biskupicach Zabarycznych koáo Ostrowa Wielkopolskiego, na zapadáej wsi w Wielkopolsce – jak pisze sam Profesor. Dom z 1812 roku, pokryty sáomą, szkoáa podstawowa z jĊzykiem angielskim, pierwsze badania áączenia rabarbaru z áopianem, to tylko wyrwane wątki wspomnieĔ tamtych miejsc i lat.

Studia przeplatane wojskiem…

Od roku 1964 pracuje w Politechnice PoznaĔskiej, na początku w Katedrze Mechaniki bĊdąc asystentem prof. E. KaraĞkiewicza, a po reorganizacji do chwili obecnej w Instytucie Mechaniki Stosowanej. Byáy to trudne lata pracy, początki badaĔ naukowych, nowe i trudne wyzwania zajĊü dydaktycznych, karkoáomne problemy organizacji laboratoriów – z czym máody pracownik I tak to siĊ zaczĊáo… zmierzyá siĊ z duĪym zapaáem i dobrymi efektami. Studia wyĪsze odbyá na Wydziale Matematyki Fizyki i Chemii Uniwersytetu Adama Mickiewicza w Poznaniu uzyskując w roku 1962 stopieĔ magistra fizyki w zakresie Drgania i Akustyka. MáodzieĔcze lata studiów przeplatane byáy róĪnymi epizodami, najczĊĞciej wytĊĪoną pracą, takĪe wojskiem jak i odpoczynkiem. Byá to trudny okres dla máodego czáowieka, naznaczony wieloma wyzwaniami, co do miejsca pobytu, czasu historii trudnych losów naszego kraju, jak i poszukiwania sposobu na przyszáe, godziwe Īycie. Wizja swojego rozwoju wyniesiona z domu, uksztaátowana przez Rodziców Profesora to drogowskaz postĊpowania, który dziĊki uporowi, czasami na przekór chwilowym trudnoĞciom, widokom i opcjom daáa oczekiwane, a nawet niespodziewanie wiĊksze efekty. WytĊĪona praca i niekiedy wypoczynek máodego… Po studiach w latach 1962 - 64 pracowaá jako asystent prof. M. Kwieka i prof. E. KaraĞkiewicza 6 DIAGNOSTYKA’ 3(47)/2008 Ku WyĪynom Nauki – Portret Profesora ...

StopieĔ naukowy doktora nauk technicznych czynienia z Czáowiekiem o duĪym formacie nadaáa mu Rada Wydziaáu Budowy Maszyn uczonego, uznanego w Ğrodowisku naukowym i od Politechniki PoznaĔskiej na podstawie rozprawy początku związanym z Politechniką PoznaĔską. doktorskiej p.t.: Drgania ukáadów prĊtowych Wymienione poniĪej odznaczenia są tylko dowodem z nieliniowymi warunkami brzegowymi w roku na to, Īe mamy do czynienia ze wspaniaáym 1968. Tutaj teĪ w marcu roku 1971 obroniá badaczem i wybitnym uczonym jednoczącym wiele rozprawĊ habilitacyjną na temat Okresowe dziedzin nauki i wiele spojrzeĔ na wspóáczesne drgania z uderzeniami w dyskretnych ukáadach problemy nauki i Īycia. Te gáówne wyróĪnienia to: mechanicznych. Od roku 1972 jest docentem x KrzyĪ Kawalerski Orderu Odrodzenia Polski w Instytucie Mechaniki i tworzy tam Laboratorium 1983 r.; DrgaĔ i Haáasu, a w roku 1974 zostaje mianowany x Medal Komisji Edukacji Narodowej 1990r.; wicedyrektorem Instytutu i Kierownikiem Zakáadu x KrzyĪ Oficerski Orderu Odrodzenia Polski 2000r. Dynamiki i Wibroakustyki Maszyn, w skáad którego Oczywistym jest, Īe trudne chwile mozolnej wchodzi utworzone poprzednio Laboratorium, i wyróĪniającej pracy Profesora mieszają siĊ ze (obecnie Zakáad Wibroakustyki i Biodynamiki sporadycznymi epizodami wesoáego Īycia Systemów). Profesor Cempel peániá funkcje zastĊpcy towarzyskiego z przyjacióámi i rodziną. dyrektora Instytutu przez wiele kadencji, bĊdąc Za wyróĪniającą pracĊ Profesor otrzymaá 12 równieĪ jego dyrektorem w kadencji 1987-90. nagród Ministra Nauki Szkolnictwa WyĪszego (lub W latach 1993 – 99 byá dziekanem Wydziaáu Edukacji w stosownym czasie zmienionego Budowy Maszyn i Zarządzania, naonczas wydziaáu nazewnictwa) indywidualnych i zespoáowych. o 6 -ciu kierunkach studiów, czterech jednostkach wydziaáowych. Zreformowaá on w tym czasie 2. OSIĄGNIĉCIA NAUKOWE strukturĊ organizacyjną Wydziaáu i wprowadziá dwie nowe specjalnoĞci; Studia Techniczno Handlowe Obszar badaĔ i osiągniĊü badawczych profesora oraz Mechatronika, obie finansowane przez Cempla moĪna zaliczyü do Dynamiki program europejski TEMPUS. Przeprowadziá takĪe i Wibroakustyki Systemów z gáĊbokimi bezpiecznie Wydziaá przez rafy zmian implikacjami i aplikacjami w Dynamice organizacyjnych, zmiany nazwy oraz wprowadziá na i Eksploatacji Maszyn, Analizie Sygnaáów wydziale skuteczną reformĊ finansowania, czyniąc Wibroakustycznych, Diagnostyce Maszyn Wydziaá samowystarczalnym, na drodze do obecnej i Systemów Technicznych, Teorii i InĪynierii rentownoĞci. Systemów i Ekologii, oraz Metodologii BadaĔ. Pewne doĞwiadczenia z tego obszaru ekonomii Celem systematycznego przedstawienia tych zdobywaá Profesor na niwie Īycia prywatnego, gdzie osiągniĊü warto je podzieliü na kilka obszarów. wspomagany i zachĊcany przez ĪonĊ KrystynĊ (związek maáĪeĔski zawarto w 1963r.) postawiá dom Akustyka maszyn i Ğrodowiska letniskowy wraz z rodziną, zadbaá o parking dla Ten obszar badaĔ otwiera pierwsza publikacja pierwszego wáasnego pojazdu i odpoczywaá tanio prof. Cempla z roku 1964, gdzie wykryto róĪnice nad polskimi jeziorami. w skáadzie widmowym haáasu máyna kulowego do Profesor Cempel uzyskaá tytuá profesora przemiaáu cementu w zaleĪnoĞci od jego zapeánienia nadzwyczajnego w roku 1977, a zwyczajnego i stopnia przemiaáu i zaproponowano wykorzystanie w roku 1985. W roku 1991 zostaá wybrany do tych cech do regulacji máyna. Ogólnie do Komitetu BadaĔ Naukowych, a ponownie na drugą najwaĪniejszych osiągniĊü naleĪy tu opracowanie kadencjĊ w roku 1994 z reelekcją na lata 2000 – korelacyjno - koherencyjnej metody identyfikacji 2004. W roku 1994, w uznaniu dorobku naukowego Ĩródeá haáasu i akustycznych wáasnoĞci pomieszczeĔ zostaje wybrany czáonkiem korespondentem produkcyjnych. ZwieĔczeniem badaĔ w tym Polskiej Akademii Nauk. W uznaniu zasáug obszarze byáo wydanie krajowej monografii naukowych i dobrej wspóápracy z Politechniką WIBROAKUSTYKA STOSOWANA w roku 1978r, SzczeciĔską, Uczelnia ta nadaje mu w roku 1996 przez PWN (II wydanie 1985r), a potem ponownie godnoĞü doktora honoris causa. Natomiast z caákowicie zmienioną zawartoĞcią w 1989r. MoĪna w uznaniu zasáug na polu innowacyjno powiedzieü, Īe ksiąĪka ta pomogáa wykreowaü wdroĪeniowym (14 patentów) i osiągniĊü w zakresie dziedzinĊ Wibroakustyki lansowanej wtedy przez diagnostyki maszyn w 1998 roku zostaje wybrany autora wspólnie z profesorem Z. Engelem z AGH. jako czáonek Akademii InĪynierskiej w Polsce. W roku 2005 Akademia Górniczo Hutnicza nadaje Wibroakustyka narzĊdzi pneumatycznych mu równieĪ godnoĞü doktora honoris causa. Na początku lat 70tych w zespole prof. Publikowany dorobek profesora Cempla liczy KaraĞkiewicza w Politechnice PoznaĔskiej ponad 400 prac opublikowanych w czasopismach rozpoczĊáy siĊ pionierskie badania identyfikacyjne krajowych i zagranicznych zagroĪenia drganiowego związanego (http://neur.am.put.poznan.pl). z uĪytkowaniem máotków pneumatycznych, Ĩródáem O licznych, doniosáych osiągniĊciach i zasáugach choroby wibracyjnej w przemyĞle. Zespoáowa Profesora Ğwiadczy juĪ ten krótki rys danych kontynuacja tych badaĔ wspólnie z M. Dobrym osobowych, z którego wynika, Īe mamy do DIAGNOSTYKA’ 3(47)/2008 7 Ku WyĪynom Nauki – Portret Profesora ...

(obecnie prof. nzw. Politechniki PoznaĔskiej), równaĔ ruchu ukáadów z uderzeniami juĪ w roku pozwoliáa zwolna pokonaü te trudnoĞci i opracowaü 1969. nieliniowy wibroizolator nie przenoszący siá W dalszych pracach z tej dziedziny skupiá siĊ na dynamicznych narzĊdzia z dokáadnoĞcią do siá tarcia tzw. wielomasowych uderzeniowych eliminatorach w jego konstrukcji. drgaĔ. Badając je zaprezentowaá w roku 1979 Poszukiwania ciszy i naleĪnego spokoju po w Journal of Sound and Vibration efektywną wyczerpującej tematyce drgaĔ i haáasu od zawsze i zgodną z eksperymentem ‘koncepcjĊ siáy towarzyszą Profesorowi. równowaĪnej’ reprezentującej ruch wielu mas w pojemniku wielomasowego eliminatora drgaĔ. Badając póĨniej z H. G. Natke (Uniw. Hannover) Ğrutowy eliminator drgaĔ zaprezentowaá równowaĪne podejĞcie energetyczne reprezentujące ruch i oddziaáywania Ğrutu i umoĪliwiające dodatkowo badanie oddziaáywaĔ i efektywnoĞci eliminatora Ğrutowego. Najnowsza koncepcja, która wykluáa siĊ na kanwie tych badaĔ to zastosowanie automatów komórkowych do modelowania tych zagadnieĔ. Bowiem zwykáe systemy symulacyjne, np. Matlab®, zawodzą w obliczu takiej záoĪonoĞci i nieliniowoĞci.

Diagnostyka wibroakustyczna maszyn Badania w tym obszarze rozpoczĊáy siĊ w zespole prof. Cempla we wczesnych latach 70- tych i byáy prowadzone szerokim frontem, o czym szczegóáowo mówi prof. Z. Engel w nastĊpnym artykule. Ta redundancja, optymalizacja, kumulanty i inne ĩmudna zmiana nocna… wypracowane czĊsto po nocach w samotnoĞci, czĊsto po dáugich dyskusjach spacerowych Wibroizolator wraz z nowym narzĊdziem z przyjacióámi wyzwalają w Profesorze potrzebĊ opatentowano w wielu krajach Ğwiata i rozpoczĊto relaksu na áonie natury. produkcjĊ seryjną narzĊdzi, które jako jedyne w Ğwiecie speániają drganiowe normy bezpieczeĔstwa ISO. Stąd teĪ miedzy innymi biorą siĊĨródáa Bio  Dynamiki uprawianej od dawna w Zakáadzie.

Natura, Īona i nowe pomysáy…

Pionierskie prace Profesora z tego obszaru zyskaáy uznanie Ğwiatowe w postaci referatów plenarnych wielu konferencji i czasopism naukowych. Idąc tak szerokim frontem badaĔ zaproponowaá NiekoĔczące siĊ dysputy z Kolegami… on ogólną MetodologiĊ Wibroakustycznej Diagnostyki Maszyn zawartą w monografiach: Dynamika ukáadów z wibrouderzeniami Podstawy wibroakustycznej diagnostyki maszyn Uderzeniowy charakter pracy wielu maszyn WNT 1982, a szczególnie w Wibroakustyczna technologicznych i brak teorii dynamiki takich  ukáadów ĞciągnĊáy uwagĊ i inicjatywĊ badawczą diagnostyka maszyn  PWN 1989, z táumaczeniem prof. Cempla. Badając te ukáady zaproponowaá niemieckim i angielskim. To przejĞcie od sztuki wpierw dystrybucyjny sposób opisu i rozwiązania pomiaru i intuicji w diagnostyce do nauki 8 DIAGNOSTYKA’ 3(47)/2008 Ku WyĪynom Nauki – Portret Profesora ... i technologii diagnozowania widaü wyraĨnie kosmicznej lub platformie wiertniczej. Narodziá siĊ w pracy zbiorowej Diagnostyka Maszyn – Zasady tu równieĪ nowy pomysá procesora energii Ogólne i przykáady zastosowaĔ, Wyd. ITE. 1982, z podukáadem samo regeneracji, na podobieĔstwo której idea wydania, wspóáautorstwo tego jak to siĊ odbywa w systemach oĪywionych. i wspóáredakcja jest jego pomysáu. Przez dáugie lata byáa ona jedynym kompendium wiedzy teoretycznej Diagnostyka wielouszkodzeniowa i praktycznej w tej szerokiej dyscyplinie. A dopiero Hardwarowe i softwarowe postĊpy w metrologii niedawno wspólnie z prof. B. ĩóátowskim w przestrzeni zjawiskowej maszyny pozwalają opracowaá  równieĪ w ramach pracy zbiorowej zmierzyü praktycznie kaĪdy proces mogący InĪynieriĊ Diagnostyki Maszyn, Wydawnictwo charakteryzowaü stan maszyny; temperatura, ITE, Radom 2005, str. 1111. drgania, moc zasilania, itd. Mamy tu zatem, dostĊpną pomiarowo przestrzeĔ moĪliwych symptomów do charakterystyki pracy i ewolucji Energetyczna Teoria Ewolucji Maszyn uszkodzeĔ nadzorowanej maszyny. JeĞli skojarzyü i Systemów to jeszcze z obecnąáatwoĞcią obliczeĔ, W pracach profesora Cempla nad modelem dekompozycji i transformacji dowolnych macierzy, ewolucji stanu maszyny, niezbĊdnego to mamy moĪliwoĞü ekstrakcji niezaleĪnej w diagnostyce, wyáoniáa siĊ koncepcja informacji uszkodzeniowej; wychodząc z nowo potraktowania rosnących uszkodzeĔ materiaáu, zdefiniowanej symptomowej macierzy obserwacji elementów i podzespoáów maszyny jako obiektu i uogólnionych symptomów uszkodzeĔ. Ta zdyssypowaną i zakumulowaną wewnĊtrznie energiĊ koncepcja ekstrakcji wielowymiarowej informacji a potem koncepcja powiązania tej energii diagnostycznej z symptomowej macierzy z dynamiką i drganiami obiektu. Tak powstaá obserwacji, w skojarzeniu z ukáadami w roku 1985 model Tribo  wibroakustyczny samouczącymi moĪe uáatwiü zaprojektowanie opublikowany pierwotnie w WEAR w Anglii, agenta diagnostycznego, jako elementu samo nastĊpnie w bardziej dojrzaáej formie w Biuletynie diagnostyki systemów mechanicznych PAN, a potem ulepszony w Journal of Mechanical i mechatronicznych. W ostatnim jednak czasie staáa Systems and Signal Processing, i innych. siĊ aktualna w Europie chiĔska metodologia Wspóápracując w tym obszarze wspólnie z prof. związana z teoria szarych systemów. DajĊ ona Natke z Hanoweru i Dr Winiwarterem z Bordalier równieĪ dobrą metodĊ prognozowania, co profesor Institute  Francja, udaáo siĊ model ten znacznie Cempel pokazaá ostatnio w swych pracach. uogólniü na inne systemy mechaniczne, a takĪe na inne typy systemów dziaáaniowych. Dla InĪynieria systemów i ekologia mechanicznych procesorów energii (materiaáy, InĪynieria systemów spoáecznych, to równieĪ maszyny, konstrukcje) moĪna pokazaü, Īe wiedza o spoáecznym przetwarzaniu i tworzeniu bezwymiarowy czas Īycia systemu to wprost wiedzy, co nabiera niezwykáej wagi w obliczu odpowiednik prawa Palmgrena  Minera, Odkwista wchodzenia w gospodarkĊ wiedzy. PrzemyĞlenia  Kaczanowa i odpowiednich praw dla innych form swoje w tym wzglĊdzie profesor Cempel zawará zuĪywania siĊ. w kilku publikacjach i w serii wykáadów dla róĪnych Koncepcja procesora energii doprowadziáa do Szkóá WyĪszych po serii spotkaĔ z róĪnymi ludĨmi, sformuáowania pojĊcia czasu Īycia i czasu róĪnymi faktami i osobami. przeĪycia (awarii) procesora, jako miary Szerokie zainteresowania profesora Cempla zdyssypowanej wewnĊtrznie energii, mierzonej od i umiejĊtnoĞü syntezy ujawniáy siĊ ostatnio urodzenia systemu aĪ do jego Ğmierci. UmoĪliwiáo w publikacjach i wykáadach popularnych na temat to z kolei wprowadzenie czasu Īycia innych Ekogospodarki przyszáoĞci i wynikających stąd systemów dziaáaniowych, w ramach, którego wyzwaniach stojących przed szkolnictwem nastĊpuje ewolucja wáasnoĞci systemów (np. zmiana wyĪszym. Bowiem stojąca przed nami zmiana masy, sztywnoĞci, táumienia) na skutek pracy paradygmatu gospodarowania, z obecnego systemu. Daáo to narzĊdzie do sformuáowania Ğrodowisko jest zasobem gospodarki, na nowy Holistycznej Dynamiki Systemów Mechanicznych, gospodarka to czĊĞü Ğrodowiska, musi byü dynamiki ujmującej dwa czasy; ewolucjĊ wáasnoĞci poprzedzone zmianą mentalnoĞci kadry naukowej, systemów w czasie ich dziaáania (Īycia) makro czas, inĪynierskiej i zarządzającej ksztaáconej w szkoáach a takĪe zjawiska dynamiczne i drgania w systemie- wyĪszych. mikro czas. Tak rozumiana holistyczna dynamika Zainteresowania procesorem energii popchnĊáy systemów jest podstawą ksiąĪki opublikowanej prof. Cempla w zagadnienie przetwarzania energii w Ğrodowisku i do koncepcji emergii wprowadzonej wspólnie z profesorem Natke pt. Model  aided diagnosis of mechanical systems, Springer Verlag w latach 80 tych przez Odum’a. Nowa ekogospodarka, a zatem gospodarka energetyczna 1997. Dalszy rozwój tej koncepcji to systemy i materiaáowa musi byü oparta na takim záoĪone z procesorów energii róĪnego bilansowaniu pierwotnie w o onej energii s o ca, przeznaczenia, co moĪe byü podstawą systemów á Ī á Ĕ czyli emergii. Z takiego rachunku bilansowego samodzielnych z reutylizacją energii, jak np na stacji DIAGNOSTYKA’ 3(47)/2008 9 Ku WyĪynom Nauki – Portret Profesora ... moĪna wyciągnąü wiele konkluzji, np. ile ludzi dziaáaniowego przetwarzającego dowolna energiĊ, moĪe utrzymaü nasza geobiosfera. ze zdefiniowanym czasem Īycia poprzez energiĊ dyssypacji, z potencjaáem destrukcji (damage 3. UDZIAà PROFESORA CEMPLA capacity) i niezawodnoĞcią symptomową. W koĔcu W ROZWOJU NAUKI posáuĪyáo to do sformuáowania holistycznej Ukazanie siĊ w roku 1978 ksiąĪki profesora dynamiki ukáadów mechanicznych, umoĪliwiającej Cempla pt.: Wibroakustyka stosowana, pierwszej badania i symulacjĊ zachowania siĊ obiektu w caáym ksiąĪki z tego obszaru w piĞmiennictwie polskim, cyklu jego Īycia, od wytworzenia aĪ do kasacji ugruntowaáo potrzebĊ i przyspieszyáo rozwój i reutylizacji. i formuáowanie siĊ nowego obszaru badaĔ To o tym rozprawiaá Profesor na jednym Wibroakustyki Maszyn. MoĪna z caáą otwartoĞcią z ostatnich Kongresów miĊdzynarodowych, stwierdziü, Īe w kraju obszar ten zostaá poszukując przy tym kawy, dyskusji i spokoju. wypromowany wspólnie z profesorem Engelem z AGH. Podobnie ma siĊ historia z wykreowaniem ksztaácenia o specjalnoĞci Diagnostyka Maszyn w latach 80 tych, jako pokáosie wielu prac z tego zakresu, a takĪe ówczesnego zapotrzebowania na tĊ wiedzĊ i umiejĊtnoĞci przez przemysá. Koncepcje te znalazáy uznanie Ğrodowisk naukowych i spoáecznoĞci akademickiej, co daáo pierwsze zaszczyty honorowe.

Ciągle konieczne wyjazdy naukowe – tu Chiny, Czeng Du …

4. WSPÓàPRACA NAUKOWA

Od wczesnych lat 70–tych, tuĪ po habilitacji, Honoris Causa dla Profesora … profesor Cempel podejmuje wspóápracĊ naukową i organizacyjną z liczącymi siĊ oĞrodkami Nowa koncepcja równaĔ ruchu ukáadów naukowymi w kraju i zagranicą. Na gruncie z uderzeniami opracowana przez profesora Cempla, krajowym jest to przede wszystkim Instytut a nastĊpnie koncepcja energetycznie równowaĪnej Podstawowych Problemów Techniki, z prof. siáy dla wielomasowego eliminatora drgaĔ S. Ziembą i doc. A. MuszyĔską na czele. Na ich z uderzeniami jest wielokrotnie cytowana przez proĞbĊ prof. Cempel podjąá siĊ organizacji róĪnych autorów. Daáo to nową znacznie prostszą MiĊdzynarodowego Kongresu DrgaĔ Nieliniowych moĪliwoĞü analizy ukáadów z wibrouderzeniami, w Poznaniu w 1972 roku. Byáa to wielka promocja gdyĪ podejĞcie ekwiwalentnej siáy zwalnia nauki polskiej a jednoczeĞnie okazja do nawiązania z rozwiązywania dziesiątków nieliniowych równaĔ roboczych kontaktów z uczonymi caáego Ğwiata. Od róĪniczkowych. tego teĪ czasu datowaáa siĊ wspóápraca profesora OsiągniĊcia profesora Cempla Cempla z prof. S. H. Crandal’em z MIT  USA, w wibroakustycznej diagnostyce maszyn są a potem z profesorem R. H. Lyon’em z tej samej widoczne zarówno w kraju jak i zagranicą. uczelni. Kolejny wspóápracujący oĞrodek to Energetyczna teoria ewolucji stanu maszyn Politechnika Warszawska z prof. Z. OsiĔskim i systemów to najĞwieĪsze i miĊdzynarodowo znane i M. Dietrichem na czele, zwáaszcza przy organizacji osiągniĊcie profesora Cempla, a rozpropagowane Letnich Szkóá z Dynamiki Maszyn, gdzie prof. w ostatniej wspóáautorskiej monografii: Model – Cempel byá wielokrotnie wykáadowcą, a materiaáy Aided Diagnosis of Mechanical Systems (Springer wykáadowe byáy publikowane przez Ossolineum. 1997). ZaczĊáo siĊ to od prostego modelu Ewolucji Z tego teĪ krĊgu wyszáa bardzo cenna inicjatywa stanu maszyny sformuáowanego przez prof. Cempla nowego czasopisma Machine Dynamics Problems, w WEAR w 1985, a zaowocowaáo modelem gdzie prof. Cempel jest w radzie redakcyjnej ewolucji Procesora Energii jako modelu systemu a przewodniczyá jej prof. Z. OsiĔski. Po Jego Ğmierci 10 DIAGNOSTYKA’ 3(47)/2008 Ku WyĪynom Nauki – Portret Profesora ... wspóápraca dalej trwa z jego wychowankami, Cempel zacząá wyjeĪdĪaü za granicĊ nie tylko na profesorami Zb. Dąbrowskim i St. Radkowskim. konferencje, ale byá zapraszany równieĪ na wykáady Efektem tego są miĊdzy innymi proszone wykáady lub cykle wykáadów oraz dalsze stypendia, na Studium Doktoranckim Wydziaáu SIMR. w szczególnoĞci DAAD. Ilustruje to poniĪsze Kolejnym oĞrodkiem intensywnej wspóápracy wyliczenie  wykáady: RWTH  Aachen 1982 prof. Cempla byá Instytut Mechaniki i 1983, Instytut Mechaniki AN NRD w Chemnitz i Wibroakustyki obecnego Wydziaáu InĪynierii 1985 i 1987, Politechnika w Birmingham 1985; Mechanicznej i Robotyki AGH. Dotyczy to Uniwersytet Kaiserslautern 1986, 1988 i 1990; szczególnie jego twórcy prof. Z. Engela i jego Uniwersytet Hanower 19861995 (corocznie), 2001; zespoáu. Zespól ten tworzy obecnie trzon MIT  Cambridge 1987; Hawana 1985 i 1987; kierowniczy kilku katedr powstaáych po podziale Shenyang Chiny 1988; Leningrad - Moskwa  Gorki Instytutu w latach 90 tych. Ta wczesna wspóápraca 1989; Oulu Finlandia 1992; Uniwersytet w ParyĪu zaowocowaáa caáym szeregiem wspólnych tematów 1991; Bordalier Institute Francja 1991; Instytut badawczych w ramach Centralnych Programów InĪynierii Mechanicznej CETIM Francja 1991, 1992 Badawczych, raportów i publikacji i innych i 1995, Techn. Universitaet Braunschweig 1998 do materiaáów, dając nowe impulsy rozwojowi 2002, corocznie. Wibroakustyki i Diagnostyki i w kraju.

Poszukiwania dyskusji naukowej i spokoju, teĪ za granicą … PodróĪe, podróĪe, podróĪe i coĞ przy okazji … Stypendia: Britiish Council w Loughborough Uniw. 3 miesiące; DAAD Hannover Uniw. po Wspóápraca ta trwa do chwili obecnej a jej 2 mieĞ.: 1987, 88, 89, 92, 93, 94, 95 (5 miesiĊcy) i rozmiar wystarczy zilustrowaü liczbami, na okoáo aĪ do roku 2001 po 1 miesiąc w roku. Profesor 200 recenzji dorobku, prac na stopieĔ i monografii, Cempel uzyskaá równieĪ stypendium TEMPUS w okoáo 40 dotyczy Wydziaáu InĪynierii Mechanicznej 1991 we Francji na uogólnienia swego procesora i Robotyki AGH. energii i drugie stypendium TEMPUS w Ilyvieska Politechnika SzczeciĔska to nastĊpny oĞrodek Institute of Technology (Finlandia 1995) dla wspóápracy w zagadnieniach Dynamiki Maszyn pogáĊbienia studiów inĪynierii systemów. w szczególnoĞci. Tutaj w nowych doktoratach Prof. Cempel jest czáonkiem szeregu towarzystw i habilitacjach zostaáy przetestowane niektóre naukowych i technicznych, krajowych koncepcje profesora Cempla, jak np. podatnoĞü i miĊdzynarodowych – áącznie 25. szerokopasmowa jako miara dynamicznoĞci ukáadu DziĊki swemu autorytetowi jest zapraszany do mechanicznego bĊdącego modelem konstrukcji, komitetów organizacyjnych konferencji krajowych miary wibroizolacji systemów, nowych definicji i zagranicznych. Jest czáonkiem rad redakcyjnych symptomów w diagnostyce maszyn, i inne. wielu czasopism krajowych i miĊdzynarodowych, Profesor Cempel byá i jest zapraszany do miĊdzy innymi: Zagadnienia Eksploatacji Maszyn, czáonkostwa róĪnych rad naukowych; poprzednio Machine Dynamics Problems, Mechanical Systems OBR áoĪysk Tocznych, WyĪsza Szkoáa Oficerska  nad Signal Processing, Maintenance Management  Piáa, KNIPT Zespóá XIII, Zespóá Dydaktyczny COMADEM, Engineering Mechanics Mechanika przy MEN (1 kadencja), Instytut i przewodniczy radzie programowej czasopisma Akustyki UAM PoznaĔ, Instytut Obróbki naukowo technicznego Diagnostyka. Plastycznej – PoznaĔ, a obecnie Instytut Maszyn Przepáywowych PAN w GdaĔsku. 6. KSZTAàCENIE

5. WSPÓàPRACA ZAGRANICZNA Profesor Cempel prowadziá zajĊcia we wszystkich formach z Mechaniki, WytrzymaáoĞci Po powrocie ze stypendium British Council Materiaáów, Mechaniki Páynów, Wibroakustyki, Diagnostyki, Teorii i In ynierii Systemów w roku 1981 i zakoĔczeniu stanu wojennego, prof. Ī DIAGNOSTYKA’ 3(47)/2008 11 Ku WyĪynom Nauki – Portret Profesora ... i Metodologii BadaĔ. Te dwa ostatnie przedmioty na nowatorski program studiów z takimi przedmiotami studium doktoranckim w Politechnice PoznaĔskiej, jak miĊdzy innymi; Technologie Informatyczne Warszawskiej i ATR, gdzie jednym z efektów jest i Symulacyjne w Badaniach i Ksztaáceniu, InĪynieria wydanie skryptu Wybrane Zagadnienia Systemów, Metodologia BadaĔ. Przedmioty te Metodologii i Filozofii BadaĔ, dostĊpnego obecnie koĔczą siĊ nowatorską forma zaliczenia, wybranym w Internecie i w postaci wydawniczej (Wyd. ITE i opracowanym samodzielnie przez studenta Radom 2003). problemem projektowym wprowadzenia na rynek wyrobu, usáugi, lub efektów swego doktoratu. Profesor Cempel kieruje obecnie Zakáadem Wibroakustyki i Bio-Dynamiki Systemów Instytutu Mechaniki Stosowanej, w ramach którego uruchomiá specjalnoĞü Eksploatacja i Diagnostyka, oraz piĊü laboratoriów badawczo ksztaáceniowych: Pomiarów WielkoĞci Mechanicznych, DrgaĔ i Dynamiki Maszyn, Dynamiki Systemów Biomechanicznych, Diagnostyki Systemów, InĪynierii Wibroakustycznej – w budowie. DąĪąc zaĞ do integracji studiów i studiowania, w obliczu niebywaáego rozwoju technologii informatyczno symulacyjnych, oraz zagroĪenia Ğrodowiska, zaproponowaá obecnie i uruchomiá koncepcjĊ zintegrowanego kierunku ksztaácenia w Politechnice Pierwszy doktorant zagraniczny… pod nazwą EkoinĪynieria, prowadzonym wspólnie przez dwa wydziaáy Politechniki od 2005r. Jest twórcą specjalnoĞci Wibroakustyka Maszyn Taka niebywaáa aktywnoĞü Profesora jest od w Politechnice PoznaĔskiej i opracowaá dla niej trzy zawsze wspierana przez najbliĪszych: ĪonĊ skrypty; Drgania Mechaniczne, Wibroakustyka, KrystynĊ, najbliĪszą RodzinĊ i niekiedy przez Diagnostyka Maszyn, a dwa ostatnie po Przyjacióá. uzupeánieniach staáy siĊ krajowymi monografiami. Byá opiekunem ponad 170 prac dyplomowych, promotorem 17 prac doktorskich, konsultantem i opiekunem wielu habilitacji z wibroakustyki i diagnostyki w kraju i w swym zespole. Byá opiniodawcą wielu wniosków i awansów profesorskich w caáym kraju.

W trudnych chwilach Īmudnej pracy naukowca zawsze moĪna na Nią liczyü…

Sylwetka profesora Cempla opisana jest wielokrotnie w amerykaĔskich i angielskich wydawnictwach biograficznych. To samo dotyczy Doktoranci krajowi, teĪ wojskowi… polskojĊzycznych wydawnictw biograficznych. Byá proszony wielokrotnie o recenzje dorobku W czasie kierowania Wydziaáem Budowy w sprawie nadawania honorowego doktoratu przez Maszyn i Zarządzania wykreowano dwa 3 letnie róĪne uczelnie: Politechnika àódzka dla profesora granty europejskie TEMPUS na ksztaácenie Z. OsiĔskiego, AGH dla profesora Z OsiĔskiego inĪynierów w kierunku Techniczno Handlowym, i profesora Z Engela, Politechnika Lubelska dla oraz w kierunku Mechatronika. Zmodernizowano profesora M. Kleibera, Politechnika SzczeciĔska dla takĪe w tym czasie caákowicie program studiów prof. J. Doerfera, dla profesora A. H., Nayfeh inĪyniersko - magisterskich na kierunku Mechanika z Virginia Polytechnic Institute USA, prof. i Budowa Maszyn dla caáego Wydziaáu. L. KobyliĔskiego z Politechniki GdaĔskiej, prof. Podczas jego kadencji ustanowiono studia K. Marchelka z Politechniki SzczeciĔskiej doktoranckie na Wydziale, ustanawiając dla nich 12 DIAGNOSTYKA’ 3(47)/2008 Ku WyĪynom Nauki – Portret Profesora ...

(dwukrotnie), prof. J. Nizioáa z Politechniki uporządkowane Jego materiaáy – bo któĪ inny lepiej Krakowskiej. ogarnie zasiĊg tych dokonaĔ. Profesor CEMPEL jest wybitną jednostką. PODSUMOWANIE Wedáug mojej oceny Jego osiągniĊcia dotychczasowe moĪna oceniü takĪe na podstawie JakĪe trudno syntetycznie napisaü dziaáalnoĞci Jego uczniów. Caáy ich zastĊp realizuje o dokonaniach wielu lat Īycia tak wybitnego idee i oczekiwania swojego Mistrza. uczonego i naszego Profesora. Zasáugi i odkrycia Pod kierownictwem i przy udziale Profesora Profesora, jakich dokonaá w swym niezwykle polska DIAGNOSTYKA, WIBROAKUSTYKA pracowitym Īyciu są wrĊcz uderzające, a opisane I TEORIA SYSTEMÓW wysunĊáy siĊ na czoáowe zostaáy w oparciu o udostĊpnione mi i bardzo miejsca w Ğwiecie. ĩycie samo dopisze dalszy scenariusz ...

Peány tekst referatu wraz ze spisem monografii, ksiąĪek i skryptów oraz wykaz waĪniejszych publikacji Profesora jest dostĊpny w materiaáach IV MiĊdzynarodowego Kongresu Diagnostyki Technicznej. DIAGNOSTYKA’ 3(47)/2008 13 ENGEL: Wkáad Profesora Czesáawa CEMPELA w rozwój wibroakustyki

WKàAD PROFESORA CZESàAWA CEMPELA W ROZWÓJ WIBROAKUSTYKI

Zbigniew Witold ENGEL

Akademia Górniczo-Hutnicza, Katedra Mechaniki i Wibroakustyki Al. Mickiewicza 30, 30-059 Kraków, e-mail: [email protected] Centralny Instytut Ochrony Pracy, ul. Czerniakowska 16, 00-701 Warszawa

1. WSTĉP maszyny, obiektu, urządzenia. NastĊpnie podaü naleĪy charakterystyki Ĩródeá, okreĞliü Nowa dziedzina wiedzy wibroakustyka, której wspóázaleĪnoĞü miĊdzy poszczególnymi Wspóátwórcą jest Profesor Czesáaw Cempel Ĩródáami, okreĞliü moc akustyczną, a takĪe podaü powstaáa w naszym Kraju okoáo 40 lat temu. Byáo to charakter generacji drgaĔ i dĨwiĊków. moĪliwe, gdyĪ w Polsce rozwijaáy siĊ róĪne dziaáy b. Identyfikacja dróg transmisji energii nauki w tym szeroko pojĊta mechanika. Powstaáy wibroakustycznej w okreĞlonym Ğrodowisku znaczące oĞrodki zajmujące siĊ problematyką (budowlach, obiektach, maszynach, urządzeniach drganiową, powstaáy silne oĞrodki akustyki. itp.). Opracowanie teorii transformacji Do powstania wibroakustyki przyczyniáa siĊ i przenoszenia energii, rozdzielenie sygnaáów polska szkoáa teorii drgaĔ oraz krakowska szkoáa wibroakustycznych, opracowanie biernych drganiowa tworzona przede wszystkim przez i czynnych metod kontroli zjawisk, opracowanie profesorów Stefana ZiembĊ, Wáadysáawa Bogusza, metod analizy na pograniczu falowego Zbigniewa OsiĔskiego, Romana Gutowskiego, i dyskretnego ujĊcia zjawisk. Janisáawa SkowroĔskiego, Kazimierza Piszczka oraz c. Diagnostyka wibroakustyczna wykorzystująca innych. Drugą dyscypliną naukową, która sygnaáy emitowane przez maszyny, urządzenia w znaczący sposób przyczyniáa siĊ do powstania itp. Sygnaáy wibroakustyczne zawierają wibroakustyki byáa akustyka. Prace Marka Kwieka, informacje o stanie zdrowia, stanie budowli, Edmunda KaraĞkiewicza, Ignacego Maleckiego, obiektu, maszyny i urządzenia. Te wáasnoĞci Stefana Czarneckiego i wielu innych doprowadziáy sygnaáów są czĊsto wykorzystywane zarówno do rozwoju tej dziedziny nauki. w diagnostyce medycznej jak równieĪ Rozwój mechaniki i jej branĪ, akustyki i innych w diagnostyce maszyn i urządzeĔ oraz dziedzin naukowych pozwoliáy na realizowanych procesów technologicznych, wykrystalizowanie siĊ nowej dziedziny wiedzy – a takĪe w badaniach nieniszczących. Zasady wibroakustyki. diagnostyki wibroakustycznej stosowane są Wibroakustyka jest nową dziedziną nauki w kaĪdej fazie istnienia maszyn i urządzeĔ: zajmującą siĊ wszelkimi problemami drganiowymi w konstruowaniu, wytwarzaniu i eksploatacji i akustycznymi zachodzącymi w przyrodzie, oraz przy sterowaniu procesami technice, maszynach, urządzeniach, Ğrodkach wibroakustycznymi. transportu i komunikacji, a wiĊc w Ğrodowisku. d. Synteza wibroakustyczna maszyn, obiektów oraz sygnaáów. Zadania podzieliü moĪna na dwie 2. WSPÓàCZESNE ZADANIA grupy zagadnieĔ: WIBROAKUSTYKI -synteza parametrów opisujących pole akustyczne, wzglĊdnie synteza wielkoĞci Znając definicjĊ wibroakustyki moĪna stosowanych metodach aktywnych, synteza w przedstawiü jej cel: „Celem utylitarnym akustyce mowy; wibroakustyki jest obniĪenie zakáóceĔ -synteza maszyn i obiektów, przez co rozumiemy wibroakustycznych maszyn, urządzeĔ, instalacji syntezĊ strukturalną, kinematyczną oraz ich otoczenia do minimum moĪliwego na i dynamiczną prowadzącą do uzyskania danym etapie wiedzy i technologii, a takĪe odpowiedniej aktywnoĞci wibroakustycznej. wykorzystanie informacji zawartych w sygnale e. Czynne zastosowanie energii wibroakustycznej. wibroakustycznym do oceny jakoĞci maszyn Procesy wibroakustyczne nie zawsze muszą byü i urządzeĔ, budowli itp. oraz realizowanych procesami szkodliwymi. Zastosowane celowo procesów technologicznych. Wykorzystanie sygnaáu przy uĪyciu odpowiednich Ğrodków dla celów diagnostyki medycznej”. zabezpieczających mogą byü efektywnym Dla tak sformuáowanego celu moĪna podaü noĞnikiem energii, która moĪe byü wykorzystana podstawowe zadania wibroakustyki. Na rysunku 1 do realizacji róĪnych procesów przedstawiono schematycznie te zadania. Są to: technologicznych. Czynne zastosowanie energii a. Identyfikacja Ĩródeá energii wibroakustycznej, wibroakustycznej związane jest z kontrolo- która polega na zlokalizowaniu Ĩródeá w obrĊbie wanym wykorzystaniem tej energii, 14 DIAGNOSTYKA’ 3(47)/2008 ENGEL: Wkáad Profesora Czesáawa CEMPELA w rozwój wibroakustyki

Rys. 1. Zadania wibroakustyki przy warunku maksymalnej efektywnoĞci szybko rozwijający siĊ i mający juĪ szereg energetycznej i minimalnych zakáóceniach praktycznych zastosowaĔ. zewnĊtrznych. Wszystkie maszyny i urządzenia, obiekty Energia wibroakustyczna wykorzystana moĪe znajdujące siĊ w Ğrodowisku tworzą záoĪony ukáad byü równieĪ dla celów medycznych przy tzw. fizyczny, który pozwala przez zastosowanie „terapii wibroakustycznej”. odpowiednich uproszczeĔ przejĞü do modelu f. Opracowanie metod kontroli emisji, propagacji mechanicznego, a nastĊpnie do modelu i eimisji energii wibroakustycznej w Ğrodowisku, wibroakustycznego. Modelowanie wibroakustyczne w tym równieĪ w maszynach i urządzeniach, naleĪy równieĪ do waĪnych zadaĔ wibroakustyki. a takĪe opracowanie metod sterowania procesami wibroakustycznymi, co siĊáączy 3. MASZYNY I URZĄDZENIA JAKO z tzw. metodami aktywnymi. Podstawową cechą PRZETWORNIKI ENERGII ukáadów aktywnych jest to, Īe zawierają one zewnĊtrzne Ĩródáo energii. Ukáady te Profesor Czesáaw Cempel swoich pracach [15, odpowiednio sterowane mogą dostarczaü lub 16] pokazaá, Īe wszelkie maszyny i urządzenia są absorbowaü energiĊ wibroakustyczną przetwornikami energii. Na rys. 2 pokazany jest w okreĞlony sposób z dowolnych miejsc ukáadu. model takiego przetwornika. Metody sterowania procesami Celowo skonstruowane obiekty (maszyny, wibroakustycznymi stanowią nowy dziaá nauki urządzenia, budowle) dla wykonania okreĞlonego zadania traktujemy jako systemy dziaáaniowe. Są to DIAGNOSTYKA’ 3(47)/2008 15 ENGEL: Wkáad Profesora Czesáawa CEMPELA w rozwój wibroakustyki systemy otwarte z przepáywem masy (materiaáu), WewnĊtrzna dyssypacja energii w kaĪdym energii i informacji. MoĪna wiĊc stwierdziü, Īe są to systemie dziaáaniowym ma charakter kumulacyjny, ukáady transformujące energiĊ z nieodáączną jej determinujący stan tego systemu. Dyssypacja energii dyssypacją wewnĊtrzną i zewnĊtrzną. WejĞciowy wynika z tytuáu zachodzących w systemie procesów strumieĔ masy, energii i informacji na energiĊ zuĪyciowych jak: zmĊczenie we wszystkich uĪyteczną w postaci innej jej formy lub produktu formach, tarcie, erozja w strumieniu cząstek oraz oraz na energiĊ niepoĪądaną, dyssypowaną, która korozje wszelkiego typu, a takĪe páyniĊcie jest czĊĞciowo emitowana do Ğrodowiska, materiaáu, zwáaszcza przy wysokich temperaturach, a czĊĞciowo akumulowana w obiekcie jako efekt áącznie z peázaniem przy wysokich obciąĪeniach. Te róĪnych procesów zuĪyciowych zachodzących procesy są przyczynkami sumarycznej dyssypacji podczas jego pracy. Zaawansowanie procesów energii. zuĪyciowych determinuje jakoĞü kaĪdego obiektu Sumaryczną dyssypacjĊ energii w systemie Ed technicznego i nosi nazwĊ stanu technicznego. Stan moĪna wyraziü nastĊpująco [15]: techniczny moĪe byü okreĞlony poprzez obserwacjĊ T przeksztaáconej energii tj. energii uĪytecznej Ed (ș) = ³ Nd [V(ș),ș] dș” Edb,V «Nd i energii niepoĪądanej. 0

gdzie: Nd - intensywnoĞü dyssypacji (moc); ș - czas dziaáania(Īycia) obiektu; V(ș) – moc dyssypacji zewnĊtrznej; Edb - pojemnoĞü dyssypacji energii przed zniszczeniem systemu; WartoĞü intensywnoĞci dyssypacji energii zaleĪy od czasu dziaáania (Īycia) obiektu ș oraz od mocy dyssypacji zewnĊtrznej. Prof. Czesáaw Cempel wykazaá, Īe caákowita moc dyssypowana, a takĪe Rys. 2. Maszyna jako system przetwarzania energii moc dyssypacji zewnĊtrznej rosną monotonicznie w funkcji czasu Īycia ș dąĪąc do nieskoĔczonoĞci Analizując energiĊ dyssypowaną obserwujemy dla czasu awarii. Pokazany na rys. 3 model ewolucji róĪnego typu tzw. procesy resztkowe róĪnego typu stanu ukáadu transformującego energiĊ moĪna np. wibroakustyczne, termiczne, elektromagne- stosowaü do opisu zmian stanu eksploatacyjnego. tyczne itp. niezamierzone przez projektanta. Ujawnia siĊ w tym fraktalna natura przeksztaácania Obserwacja wyjĞü daje duĪe moĪliwoĞci energii. diagnozowania stanu technicznego z jednej strony, zaĞ z drugiej minimalizacjĊ czynników ujemnie wpáywających na Ğrodowisko, ale takĪe na sam obiekt.

zwyĪkująca lub uĪytkowa moc Nu wejĞciowa dla nastĊpnego Ni wyĪszego poziomu transformacji Nd moc wejĞciowa caákowita lub wartoĞü moc energii rozproszona E Ğmierü v zewnĊtrznie T sprzĊĪenie 4 eksportowana s zwrotne WEWNĉTRZNA moc MOC dostrzegalne rozproszona symptomy ROZPROSZENIA tkowy ZGROMADZONA ą lub wektor JAKO ED symptomów Ed czas pocz 4 narodziny

Rys. 3. Model energetyczny systemu dziaáaniowego 16 DIAGNOSTYKA’ 3(47)/2008 ENGEL: Wkáad Profesora Czesáawa CEMPELA w rozwój wibroakustyki

Koncepcja procesora energii doprowadziáa do Springer staáy siĊ one podstawowymi podrĊcznikami sformuáowania pojĊcia czasu Īycia i czasu z tego zakresu w Ğwiecie naukowym. przeĪycia, jako miary zdyssypowanej wewnĊtrznie C. Cempel stwierdzaá, Īe diagnostyka to energii, mierzonej od urodzenia systemu aĪ do jego zorganizowany zbiór metod i Ğrodków do oceny Ğmierci. UmoĪliwiáo to z kolei wprowadzenie czasu stanu technicznego obiektów, maszyn, urządzeĔ. Īycia innych systemów dziaáaniowych, w ramach W wiĊkszoĞci przypadków są to systemy którego nastĊpuje ewolucja wáasnoĞci systemów. dziaáaniowe generujące i przenoszące procesy Daáo to narzĊdzie do sformuáowania przez wibroakustyczne. Profesora Cempela „Holistycznej Dynamiki Systemów Mechanicznych”, dynamiki ujmującej dwa czasy: ewolucjĊ wáasnoĞci systemów w czasie ich dziaáania – makro czas, a takĪe zjawiska dynamiczne i drgania w systemie mikro czas.

4. IDENTYFIKACJA ħRÓDEà ENERGII WIBROAKUSTYCZNEJ

Jednym z podstawowych zadaĔ wibroakustyki jest identyfikacja Ĩródeá generacji energii wibroakustycznej. Identyfikacja ta obejmuje miĊdzy innymi lokalizacjĊ poszczególnych Ĩródeá, okreĞlenie charakterystyk i wspóázaleĪnoĞci Rys. 5. Rodzaje i cele diagnostyki wibroakustycznej. pomiĊdzy poszczególnymi Ĩródáami.. W wielu przypadkach przyjmowaáo siĊ, Īe maszyna lub Procesy te są noĞnikami informacji o stanie urządzenie jest pojedynczym Ĩródáem tej energii. technicznym. Stan techniczny obiektu podawany jest Takie przyjĊcia byáy duĪym uproszczeniem, gdyĪ w kategoriach jakoĞci i bezpieczeĔstwa jego nawet prosta maszyna czy urządzenie posiada od dziaáania poprzez wektor miar bezpoĞrednich lub kilku do kilkuset elementarnych Ĩródeá. poĞrednich. Miary bezpoĞrednie są np. wymiary, Zagadnieniami identyfikacji Ĩródeá energii parametry technologiczne itp. Miary poĞrednie stanu wibroakustycznej zajmowaáo siĊ szereg osób technicznego noszą nazwĊ symptomów, czyli w naszym kraju, miĊdzy innymi Prof. Cz. Cempel, wielkoĞci wspóázmienniczych z miarami stanu który na drodze teoretycznej i doĞwiadczalnej technicznego. prowadziá tego typu badania. Na podkreĞlenie W diagnostyce maszyn naleĪy wyróĪniü cztery zasáuguje stosowanie przez niego do identyfikacji rodzaje zastosowaĔ: diagnostykĊ konstrukcyjną, metod koherencyjnych. Na rys. 4 przedstawiony jest diagnostykĊ kontrolną, diagnostykĊ eksploatacyjną schemat ukáadu do identyfikacji Ĩródeá przy oraz diagnostykĊ procesów (rys. 5). Celem zastosowaniu metod koherencji, zastosowany przez diagnostyki konstrukcyjnej dokonywanej na etapie Profesora Cempela. badaĔ prototypu jest identyfikacja Ĩródeá zakáóceĔ wibroakustycznych jako zjawisk Ğwiadczących o niedociągniĊciach projektowych i konstrukcyjno – montaĪowych, a takĪe identyfikacja wáasnoĞci dynamicznych. Celem diagnostyki kontrolnej jest ocena jakoĞci wytworzonych elementów i podzespoáów maszyn i urządzeĔ, natomiast diagnostyka eksploatacyjna ma za zadanie ocenĊ bieĪącego i przyszáego stanu eksploatacyjnego maszyn i urządzeĔ w trakcie ich eksploatacji. Rys. 4. Schemat zastosowanej procedury Wreszcie diagnostyka procesów technologicznych ma na celu ocenĊ jakoĞci i etapu procesu. 5. DIAGNOSTYKA WIBROAKUSTYCZNA Problemy badawcze i aplikacyjne diagnostyki wibroakustycznej są wielorakie i coraz lepiej Na specjalne podkreĞlenie zasáuguje wkáad rozwijane za pomocą wspóáczesnych Ğrodków Profesora Cempela w rozwój diagnostyki pomiarowych, technologii informatycznych oraz wibroakustycznej maszyn i urządzeĔ. Czesáaw sztucznej inteligencji. Do nich zaliczyü moĪna: Cempel stworzyá podstawy teoretyczne diagnostyki - wybór miejsca pomiaru sygnaáu wibroakustycznej, pokazaá jej praktyczne wibroakustycznego, co decyduje o zawartoĞci zastosowania. Wyniki swoich badaĔ opublikowaá sygnaáu uĪytecznego; w kilku monografiach i publikacjach - wybór sposobu przetwarzania sygnaáu dla w renomowanych czasopismach naukowych. uzyskania symptomu stanu; Monografie te byáy przetáumaczone na jĊzyki obce - ekstrakcja caáoĞciowej informacji diagnostycznej i opublikowane przez takie wydawnictwa jak ze zbioru symptomów stanu i selekcja DIAGNOSTYKA’ 3(47)/2008 17 ENGEL: Wkáad Profesora Czesáawa CEMPELA w rozwój wibroakustyki

rozwijających siĊ niezaleĪnych uszkodzeĔ LITERATURA w obiekcie; - ocena zaawansowania uszkodzeĔ i podjĊcie decyzji 1. Cempel C.: Podstawy wibroakustycznej o stanie obiektu. diagnostyki maszyn. WNT Warszawa,1982. KaĪdy z tych cząstkowych celów diagnostyki 2. Cempel C. Wibroakustyka stosowana. PWN wibroakustycznej moĪe byü rozwiązywany na wiele Warszawa 1978. sposobów i w róĪnych zakresach wypeánienia 3. Cempel C.: Diagnostyka wibroakustyczna zadania. maszyn. Wyd. Politechniki PoznaĔskiej 1985. Reasumując, naleĪy stwierdziü, Īe identyfikacja 4. Cempel C. Wibroakustyczna diagnostyka maszyn. uszkodzeĔ w diagnostyce wibroakustycznej odbywa PWN Warszawa,1989. siĊ poprzez pomiar specjalnych uszkodzeniowo 5. Cempel C.: Vibroakustische Maschinen- zorientowanych symptomów. Profesor Cempel jako Diagnostik. Verlag Technik Berlin 1990. pierwszy zdefiniowaá te wielkoĞci oraz dodaá kilka 6. Cempel C. Vibroacoustic Condition Monitoring. áatwo mierzalnych jak np. wspóáczynnik Horwood, Chichester, New York,1991. impulsowoĞci, luzu, czĊstoĞci Rice’a, wspóáczynniki 7. Cempel C:. The Tribovibroacoustical Model of harmonicznoĞci. Zaproponowaá softwarową Machines. Wear 105, str. 297-305, 1985. procedurĊ dyskryminacji i klasyfikacji uszkodzeĔ 8. Cempel C.: The Wear Process and Machinery maszyn poprzez wybór zbioru symptomów Vibration – a Tribovibroacoustic Model. Bull.of o minimalnej redundancji opisujących stan maszyny. the Pol. Acad.of Sc. vol. 35, nr. 7-8, str. 347-363, Procedura ta oparta byáa o metody rozpoznawania 1987. obrazów, a w szczególnoĞci o metodĊ skáadowych 9. Cempel C.: Theory of Energy Transforming gáównych macierzy uszkodzeĔ. Prof. C. Cempel Systems and their Application in Diagnostics of wprowadziá równieĪ do zastosowaĔ w diagnostyce Oprerating Systems. Applied Mathematics and wibroakustycznej pojĊcie niezawodnoĞci Computer Science, vol. 3, nr. 3, str. 533-548, symptomowej. 1993. Omówione problemy diagnostyki byáy i są 10. Cempel C.: Energy Processors in Systems przedmiotem prac badawczych Profesora Cempela, Engineering and their Evolution. Bull. Of the a takĪe licznych Jego publikacji w tym monografii. Pol. Academy of Sc. Vol. 45, nr 4, str. 495-511, 1997. 6. PODSUMOWANIE 11. Cempel C.: Diagnostically Oriented Measures of Vibroacoustical Processes. Journal of Sound and Profesor Czesáaw Cempel jest wspóátwórcą Vibration 73(4), str. 547 -561, 1980. wibroakustyki – nowej dziedziny wiedzy. OkreĞliá 12. Cempel C. Modele ewolucji systemów. nie tylko podstawowe zadania i metody Problemy Eksploatacji nr. 3, 47- 69, 1997. wspóáczesnej wibroakustyki, lecz równieĪ pokazaá 13. Cempel C., Kosiel U.: Noise Sources praktyczne zastosowania. DziaáalnoĞü naukowa Identification in Machines and Mechanical i publikacyjna Czesáawa Cempela z zakresu Devices. Archives of Acoustics 1,1, str. 5-19, wibroakustyki moĪe byü podzielona na kilka grup: 1976. -wibroakustyka maszyn i Ğrodowiska; 14. Cempel C., Engel Z.: Wibroakustyka maszyn, -wibroakustyka narzĊdzi rĊcznych, gáównie geneza dziedziny i jej podstawowe problemy. pneumatycznych; Zeszyty Naukowe AGH, nr. 728, str. 13-17, -dynamika ukáadów wibrouderzeniowych; 1979. -diagnostyka wibroakustyczna. 15. Cempel C., Lotz G.: Efficiency of Vibrational W ramach tych grup zagadnieĔ, Prof. Czesáaw Energy Dissipation by Moving Shot. Journal of Cempel opublikowaá duĪą iloĞü prac. Wyniki swoich Structural Engineering, Vol 119, no 9, str. 2642- badaĔ przedstawiaá na licznych kongresach, 2652, 1993. konferencjach i sympozjach naukowych. Byá 16. Cempel C., Natke H. G.: The Modeling of wielokrotnie zapraszany do wygáaszania referatów Energy Transforming and Energy Recycling plenarnych na Ğwiatowych kongresach. Systems. Journal of Systems Engineering. Vol. 6, Szerokie, róĪnorodne Jego osiągniĊcia naukowe str. 79-88, 1996. i wdroĪenia z zakresu wibroakustyki przyczyniáy siĊ 17. Engel Z., Cempel C.: Vibroacoustics and its do rozwoju tej dziedziny wiedzy nie tylko w Polsce, Place in Science. Bull. Of the Pol. Acad. of Sc. lecz równieĪ na caáym Ğwiecie. Vol 49, nr 2, str. 185-198, 2001. 18 Natke H. G., Cempel C.: Model-Aided Diagnosis of Mechanical Systems. Springer Verlag, 1997. 18 DIAGNOSTYKA’ 3(47)/2008 ENGEL: Wkáad Profesora Czesáawa CEMPELA w rozwój wibroakustyki

Prof. zw. dr hab. inĪ. Zbigniew ENGEL jest wieloletnim profesorem AGH w Krakowie i Centralnego Instytutu Ochrony Pracy w Warszawie. Jest specjalistą zakresu mechaniki i inĪynierii Ğrodowiska, twórca nowej dziedziny nauki – wibroaku- styki. Dorobek publikacyjny prof. Engela obejmuje ponad 550 publikacji, w tym monografie, skrypty, podrĊczniki, artykuáy w znanych czasopismach zagranicznych i krajowych, referaty na konferencjach naukowych. Jest doktorem honoris causa Akademii Górniczo- Hutniczej i Politechniki Krakowskiej. DIAGNOSTYKA’ 3(47)/2008 19 DOBRY: Rozwój Diagnostyki I Kongresy Diagnostyki Technicznej W Polsce

ROZWÓJ DIAGNOSTYKI I KONGRESY DIAGNOSTYKI TECHNICZNEJ W POLSCE

Marian W. DOBRY

Politechnika PoznaĔska, Instytut Mechaniki Stosowanej 60-965 PoznaĔ, ul. Piotrowo3, fax.: 061 665 23 07, [email protected]

Streszczenie W artykule przedstawiono rozwój diagnostyki i MiĊdzynarodowych Kongresów Diagnostyki Technicznej w Polsce. Przedstawiony rozwój dotyczy okresu od początku kumulowania siĊ wiedzy praktycznej i technologii okreĞlania stanu technicznego obiektów technicznych do uzyskania przez diagnostykĊ poziomu naukowego jako nowej odrĊbnej dyscypliny.

Sáowa kluczowe: diagnostyka techniczna, kongresy, historia.

DEVELOPMENT OF DIAGNOSTICS AND DIAGNOSTIC CONGRESSES IN POLAND

Summary Development of machinery diagnostics and International Congresses on Machinery Diagnostics in Poland are presented in the paper. Presented development is concerned with a period since beginning of cumulating practical knowledge and technology defining of technical objects condition to obtainment by Machinery Diagnostics of scientific level as a new independent discipline. Keywords: technical diagnostics, diagnostic congresses, history.

Początek rozwoju diagnostyki technicznej, jako myĞlenie, dlaczego jest tak i jakie to pociąga skutki nauki zaliczanej do inĪynierii mechanicznej, oraz co moĪna udoskonaliü. przypada na lata siedemdziesiąte. Diagnostyka AktywnoĞü w obszarze diagnostyki w przemyĞle techniczna jest zatem stosunkowo máodą nauką. i na uczelniach wyĪszych odnotowano równieĪ W pierwszej poáowie lat siedemdziesiątych w Polsce. Pierwsze warsztaty naukowe na temat wyksztaáciáa siĊ ona w efekcie kumulacji praktycznej diagnostyki maszyn odbyáy siĊ w 1973 roku i byáy wiedzy dotyczącej eksploatacji maszyn oraz metod zorganizowane przez zespól Prof. L. Muellera. Od badawczych pozwalających okreĞliü stan techniczny tego momentu dziedzina diagnostyki maszyn obiektów technicznych w czasie eksploatacji. Nauka w Polsce zaczĊáa gwaátownie wzrastaü. Wystąpiáa ta pojawiáa siĊ na początku, jako naturalna potrzeba potrzeba organizowania wiĊcej kursów, szkoleĔ mierzenia róĪnych parametrów pracy maszyn, które i spotkaĔ, a w 1977 roku powoáano oficjalnie Zespóá mogáyby sygnalizowaü niepoprawną pracĊ maszyny Diagnostyki przy Komitecie Budowy Maszyn PAN. lub jej awariĊ. Zaliczyü do nich moĪna pomiar W organizowaniu nowej dziedziny nauki – czyli temperatury, ciĞnienia oleju w systemach diagnostyki, od początku aktywnie uczestniczyá smarowania, drgania i haáas. Dla tych wielkoĞci Prof. C. Cempel. Jako máodemu profesorowi, fizycznych ustalono wartoĞci dopuszczalne zaproponowano mu prowadzenie Zespoáu i opracowano normy zalecane do stosowania. Diagnostyki. Spotkania odbywaáy siĊ co kwartaá Wszystkim nam znane są np. normy VDI 2056 kaĪdego roku, na których wygáaszano kilka i normy IRD Mechananalysis Inc. PamiĊtamy referatów i wymieniano róĪne idee oraz wiedzĊ. równieĪ pierwszą znaną amerykaĔską ksiąĪkĊ Natomiast, co dwa lata, organizowano tzw. „Szkoáy Bleke’a M. P. i Mitchell W. S.: Vibration and Diagnostyki”. Pierwsza Szkoáa Diagnostyki Acoustic Measurements Handbook (Spartan Books zorganizowana byáa w 1977 r. w Biaáym Borze, 1972), która dotyczyáa bardzo praktycznych spraw zatytuáowana: Diagnostyka UrządzeĔ związanych z pomiarami, ale bez ich postaw Mechanicznych. Kolejne cztery Szkoáy Diagnostyki teoretycznych. Podstawy teoretyczne diagnostyki – II, III, IV i V odbyáy siĊ równieĪ w Biaáym Borze technicznej nie byáy wtedy jeszcze znane. Pierwszą w latach: 1978, 1979, 1980 i w 1981, a ich tematyka ksiąĪką, która w tytule zawieraáa sáowo diagnostyka, byáa nastĊpująca: Podstawy Diagnostyki UrządzeĔ byáa ksiąĪka rosyjska zatytuáowana: Akustyczna Mechanicznych, Metody Cyfrowej Analizy Sygnaáów Diagnostyka Maszyn autorstwa Pavlov’a B. V., Wibroakustycznych, Diagnostyka àoĪysk Tocznych wydana w Moskwie w 1971 roku. W ksiąĪĊce tej oraz Diagnostyka Pojazdów. rozpoczĊto wyjaĞniaü co, dlaczego i kiedy naleĪy Kolejne cztery Szkoáy Diagnostyki VI, VII, VIII zrobiü w diagnostyce maszyn. Byá to początek i IX zorganizowane przez zespóá z Politechniki monitorowania stanu technicznego maszyn PoznaĔskiej odbyáy siĊ w Rydzynie koáo Leszna w praktyce, który rozpocząá racjonalne i naukowe w latach 1983, 1985, 1987 i 1989. PoĞwiĊcone one byáy w kolejnych latach nastĊpującej tematyce 20 DIAGNOSTYKA’ 3(47)/2008 DOBRY: Rozwój Diagnostyki I Kongresy Diagnostyki Technicznej W Polsce sygnalizowanej w ich tytuáach: Komputerowe równieĪ pojĊcie niezawodnoĞci symptomowej. Przetwarzanie Sygnaáów oraz w trzech kolejnych Pozwoliáo ono póĨniej pokazaü jej prosty związek latach – Wnioskowanie Diagnostyczne. z czasem awarii systemu mechanicznego. Byá to Ostatnia X Szkoáa Diagnostyki odbyáa siĊ moment w badaniach, aby zaproponowaü ogólną w 1992 r. w Zajączkowie koáo Poznania. MetodologiĊ Wibroakustycznej Diagnostyki W miĊdzyczasie, przeprowadzono kilka innych Maszyn, która ukazaáa siĊ w monografiach: konferencji dedykując je dla poszczególnych branĪ Podstawy Wibroakustycznej Diagnostyki Maszyn przemysáu lub wyposaĪenia, jak np. silników w 1982 [2] oraz w Wibroakustycznej Diagnostyce wysokoprĊĪnych czy maszyn roboczych ciĊĪkich. Maszyn w 1989 roku [3, 4]. Wymienione wyĪej Cel do osiągniĊcia i zakres diagnostyki byá juĪ koncepcje i rezultaty badaĔ uznano w nauce jako bardzo duĪy, co doprowadziáo do powoáania podstawowe dla diagnostyki WA, co daáo początek Polskiego Towarzystwa Diagnostyki Technicznej nowej nauce w momencie, w którym „sztuka w roku 1990. Liczyáo ono wtedy 150 czáonków pomiaru i intuicja wnioskowania” byáa juĪ dobrze krajowych. znana. Fakt ten, przejĞcia do nauki i technologii W historii rozwoju diagnostyki technicznej diagnozowania jest widoczny szczególnie w pracy w Polsce, Diagnostyka Wibroakustyczna Maszyn zbiorowej zatytuáowanej Diagnostyka Maszyn – (DWA) zajmuje swoje początkowe miejsce. Zasady ogólne i przykáady zastosowaĔ wydane Sformuáowanie zapisu matematycznego uĞredniania drukiem w 1992 [10]. Przez wiele lat byáa ona synchronicznego sygnaáów wibroakustycznych jedynym zasobem wiedzy teoretycznej i praktycznej miaáo miejsce wáaĞnie na początku lat 70-tych. w tej szerokiej dyscyplinie. Taką rolĊ peáni obecnie W celu identyfikacji uszkodzeĔ maszyn ksiąĪka – poradnik liczący 1111 stron, wspólne w diagnostyce WA wprowadzono pomiary dzieáo wielu autorów zajmujących siĊ diagnostyką specjalnych uszkodzeniowo zorientowanych w Polsce pod tytuáem: InĪynieria Diagnostyki symptomów. Do najbardziej znanych zaliczyü Maszyn – Poradnik, wydane w 2005 r. [24]. moĪna zdefiniowanie takie wielkoĞci jak: np. W rozwoju diagnostyki naleĪy równieĪ wspóáczynnik impulsowoĞci, luzu, czĊstoĞci Rice’a, wymieniü prace nad modelem ewolucji stanu wspóáczynniki harmonicznoĞci i róĪnego typu maszyny, niezbĊdnym w diagnostyce, które kumulanty [2]. WielkoĞci te są stosowane do dnia doprowadziáy do koncepcji energetycznej ewolucji dzisiejszego w diagnostyce wibroakustycznej maszyn i systemów spowodowanej przez rosnące maszyn. uszkodzenia materiaáu, elementów i podzespoáów W roku 1980 opracowano softwarową procedurĊ maszyny. Za miarĊ tych uszkodzeĔ przyjĊto dyskryminacji i klasyfikacji uszkodzeĔ maszyn, zdyssypowaną i zakumulowaną wewnĊtrznie poprzez wybór zbioru symptomów o minimalnej energiĊ. redundancji opisujących stan maszyny. Procedura ta W efekcie ww. prac powstaá w roku 1985 model oparta byáa o metody rozpoznawania obrazów, tribo-wibroakustyczny opublikowany pierwotnie a w szczególnoĞci o metodĊ skáadowych gáównych w WEAR [6] w Anglii, a nastĊpnie w Biuletynie (PCA) macierzy obserwacji diagnostycznej i ich PAN i w Journal of Mechanical Systems and Signal odpowiednich wektorów, jako nowych niezaleĪnych Processing. Model ten uogólniono na inne systemy symptomów uszkodzeĔ. Po dwudziestu latach, wraz mechaniczne, a takĪe na inne typy systemów z niebywaáym rozwojem systemów obliczeniowych, dziaáaniowych. W opracowanej w ten sposób teorii które same w sobie zawierają podobne procedury; Procesora Energii poáączono obserwowany np. SVD (Singular Value Decomposition), moĪna symptom stanu (Īycia) procesora z jego byáo powróciü do tej koncepcji proponując nową wewnĊtrznym zaawansowaniem (akumulacją) metodĊ wielowymiarowej i wielouszkodzeniowej uszkodzeĔ z tytuáu dziaáania (Īycia). diagnostyki maszyn, formuáując nową symptomową TeoriĊ tĊ poáączono z wczeĞniej sformuáowanym macierz obserwacji i uogólnione symptomy pojĊciem niezawodnoĞci symptomowej. Dla niezaleĪnych uszkodzeĔ. W klasyfikacji stanu mechanicznych procesorów energii (materiaáy, maszyny pojawiáa siĊ koniecznoĞü okreĞlenia stanu maszyny, konstrukcje) pokazano, Īe bezwymiarowy granicznego w przestrzeni symptomów stanu czas Īycia systemu jest odpowiednikiem prawa obiektu. W tym celu zaproponowano kilka metod Palmgrena – Minera, Odkwista – Kaczanowa opartych o rozkáady wartoĞci obserwowanych i odpowiednich praw dla innych form zuĪywania siĊ. symptomów; np. typu Pareto, Weibulla, Frecheta. Koncepcja procesora energii umoĪliwiáa Opis tych metod moĪna znaleĨü w materiaáach wielu sformuáowanie pojĊcia czasu Īycia1 i czasu konferencji miĊdzynarodowych oraz przeĪycia (awarii) procesora. Są to miary w czasopismach naukowych takich jak: Mechanical zdyssypowanej wewnĊtrznie energii, mierzonej od Systems and Signal Processing, Journal of Sound zaistnienia systemu aĪ do jego likwidacji. Koncepcja and Vibration, Bulletin PAN, i inne krajowe to pozwoliáa na wprowadzenie czasu Īycia innych czasopisma naukowe. systemów dziaáaniowych, w których nastĊpuje W klasyfikacji stanu maszyny nieodzowne jest okreĞlenie stanu granicznego w przestrzeni 1 Taka energetyczna definicja C Cempla zostaáa zamieszczona symptomów stanu obiektu. W tym celu w International Encyklopedia of Systems and Cybernetics, do zastosowaĔ w diagnostyce WA wprowadzono K G Saur, Muenchien, 1997 DIAGNOSTYKA’ 3(47)/2008 21 DOBRY: Rozwój Diagnostyki I Kongresy Diagnostyki Technicznej W Polsce ewolucja wáasnoĞci systemów (np. zmiana masy, Maszyn Przemysáowych w GdaĔsku. sztywnoĞci, táumienia) w czasie pracy systemu. Przewodniczącym KO byá Wojciech CHOLEWA, W ten sposób sformuáowano Holistyczną DynamikĊ wiceprzewodniczącym Jan KICIēSKI. Systemów Mechanicznych, która jest przedmiotem W Kongresie wziĊáo udziaá ponad 350 uczestników ksiąĪki pt. MODEL - AIDED DIAGNOSIS OF i 11 wystawców, wygáoszono 8 wykáadów MECHANICAL SYSTEMS, Springer Verlag 1997 zaproszonych i 24 plenarnych. Opublikowano 140 [10]. Rozwój tej koncepcji prowadzi do systemów artykuáów w 3 tomach materiaáów kongresowych záoĪonych z procesorów energii róĪnego i zaprezentowano 108 prac w formie posterów. przeznaczenia. Podczas trwania Kongresu odbyáo siĊ 7 specjalnych Rozwój informatyzacji diagnostyki technicznej, kursów diagnostycznych, na których byáo a w niej szczególnie informatycznych technik prezentowane wyposaĪenie specjalistyczne. Program pomiarowych, umoĪliwiá precyzyjne pomiary socjalny Kongresu dla uczestników i osób róĪnych wieloĞci fizycznych charakteryzujących towarzyszących byá bardzo dobrze zorganizowany stan maszyny. Zaliczyü do nich moĪna moc w kilku interesujących miejscach wybrzeĪa zasilania, moc obciąĪenia, temperaturĊ, drgania – baátyckiego i w GdaĔsku. amplitudy przyspieszeĔ, prĊdkoĞci i przemieszczeĔ, DuĪy rozgáos dotyczący I Kongresu Diagnostyki haáas, produkty procesów tarciowych w maszynach Technicznej spowodowaá, Īe kolejny itp. W efekcie prowadzonego on-line nadzoru II MiĊdzynarodowy Kongres Diagnostyki diagnostycznego uzyskuje siĊ przestrzeĔ moĪliwych Technicznej zostaá zorganizowany wspólnie przez symptomów sáuĪących do scharakteryzowana pracy zespóá skáadający siĊ z przedstawicieli i ewolucji róĪnych uszkodzeĔ nadzorowanej Politechniki Warszawskiej i Instytutu Energetyki maszyny. Dalsza obróbka uzyskanej w ten sposób w Warszawie w dniach 19-22 wrzeĞnia 2000 roku. bazy danych w nowoczesnych metodach Przewodniczącym KO byá Stanisáaw transformacji i dekompozycji prowadzi do RADKOWSKI, a wice-przewodniczącym Zenon wydobycia niezaleĪnych informacji ORàOWSKI. W Kongresie wziĊáo udziaá ponad uszkodzeniowych. Uzyskuje siĊ w ten sposób 110 uczestników, w tym kilku z zagranicy. Obrady symptomową macierz obserwacji badanego odbywaáy siĊ w 7 sesjach plenarnych z 36 obiektu i uogólnionych symptomów uszkodzeĔ. proszonymi wykáadami, w 4 sesjach posterowych Jest to tematyka wspóáczesnych badaĔ, o których z 87 prezentacjami. Wydrukowano dwa tomy moĪna dowiedzieü siĊ z materiaáów konferencyjnych materiaáów kongresowych i zaáączono do nich dyski np. IMEKO World Congress, June, 2003, Dubrovnik CD. Program socjalny kongresu byá równieĪ [13] i czasopism naukowych takich jak np. interesujący, w którym uwzglĊdniono kilka Mechanical Systems and Signal Processing. najpiĊkniejszych miejsc w Warszawie, jako stolicy Koncepcja wyodrĊbniania wielowymiarowej Polski. informacji diagnostycznej z symptomowej macierzy Kolejny, III MiĊdzynarodowy Kongres obserwacji, w skojarzeniu z ukáadami Diagnostyki Technicznej odbyá siĊ w dniach 6-9 samouczącymi siĊ moĪe uáatwiü zaprojektowanie wrzeĞnia 2004 roku w Poznaniu. Organizatorem agenta diagnostycznego, jako elementu Kongresu byá zespóá skáadający siĊ samodiagnostyki systemów mechanicznych z przedstawicieli Politechniki PoznaĔskiej i mechatronicznych. i Akademii Techniczno – Rolniczej w Bydgoszczy. W zakresie prognozy stanu przyszáego proponuje Przewodniczącym KO byá Marian W. DOBRY, siĊ w diagnostyce technicznej metody zaczerpniĊte a wiceprzewodniczącym Bogdan ĩÓàTOWSKI. z teorii szeregów czasowych i z ekonometrii. Uczestniczyáo w nim ponad 110 uczestników Metody te są obecnie wypierane przez sieci równieĪ kilku z zagranicy, wygáoszono 108 neuronowe. referatów oraz 6 prezentacji dotyczących aparatury W diagnostyce technicznej pojawiáa siĊ równieĪ diagnostycznej wystawianej przez wystawców. Na w ostatnim czasie w Europie chiĔska metodologia III Kongresie wygáoszono 16 referatów plenarnych, związana z teorią szarych systemów. DziĊki niej 92 referaty sesyjne oraz odbyáy siĊ dwie sesje uzyskuje siĊ dobrą metodĊ prognozowania, co panelowe na temat waĪnych zagadnieĔ diagnostyki. wykazano w pracach Cempela. Program socjalny związany byá z prezentacją Polityczne i gospodarcze zmiany w Polsce po lokalnego folkloru oraz zwiedzaniem ciekawych dla roku 1990 nie sáuĪyáy rozwojowi Diagnostyki. uczestników miejsc regionu Wielkopolski. Mimo tego, dalsza edukacja i badania w tej Organizowane co cztery lata MiĊdzynarodowe dziedzinie byáy kontynuowane. Potrzeby Kongresy Diagnostyki Technicznej stają siĊ dobrą zintegrowania Ğrodowiska i wymiany wiedzy tradycją. Pozwalają one dokonaü podsumowania dotyczącej diagnostyki technicznej spowodowaáy, Īe dotychczasowych badaĔ, przeglądu aktualnych Polskie Towarzystwo Diagnostyki Technicznej trendów i nowoĞci w diagnostyce technicznej. zorganizowaáo w dniach 17-20 wrzeĞnia 1996 roku Zadaniem Kongresu jest równieĪ wytyczaü nowe I MiĊdzynarodowy Kongres Diagnostyki kierunki dalszych badaĔ w zakresie diagnostyki Technicznej w GdaĔsku. OrganizacjĊ Kongresu technicznej. Mam nadziejĊ, Īe obecny, powierzono zespoáowi skáadającemu siĊ organizowany przez zespóá z Uniwersytetu z przedstawicieli Politechniki ĝląskiej i Instytutu WarmiĔsko-Mazurskiego w Olsztynie oraz 22 DIAGNOSTYKA’ 3(47)/2008 DOBRY: Rozwój Diagnostyki I Kongresy Diagnostyki Technicznej W Polsce z Akademii Marynarki Wojennej w Gdyni, IV 18. IRD Mechanalysis Inc.: Special treatments of MiĊdzynarodowy Kongres Diagnostyki vibration sources to reduce plant noise. Technicznej w Olsztynie w dniach 9-12 wrzeĞnia Technical Pap. No. 110, 1976 2008 roku, speáni oczekiwania wszystkich 19. KiciĔski J.: Modelowanie i diagnostyka uczestników. Przewodniczącym KO jest Stanisáaw oddziaáywaĔ mechanicznych, aerodyna- NIZIēSKI, a wiceprzewodniczącym Zbigniew micznych i magnetycznych w turbozespoáach KORCZEWSKI. energetycznych. Wyd. Instyt. Maszyn Kierując do Wszystkich Īyczenia owocnych Przepáywowych PAN, GdaĔsk 2005. obrad i miáych spotkaĔ w czasie IV Kongresu – 20. Michalski R.: Diagnostyka maszyn roboczych. pozostając z wyrazami powaĪania ITE. Radom 2004 – Marian W. DOBRY 21. Moczulski W.: Diagnostyka Techniczna. Metody pozyskiwania wiedzy. Wyd. LITERATURA Politechniki ĝląskiej, Gliwice 2002 1. Bartelmus W.: Diagnostyka Maszyn, Górnictwo 22. NiziĔski S., Michalski R.: Diagnostyka Odkrywkowe, Wyd. ĝląsk, Katowice 1998 obiektów technicznych, Wyd. ITE, Radom 2002 2. Cempel C.: Podstawy wibroakustycznej 23. Oráowski Z.: Diagnostyka w Īyciu turbin diagnostyki maszyn, WNT, Warszawa 1982 parowych, WNT, Warszawa 2001 3. Cempel C.: Wibroakustyka Stosowana. PWN 24. ĩóátowski B., Cempel C.: InĪynieria Warszawa-PoznaĔ 1978, wyd. II-gie 1985 Diagnostyki Maszyn - Poradnik, Wyd. ITE 4. Cempel C.: Wibroakustyka Stosowana, wyd. II Radom 2004 zmienione, PWN, Warszawa, 1989 25. ĩóátowski B.: Podstawy diagnostyki maszyn, 5. Cempel C.: Diagnostyka Wibroakustyczna Wydawnictwo Uczelniane Akademii Maszyn. Wyd. Pol. PoznaĔskiej 1985 Techniczno-Rolniczej w Bydgoszczy, 6. Cempel C., The Tribovibroacoustical Model of Bydgoszcz 1996 Machines. Wear 105, 1985, s. 297-305. 26. ĩóátowski B., Jankowski M., ûwik Z.: 7. Cempel C.: Wibroakustyczna Diagnostyka Diagnostyka techniczna pojazdów, Wyd. ATR, Maszyn, PWN Warszawa 1989, Bydgoszcz 1994 8. Cempel C.: Vibroakustische Maschinen- Diagnostik. Verlag Technik, Berlin, 1990, Dr hab. inĪ. Marian Witalis 9. Cempel C.: Vibroacoustic Condition DOBRY, profesor nadzw. - Monitoring, E. Horwood, Chichester - New jest pracownikiem naukowo- York, 1991 dydaktycznym Politechniki 10. Cempel C., Tomaszewski F.: Diagnostyka PoznaĔskiej. Peániá funkcjĊ Maszyn; Zasady Ogólne - Przykáady Prodziekana ds. Ksztaácenia ZastosowaĔ - Poradnik, Wyd. MCNEMT - na Wydziale Budowy Maszyn Radom, 1992 i Zarządzania (2002-2005) 11. Natke H. G., Cempel C.: Model - Aided oraz jest kierownikiem Lab. Diagnosis of Mechanical Systems, Springer Dynamiki i Ergonomii Verlag, 1997 Metasystemu: Czáowiek – Techniczny Obiekt – rodowisko. Jego dziedzina 12. Cempel C.: Fundamentals of Vibroacoustical ĝ Condition Monitoring, chapter 13, (pp. 324 – aktywnoĞci naukowej to: Mechanika, a specjalnoĞci  Mechanika stosowana, Dynamika maszyn i systemów 253) in: Handbook of Condition Monitoring, biologiczno-mechanicznych, Wibroakustyka, Przepáyw edited by A. Davies, Chapman and Hall, energii i rozkáad mocy w systemach mechanicznych, London, 1998 biologicznych i biologiczno-mechanicznych, 13. Cempel C.: Multi fault condition monitoring of Diagnostyka energetyczna wyĪej wymienionych mechanical systems in operation, XVII systemów, Biomechanika, Ergonomia, Ochrona IMEKO World Congress, Dubrovnik June czáowieka i Ğrodowiska przed drganiami i haáasem 2003, pp.1422-1425. w ujĊciu konwencjonalnym i energetycznym. Jest 14. Cholewa W., KiciĔski J.: Diagnostyka Czáonkiem: Sekcji Dynamiki Ukáadów i Biomechaniki techniczna. Metody odwracania nieliniowych Komitetu Mechaniki PAN, Polskiego Komitetu Teorii modeli obiektów. Zeszyt KPKM nr 120, Pol. Maszyn i Mechanizmów PAN oraz Sekcji ĝląska, Gliwice 2001 Technicznych ĝrodków Transportu Komitetu 15. Chudzikiewicz A.: Elementy diagnostyki Transportu PAN. Jest autorem: 1 monografii, kilku pojazdów szynowych, Wyd. ITE, Politechnika rozdziaáów w monografiach, ponad 150 publikacji, Warszawska, Radom 2002 kilkudziesiĊciu opracowaĔ wdroĪonych do przemysáu, 16. Dybaáa J., Mączak J., Radkowski S.: 29 patentów krajowych i zagranicznych (w Polsce, Wykorzystanie sygnaáu wibroakustycznego Europie, i USA) chroniących konstrukcjĊ w analizie ryzyka technicznego, Wyd. ITE, wibroizolatorów WoSSO i wdroĪonych do produkcji Warszawa-Radom 2006 drganiowo i energetycznie bezpiecznych, 17. Engel Z., Plechowicz J., Stryczniewicz L.: ergonomicznych (10 cech) rĊcznych narzĊdzi uderzeniowych (4 wielkoĞci). Podstawy wibroakustyki przemysáowej. Wyd. Wydz. InĪ. Mech. i Rob., AGH, Kraków 2003 DIAGNOSTYKA’ 3(47)/2008 23 CEMPEL, Generalized Singular Value Decoposition In Multideimensional Condition Monitoring Of Systems

GENERALIZED SINGULAR VALUE DECOPOSITION IN MULTIDEIMENSIONAL CONDITION MONITORING OF SYSTEMS

Czesáaw CEMPEL1

PoznaĔ University of Technology, Applied Mechanics Institute, ul. Piotrowo 3, 60-965 PoznaĔ, Poland, email: [email protected]

Summary With the modern metrology, we can measure almost all variables in the phenomenon field of a working machine, and much of measuring quantities can be symptoms of machine condition. On this basis, we can form the symptom observation matrix (SOM) for condition monitoring. From the other side we know that contemporary complex machines may have many modes of failure, so called faults, which form the fault space. This multidimensional problem is not a simple one, even if we apply some modern tool like SVD for the fault extraction purpose. So the question remains if one can learn considering similar problem when having SOM of similar machine observed just before. In this way, we can consider the application of generalized GSVD to the machine condition monitoring problems, and uncover some new possibilities.

Keywords: machine condition, multidimensional, generalized SVD, observation space, fault space, condition similarity.

UOGÓLNIONY ROZKàAD WARTOĝCI SZCZEGÓLNYCH W WIELOWYMIAROWEJ DIAGNOSTYCE STANU SYSTEMÓW

Streszczenie Obecnie potrafimy mierzyü wiĊkszoĞü procesów pola zjawiskowego pracującej maszyny, a wiele z tych procesów moĪe dostarczyü symptomów jej stanu technicznego. Wychodząc stąd moĪemy tworzyü symptomową macierz obserwacji (SOM) do celów diagnostyki maszyn, czyli oceny ewolucji jej stanu technicznego w czasie Īycia ș. Ale wspóáczesne maszyny mają wiele uszkodzeĔ rozwijających siĊ wspóábieĪnie, stąd tez propozycja diagnostyki wielowymiarowej i zastosowania rozkáadu (SVD), co pokazano juĪ w wielu pracach. Powstaje pytanie czy potrafimy uzyskana wiedzĊ wykorzystaü i nauczyü siĊ diagnozowaü lepiej maszyny, które juĪ są rozpoznane diagnostycznie za pomocą SVD. Taki wáaĞni problem postawiono stosując uogólniony rozkáad SVD, umoĪliwiający porównanie dwu macierzy obserwacji, znanej uprzednio i wáaĞnie rozwijającej siĊ. Tak moĪliwoĞü istnieje, a stawia przed nami nowe wymogi nauczenia siĊ nowej semantyki wspólnego jĊzyka GSVD.

Sáowa kluczowe: stan techniczny, wielowymiarowoĞü, uogólnione SVD, przestrzeĔ uszkodzeĔ, przestrzeĔ obserwacji, podobieĔstwo stanu.

1. INTRODUCTION possible application of generalization of SVD method, which takes into account the other SOM of The multidemsionality of fault space in machine the similar object, with the same number of condition monitoring is nowadays well formulated symptoms (columns), but the number of rows and explored, for example by the application of (observation) may differ. This may be the situation neural nets [4], singular value decomposition [6], or of learning from the previous usage of the same principal component analysis [2]. Much worse it object or even similar one. The accessible list of looks when considering the decision making process references to GSVD application is not big one. One in multidimensional case, where we have some can see a few in connection with engineering, but method of data fusion and the concept of symptom there are some papers of GSVD application in reliability applied to generalized fault symptom physics and biology, as we will see later on. In such obtained from the application of SVD [10]. So, there situation the paper introduce the GSVD concept on is a room for looking to other promising methods of the basis of previous SVD application, and from condition symptoms processing and decision making these introductory results one can notice the possible in a multidimensional case. This paper looks for the application of GSVD in machine condition monitoring, particularly when looking for the similarity of machine wear symptoms and indices.

1 GeneralSVD06 - Intended to Diagnostic Congress 08. 24 DIAGNOSTYKA’ 3(47)/2008 CEMPEL, Generalized Singular Value Decomposition in Multidimensional Condition…

2. SINGULAR VALUE DECOMPOSITION Ȉpr = diag ( ı1, …, ıl ), and ı1 > ı2 >…> ıu >0, AND EXTRACTION OF FAULT (3)

SYMPTOMS Vu+1 =… ıl =0, l= max (p, r), u d min (p, r), u <.r < p. Having in mind the above, let us take into Mathematically it can be shown also, that every consideration a critical machine in operation, where perpendicular matrix has such decomposition (2), we have the possibility to observe several ‘would be’ and it may be interpreted also as the product of the symptoms1 of condition. During its working life 0 < three matrices [15], namely T < Tb , (Tb –anticipated breakdown time), several Opr = (Hanger) X ( Stretcher) X (Aligner). (2a) independent faults (usually a few); Ft(T), t = 1,2,..u, This is very metaphorical description of SVD are evolving and growing. Hence, we would like to transformation, but it seems to be useful analogy for identify and assess the advancement of these faults statistical reasoning and diagnostic decision making by forming and measuring the symptom observation in our case. vector; [Sm] = [S1,...,Sr], which may have In terms of machine condition monitoring the components different physically, like vibration above decomposition means, that from the r amplitudes (displacement, velocity, acceleration), primarily measured symptoms (dimension of the temperature, machine load, life time T, etc. observation space) we can extract only z d r In order to track machine condition (faults independent sources of diagnostic information evolution) by these observations, we are making describing evolving generalized faults Ft , creating in equidistant reading of the above symptom vector in this way fault space (see Fig. 1). As it is seen from the lifetime moments; Tn, n = 1, ... p, Tp d Tb , Fig. 1 upper right picture, only a few developing forming in this way the rows of a rectangular faults are making essential contribution to total fault symptom observation matrix (SOM). From the information, the rest of generalized faults are below previous research and papers [6], we know, that the the standard 10% level of noise. What is important best way of SOM preprocessing is to center it here, that such SVD decomposition can be made (subtract), and normalize (divide it) to the symptom currently, after each new observation (reading) of initial value; Sm(0) = S0m , of each given symptom the symptom vector [Sm]; n = 1, … p, and in this (column of SOM). way we can trace the fault life evolution in any It is also known from this research, that amount operating mechanical system. of diagnostic information in SOM increases if we append the lifetime T column, as the first Diagnostic interpretation of SVD results approximation of system logistic vector L and the From the current research and implementation of load [7]. Finally, in order to minimize stochastic this idea [11], we can say, that the most important disturbances in readings we will apply also the three fault oriented indices obtained from SVD is the first points moving average procedure to the successful pair: (SDt , Vt ), t=1,2. This pair presents the lifetime symptom readings, as it was shown and validated in evolution of all independent sources of information the last paper [14]. contained in our SOM. We interpret them as the So, after such preprocessing we will obtain the fault development life curves Ft (T). From the other dimensionless symptom observation matrix (SOM) side we need also some measure of total damage in the form: advancement in diagnosed object in a form Vt(T). The first fault indices SDt can be named as Snm SOM { Opr = [Snm], Snm = 1, (1) discriminant or the generalized symptom of the fault S0m t, and one can get it as the SOM product and the first where bold non italic letters indicate primary singular vector of the matrix V , as below measured and averaged dimensional symptoms. SD = Opr * Vrr = Upp * Ȉpr , (4) As it was already said in the introduction, we and for the one column component of this matrix we apply now to the dimensionless SOM (1), the will have simply

Singular Value Decomposition (SVD) [9, 15], to SDt = Opr * vt = ıt ˜ ut . t=1,...z. (5) obtain singular components and singular values in We know from SVD theory [9, 15], that all the form of matrix formulae; singular vectors vt , ut are normalized to one, so the T Opr = Upp * Ȉpr * Vrr , (2) energy norm of this new discriminant (vector) is (T- matrix transposition ), simply where Upp is p dimensional orthonormal matrix of Norm (SDt) { ¨¨SDt ¨¨= ıt. , t = 1, ...,z. (6) left hand side singular vectors, Vrr is r dimensional If the number of observation is growing in the orthonormal matrix of right hand side singular life time, so the above discriminant SDt(T) can be vectors, and the diagonal matrix of singular values also named as lifetime fault profile, and in turn is as below 6pr singular value Vt(T) as a function of the lifetime seems to be its damage advancement (energy norm).

1 Symptom is a measurable quantity taken from the phenomenal field of the machine, which sees to be correlated to machine condition, we are looking for. DIAGNOSTYKA’ 3(47)/2008 25 CEMPEL, Generalized Singular Value Decomposition in Multidimensional Condition…

Symp.Observ.Matrix with movavg.of A0=sil24d2 Inform.Contrib.of singular values V 300 80 i m %i S Vf 60 s- 200 o p n my oti 40 S u f bi o 100 rt t. n 20 li o p C Sing.Val.Numb.V m 0 .l 0 i A e 0 5 10 15 20 25 R 0 5 10 15 m Sympt.Observ.Matrix;cen&norm.init; A1=sil24d2 1:3 Singul.Compon.-contrib.of primary symptoms S 1.5 4 i s D m S ot 1 f p o 2 m Stright line = life e y d S 0.5 u f ti o pl 0 t. li 0 m p A m .l A -0.5 e -2 .l R e 0 5 10 15 20 25 0 5 10 15 R svdavgopt: Life evolution of SumSD &SD discr. Life Evolut.of Symptom Limit Value S ; for A1=sil24d2 s 4 i i 3.5 l p- SumSD 1 i S +1=3.4811 + l l m SD o i S 3 C. 2 e g ul ni a 2.5 S Vt f 0 i o. m pl i 2 L. m pt A.l -2 m 1.5 e 0 5 10 15 20 25 y 5 10 15 20 25 R S Number of observations used for S calculation Lifetime obs-s l Fig. 1. Illustration of inference possibilities in multidimensional observation by the application of SVD

The similar fault inference can be postulated to If we define; gsvd(A,B) { [U,V,X,C,S], the meaning, and the evolution, of summation than it gives the following decompositions of: quantities, what can mean the total damage profile A = U*C*XT , (8) T SumSDi(T), and total damage advancement and ; B = V*S*X , (9) SumVi(T), as follows; with singular values diagonal matrix: z z -1 SumSD ( ) SD u P( ) , 6 = C * S , ascending ordered, i T ¦ i T ¦V i T ˜ i T T i 1 i 1 (10)

Norm (SumSDi(ș)) {¨¨Ȉ SDi (ș)¨¨d ¨¨Ȉ and additional identity relation: ıi(ș)˜ui(ș)¨¨= Ȉıi(ș) , CT *C + ST * S = I. (11) hence; It maybe important to show, when the GSVD z z Sum ( ) ~ F F . (7) becomes SVD, because we know already diagnostic V i T ¦¦V i T T i T i 11i interpretation of the second decomposition. But it is worthwhile to add, that the meaning of We have from (9); S-1* VT*B = XT . (12) the last relation with Vt(T)eems to be not fully validated experimentally, as yet. It seems to be also, Using it with the decomposition of A as in the that the condition inference based on the first relations (8) one can get; summation measure; Sum(SDi) may stand for the T -1 T T first approach to multidimensional condition A = U*C*X = U*C *S V *B = U*6 *V *B .(13) inference, as it was clearly shown in the previous So, if the auxiliary matrix is the identity matrix, i. e. papers (see for example [17]), and shortly illustrated B = I , on Fig. 1 below. we have the already known SVD, with the properties shown in the paragraph of 2.1 above. 3. GENERALIZED SINGULAR VALUE But using A as a self reference matrix, when DECOMPOSITION GSVD B=A, we obtain from (8) and (9) immediately,

T T The diagnostic application of generalized SVD A = U*C*X = V*S*X . (14) [9], as far as for today, is not known at all. However, it seems to be different from the ordinary SVD. This And it can be possible only if : C=S and U=V. is because we have not one but two symptom Finally, with this assumption we have: observation matrices A and B. Let us assume we 6 = C * S-1 = I . (14a) have primary SOMp = A and auxiliary SOMa = B, Hence, if both matrices primary and auxiliary are and we will try to align SVD decomposition to both identical they singular vales obtained from GSVD matrices, treating the first as primary SOM and the are all equal one! This means that using this property second as auxiliary. Even so, the matrices may we can investigate the similarity between two differ, having different number of rows SOMs, so between respective diagnosed objects. (observations), but they must have the same number From the other side we can investigate if there is of columns (symptoms). some possibility to learn, using already known Going to the definition of GSVD we have knowledge from one object to diagnose the other, following relationships [16]. not known already. 26 DIAGNOSTYKA’ 3(47)/2008 CEMPEL, Generalized Singular Value Decomposition in Multidimensional Condition… But before deep penetration of this possibility, let (U, V) for the matrices defined by GSVD. Using the us create the similar condition related (life) commonly shared numbers of rows of both matrices quantities for both matrices SOMA and SOMB .We (observations) l=min(n,p) for a current life time may try to obtain fault related discriminants from moment T one can write it in a Matlab® notation, GSVD in the same manner as in case of one SOM with some weighting matrix 6; used with usual SVD (see (4) and (5). Cuv = corcoef(U(1:l,: ),V(1:l,: )) *6 (17) Several approaches to accomplish this task was In general, this means, that as for now we have made, and one of the best which gives similar results two independent measures of similarity between to the case of single SOM, (i.e B=I ) is proposed primary and auxiliary SOM. The first measure (16), here as below: shows the angles in the observation spaces (A, B), -1 T -1 Di SDA = A *X *S = U * C * X *X*S , between axes defined by matrices C and S of -1 T -1 GSVD. SDB = B X*S = V * S * X * X *S , (15) and of course matrices have ascending column The second measure (17) shows the same norms, in contrary to ordinary SVD. similarity but seen here as the column by column The above-proposed relations give sometimes correlation coefficients of (U, V) matrices defined a little greater numeric results than for one SOM by GSVD, and weighted by multiplication of case, but the qualitative life course of singular singular value matrix 6.. Such weighting gives much vectors and symptom limit value Sl is much similar. better differentiation of values of similarity Also this is in some agreement with information measures for different objects. contribution (ıt) of the singular vectors vt , ut . There is another possibility of similarity measure From the relation (13) it is seen that in a case calculation, using the vector of singular values taken B z I all matrices of GSVD, that means U, 6 , V , from diagonal matrix 6 as in (10). Extracting its must be aligned to the properties of A and B diagonal, and treating it as the vector we will have it symptom observation matrices. Hence, we can infer in the form Sig=diag(6 ). We know that in the case that application of GSVD in diagnostic may be of identity all singular vales are equal 1. So, it will interpreted also as some kind of learning process. be good if we subtract this identity value from the That mean for example that, based on previous previous vector. Calculating now the norm of such observation (auxiliary SOM =B) we are trying to new vector and normalizing it to the not subtracted compare the current wear process observed by value we can define the index of similarity of SOM primary SOM =A. That is we should look now for matrices in GSVD, as below some measures of similarity between matrices SI = 1 - (Sig -1)T *(Sig-1)/(SigT * Sig) (18) A and B. One can see from the above, that such similarity Having now the possibility of real application of index ranges from zero to unity, being one if both GSVD in condition monitoring let us look for the matrices in GSVD are identical help at the other branches of science, namely We will see these possibilities of inferring from bioinformatics, where one can find already the the previous observations (auxiliary SOM) on the application of GSVD [18]. Following this paper and examples below. This will indicate how these relation (10) with the additional condition of (14) we measures of similarity of matrices A and B behaves, can find, that for identical observation matrices A=B and how sensitive they are to the abbreviated data in all singular values of GSVD are equal unity; 6 = C * a SOM, and to the data taken from the another S-1 = I. If we interpret this as the tangent of the angle object. D between two SOMs (A,B). So in the case of their o 4. COMPARATIVE PROPERTIES OF GSVD identity we have Do = S/4 =45 , and in all other cases we will have some angular measure of similarity IN CONDITION MONITORING, AND differing from the angle 45o . Centering it to zero, POSSIBLE DIAGNOSTIC APPLICATION for the general case of similarity, we can write in AND MEANING OF GSVD a matrix notation, -1 As we have mentioned earlier, generalized SVD 6D = C * S – I = tg D ; and D = arc tg ( 6D ) . (16) can use auxiliary diagnostic observation. So it is Once more, we may suppose from the above, that possible to use another SOM obtained previously if the defined measure is equal zero for a some from the same object, or from the similar diagnosed object. This statement is by analogy to other singular value Vi of primary SOM it may mean that, that the wear process associated with the given applications in computational biology [18], as for the author knowledge, no condition monitoring singular value is similar as it was previously, for the known already case of auxiliary SOM = B . (CM) application is known to this date. Starting at We can invent the other similarity measures. beginning let us take the simplest possible case of the same object treated by specially elaborated Moreover, for the SOM identity case as in (14) we have U = V, and the correlation coefficient program written in Matlab® called gsvdavg.m. In calculated between columns or rows is equal one. addition, we have shown earlier, that for the real CM industrial data with some instability of So, for the general case of A, B matrices we can define correlation coefficient between columns DIAGNOSTYKA’ 3(47)/2008 27 CEMPEL, Generalized Singular Value Decomposition in Multidimensional Condition…

gsvdavg.m; Main Symp.Obs.Matrix; A1=sier1 Auxiliary Symp.Obs.Matrix; B1=sier1 1 1 m m S S s s p- p- my 0 my 0 S S f f o. o. ti ti pl pl m -1 m -1 A 0 5 10 15 20 25 30 A 0 5 10 15 20 25 30 1-4 General.Sympt.of Main SOM; A1=sier1 1-4 General.Sympt.of Auxiliary SOM; B1=sier1 5 5 s s p- p- my my S. S. n n e 0 e 0 Observ-s / Life G G f f o o t. t. li Observ-s / Life li p -5 p -5 m m A 0 5 10 15 20 25 30 A 0 5 10 15 20 25 30 Relat.informat.contrib.of GSC for A0=sier1 Life Evolut.of Sympt.Limit Value S ; for A1=sier1 20 i 1 4 l n + l o S +1=3.9514 ti S l u Auxiliary SOM=sier1 e 3 bi ul rt 10 a n V o t c i i 2 mi Vf L o Ordering numb.of singular value -ascending t. Number of observat-s used for S calculation 0 Vi p 1 l % 1 2 3 4 5 my 10 15 20 25 30 Sing.value similar.meas.of Main/Auxiliary SOM S (UV*S)Corr.similar.measure of Main/Auxiliary SOM 1 1 s e er e u gr s Similarity Index=1 a e e d 0 0.5 ni m. y 0 degr.=>SOM ident.; Auxiliary SOM=sier1 il t if CS=1=>SOM identity ri mi a s. il r mi -1 o 0 S 1 2 3 4 5 C 1 2 3 4 5 Order.number of singul.value -ascending Order.number of generaliz.symptom-ascending Vi Fig. 2. Industrial fan as an example of application of GSVD applied to the same symptom observation matrix symptom readings it is good to apply the moving notice it seems to be quite good evolution of this average operation (avg) for the primary SOM [14]. diagnostically important quantity. In the presented example here, we have taken the The last row of pictures, at the left, shows the huge industrial fan, which pumps the air to shaft of same singular values as above but treated as the the copper mine, and has been working 32 weeks tangent of the angle of similarity between the two with one reading per week of the vibration symptom spaces of symptom observation matrices (16), vector (5 components). Fig. 2 present this example primary A and auxiliary B. And of course for the elaborated by special GSVD program and identical matrices the D angle is equal zero. The subdivided into the 8 pictures described below. bottom right picture gives another measure of The first left top picture presents averaged similarity (17), the transformed correlation centered and normalized primary symptom coefficient between generalized fault symptom of observation matrix. As it can be seen, the variability primary and auxiliary matrices, multiplied of observed symptoms is not a great one, ranging additionally by the respective part of singular value from zero up to r 1, although some of the symptom (S matrix). This is in order to make the measure life curve changes the sign of values, their more sensitive. Of course, for the case of identical oscillations is not a big one, as a result of an matrix this measure is also equal one for every introductory performed averaging operation (avg). singular vale. Therefore, it seems to that in this way The same is shown on the top right picture for the as above, we can investigate the similarity between auxiliary B matrix, and as both were assumed two SOM, and to decide if the fault development identical, it is the same picture as at the top left. The during the machine operation is the same as next two pictures, the second row from the top, previously or only slightly similar to the previous present us the results of GSVD calculation, and here case. we show only the biggest four generalized fault Knowing this let us take not identical SOMs but symptoms Ft(T), t=(1,4), calculated according to similar one, like for example auxiliary SOM the formula (15). They are of course the same, due to same as primary but with smallest number of rows. our identity assumption. The next row of pictures is For the good illustrative purposes it can be the same quite different, from the left one can see singular SOM sier1, but with the smaller number of values, and they are equal each other due to assumed observations. Fig. 3 illustrates this case, and the matrix identity. Further on the right picture presents organization and the meaning of the individual calculated symptom limit value Sl , and one can pictures are the same as previously. 28 DIAGNOSTYKA’ 3(47)/2008 CEMPEL, Generalized Singular Value Decomposition in Multidimensional Condition…

gsvdavg.m; Main Symp.Obs.Matrix; A1=sier1 Auxiliary Symp.Obs.Matrix; B1=sier0 1 1 m m S S s 0.5 s 0.5 p- p- my 0 my 0 S S f f o. -0.5 o. -0.5 ti ti pl pl m -1 m -1 A 0 5 10 15 20 25 30 A 0 5 10 15 20 25 30 1-4 General.Sympt.of Main SOM; A1=sier1 1-4 General.Sympt.of Auxiliary SOM; B1=sier0 5 5 s s p- p- my my S. S. n n e 0 e 0 G G f f o. o. ti ti pl Observ-s / Life pl Observ-s / Life m -5 m -5 A 0 5 10 15 20 25 30 A 0 5 10 15 20 25 30 Relat.informat.contrib.of GSC for A0=sier1 Life Evolut.of Sympt.Limit Value S ; for A1=sier1 30 i 1 2.5 l n + o l S +1=2.361 ti S l u 20 e 2 bi ul rt a n V o Auxiliary SOM=sier0 t c i 10 i 1.5 mi Vf L o Ordering numb.of singular value -ascending t. Number of observat-s used for S calculation 0 Vi p 1 l % 1 2 3 4 5 my 10 15 20 25 30 Sing.value similar.meas.of Main/Auxiliary SOM S (UV*S)Corr.similar.measure of Main/Auxiliary SOM 20 2 s e er e u gr s Similarity Index=0.88232 a e e d 10 1 ni m. y il t if CS=1=>SOM identity ri 0 degr.=>SOM ident.; Auxiliary SOM=sier0 mi a s. il r mi 0 o 0 S 1 2 3 4 5 C 1 2 3 4 5 Order.number of singul.value -ascending Order.number of generaliz.symptom-ascending Vi Fig. 3. GSVD comparison of the same SOM’s but with the different number of observations (the last sixth cancelled)

Comparing now Fig. 2 and 3 one can notice the indicates also one singular value close to unity essential difference at the last two rows of pictures (picture bottom right). Finally, the course of only. That means, that GSVD singular values are not symptom limit value (picture second right) is identical, the value and the course of symptom limit evolving gradually, and growing rapidly at the end values Sl is also not the same. What is more, the of the life of both systems. similarity measures at the last row of pictures are not Let as now compare another class of diagnosed as previously, because the last two singular values objects namely rolling bearings at durability-test are not identical. We can infer that shorter SOM stand. Fig. 5 gives here the results of comparison in produce such differentiation, although there is no the same way of pictures organization as before for other difference, only in the number of the rows the diesel engines. (observations). Almost the same situation is As one can notice from the first row of pictures noticeable when we exchange the calculation the durability (expected lifetime) of bearings is sequence of primary and auxiliary matrix. Above, different here, but the measured life curves are the first three singular values give the measures of similar, and the same can be said with respect of identity 0 (left picture) and 1 (right picture), and in generalized symptoms at the second row of pictures. case of the matrix exchange the last three singular Again one can notice, that there is one singular value values gives the sign of matrix identity. essentially different in quantity from the others Let us now pass to the diagnostic objects of the (third row-left picture), and the symptom limit value same type but different copies of it. Fig. 4 present Sl evolving gradually. The last row of pictures is the comparison of two different exemplars of similar, like for the diesel engines, namely the railroad diesel engines with the different primary angular measure of similarity is spread from -50 to and auxiliary SOM. This gives of course the +50 degrees, and correlation measure of similarity difference in generalized symptoms (second row of indicates one singular value greater than 1. It may pictures), with the last two singular values mean that there is only one way of degradation, essentially different from zero. The angular measure common to both tested bearings. But if we reverse of similarity (picture bottom left) is spread here from the matrix order (primary-auxiliary) the last -50o to +50o degrees, giving no essential message to indication on similarity changes a little giving one us, but the correlation measure of similarity singular values close to 1, and the second close to -1. DIAGNOSTYKA’ 3(47)/2008 29 CEMPEL, Generalized Singular Value Decomposition in Multidimensional Condition…

gsvdavg.m; Main Symp.Obs.Matrix; A1=sil54d2 Auxiliary Symp.Obs.Matrix; B1=sil24d2 1 1.5 m m S S 1 s s p- 0.5 p- my my 0.5 S S f 0 f o. o. 0 ti ti pl pl m -0.5 m -0.5 A 0 5 10 15 20 25 A 0 5 10 15 20 25 1-4 General.Sympt.of Main SOM; A1=sil54d2 1-4 General.Sympt.of Auxiliary SOM; B1=sil24d2 400 600 s s p- p- m m 400 y 200 y S. S. n n e e 200 G G f 0 f o. o. 0 tli tli p Observ-s / Life p Observ-s / Life m -200 m -200 A 0 5 10 15 20 25 A 0 5 10 15 20 25 Relat.informat.contrib.of GSC for A0=sil54d2 Life Evolut.of Sympt.Limit Value S ; for A1=sil54d2 i l 60 1 440 n + l S +1=435.3871 o Auxiliary SOM=sil24d2 l ti S 420 u 40 e bi ul rt a 400 n V o t c i 20 i mi 380 Vf L o t. p Number of observat-s used for S calculation 0 360 l % my 0 Ordering2 numb.of4 singular6 value8 -ascending10 12 5 10 15 20 25 Vi S (UV*S)Corr.similar.measure of Main/Auxiliary SOM 50 Sing.value similar.meas.of Main/Auxiliary SOM 6 s e er e u 4 gr s Similarity Index=0.058679 a e e d 0 2 ni 0 degr.=>SOM ident.; Auxiliary SOM=sil24d2 m. y il if CS=1=>SOM identity t ri mi 0 a s. il r mi o S -50 C -2 0 2 4 6 8 10 12 0 2 4 6 8 10 12 Order.number of singul.value -ascending Order.number of generaliz.symptom-ascending Vi Fig. 4. GSVD comparison of two different exemplars of the same diesel engines

gsvdavg.m; Main Symp.Obs.Matrix; A1=krak1 Auxiliary Symp.Obs.Matrix; B1=krak7 20 15 m m S S s s 10 p- 10 p- my my 5 S S f 0 f o. o. 0 ti ti pl pl m -10 m -5 A 0 10 20 30 40 50 A 0 10 20 30 40 50 1-4 General.Sympt.of Main SOM; A1=krak1 1-4 General.Sympt.of Auxiliary SOM; B1=krak7 10000 10000 s s p- p- m m y 5000 y 5000 S. S. n n e e G G f 0 f 0 o. o. ti ti pl Observ-s / Life pl Observ-s / Life m -5000 m -5000 A 0 10 20 30 40 50 A 0 10 20 30 40 50 Relat.informat.contrib.of GSC for A0=krak1 Life Evolut.of Sympt.Limit Value S ; for A1=krak1 i l 80 1 8000 n + l o ti 60 S 6000 S +1=6190.5237 u e l bi Auxiliary SOM=krak7 ul rt 40 a 4000 n V o t c i i 20 mi 2000 Vf L. o Ordering numb.of singular value -ascending t Vi p Number of observat-s used for S calculation % 0 m 0 l 1 2 3 4 5 6 7 y 10 20 30 40 50 S Sing.value similar.meas.of Main/Auxiliary SOM (UV*S)Corr.similar.measure of Main/Auxiliary SOM 40 5 s Similarity Index=0.1925 e er e 20 u gr s 0 a e e d 0 0 degr.=>SOM ident.; Auxiliary SOM=krak7 if CS=1=>SOM identity ni m. y il t -5 ri -20 mi a s. il r mi o S -40 C -10 1 2 3 4 5 6 7 1 2 3 4 5 6 7 Order.number of singul.value -ascending Order.number of generaliz.symptom-ascending Vi Fig. 5. GSVD similarity of rolling bearings at the durability test stand 30 DIAGNOSTYKA’ 3(47)/2008 CEMPEL, Generalized Singular Value Decomposition in Multidimensional Condition…

Altogether 10 ball bearings were tested at the DAMAS'99, Key Engineering Materials, Vols. durability stand, and they damage advancement were 167-168, 1999, pp. 172-189. described by the same symptom observation vector [7]. Natke H. G., Cempel C.: The symptom and SOM with different numbers of rows only. In observation matrix for monitoring and addition, in every case the measures of similarity diagnosis, Journal of Sound and Vibration, between bearings behave like on the last row of the 248, 597 – 620, 2002. fig. 5. In particular, the angular measure of similarity [8]. Natke H. G., Cempel C.: Model Aided is spread 50 to + 50 degrees, and correlation Diagnosis of Mechanical Systems, Springer measure indicates all singular values close to zero Verlag, Berlin,1997, p. 248. with the exception of the last one being close to [9]. Golub G. H., VanLoan C. F.: Matrix unity or much bigger. Computation, Oxford, North Oxford Well, these are some introductory diagnostic Academic, 1996. meaning and possible application of Generalized [10]. Cempel C., Natke H. G., Yao J. P. T.: Symptom SVD, and the question now remains, is that all what Reliability and Hazard for Systems Condition can be done in condition monitoring? I am sure not, Monitoring, Mechanical Systems and Signal we should investigate the other possible application Processing, Vol. 14, No 3, 2000, pp 495-505. not only in condition monitoring but also in quality [11]. Cempel C.: Multidimensional Condition monitoring and comparison, for example. Monitoring of Mechanical Systems in Operation, Mechanical Systems and Signal 5. CONCLUSIONS AND FURTHER Processing, 17(6), 2003, p. 1291 – 1303. PROBLEMS [12]. ĩóátowski B. Cempel C. (eds.): Engineering of Machine Diagnostics, (in Polish), ITE Press, As one can infer from the above consideration Radom, 2004, p 1308. and examples of application, there is some [13]. Cempel C., Tabaszewski M.: Multidimensional possibility of GSVD application in machine vibration condition monitoring of nonstati- condition monitoring. This is based mainly on onary systems in operation, Mechanical looking at similarities in a machine wear processes Systems and Signal Processing, Vol. 21, 2007, and symptoms of its condition. For this purpose, pp 1233-1247. several measure of similarity were defined and [14]. Cempel C., Tabaszewski M.: Averaging the calculated for the cases of examples taken from the Symptoms in Multidimensional Condition real monitored objects. There are some promising Monitoring for Machines in Nonstationary results. However, it seems to be too early to Operation, (paper No519 on CD) 13 Interna- formulate some solid conclusions concerning GSVD tional Congress on Sound and Vibration, use in machine condition monitoring. Some more Vienna, July 2006. approaches and trials seem to be needed to formulate [15]. Will T.: Hanger matrix, two-thirds theorem, such conclusions. Internet, (2005); http://www.uwlax.edu/faculty/will/svd/svd/ind 6. REFERENCES ex.html . [16]. Matlab, GSVD Function Reference, [1]. Korbicz J., et all, (eds.): Fault Diagnosis – MathWorks, 2006. Models, Artificial Intelligence, Applications, [17]. Cempel C.: Multi fault vibrational diagnostics Springer Verlag, Berlin - Heidelberg, 2004, p. of critical machines, Zagadnienia Eksploatacji 828. Maszyn, 2005, pp135-142, (in Polish) [2]. Tumer I. Y., Huff E. M.: Principal component [18]. Alter O., at all: Generalized singular value analysis of tri-axial vibration data from decomposition for comparative analysis of helicopter transmission, 56th Meeting of the genome-scale expression data sets of two Society of Machine Failure Prevention different organisms, PNAS, March 18, 2003, Technology, 2002. vol.100, no. 6, pp 3351-3356. [3]. Bartelmus W., Zimroz Z., Batra H.: The [19]. Gui L., Wu X.: KPCA based multisensor data gearbox vibration signal preprocessing and fusion for the machine fault diagnosis, input values choice for neural network Proceeding of ICME06, Chengdu, China training, AI Methods Conference, Gliwice, October 2006, pp 342-352. Poland, 2003. [4]. Pantopian N. H., Larsen J.: Unsupervised condition detection in large diesel engines, Proceedings of the IEEE Workshop on Neural Networks,1999 [5]. Cempel C.: Vibroacoustic Condition Monitoring, Ellis Horwood, London, 1991, p 212. [6]. Cempel C.: Innovative developments in systems condition monitoring, Keynote Lecture DIAGNOSTYKA’ 3(47)/2008 31 KOĝCIELNY, SYFERT, DZIEMBOWSKI, The Issue Of Symptoms Arising Delays During Diagnostic Reasoning

THE ISSUE OF SYMPTOMS ARISING DELAYS DURING DIAGNOSTIC REASONING

Jan M. KOĝCIELNY, Michaá SYFERT, Bolesáaw DZIEMBOWSKI

Institute of Automatic Control and Robotics Warsaw University of Technology ul. Ğw. A. Boboli 8, 02-525 Warszawa, Poland e-mail: {jmk, m.syfert, b.dziembowski}@mchtr.pw.edu.pl

Summary The paper shows the influence of the dynamics of the symptoms forming on the correctness of generated diagnoses. There are given a few approaches that allow to take into account the symptoms delays in the algorithms of diagnostic reasoning. Finally, there is presented the algorithm of proper reasoning while the information about symptoms delays is omitted. Key issues are illustrated with simple examples.

Keywords: fault isolation, diagnostics, dynamic systems.

PROBLEM UWZGLĉDNIENIA DYNAMIKI POWSTAWANIA SYMPTOMÓW WE WNIOSKOWANIU DIAGNOSTYCZNYM

Streszczenie W artykule rozwaĪany jest wpáyw dynamiki powstawania symptomów na poprawnoĞü formuáowanej diagnozy. Zaprezentowanych jest kilka algorytmów pozwalających na uwzglĊdnienie opóĨnieĔ symptomów w procesie wnioskowania. Zaprezentowane są takĪe mechanizmy prawidáowego wnioskowania (pod wzglĊdem formuáowanych diagnoz) pomijające bezpoĞrednio informacjĊ o opóĨnieniach symptomów. Kluczowe zagadnienia zilustrowane są na prostym przykáadzie.

Sáowa kluczowe: lokalizacja uszkodzeĔ, diagnostyka, systemy dynamiczne.

1. PROBLEM FORMULATION S/F f1 f2 f3 f4 f5 f6 f7 f8 In most of the diagnostic strategies, to be able to s1 1 1 proceed with diagnosis, the mapping of the s2 1 1 1 1 1 diagnostic signal space (residual values) onto the s3 1 1 1 1 fault space is necessary. Different forms of this s4 1 1 1 1 representation are known [3, 4, 8]. Most of them Fig. 1. Example of diagnostic binary matrix have static nature. However, the diagnosed processes are dynamical systems. Therefore, from x Time 2-4 [s]: achieved diagnostic signals - the moment of fault occurrence to the moment when S={0,1,0,0}; diagnosis - DGN2-4={f6}. one can obtain measurable symptoms the particular x Time 4-6 [s]: achieved diagnostic signals - time period elapses. In general, this time period is S={0,1,1,0}; diagnosis - DGN4-6={f2}. different for each fault and each diagnostic signal x Time •6 [s]: achieved diagnostic signals - which detects that fault. Only after some period of S={0,1,1,1}; diagnosis - DGN6={f3}. time all symptoms are observed. Just a few Only diagnosis DGN is proper and has final approaches reference this problem [2]. 6 nature. The earlier ones were false. The wrong diagnosis could be generated if one The following problems appears: How to take didn’t take the dynamic of symptoms into into account the symptoms delays in the fault consideration. This is illustrated by the following isolation algorithms in order to eliminate possibility example. of formulating false diagnosis? Is it possible to Example 1. Let us consider the diagnostic binary develop faults isolation algorithm insensitive to matrix presented in Fig. 1. Let us assume, that fault symptoms delays? The problems presented above are f . occurred. It is detectable by diagnostic signals s , 3 2 the subject of this paper. Firstly, the formal description s and s . Assume that symptoms arising times are 3 4 of symptoms delays is introduced together with the different for each diagnostics signal and equal theoretical way of its calculation. Then, several respectively: T =1, T =2, T =4, T =6 [s]. Parallel 1 2 3 4 diagnostic reasoning strategies that take these delays diagnostic reasoning runs as follows: into account are presented. There are shown complex Time 0-2 [s]: the fault is not detected. x as well as simplified approaches. 32 DIAGNOSTYKA’ 3(47)/2008 KOĝCIELNY, SYFERT, DZIEMBOWSKI, The Issue Of Symptoms Arising Delays During Diagnostic Reasoning

2. SYMPTOMS DELAYS forming times is difficult in practice because it requires the modeling of fault influence on The delays of symptoms forming depend on the measurable outputs. The fault development function dynamic characteristic of the process, fault type as well as residual threshold value must be assumed (abrupt, incipient), its time development arbitrary. Because of modeling errors the precision characteristic, the applied method and detection of analytical estimation of symptom times is poor. algorithm parameters. It’s possible to calculate In practice, based on the knowledge about the analytically these times basing on the dynamic process and detection algorithms, it is possible to description (e.g. transmittance) of the controlled part estimate symptom times by giving their minimum of the process (where the fault is an input and the and maximum values [4, 6]. Let us use the following process value is an output) and the transient notation: 1 th response of fault appearance. We make an x ș k,j – minimal time period from k fault assumption that the limitation function parameters occurrence to jth symptom appearance, 2 th are known and there is no influence of the x ș k,j - maximal time period from k fault diagnostic test methods on the process operation. occurrence to jth symptom appearance. The mathematical process description can be These parameter can be expressed in seconds or achieved based on the equations describing the dimensionless units equal multiple of the smallest physical effects taking place in the process. In this process value sampling time. They are assigned to case, it is necessary to treat all the possible faults as each ordered pair (fault, diagnostic signal) separate inputs in the system of equations. After the satisfying the relation SŸF. The actual symptom 1 2 linearization at the operating point and applying time belongs to interval < ș k,j , ș k,j >. Laplace transformation one achieves linear model in During process state monitoring the diagnose the form [3, 4, 8]: should be formulated after all of the symptoms are F time invariant. The approach that takes into account y(s) G(s)u(s)  G (s)f (s) . (1) 2 only the maximum symptom times ș k,j allows to Each constituent equation has the following avoid generating false diagnosis. form: The problem can be additionally simplified, by

F assigning the cumulative symptom time Tj to each y j (s) G j (s)u(s)  G j (s)f (s) , (2) diagnostic signal [5, 7]. The cumulative symptom while Gj(s) (j=1,…,J – number of diagnostic signals) time is defined as the maximum interval from the denotes input-output transmittance: appearing of any of the faults controlled by this test to the moment when the symptom is detected: G p,j (s) y j (s) / u p (s); p 1,...,P , (3) 2 F T max T ,k : f  F(s ) . (10) and G j(s) denotes the transmittance for each fault- j ^ k,j` k j output couple: The use of cumulative symptom times simplifies F the way of describing the process dynamic G k,j (s) y j (s) / f k (s); k 1,...,K . (4) properties and, especially, the reasoning algorithm. If there are no existing faults in the process, then It is easier to define these parameters, however, it is the following dependence is satisfied: still not an easy task. y (s)  G (s)u(s) G F (s)f (s) 0 . (5) The use of cumulative symptoms times was j j j implemented in DTS and F-DTS methods presented The residuals are calculated based on the by KoĞcielny (1995). The full description of the following equation called calculation form: dynamic properties was applied in method i-DTS [7]. rj (s) y j (s)  G j (s)u(s) . (6) Equation (7) (internal form) reflects general 3. DIAGNOSTIC REASONING TAKING INTO relation between particular residual and faults: ACCOUNT CUMULATIVE SYMPTOMS F TIMES rj (s) G j (s)f (s) (7) F F F G j,1(s)f1(s)...  G j,k (s)f k (s)...  G j,K (s)f K (s). Presented below diagnostic reasoning is based on the analysis of successive diagnostic signals and its If r is sensitive for fault f and no other faults are j k cumulative symptoms times T introduced in section 2. present (f =0) then one achieves: j m The diagnosis is formulated in several steps, in which F rj s G k,j s f k s ,f m 0 :m 1,2,...,K, m z k . (8) the set of possible faults is gradually constrained [4]. In fk the case of such reasoning (serial approach), the For so defined residual its time development diagnostic relation RFS is defined by attributing to each function is defined by the following relation: diagnostic signal the subset of faults detectable by this 1 F signal: rj t L Gk, j s fk s . (9) F(s j ) ^f k  F :f k R FS s j`. (11) The analytical calculation of the symptom DIAGNOSTYKA’ 3(47)/2008 33 KOĝCIELNY, SYFERT, DZIEMBOWSKI, The Issue Of Symptoms Arising Delays During Diagnostic Reasoning

The isolation procedure is started after the first faults undetectable by that signal. The new set of symptom is observed. Its occurrence indicates that possible faults is a product of past possible faults one of the fault from the set F(sx) of the faults and the set of faults detectable by that signal F(sj): detectable by that diagnostic signal had arisen. Such s 1 Ÿ DGN DGN ˆ F(s ) . (20) a subset of possible faults is indicated in the primary j r r1 j diagnosis: During the diagnostic reasoning the preliminary diagnosis is formulated after the first symptom is (t t1) š (s 1) Ÿ DGN { F1 F(s 1) . (12) x 1 x observed and then constrained when further, The subset of diagnostic signals S1 useful for consecutive diagnostic signal values are taken into isolation of faults from the set F1 is created: account. Usually, there is no need to analyze all the

1 1 signals to be able to formulate the final diagnosis. S ^s j  S: F ˆ F(s j ) z ‡` . (13) Such situation takes place when the diagnosis The values of the diagnostic signals from the set consists of only one fault or the set of S1 are interpreted, step-by-step, according to the indistinguishable faults. sequence determined by the attributed symptoms Example 2. The serial reasoning in the case of times. The jth diagnostic signal is used under the fault f6 appearance (example form Fig. 1) is show following condition that protects against formulating below. The first observed symptom is s2=1. As false diagnosis: a result, the following sets are created: the subset of 1 1 possible faults F ={f1,f2,f3,f5,f6} and the subset of (t  t ) ! T . (14) 1 j useful diagnostic signals: S ={s1,s2,s3,s4}. The time instants of consecutive diagnostic The time instants when the successive diagnostic signals interpretations are determined. They create signals should be analyzed are determined: 2 3 4 the following series: T T1 1; T T3 4; T T4 6 . 1 2 r p Then, in the following steps, the diagnosis is T d T d ... d T d ... d T , (15) constrained: where Tr {T :s S1}, while r defines the 1 j j x t t 1;s1 0 Ÿ DGN 2 {f 2 ,f3 ,f6} sequence of analysis of the diagnostic signals from 1 x t t  4;s3 0 Ÿ DGN 3 f6 . the set S1. 1 r After the second step the process of diagnosing Successively, in the time instant t t  T for is stopped. Finally, the same diagnosis as in the r=1,...p, the values of particular diagnostic signals case of parallel reasoning is assumed but it is are analyzed and the reduction of the set of possible concluded basing on only three diagnostic signal faults takes place. The process state z(fk) is values after 4 seconds when the first symptom was attributed to each of the faults fk from the set F. It is observed. The value of diagnostic signal s4 was not defined in the following way: needed for final diagnosis formulation.

­0  the state without fault f k z f k ® (16) 4. SYMPTOMS BASED REASONING ¯ 1 the state with fault f k The “0” value of the diagnostic signal testifies, In the above described diagnostic reasoning that none of the faults controlled by that diagnostic methods the information about the symptoms delays signal had occurred: was used to avoid formulating false diagnosis before all the symptoms occur. However, achieving the s j 0 œ  z(f k ) 0 . (17) data concerning the times of symptoms arising is not k:fkF(s j) easy. The following question appears: Is it possible The “1” value testifies, that at least one of the to formulate proper diagnosis without taking into faults from the set F(sj) had occurred: account the symptom arise times? It is shown below, that is it possible. s 1 z(f ) 1 j œ  k . (18) In the described reasoning rules the information k:fkF(s j) about the appearance of the particular symptoms, in When single fault occurrence is assumed, the the predefined interval, as well as the lack of other following rules of reducing the set of possible faults ones was used during diagnosis formulation. While indicated in the consecutive steps of diagnosis the symptom appearing is easy to observe, one must formulation are used: wait for proper time period, when the symptom xThe value of „0” of the diagnostic signal causes should appear, to be able to take into account its the reduction of the set of possible faults by the lack. It is possible to simplify the reasoning faults detectable by that signal: procedure by taking into account only the observed symptoms and rejecting the information carried out s 0 Ÿ DGN DGN  DGN ˆ F(s ) . (19) j r r1 r1 j by the lack of symptoms. It means the use of the rule xThe value of „1” of the diagnostic signal causes (18) and rejecting the rule (17). The set of possible the reduction of the set of possible faults by the faults is reduced only according to the rule (20). The 34 DIAGNOSTYKA’ 3(47)/2008 KOĝCIELNY, SYFERT, DZIEMBOWSKI, The Issue Of Symptoms Arising Delays During Diagnostic Reasoning

diagnosis, in each reasoning step, is proper and sequence if their sequence signatures (21) are points out such faults, for which the observed identical: symptoms are consistent with those ones defined in f R f œ SK(f ) SK(f ) . (22) the signatures. However, the fault isolability can be k N n k n lower. In this case, the reasoning consists of comparing Example 3. Let us assume, that the fault f2 registered symptom sequence with pattern ones appears (example form Fig.1). It is detected by the describing particular faults: 1 symptom s2=1, so: t 0 ; DGN {f k : SK(f k ) SK} , (23) 1 1 where SK denotes currently registered symptoms DGN1 F {f1,f2 ,f3,f5 ,f6}, S {s1,s2 ,s3,s4} . sequence. The diagnosis is modified after each new It is sufficient to be able to isolate any pair symptom appears. In this case, only one symptom of faults for which the sequence of any pair of fault appears: s3=1. According to (20) one achieves: symptoms is different: s =1Ÿ DGN ={f , f }. It is a final diagnosis, 3 2 2 3 SK(f ) s ,s ! , SK(f ) s ,s ! . (24) because none other symptoms will appear. In k j p n j p comparison, the serial reasoning which takes into account symptom times was finished after taking 6. DIAGNOSTICS BASED ON into account the diagnostic signal s4=0, in time THE KNOWLEDGE ABOUT SYMPTOM 1 moment t=t +6. It leads to more precise diagnosis: INTERVAL DELAYS DGN2=f2.

One must also notice, that in the case of fault f6 This section presents the fault isolation algorithm considered in Example 2, the serial reasoning based that utilizes the knowledge about the diagnostic on symptoms is finalized in the first step: relation and the values of the minimal and maximal s2=1Ÿ DGN1={f1, f2, f3, f6} with the diagnosis that symptoms forming delays. It assumes single fault pointing out four faults that are unisolable in scenarios, however, the multiple faults issue is also respect to only one observed symptom (so far). addressed. The algorithm implements serial diagnostic reasoning. The following notation is th 5. DIAGNOSIS BASED ON THE SYMPTOMS used: DGNr – final diagnosis elaborated in r step * ** SEQUENCE of reasoning; DGN r, DGN r – intermediate diagnosis. The sequence of symptoms arising is an Three main stages of reasoning algorithm can be important information which is worth of using in the distinguished: initialization, diagnosis specifying, diagnostic process. The different sequence of and final diagnosis formulation. symptoms arising can allow to isolate Initialisation of isolation procedure. The undistinguishable faults with identical fault isolation algorithm starts in the time t1=0 when the 1 signatures. first symptom s x=1 is observed (fault detection). The symptoms sequence for particular fault does The following steps are conducted: not depend on the fault time development x Determining the set of possible faults. characteristic fk(t). Based on (9), assuming particular The primary set of possible faults is determined form of a function fk(t) (e.g. step function) and the based on diagnostic relation. It consists of all the threshold residual value it is possible to calculate the faults, for which the diagnostic signal with the time, after which the symptom of a kth fault will observed symptom is sensitive for: appear. Such calculations must be done for all the (s1 1) DGN* {f : [q(f ,s1 ) 1]} (25) residuals sensitive for kth fault. x Ÿ 1 k k x * The sequence of symptoms forming can be done where: DGN 1 denotes temporary diagnosis, by arranging the values of symptoms delays for elaborated under the condition of use of the first a particular fault in ascending order. Finally, the 1 diagnostic signal s x but without taking into signature of the symptoms forming sequence for account the intervals of symptoms delays; particular set of residuals for each fault is achieved: 1 1 q(fk, s x)=1 denotes that diagnostic signal s x SK(f k ) si ,s m ,s p ,... ! . (21) detects the fault fk according to the diagnostic relation. The signature consists of the series of symptoms Reduction of primary set of possible faults. Let us s for particular faults f written down in the order of x j k 1 2 appearing. introduce the notations Tk,x and Tk,x The different symptoms sequence can for minimal and maximal periods from kth fault characterize faults that are unisolable based on occurring to the first, detected symptoms s1 1 binary diagnostics matrix (fault with the same x signatures). The symptoms are not isolable formulation, respectively. The faults, which occurrence should cause another symptoms to be (in respect to the relation RN) based on symptoms DIAGNOSTYKA’ 3(47)/2008 35 KOĝCIELNY, SYFERT, DZIEMBOWSKI, The Issue Of Symptoms Arising Delays During Diagnostic Reasoning

1 known intervals of symptoms delays, can be observed before the symptom s x , in respect eliminated from the diagnosis elaborated to known intervals of symptoms delays, are 1 in previous step: eliminated from the set DGN(s x ) : DGN** {f  DGN* : E2 E1 } (31) * 2 1 r k r k,p k,j DGN1 {f k  DGN1 : Tk,j Tk,x } , (26) 1 The reduction of the set of possible faults after s jzsx x the analysis of the maximal times of the while DGN1 denotes first, temporary diagnosis symptoms delays. The lack of symptom after elaborated while taking into account the intervals predefined time period, t ! E 2 , allows for the of the symptoms delays. jk x Determining the set of diagnostic signals useful reduction of the set of possible faults due to the for further fault isolation in the following form: formula: * 1 * 2 S {s : F(s ) DGN } s . (27) (s j 0) š (s j S ) š (t ! Ek,j ) Ÿ j j ˆ 1 z ‡  x (32) ** x Defining the intervals of symptoms possible DGNr {fk  DGNr } fk consecutive forming. Due to the fact that the real The end of fault isolation. The algorithm stops time of fault occurring is unknown (only the time when all the diagnostic signals from the set S* are of the first symptom detection is registered) the taken into account. time intervals of the appearing of the consecutive Taking into account the symptoms forming symptoms of the diagnostic signals from the set delays can increase faults distinguishability S* must be recalculated in respect to the moment comparing with the diagnosis elaborated basing only of the first symptom detection. Such calculations on binary diagnostic matrix. In some cases it reduces must be conducted for the faults pointed out in the diagnosing time. the diagnosis in the following way: Example 4. On the base of binary diagnostic matrix from Fig.1 faults f and f as f and f are ­ 0 if T1  T2 d 0 1 5 4 7 1 ° k,j k,x undistinguishable. Let assume the symptoms delay Ek,j ® 1 2 1 2 (28) ¯°Tk,j  Tk,x if Tk, j  Tk,x t 0 intervals for the first pair of undistinguishable faults as follows: 2 2 1 1 2 Ek,j Tk,j  Tk,x . (29) 1 2 [T1,1,T1,1] [1,3] , [T1,2 ,T1,2 ] [4,5] , 2 1 2 1 2 The parameters E k,j for the faults fkDGN1 and [T5,1,T5,1] [6,8] ,[T5,2 ,T5,2 ] [2,5] . * the diagnostic signals sjS are arranged in ascending order. This implicates that in case of f1 fault the Iterative diagnosis specifying. The second part symptom s1=1 always appears before the symptom of the reasoning has iterative nature. The elaboration s2=1, whereas f5 fault occurrence will cause reverse of the following diagnosis takes place: sequence of the symptoms. It makes their unique recognition possible. x after the detection of each, successive faults Let us assume that delays intervals symptom, for undistinguishable faults f i f are as follows: each time when the maximal period of the 4 7 x 1 2 1 2 2 [T ,T ] [2,3] , [T ,T ] [4,5] , symptom delay E k,j from the ordered series of 4,3 4,3 4,4 4,4 these parameters passes. 1 2 1 2 [T7,3 ,T7,3 ] [1,3] , [T7,4 ,T7,4 ] [7,9] . During this stage, the following steps are conducted iteratively: Both faults results with the same symptoms x The reduction of the set of possible faults based sequence but in spite of this they are distinguishable. Maximal s =1 symptom delay after s =1 symptom o on diagnostics relation. If the symptom sj=1 4 3 * for f4 fault equals 5-2=3, whereas minimum time (sjS ) was detected in the proper period of delays than the set of possible faults is interval between these symptoms for f7 fault equals reduced according to formula: 7-3=4. Therefore if s4=1 symptom is delayed relatively to s =1 symptom less than 3 seconds that r * * 3 (s j 1) š (s j S ) Ÿ DGNr this indicates f , fault, if delay is bigger we are (30) 4 1 2 inferring about f7 fault occurrence. {f  DGN : q(f ,s ) 1š (t [E ,E ]} k r1 k j k,j k,j The algorithm has also limited ability to isolate Such an operation is realised for all the faults multiple faults. In general, if the new symptom

from the set fk  F(s j ) . s j 1 is observed, and if it does not appeared in x The reduction of the set of possible faults based the predefined period of symptoms delays in respect on the analysis of delays interval. The faults, to the first observed symptom, than the new fault which occurrence should cause another isolation thread is started. In that case it is assumed that this symptom is caused by another fault than the symptoms s p 1 to be observed before the faults pointed out in the previous steps. In this case currently observed symptom, in respect to the it is sometimes possible to formulate final diagnosis 36 DIAGNOSTYKA’ 3(47)/2008 KOĝCIELNY, SYFERT, DZIEMBOWSKI, The Issue Of Symptoms Arising Delays During Diagnostic Reasoning

about multiple faults if the proper sets of diagnostic [6] KoĞcielny J. M. and K. Zakroczymski.: Fault signals used in each isolation thread fulfill some Isolation Algorithm that Takes Dynamics of necessary conditions. The detailed description of Symptoms Appearances Into Account. Bulletin that problem is not in the scope of this paper. Some of the Polish Academy of Sciences. Technical information about creating fault isolation threads Sciences. 2001, Vol. 49, No 2, 323-336. can be found in [7]. [7] KoĞcielny, J. M. and M. Syfert: On-line fault 7. FINAL REMARKS isolation algorithm for industrial processes. Preprints of: 5th IFAC Symposium The Section 3 presents the reasoning algorithm SAFEPROCESS`2003, Washington D. C., that takes into account the simplified information USA, 9-11.VI, 777-782. about symptoms delays and enables to elaborated [8] Patton R., P. Frank, R. Clark (Eds.): Issues of proper diagnosis for dynamic systems. fault diagnosis for dynamic systems. It was also shown, in Section 4, that one can Springer, 2000. achieve proper diagnosis without taking directly into account the information about symptom forming Prof. dr hab. in . Jan Maciej times, however, it leads to lower fault isolability. Ī KO CIELNY is employed The knowledge about symptoms interval delays ĝ in the Institute of Automatic enables, in many cases, to isolate the fault that are Control and Robotics, unisolable based on binary diagnostic matrix. Warsaw University of However, to be able to determine the symptoms Technology. He conducts intervals we need the residual equations in the inner researches in the field of form or very precise expert knowledge. This is very diagnostics of industrial and difficult to obtain in practice. Such algorithm with mechatronics systems and detailed description was presented in Section 6. fault tolerant control The alternative approach was shown in Section systems. He leaded research works which resulted 5. The knowledge about the sequence of symptoms in elaboration and industrial implementation of generation also enables, in many cases, to isolate the advanced monitoring and diagnostic systems. He is fault that are unisolable based on binary diagnostic a member of the Automatic Control and Robotics matrix. It is easier to define such a sequence than Committee of the Polish Academy of Science, the precise symptoms interval delays. International Editorial Board of the science journal International Journal of Applied Mathematics and AKNOWLEDGMENTS Computer Science and a branch editor of Pomiary- Automatyka-Kontrola journal. The authors acknowledge partial funding support of Polish Ministry of Education and Higher Education under the grant R01 012 02 DIASTER: Dr inĪ. Michaá SYFERT is Intelligent diagnostic system supporting control employed in the Institute of of industrial processes. Project financed from the Automatic Control and fund supporting research in the years 2007-2009. Robotics, Warsaw University of Technology. REFERENCES He conducts researches in [1] Blanke M., Kinnaert M., Lunze J., the field of diagnostics of Staroswiecki M.: Diagnosis and Fault- industrial processes and the Tolerant Control. Springer 2003. applications of fuzzy logic. [2] Combastel C., Gentil S., Rognon J. P.: Toward He is main author of a better integration of residual generation and diagnostic system DIAG and one of the main diagnostic decision, Preprints of: 5th IFAC authors of the advanced monitoring and diagnostic Symposium SAFEPROCESS`2003, system AMandD. Washington D. C., USA, 9-11.VI [3] Gertler J.: Fault Detection and Diagnosis in Engineering Systems. Marcel Dekker, Inc. New York - Basel  Hong Kong 1998. [4] Korbicz, J., KoĞcielny J. M., Kowalczuk Z., Cholewa W.: Fault Diagnosis: Models, artificial intelligence methods, applications. Springer, 2004. [5] KoĞcielny J. M.: Fault Isolation in Industrial Processes by Dynamic Table of States Method. Automatica, 1995, Vol. 31, No.5, 747-753. DIAGNOSTYKA’ 3(47)/2008 37 TAYMANOV, SAPOZHNIKOVA, Automatic Metrological Diagnostics Of Sensors

AUTOMATIC METROLOGICAL DIAGNOSTICS OF SENSORS

Roald TAYMANOV, Ksenia SAPOZHNIKOVA

D. I. Mendeleyev Institute for Metrology (VNIIM) 19 Moskovsky pr., St. Petersburg, Russia, 190005, fax: (+7-812) 251-99-40, e-mail: [email protected]

Summary As the number of sophisticated technical complexes with automatic control systems grows, the number of embedded sensors increases. The specific scientific problems that emerge while developing the sensors characterized by fault tolerance and long-term lifetime without metrological maintenance are considered. The possible ways of solution of these problems are outlined. The features of metrological diagnostics are demonstrated.

Keywords: automatic metrological diagnostics, sensor, fault tolerance.

AUTOMATYCZNA DIAGNOSTYKA METROLOGICZNA CZUJNIKÓW

Streszczenie W miarĊ wzrostu liczby skomplikowanych technicznych kompleksów z automatycznymi systemami sterowania zwiĊksza siĊ liczba wbudowanych czujników. Rozpatrywane są problemy naukowo-techniczne związane z konstrukcją czujników charakteryzujących siĊ dáugim czasem pracy bez obsáugi metrologicznej i odpornoĞcią na uszkodzenia. Zostaáy zarysowane drogi rozwiązania tych problemów i mozliwosci diagnostyki metrologicznej.

Sáowa kluczowe: automatyczna diagnostyka metrologiczna, czujnik, odpornoĞü na báĊdy.

1. INTRODUCTION First of all, such sensors are necessary for technical systems with long-term technological Development of industrial equipment and cycle. However, in the near future, the other high increasing number of sophisticated technical duty objects including transport equipment, power complexes goes with complication of automatic units, etc., will need the sensors with enhanced control systems (ACS) and with increase in the metrological reliability. Moreover, the development number of in-built sensors. At the same time: of diagnostic systems depends, to a great extent, on x participation of personnel in the equipment the possibilities of their application. control decreases; The distinctive features of such sensors are: x expenditures on metrological assurance of x specified lifetime of many years without ACS grows; metrological maintenance (the lifetime should be x intervals between scheduled outages become proved experimentally); longer; x ability to self-diagnosing and revealing the x probability that invalid information can pass to growing uncertainty; ACS increases; x the ability of a sensor to keep the measurement x risk of accident or failure grows. uncertainty for the most of single sensor defects More and more, trouble-free operation and within an enlarged, but permissible, range as production quality depend on sensor condition. defined by the user (fault tolerance), as well as to In many cases in order to carry out metrological correct the uncertainty automatically (in assurance procedures such as calibration or a number of cases). verification (hereinafter referred to as calibration), it The set of these features affords ground for is necessary to interfere in a technological process. considering the corresponding sensors as Meanwhile, the experience shows that calibration “intelligent” [2]. does not ensure the sufficient credibility of measurements over calibration period. 2. ABOUT ASSESSMENT OF QUALITY OF In current situation, modern technological SENSORS INTENDED FOR LONG-TERM processes require industrial sensors which are OPERATION expected to provide many years of operation without metrological maintenance while ensuring a high The first problem the solution of which level of confidence in their measurement data [1]. predetermines the possibility of intelligent sensors development is elaboration of the methods for 38 DIAGNOSTYKA’ 3(47)/2008 TAYMANOV, SAPOZHNIKOVA, Automatic Metrological Diagnostics Of Sensors sensors quality control with respect to their design At present, approaches for the plan of the tests to and technology, taking into account the long-term evaluate Ɍ are different in different companies. lifetime. There is no point in complicating sensors Therefore, the results of Ɍ evaluation can be by introduction of the diagnostic features if the different. In order to obtain comparable results, it is sensors fail in a short period. Customers need to necessary to work out international guides that will know the estimate of the sensor lifetime Ɍ during include corresponding test procedures. Before the which the sensor uncertainty is kept within the necessary documents become operational, it is specified limits. This value is equal to the sensor expedient to specify the test procedure on the basis calibration interval. We could not find any special of which the lifetime of sensors would be assigned. requirements for quality check of the sensors with a specified long-term lifetime neither in current 3. METHODS OF AUTOMATIC standards, nor in any guides, including ISO DIAGNOSTICS OF SENSORS documents. Fridman in [3] proved that it is inexpedient to In general, there are the following differences apply fundamental assumptions of the classical between metrological diagnostics and conventional reliability theory (independence of failure rates and procedures of metrological assurance of sensors: failure rate stability) to measuring instruments. x the value estimated in diagnostic procedure is the Therefore, the lifetime estimation should be based parameter defined by a set of uncertainty on certification tests. components (not the overall instrumental Taking economical reasons into account, sensor uncertainty); lifetime tests should be chosen at least 50 – 150 x diagnostics conditions are identical with the times shorter than the planned sensor lifetime technological process conditions (not the without maintenance, while the number of tested reference conditions); sensors should be minimal. Test influencing factor x the sensor diagnosed remains in-situ (not in values are limited: they are to be within the limits a calibration laboratory); which provide that the sensor degradation x metrological diagnostics envisage on-line mechanisms under test conditions and during actual method which does not require interruption of operation are adequate. The plan of the certification technological parameter measurements, while the tests should take the sensor design, technology and conventional procedures are out-of-process operation conditions into consideration. methods. Taking into account [4], in order to evaluate Ɍ, it The first way of organizing the metrological is possible to recommend a technique which diagnostics is integration of sensors in a system with includes: determination of influencing factors that common diagnostic means. This system can be characterize sensor operating conditions; study of organized by several methods. degradation processes; detection of probable reasons The most wide-spread method is application of for the uncertainty increase; ranging of the a number of identical sensors and comparison of uncertainty components according to their their output signals [7]. contribution; after that, certification tests However, for mass-produced sensors of the same themselves. type, a drift of metrological parameters in the same For the certification tests plan, one should direction and with a close speed is the most probable choose only those influencing factors that give rise [2, 8]. If integration of such sensors is used, this to the most “dangerous” (significant) uncertainty drift cannot be revealed. The drawback is also in the components of the sensor, i.e. predominant necessity to place several (e.g., three) sensors in the components or those tending to rise quickly. equipment, which, in many cases, is impermissible Estimate of Ɍ can be obtained by processing the by an argument of engineering limitations. results of complex tests consisted of a simulation The second method implies integration of test and accelerated test. sensors that measure various quantities In simulation test, harsh operation conditions characterizing physical field parameters that that may take place during operation are simulated. correlate between each other. At the stage of the accelerated test, it is expedient to The drawbacks of this method are: expose the sensors to influencing factors by equal x presence of the uncertainty components due to cycles. One cycle may include exposure to one or the diagnostic method, which depends on the several factors, e.g. vibration and temperature of accuracy of relationship between the a maximum permissible level [5]. measurements, Stability of sensor manufacturing technology x necessity to apply more accurate sensors. should be proved by periodic and extraordinary (in The third method involves application of a more case of modification of the sensor design or accurate sensor in addition to other sensors. technology) tests. For these tests, the cycles like However, in order to organize the effective those that were chosen for the accelerated test can be diagnostics using this method, it is necessary to set applied [6]. a stationary state of technological equipment, on one DIAGNOSTYKA’ 3(47)/2008 39 TAYMANOV, SAPOZHNIKOVA, Automatic Metrological Diagnostics Of Sensors hand, and on the other hand, to keep in the WWER-1000 nuclear reactor. The sensor of a comparatively short calibration interval for the the MS realizes a combinatory code chain. Besides most accurate sensor of the system. measuring the control rod position, the MS performs Nevertheless, in a number of cases, metrological metrological and technical diagnostics of the sensor diagnostics of mass-produced sensors by their and microprocessor unit. The MDC generally integration in a system ensures a higher confidence consists in comparison between: in measurements. x the code combinations identified and the code The other way is development of sensors with combinations specified; the in- built capability of self-checking [2]. x the code combinations related to consequent There are two different ways of realization of control rod positions and the specified code this function. The first consists in embedding combinations; a reference standard (a reference measure or x the control rod position and the number of steps additional sensor, which is more accurate than the made by the rod. sensor under check) in the equipment and In various countries, new self-diagnosing, self- comparing output signals of the reference standard checking, or self-validating devices have been and checked sensor. The second way is associated developed, for example [11-20]. Since recently, with sensor intellectualization. It consists in along with conventional devices, a number of comparison of several signals or parameters, which companies have started mass-production of are close in accuracy, i.e. metrologically equivalent. measuring instruments which can perform We call the latter method metrological diagnostic metrological self-diagnostics. Such instruments are check (MDC) [1, 2, 5]. double thermocouples or resistance thermometers in If the metrological self-check is accompanied by the same housing, a temperature transmitter which a quantitative evaluation of measurement quality, it can accept 2 independent temperature sensor inputs is usually called self-validation [9]. (either Pt100 or thermocouples), as well as self- Development of the intelligent sensor requires diagnosing electromagnetic and ultrasonic flow a deep metrological investigation to be carried out. meters. This investigation should result in determination of Some metrological self-check modes (under the limited number of the most dangerous various names) have been used in industry for many uncertainty components and formation of the years. The first national regulations [21, 22] were required dependence of diagnostic parameters on published in Russia (1989) and the UK (2001). influencing factors. The distinguishing features of Their main statements were developed in new sensors capable of performing the metrological documents [23-25]. Attempts to systematize such diagnostic checking are presence of: methods were made, for instance, in [9, 26-28]. x structural and/or information redundancy which afford ground for obtaining an additional signal 4. SELF-CORRECTION AND FAULT regarding ambient conditions and/or sensor TOLERANCE “health”, x microprocessor that provides processing of The capability of self-diagnostics which is the measurement and diagnostic data. inherent feature of the intelligent sensor allows to Metrological diagnostics allows to determine increase metrological reliability significantly by the whether the sensor uncertainty is being kept within application of measures of active character. These the specified limits. If the uncertainty exceeds the measures are aimed at self-correction of the external specified limits, it is possible to diagnose the influences and ageing of components as well as at uncertainty variation specifics and to localize the support of fault tolerance. a defect. Structural and Information redundancy of sensor As a rule, on the basis of metrological self- as well as computer technology afford ground for diagnostics, the specified calibration interval can be these measures. considerably increased in comparison with its value The simplest example for self-correction in the which can be set on the basis of conventional presence of transient errors is frequency filtration or method of metrological assurance. time filtration. In order to apply such a correction, it The method requires the sensor sensitivity to be is sufficient to use a priori information on the higher than it is required for usual measurements in parameters of a signal coming from the sensor and the technological process. However, fulfilling this on the specified limits of the signal. requirement, as a rule, does not cause serious If a fault occurs, the intelligent sensor shell: difficulties. x register the fact of fault, The efficiency of the method is proved by the x inform the human operator that the measurement measuring and diagnostic system with an eddy reliability decreased, current sensor of the DPL-KV type (MS). The MS x correct the measurement result and continue was developed at the VNIIM [10] in order to apply operating. it in the linear stepping drive that move a control rod 40 DIAGNOSTYKA’ 3(47)/2008 TAYMANOV, SAPOZHNIKOVA, Automatic Metrological Diagnostics Of Sensors

An example of how the sensor fault tolerance is 5. CONCLUSION assured is the MS [10] mentioned above. The sensor In order to provide reliable functioning of of the MS contains a set of inductance coils. If sophisticated technical complexes with automatic a fault occurs, e.g. any signal wire breaks or any coil control systems, besides technical diagnostics of the fails, the sensor is keeping operation. equipment, it is necessary to apply high-reliability In the most part of the position range, due to sensors. They must provide long-term operation and incorporating a redundant number of coils into the realization of automatic metrological diagnostics. sensor (in comparison with minimally needed), the Development of such sensors should go with code combination is distorted, but it keeps development of specific methods for assessment of information regarding the control rod position. In quality of sensors in respect of their design and addition, the position of the rod can be corrected on technology. Automatic diagnostics should be the basis of the known number of steps made by the provided on the basis of structural and/or rod from the nearest position that was reliably information redundancy and computer technologies. measured. Success in developing such sensors depends, to In many cases, it is not possible to obtain a great extent, on elaboration of international a quantitative estimation of the uncertainty. For regulatory documents which can establish the a considerable part of applications, it is expedient to requirements for sensors characterized by long- term estimate the quality of measurement results using lifetime. measurement value status (MVS) [9]. In [9] the following status values were recommended: secure, REFERENCES clear, blurred, dazzled, and blind. In the joint paper of Oxford and St.Petersburg [1] Tarbeyev Yu., Kuzin A., Taymanov R., scientists [29] a comprehensive justification of the Lukashev A.: New Stage of Development of necessity to introduce the measurement value status Sensors Metrological Assurance, Measurement was given and some details were proposed. It was Technique, 3, 2007. p. 69-72. noted that the number of status states should depend [2] Taymanov R., Sapozhnikova K.: Problems of on the number of human operator’s actions required Developing a New Generation of Intelligent in response to information about the measurement Sensors, Proc. of the 10th IMEKO TC7 value status. The number of MVS states is Symposium, St.Petersburg, 2004, p. 442-447. comparatively small: [3] Fridman A. E.: Theory of Metrological A) Firm confidence, that a measurement value is Reliability, Measurement Technique, 1991, 11, reliable, corroborated by an additional p. 3-10. information source, i.e. it is based on redundant [4] Efremov L., Sapozhnikova R.: Assessment of information. The sensors are fault-free. the Sensors Lifetime on the Basis of Test B) Assumption that a measurement value is reliable, Results, Proc. of the 10th IMEKO TC7 Symp., but there is no corroboration from any additional St. Petersburg, 2004, p. 448-453. information sources. [5] Sapozhnikova K., Taymanov R.: Development C) Understanding that measurement confidence has of Sensors Distinguished by Enhanced been decreased due to some fault. Measurement Reliability, Proc. of ɋonf. “Nuclear Power value was corrected for this fault condition, but Instrumentation, Controls, and Human Machine the uncertainty is not too great. The Interface Technology“, Albuquerque, New measurement confidence is sufficient to get Mexico, USA, 2006. operator's bearings in the technological process [6] Taymanov R., Moiseeva N.: Application of the and technological equipment condition. Metrological Investigation to the Development D) Conception that measurements are not reliable of the Manufacturing Technology of High- for a short period of time. This relates to carrying reliable Sensors, Proc. of the 10th IMEKO TC7 out a special test, technological operation, or Symp., St. Petersburg, 2004, p. 433 - 436. data filtration. In such a situation the [7] Hashemian H. M.: Maintenance of Process measurement values are projected from the past Instrumentation in Nuclear Power Plants, history. Springer - Verlag, Berlin Heidelberg, 2006. E) Confidence that measurement values are not [8] Reed R. P.: Possibilities and Limitations of reliable. This situation obliges, if necessary, to Self-validation of Thermoelectric Thermometry, stop the technological process. If the substituted Proc. of the Conf. “Temperature: Its data concerning the measurements are available, Measurement and Control in Science and step must be taken to move the process away Industry”, 7, D. C. Ripple et al. eds., AIP from any critical technological constraints. Conference Proceedings, Melville, New York, 2003, p. 507-512. [9] Henry M. P., Clarke D. W.: The Self-validating Sensor. Rationale, Definitions, and Examples, Control Eng. Practice, v. 1, 1993, 4, p. 585-610. DIAGNOSTYKA’ 3(47)/2008 41 TAYMANOV, SAPOZHNIKOVA, Automatic Metrological Diagnostics Of Sensors

[10] Taymanov R., Sapozhnikova K., Druzhinin I.: [24] BS 7986:2005. Specification for Data Quality Measuring Control Rod Position, Nuclear Plant Metrics of Industrial Measurement and Control Journal, 2007, 2, USA, p. 45-47. Systems, British Standards Institute, London, [11] Henry, M.P., Clarke, D. W., et al.: A Self- 2005. validating Digital Coriolis Mass-flow Meter: an [25] NAMUR Recommendation NE 107, Self- Overview, Control Engineering Practice, 8, monitoring and Diagnosis of Field Devices, 2000, 5, p. 487-506. 2005. [12] Yang J. C.-Y., Clarke D. W.: A Self-validating [26] Sapozhnikova K. V.: Metrological diagnostic Thermocouple, IEEE Transactions on Control check, Metrological Service in the USSR, 1991, Systems Technology, v. 5, 1997, 2. 2, p.18-24. [13] Bernhard F., Boguhn D., et al.: Application of [27] Werthschutzky R., Muller R.: Sensor Self- Self-calibrating Thermocouples with Miniature Monitoring and Fault-Tolerance, Technisches Fixed-point Cells in a Temperature Range from Messen, v.74, 2007, 4, p. 176-184. 500 oC to 650 oC in Steam Generators, Proc. of [28] Feng Z., Wang Q., Shida K.: A review of self- the XVII IMEKO World Congress, 2003, validating sensor technology, Sensor Review, Dubrovnik, Croatia, p. 1604-1608. 27, 2007, 1, p. 48-56. [14] Feng Z.-G., Wang Q., Shida K.: Structure [29] Sapozhnikova K., Henry M., Taymanov R.: Design and Finite Element Analysis of Self- Proc. of the Joint IMEKO TC1+ TC7 Symp., Validating Pressure Sensor, Chinese Journal of Ilmenau, Germany, 2005, p. 71-72. Sensors and Actuators, 20, 2007, 2, p. 279-282. [15] Belevtczev A. B., Karzhavin A. V., et al.: In- Dr. Roald TAYMANOV. built Method for Estimation of Confidence in Dr. Taymanov is head of Thermocouple Measurements, Mir Izmerenii, laboratory at the VNIIM, 2005, 3, p. 12-15. academician of the Russian [16] Hans V., Ricken O.: Self-monitoring and Self- Metrological Academy. He has Calibrating Gas Flow Meter, Proc. of 8th been the scientific leader of International Symposium on Measurement many various projects on Technology and Intelligent Instruments, Sept signals transformation and 24-27, 2007, p. 285-288. measurements. Dr. Taymanov [17] Barberree D.: Dynamically Self-validating is the author of more than 160 publications. Listed Contact Temperature Sensors, Proc. of the in “Marquis Who’s Who in Sci&Eng 2006-2007” Conf. “Temperature: Its Measurement and “Who’s Who in the World 2008”, and “Outstanding Control in Science and Industry”, 7, D. C. Scientists of the 21st Century”. Ripple et al. eds, AIP Conference Proceedings, Melville, New York, 2003, p.1097-1102. Dr. Ksenia SAPOZHNIKOVA. Dr. Sapozhnikova [18] WO2007097688, Redundant Level is deputy head of laboratory at the VNIIM, lecturer Measurement in Radar Level Gauging System, at the Courses for Power G01F23/284; G01F23/14; G01F25/00, Haegg Plants Chief-metrologists. L., Larsson L.O., et al, Rosemount Tank Radar Main research area: AB, 2007.08.30. measurements under harsh [19] Kao I., Zhang K.: Miniaturized Sensors for environment and metrolo- Intelligent System Fault Detection and gical diagnostic check of Diagnosis (FDD), Proceedings of SPIE, 2007, sensors. Dr. Sapozhnikova part 1, art. No. 65290U. is the author of more than 80 [20] Frolik J., Abdelrahman M.: A confidence – publications. based Approach to the Self-validation, Fusion and Reconstruction of Quasi-redundant Sensor Data, IEEE Transactions on Instrument and Measurement, v.50, 2001, 6, p. 1761-1769. [21] MI Recommendation 2021-89, Metrological Assurance of Flexible Manufacturing Systems. Fundamentals, Committee on Standard. and Metrology, Moscow, 1991. [22] BS 7986:2001, Data Quality Specification for Industrial Measuring and Control Systems, British Standards Institute, London, 2001. [23] MI Recommendation 2233-00. Assurance of Measurement Efficiency in the Control of Technological Processes. Fundamentals, VNIIMS, Moscow, 2000.

DIAGNOSTYKA’ 3(47)/2008 43 SKALSKY and others, Portable Multi-Channel Device For Acoustic Emission Monitoring Of Structures …

PORTABLE MULTI-CHANNEL DEVICE FOR ACOUSTIC EMISSION MONITORING OF STRUCTURES AND PRODUCTS

Valentyn SKALSKY, Oleg SERHIYENKO, Yevhen POCHAPSKY, Roman PLAKHTIY, Roman SULYM

Physico-Mechanical Institute of National Academy of Sciences of Ukraine Lviv, Ukraine, [email protected]

Summary In the paper eight channel acoustic emission (AE) device for selection, recording and processing signals of AE intended for use in the sphere of non-destructive testing of materials, products and structures of various forms and functional applications is described. The device is designed using a wide range of SMD elements and is adapted to work with Windows family operating systems. Specially developed software realizes functions of input data processing and its visualization, determination of defect position and storing the obtained results into computer memory. The above-listed characteristics as well as high rate of data exchange between the device and PC (12 Mbit/s), enables to work in real-time mode, and efficient software allow to compete with developments of such world leading manufacturers as PAC, Vallen Systeme, Interunis etc.

Keywords: acoustic emission, destruction, non-destructive testing, technical diagnostics.

1. TOPICALITY OF PROBLEM 2. STATE OF PROBLEM

Presently in most production spheres new If all the developments of AE devices known in materials and technologies are launched, experience literature are generalized, they can be classified into of using which under conditions of intensive strain the following groups: for complex researches, and activity of aggressive working environment is specific purpose, for control over the state of large- insufficient. Therefore the objects using new sized objects and portable one- and multichannel materials or technologies need to be monitored to ones. provide their reliable maintenance. With this Facilities for complex researches are intended for purpose new progressive methods and means of the reception of AE signals during defect diagnosing firmness and durability of the mentioned development, which initiate in materials, wares and items, constructions and buildings are used. constructions. Equipped, as a rule, by devices are Analysis of reasons for construction materials able to distinguish the signals from background fail, which are widely used in mechanical and power noises and hindrances, they allow to estimate engineering, pipeline transport, aircraft building, various parameters of Ⱥȿ signals and determine the chemical and oil industry etc. showed that in the state of the object under control. majority of cases it is the result of initiation and Facilities for the specific purpose are developed development of crack-like defects [1, 2]. Therefore mostly to solve particular NT tasks: to reject as the research of these processes has become a topical defective wares during their mechanical testing; to task. record by AE signals the moment of initiation in Experience of the last decades proved great construction machine-building materials the tensions potential possibilities of the acoustic emission (AE) which correspond to the physical limit of fluidity; to method. Its application is especially relevant under estimate plastic volume of material; to research the conditions when visual control is impossible or phenomenon of corrosion under tension; to record access to the object under control is complicated [3]. and analyze AE signals during the friction of solids Distance control, high sensitivity, possibility of etc. remote defect detection, which considerably exceed Facilities of large-sized objects AE control allow their sizes, possibility to obtain information determining the location of developing defects. regardless of form and sizes of the object under Knowledge of coordinates of AE sources allows control, recording real time destruction development estimating distribution of defects within control area etc. are the advantages that have put the method of and taking into account power parameters of the AE in a leading place among the known perspective radiation to estimate the level of damage risk. Such methods of non-destructive testing (NT) [4]. Ⱥȿ systems serve to reveal danger of initiation and For its successful implementation new effective accumulation of defects, location of AE sources and portable devices based on modern achievements of as preventive control systems for emergency electronics have to be developed. situations of responsible buildings. When testing the state of overall objects of complicated configuration, the determination of 44 DIAGNOSTYKA’ 3(47)/2008 SKALSKY and others, Portable Multi-Channel Device For Acoustic Emission Monitoring Of Structures … coordinates or areas of emission sources location is ƒthe price of the device is significantly lower in performed using various methods. The most comparison with similar AE equipment of this widespread is the method of calculating coordinates class of other well-known manufacturers. by means of triangulate calculations and area method of locating AE sources. Both are based on 4. CONDUCTING EXPERIMENTAL registering the difference in arrival times of AE RESEARCHES signals perceived by the group of primary transformers. An approbation of the device when monitoring Purpose of work – to develop and do the state of several bridges, overpasses, tunnel experimental approbation of the portable eight- transitions etc. was conducted, that is represented in channel acoustic emission device. proper publications and acts of NT applied methods implementation (fig. 1). 3. DESCRIPTIONS AND BASIC ADVANTAGES

Eight channel acoustic emission (AE) device for selection, recording and processing signals of AE is intended for use in the sphere of non-destructive testing of materials, products and structures of various forms and functional applications. The device is designed using a wide range of SMD elements and is adapted to work with Windows family operating systems. Specially developed software realizes functions of input data processing and its visualization, defect location and storing the obtained results into computer memory. The above-listed characteristics as well as high rate of data exchange between the device and PC (a) (12 Mbit/s), enabling to work in real-time mode, and efficient software allow to compete with developments of such world leading manufacturers as PAC, Vallen Systeme, Interunis etc. In comparison with developments of world- famous manufacturers the designed device has a number of advantages: ƒportability allows the device to be used not only in field conditions of inspected objects diagnosing but also in hard-to-reach, high- altitude and other difficult conditions; ƒself-contained power supply enables to use the device under conditions of limited or unavailable network power supply; ƒsupply current 120 mȺ; (b) ƒsensitivity to tested surface displacement Fig. 1. Application of the developed Ⱥȿ system -14 -12 10 ...10 m; when diagnosing heat-and-power engineering ƒUSB interface connection provides high rate of equipment (a) and bridge transition (b) data exchange between the device and a PC; ƒprogrammed control capabilities: choice of 5. AREA OF APPLICATION number of working channels, variability of sampling duration etc. The device can be used for monitoring and ƒinaccuracy in determination of AE source technical diagnostics of long-term operation objects: location depending on inspected object testing ƒbridges; conditions does not exceed 10 %; ƒtanks, pressure vessels; ƒuser-friendly software interface together with ƒpipelines; convenient help system allow users to quickly ƒelements of bridge, frame and tower cranes; acquire skills of working with the device; ƒport lift-and-carry mechanisms; ƒoverall dimensions: 370×256×30 mm, weight – ƒother components and mechanisms, 2,1 kg; as well as in laboratories for fundamental and ƒcompactness and good constructional solution of applied researches of structural materials: the device conduce to easy and convenient ƒstatic and cyclic crack growth resistance; transportation; ƒcreeping, plastic deformation; DIAGNOSTYKA’ 3(47)/2008 45 SKALSKY and others, Portable Multi-Channel Device For Acoustic Emission Monitoring Of Structures …

ƒnucleation and propagation of cold - and Oleh SERHIYENKO, PhD. autocracks while welding; (mathematics and physics), ƒthreshold value of stress intensity factor for Senior Researcher of the hydrogen-induced and stress corrosion cracking Department of Acoustic of materials; emission diagno-stics of ƒcomposites research etc. structural elements, Physico – mechanical Institute, LITERATURE National academy of Sci- ences of Ukraine. Authored [1] Skalsky V. R., Sergienko O. M., Oliyarnyk B. more than 140 papers in the O., Plakhtiy R. M., Sulym R. I.: Development of areas of fracture mechanics, technical diagnostics methods and means for detecting initiation and and non-destructive testing. development of cracks in large-sized objects under the influence of loading and working environment. Sc. Art. Col “Problem of resource Yevhen POCHAPSKIY, and safety of constructions, buildings and PhD. (mathematics and machines maintenance”, Kyiv, 2007, P. 48. physics), Senior Researcher [2] Skalsky V. R., Koval P. M.: Some of of the Department of methodological aspects of acoustic emission Acoustic emission diagno- application. Lviv, Spolom, 2007. stics of structural elements, [3] Skalsky V. R., Koval P. M., Stashuk P. M., Physico – mechanical Plakhtiy R. M., Sulym R. I.: Operative acoustic Institute, National academy emission control of oil tank defectiveness. of Sciences of Ukraine. His Methods and devices of quality control. ʋ15, reserch area – selection and 2005, processing of random signals, development of the P. 3. information-measuring systems. Authored more [4] Skalsky V. R., Andreykiv O. E., Sulym G. T.: than 100 scientific publications. Theoretical bases of acoustic emission method in mechanics of destruction. Lviv, Spolom, 2007. Roman PLAKHTIY, [5] Skalsky V. R., Oliyarnik B. O., Plakhtiy R. M., Principal Engineer of the Sulym R.I.: The portable eight-channel acoustic Department of Acoustic emission device. Lviv Polytechnic National emission diagnostics of University Radio electronics and structural elements, Physico telecommunications Bulletin. ʋ534, 2005, P. –mechanical Institute, 55. National academy of [6] Sulym R. I.: Software of acoustic emission Sciences of Ukraine. His device SKOP-8. Proceedings of XIX conference field of study – development of KMN–2005. Lviv, 2005, P. 427. of electronic devives for acoustic-emission diagnostics. Authored more than Valentine SKALSKY, 20 scientific publications. Dr. Sci (Engineering Sciences), the Head of the Roman SULYM, Junior Department of Acoustic Resercher of the emission diagnostics of Department of Acoustic structural elements, Physico emission diagnostics of –mechanical Institute, structural elements, Physico National academy of –mechanical Institute, Sciences of Ukraine. National academy of Authored more than 230 Sciences of Ukraine. papers in the area of technical diagnostics and non- National academy of destructive testing of materials and structural Sciences for young elements. scientists 2007 prize- winner. Research area – development and software implementation of methods and means of acoustic emission diagnostics. Authored more than 20 papers.

DIAGNOSTYKA’ 3(47)/2008 47 DIAZ, GARCÍA-POZUELO, BOADA, GAUCHIA: Steering Inspection By Means Of Tyre Force Measure

STEERING INSPECTION BY MEANS OF TYRE FORCE MEASURE

V. DIAZ, D. GARCÍA-POZUELO, M. J. L. BOADA, A. GAUCHIA

* Universidad Carlos III de Madrid, Department of Mechanical Engineering Avda. Universidad nº30. 28911 Leganes. Madrid. Spain. E-mail: {vdiaz; dgramos; mjboada; agauchia}@ing.uc3m.es

Summary

A few systems, such as steering, brakes or suspension, critically affect the vehicle’s safety. In light of this, it is necessary to check these elements up to a certain age in order to maintain the vehicle in optimal safety conditions. This study is part of a project about effectiveness of Periodic Motor Vehicle Inspections (PMVI), in order to suggest metodologies, instruments and criterions to inspect the vehicle´s safety. It´s well known the relationship between steering geometries and the force in contact patch. The measurement of these geometries is important to evaluate the vehicle dynamic performance. The main objective is to evaluate the steering inspection method at low speed, current rejection criteria in order to analyze the effect of these results in vehicle´s safety behaviour in real dynamic conditions. The angle that has more influence on the PMVI inspection is the toe angle, being this one the principal parameter of analysis. Experimental force in contact patch with dynamometer plate and simulation results with CarSimTM have been carried out in order to validate the methodology of steering inspection and to check the information obtained in these tests.

Keywords: Vehicle’s safety, steering inspection, dynamometer plate, lateral tyre force, PMVI.

1. INTRODUCTION have been carried out by means of CarSimTM, a well known simulation tool [2]. The recent need to increase vehicle safety leads The main objective of this study is to evaluate to study every one of the parameters that influence the steering inspection method at low speed, current traffic accidents. Different authors such as Van rejection criteria in order to analyze the effect of Schoor, 2001 [17], or G. Rechnitzer, 2000 [13], have these results in vehicle´s safety behaviour in real investigated PMVI influence on traffic safety, dynamic conditions. The angle which has more concluding that vehicle defects are a contributing influence on the PMVI inspection is the toe angle factor over 6% of crashes. (toe describes the angle between the tire's centerline Nowadays, PMVI analyses the vehicles steering and the vehicle’s longitudinal plane), being this one system. Several studies show that steering the principal parameter of analysis in this study. disalignment could be an important factor in traffic Therefore, not only the available adherence in the accidents. However, very little statistics are contact tyre-road for different toe angles will be available. It is very complicated to check that an investigated [3], but also the consequences over the accident has been produced by steering disalignment final steering characteristics of the vehicle [18]. and generally other factors influence the accident. International Motor Vehicle Inspection 2. THEORETICAL ANALYSIS Committee (CITA) does not establish a reject value for the steering inspection by means of standard Driving dynamics deal with the mechanical laws alignment plate. Moreover, spanish PMVI program that govern a vehicle’s motion with respect to the establishes steering disalignment, measured by vehicle’s properties and the ones of the road. The standard alignment plate, as a minor defect and no description of the vehicle performance is very vehicle is rejected by this measure. In this article the complex and mathematical models are required for need of measuring effectively the steering condition the design and construction of the vehicles is presented, therefore, in order to solve this problem themselves [2, 3, 4, 11, 12]. the dynamometer plate is proposed [1, 5]. For instance, the forces acting in the contact area A dynamometer plate has been used to carry out between the tyre and the road are the longitudinal or this investigation, because it allows measuring in the tangential force and the lateral or side force. The contact patch the force in the three spatial directions. longitudinal force in a straight path causes a driving Therefore, not only the lateral deviation will be tractive or braking movement. During cornering, the obtained by means of the lateral force, as carried out front wheels are at an angle to the longitudinal axis in PMVI, but also the longitudinal and vertical force. of the vehicle, thereby causing the development of In addition, to characterise the vehicle steering system in different dynamic situations, simulations 48 DIAGNOSTYKA’ 3(47)/2008 DIAZ, GARCÍA-POZUELO, BOADA, GAUCHIA: Steering Inspection By Means Of Tyre Force Measure

a lateral force FR that is responsible of lateral or side friction f R , defined as: F f R (1) R Q where Q is the weight vertical force on the wheel. When both the radial and longitudinal forces are present, the resultant force should not exceed an upper limit that could lead to the vehicle sliding off the road.

Driving Fig. 2. Two axles vehicle model Avance

V 2 FPyd= 2 d (2) BrakingFrenado gR v2 FPyt= 2 t gR (3) Fig. 1. Ellipse of adhesion limits, Kiencke U. Taking into account that Į = Fy/KĮ, where KĮ is and Nielsen L., 2000 [9] the cornering stiffness of one wheel, substituting in Eq. (2) and Eq. (3), and considering that both wheels In order to provide a safe and accurate of the same axle have two times the stiffness of one directional control, the steering and suspension of them: design on the front axle of the vehicle has lead to 2 sophisticated steering geometries where the wheel F yd V P d D d = = (4) alignment is governed by the following inclination 2gRK DDddK angles which play an important role in the vehicle 2 F yt V t dynamic performance: camber, toe, caster and = = P D t (5) steering axis angle or kingpin [8, 14]. 2gRKKDDtt

2.1. Cornering performance These equations for the slip angles for the front and rear axle can be introduced in Eq. (6), obtaining When a vehicle supports a certain lateral load Eq. (7): (wind, centrifugal force, etc.) a force is generated in L G - DDdt + (6) the tyre contact to counteract that effect. The tyre R ability to generate the needed forces so as to follow 2 the correct path is very important, as it allows L §·dtV = + PP - evaluating the safety vehicle condition. Therefore, G ¨¸ (7) RgR©¹KKDDdt the forces generated in the contact patch depend on the steering angles, and this relationship is the one or: analysed in this investigation. It has been considered 2 L V as a reference case that, in which the vehicle follows G = + K V (8) a circular path and it is subjected to a centrifugal R gR force [8,16]. = Pdt - P For low values of the steering angles (the bend K V KKDDdt (9) radius is much bigger than the vehicle wheelbase) and if the centrifugal force is applied in K , is the understeer increment and its value is a perpendicular direction to the vehicle longitudinal V of great importance to analyse the vehicle steering plane the lateral forces on the front tyres F and on yd performance. The vehicle cornering performance is the rear tyres F are: yt measured by the understeer increment sign. From Eq. (8) it is obtained the the steering angle required for a constant radius turn varies with the speed, or it will be independent, depending on the sign of KV. Thus, a vehicle can be neutral steer, understeer or oversteer. Considering R = cte: DIAGNOSTYKA’ 3(47)/2008 49 DIAZ, GARCÍA-POZUELO, BOADA, GAUCHIA: Steering Inspection By Means Of Tyre Force Measure

Neutral steer: Understeer: Oversteer: (longitudinal and lateral forces). The dynamometer KV = 0 KV > 0 KV < 0 plate allows a complete force and momentum (independant of V) (į decreases when V (į decreases when V measurement which is essential to obtain an decreases) increases) objective reject value, whereas the standard plate, by 2 2 L L V L V itself, does not provide it. The standard plate does G = G = + K V G = - | K V | R RgR RgRnot register the vertical force which is a fundamental parameter to be taken into account. The fig. 3 shows the force mesurement in this Calculating Kv allows knowing the vehicle cornering performance. It is important to highlight test, when the vehicle cross over the dynamometer that usually vehicles are designed neutral or plate. The results of the same test for different toe understeer, due to the fact that in an oversteer angles are depicted in fig. 4. vehicle the guidance angle will become negative for a certain value of V, called the critical speed (view 2500 Eq. (10)). From this point on, the steering wheel will 2000 15 0 0 have to be rotated in the opposite direction to the ) vehicle turning and the vehicle will become 10 0 0 unstable. 500 0

gL -500 V cri = (10) | K V | -1000

Lateral Force (N -1500

3. EXPERIMENTAL ANALYSIS -2000

-2500 In this study a dynamometer plate has been used -5 -4 -3 -2 -1 0 1 2 3 4 5 to obtain the instantaneous force and momentums Toe Angle (deg) values in the contact patch. The experimental data obtained by means of Fig. 4. Lateral forces variation obtained in diferents dynamometer plate has been registered in PMVI test test for nine different toes angles (PMVI conditions). conditions: a vehicle cross over the dynamometer plate at low speed (1-5 Km/h). It’s important to The data obtained by means of dynamometer emphasize that the forces measured in this speed plate has been completed with other experimental range have a very little variations (less than 2 %), results for diferent vertical loads and slips [18]. therefore the dynamometer plate shows a great These measures have been obtained by means of robustness for speed variations in the considered dynamometer drum and trailer. speed range. The input parameters of the simulation application are the measurements obtained by mean Fz 4000 250 of the dinamometer plate, dynamometer drum and Fx trailer. Fy 200 3500 Fig. 5 shows the experimental data of the lateral tire force in contact patch that was introduced in 15 0 TM 3000 simulation tool CarSim . Similar data about

10 0 longitudinal tire force and aligning moment have 2500 been introduced. N 50

) 2000

0

Fz (N 15 0 0

-50 Fx and Fy ( Fy and Fx

10 0 0

-100

500 -150

0 -200 0 5 10 15 20 25 30 35 40 45

-500 -250 Time (s) Fig. 3. Force measurement in contact patch (PMVI conditions). A dynamometer plate consists of a metallic plate Fig. 5. Experimental three-dimensional graphics of supported by eight load cells that measure the forces the relationship between longitudinal tire force and in the vertical direction and in the contact patch slip ratio for diferent vertical loads. 50 DIAGNOSTYKA’ 3(47)/2008 DIAZ, GARCÍA-POZUELO, BOADA, GAUCHIA: Steering Inspection By Means Of Tyre Force Measure

4. SIMULATION PROGRAM the geared-down steering wheel angle to the road An existing simulation program CarSimTM [10], steer angle, with Ackerman and others effects. [15] was used for this particular study. The program An important item to take into account is the is a vehicle industry standard, specifically developed performance of the tire. Parameters from a 175/65 for simulating the dynamics of vehicles with tires. It R14 tire have also been used. shows how vehicles respond dynamically to inputs from the driver and the immediate environment Table 1. Vehicle characteristics summary. (road and wind). Item Unit Value It produces the same kind of outputs that might Wheelbase m 2.49 be measured with physical tests involving instrumented vehicles. The program is based in part Front Track m 1.47 on technologies developed by the University of Rear Track m 1.445 Michigan Transportation Research Institute. Total Weight N 11368 The model of a vehicle is built up by using Front Weight N 6595.4 different components that are mathematically Rear Weight N 4772.6 described by mass and inertias. The movements of Height of centre gravity m 0.54 the different components are restricted relative to each other by connection elements (ball joints, links Front Sprung mass kg 603 and swing axels) which are also described Rear Sprung mass kg 387 mathematically. Force elements as coil springs, anti- Front Unsprung mass kg 70 roll bar, and dampers are placed between the Rear Unsprung mass kg 100 components. The force characteristics of these Roll inertia (Ixx) kg·m2 288.0 elements may be nonlinear. A tire model based on Pitch inertia (Iyy) kg·m2 1152.0 Pacejka 5.2 [12] version of the Magic Formula is Yaw inertia (Izz) kg·m2 1152.0 usually used by the program to simulate the behaviour of the tire. In this study the experimental Total length m 3.925 measurements have been used in order to achieve Width m 1.68 the desired accuracy, so that safety conditions due to Height m 1.545 steering geometry are correctly evaluated. Tires 195/60 R15 The input to the system consists of a path that the vehicle has to cross at a certain speed. Other inputs 6. TEST CONDITIONS are external forces acting on the vehicle (wind forces), the steering angle of the wheel, torque on In order to evaluate the influence of wheel the driver wheel and road excitation. For the alignement (front axel) on handling characteristics simulation of the transition closed loop handling test of the vehicle, a double lane change manoeuvre has (double lane change), it is necessary to use a driver been tested. Three vehicle conditions have been model that steers the vehicle along the prescribed compared: path. Vehicle Condition A Æ (Toe angle = 0º) Usual Outputs of the simulation program, which can be design specification. extracted against time or other variable, include over Vehicle Condition B Æ (Toe angle = 2º) Bad 500 parameters as: condition. xDisplacement, velocity and acceleration in any of Vehicle Condition C Æ (Toe angle = 4º) Very the six degrees of freedom of the sprung mass. bad condition. xTire force and moments. The double lane change (DLC) is a well-known xSpring and damping forces and displacements. and commonly used test that has been prescribed in a concept standard ISO TR-3888-1 [7]. This test 5. VEHICLE CHARACTERISTICS allows for the evaluation and comparison of the handling characteristics of vehicles through some A typical model for a small class European objective parameters such as roll angle, roll rate, vehicle has been used for the purpose of this study. yaw rate, lateral acceleration (ay) and the Dynamic The main vehicle parameters which have been used Stability Index (DSI) [6]. are listed in Table 1. The rest of parameters that are Double lane change tests are implemented with not listed in Table 1 have less influence on side a transition length of 35 m and a width of 3.5 m, behaviour and medium values obtained from following the path illustrated in fig. 5. Tests have CarSimTM database have been used. been done following the suggestion of the standard The steer due to the steering system is obtained ISO TR-3888-1 [7], which involves beginning at 40, by combining the steering wheel control with 60, 80 km/h and increasing the speed until the nominal gear ratio, while nonlinear tables combines vehicle fails the test. DIAGNOSTYKA’ 3(47)/2008 51 DIAZ, GARCÍA-POZUELO, BOADA, GAUCHIA: Steering Inspection By Means Of Tyre Force Measure

The article simulations have been carried out at the relationship between the toe angle and the lateral 60 km/h. force is approximately equal when comparing the simulation results (Fig. 7, 8 and 9) and the experimental tests (Fig. 4).

150

100

50

0 0123456Front left tire (toe: 0º) -50 Fig. 6. Asymptotic driving course for a DLC test. Rear right tire (toe 0º) -100 Stationary tests have also been considered to Lateral forces (N) Front right tire (toe 0º) -150 study the vehicle steering characteristics [16]. These Rear left tire tests allow obtaining the cornering stiffness, Kv, and Time (s) (toe: 0º) how it changes when dynamic conditions are also altered, as well as its relationship with the toe angle. Fig. 7. Vehicle” A” lateral tire forces evolution The proposed tests are detailed below: in a simulation (PMVI conditions).

- Constant radius tests: During this test the vehicle 1500 travels along a constant radius turn, at different speeds, measuring either the speed or the lateral 1000 2 acceleration ay = V /R, and controlling the 500 steering angle by means of the steering wheel

0 Front left tire (toe: angle įV, which allows knowing the wheel steering, allowing us to characterise the 01234562º) -500 Rear left tire (toe: cornering stiffness as a function of the 0º)

centrifugal force. -1000 Front right tire - Constant speed tests: Now the vehicle travels at Lateral forces (N) (toe: 2º) -1500 Rear right tire a constant speed for different steering angles, the (toe: 0º) path will be of different radius and the vehicle Time (s) will also be subjected to different lateral accelerations (V2/gR), obtaining the cornering Fig. 8. Vehicle “B” lateral tire forces evolution. stiffness as a function of the centrifugal force. 2500 - Constant steering angle test: In this case, the 2000 steering wheel angle įV is constant, and when the 15 0 0 vehicle travels at different speed different paths 10 0 0 and lateral accelerations will be obtained. Thus, the relationship between the cornering stiffness 500 0 and the centrifugal force is obtained. 012 3456 -500 Front left tire (toe: 4º) -1000 7. TEST RESULTS Rear left tire -1500

Lateral forces (N) forces Lateral (toe: 0º)

-2000 Front right tire Preliminary simulations show the lateral forces (toe: 4º) in a test for PMVI conditions (straight movement at -2500 Rear right tire 3 km/h) as fig. 7, 8 and 9 show. This information is Time (s) (toe: 0º) important to compare the simulations with Fig. 9. Vehicle “C” lateral tire forces evolution. experimental test in the same conditions (fig. 4). The results indicate the enourmous resemblance If it is taken into account that the maximum between both kinds of data. It is a normal conclusion forces that can be transmitted to the ground are those because in these conditions, at low speed, there is no limited by adherence in the longitudinal and lateral load transference neither other dynamic effects, and direction, as shown by the adherence ellipse shown the results are a direct consequence of tire data in Figure 1, it can be observed that the margin for introduced. case A is bigger than for case B, and for this one It can be seen that the introduced toe angle in bigger than case C. Thus, a vehicle travelling along each of the vehicles results in different lateral forces a straight line uses more lateral adherence when the linear proportional for the studied range (0º - 4º). toe angle is bigger, reducing its lateral adherence This implies, that the lateral force for vehicle A that margin for dynamic conditions. travels in a straight path is lower than vehicle B and Without taking into account the load transfer and this one lower than vehicle C. It can be checked that the slip angle, and considering a good pavement condition with lateral adherence value of ȝy = 0.64 52 DIAGNOSTYKA’ 3(47)/2008 DIAZ, GARCÍA-POZUELO, BOADA, GAUCHIA: Steering Inspection By Means Of Tyre Force Measure the 3500 N that each wheel of the front axle support The double lane change ISO TR-3888-1 [7] is leads to a maximum transmitted lateral force of 2240 a very hard dynamic situation, and is difficult even N. This value is very near to that achieved for a toe for the vehicle A. The vehicle B keeps the trajectory angle of 4 degrees (2100 N), that is, vehicle C will until the last corner, and the vehicle C loses the therefore have less adherence capacity than vehicles trajectory on the third corner. As it has been A or B before beginning to slip in the lateral described previously, the vehicle with a toe angle direction and less capacity to transmit longitudinal included between ±0.25º has more adherence forces. Therefore, the proposed assumptions, available than vehicles with upper toe angles. although severs, allow carrying out a first order Yaw rate, vehicle A approximation to how the system is conditioned and Yaw rate, vehicle B to justify the need to have a good steering system 60 Yaw rate, vehicle C inspection. 40 To complete this argument a simulation was done 20 in a demanding dynamic situation, double lane change 0 ISO TR-3888-1 [7], to compare the differences 012345678 introduced for three differents toe angles. -20 Initially there were obtained qualitative outputs to -40

-60 compare the different behaviours for these vehicle Yaw rate (deg/s) configurations. -80 Time (s) Fig. 12. Yaw rate (sprung masses). Simulation time: 7s

From Figure 12 it can be observed the progressive vehicle B held up concern the vehicle A, and the vehicle C concern vehicle B. The maximum values for yaw rate are similar for three vehicles. This maximum values for the vehicle A fit in with the corner positions for the double lane change test fixed.

4000 Front left tire Fig. 10. Three different trajectory described for the (V ehicle A ) 2000 vehicles: A (green), B (blue) and C (red). Rear left tire (V ehicle A ) 0 The fixed trajectory is correctly described by 01 23 456 78 Front left tire vehicle A, however, vehicles B and C could not -2000 (V ehicle B ) Rear left tire describe it correctly, as depicted in fig. 10, 11. -4000 (V ehicle B ) Vehicles that are not able to follow the trajectory Front left tire describe the most nearer trajectory to the reference -6000 (V ehicle C) Lateral forces(N) Rear left tire one under limit adherence conditions. It is defined -8000 (V ehicle C) a performance measure to evaluate the ability of the Time (s) vehicle to follow a predetermined path as the maximum lateral deviation with respect to the Fig. 13. Tire lateral forces (Fy), left side. reference one. For vehicle condition A, the Simulation time: 7s maximum lateral deviation is 1.5 m, for vehicle B, 2.5 m and 7 m for vehicle C. It is very interesting to observe how at the beginning of the test the lateral forces are very Traject o ry vehicle A Traject o ry vehicle B different, before the first corner, as shows fig. 13.

) 15 Traject o ry vehicle C Before the first corner, vehicle C already consumes almost all force necessary to describe the corner and 10 therefore, it does not trace the correct trajectory. 5 From that moment on the vehicles with the three configurations cannot describe the same trajectory; 0 020406080100120 their initial conditions do it impossible. -5 From above, it arise the need to characterise in

-10 an objective and precise procedure the steering Y: Lateral position (m position Lateral Y: characteristics of a vehicle. Therefore, simulations X: Longitudinal position (m) of the stationary tests were carried out in order to Fig. 11. Trajectory for the three different toe angles. compare the values of the cornering stiffness for Simulation time: 5s. different vehicle configurations. DIAGNOSTYKA’ 3(47)/2008 53 DIAZ, GARCÍA-POZUELO, BOADA, GAUCHIA: Steering Inspection By Means Of Tyre Force Measure

Next, different figures show the proposed tests (Fig. 14, 15, 16 and 17), and the influence of the Speed=70Km/h and Steer angle=10º/s cornering stiffness is analysed for each of them. 400 350 300 RADIUS=200m Vehic. A 250 80 Vehic. B 200 Vehic. C ) 60 150 Line Kv=0

Steer angle (º) 100 Vehic. A 40 50 Vehic. B 0 Vehic. C 20 00,51 Steer angle (º angle Steer Lateral acceleration (g's) 0 00,51 Fig. 17. Constant speed of 70km/h constant steer angle variation 10º/s. Lateral Acceleration (g's)

Fig. 14. Stationary test: constant radius In Fig. 14, 15, 16 and 17 it can be seen that the turns of 200m. vehicle with bigger understeering behaviour is the vehicle with bigger toe angle. Specifically, in Fig. RADIUS=100m 14 and 15 it can be seen that for different radius 400 turns the graphic slope is more positive for bigger values of the toe angle. If it is taken into account that ) 300 the slope of the graphic has a direct proportion with Vehic. A KV, which will be negative when the slope is 200 Vehic. B negative (oversteer), equal to zero when the slope is Vehic. C zero (neutral) and positive when the slope is positive 100 (understeer), it is clearly observed that there is an Steer angle (º angle Steer increase of the understeer behaviour for bigger 0 values of the toe angle. On the other hand, Figures 00,51 16 and 17, also show that there a relationship between the graphic slope and Kv, but in this case Lateral Acceleration (g's) the slope of 45º corresponds to Kv =0, slopes bigger than 45º will be for understeer vehicles and for Fig. 15. Constant radius turns of 100m. lower slopes the vehicle will be oversteer. For the In Fig. 14, 15, 16 and 17 the most representative case of constant speed the vehicle will have an results for the stationary tests are depicted. unstable steering behaviour for negative slopes, Therefore, although other simulations with constant which is first arrived for bigger values of the toe steering angle and many others for turns with angle, as shown in Fig. 17. All of these results different radius, different speeds, etc. because the indicate the due to the fact that the available four figures show the most important information adherence capacity is decreased due to the increase about steering behaviour. of the toe angle, the vehicle suffers a slip of the front axle leading to an increase of the understeer Speed=110Km/h and Steer behaviour. However, this analysis has been carried angle=1,5º/s out for vehicles with different types of suspension 25 configurations and it has been seen that understeer 20 behaviour strongly depends on the type of suspension employed. Vehic. A 15 Nowadays, and taking into account the tests and Vehic. B simulations shown in this investigation, a possible Vehic. C 10 reject limit for the dynamometer plate of ȝ = Fy/Fz Line Kv=0 y

Steer angle (º) = 0.15 is proposed. However, no experimental tests 5 have been carried out to validate that value. The 0 possible range to establish a reject value of ȝy is between 0.12 and 0.2. For the case of the analysed 0 0,1 0,2 0,3 Lateral Acceleration (g's) vehicles: A, B and C, ȝy would be very near to 0.065, 0.57 to 1.14 respectively. However, it has to Fig. 16. Stationary test: constant speed of 110km/h be taken into account that extreme cases have been constant steer angle variation 1.5º/s. considered; a value of 0.15 for this vehicle implies 54 DIAGNOSTYKA’ 3(47)/2008 DIAZ, GARCÍA-POZUELO, BOADA, GAUCHIA: Steering Inspection By Means Of Tyre Force Measure

a toe angle of 0.5º in each test, which is bigger than Influence on Vehicle Dynamic Performance, the conventional values fixed during steering design. SAE Paper No. 700377, (1970). [4] Gim G. and Nikravesh P E.: An Analytical 8. CONCLUSION Model of Pneumatic Tires for Vehicle Dynamic Simulations. Part 1. Pure Slips, International x Measurements with dynamometer plate in Journal of Vehicle Design, 11(2):589-618, periodic inspection conditions, double lane (1990). change simulations and stationary conditions [5] Gobbi M., Mastinu G., Giorgetta F.: “Sensors simulations have been carried out. The results for measuring forces and moments with obtained show the need to carry out some kind of applications to ground vehicle design and steering inspection to guarantee the safety engineering”, 2005 ASME IMECE, November vehicle conditions. 5-11, 2005, Orlando, Florida USA (2005). x Simulations show the toe angle influence, and [6] Heydinger G. J., Howe J. G.: “Analysis of the important behaviour dynamic variations. vehicle response data measured during severe Also it has been analyzed the camber angle, but manoeuvres”, SAE Paper 2000-01-1644, 2000. its influence is notably fewer than the toe angle [7] ISO TR-3888-1 Passenger Cars. “Test track for influence, and for this reason it has not been a severe lane-change manoeuvre”, included in this paper. International Organization for Standardization, x Other steering angles have not been studied, 1999. because they should not change in the vehicle [8] Dixon J. C.: Tires, Suspension and Handling, life, except if the vehicle suffers an accident or Society of Automotive Engineers, (1996). an important reform. These situations are out of [9] Kiencke U. and Nielsen L.: Automotive Control the objectives of a periodic inspection and out of System for Engine, Driveline and Vehicle, this study. Springer-Verlag, Berlin, 412 p., 2000. x For the cornering vehicle behaviour, it has been [10] Mechanical Simulation Corporation, shown that the vehicle becomes understeer for “CarSimTM Reference Manual”, 2005. bigger values of the toe angle. This is due to the [11] Muñoz B.: Modelo de comportamiento de increase of the slip of the front axle because the placa alineadora. ThD Thesis, EPS available adherence is diminished for bigger Universidad Carlos III de Madrid (2001). values of the toe angle. [12] Pacejka H. B., Sharp R. S.: “Shear force x Dynamometer plate measurements have shown development by pneumatic tyres in steady state to be representative of vehicle safety. conditions: A review of modelling aspects”, x In future studies other kind of dynamic tests will Vehicle System Dynamic, Vol. 20, pp. 121- be carried out, and dump conditions, toe angle 176, 1991. influence in emergency brake, etc. will be [13] Rechnitzer G., Haworth N., Kowadlo N. 2000. included. Furtheremore the dynamometer plate The effect of vehicle roadworthiness on crash system will be analyzed as an instrument used to incidence and severity. Monash University carry out periodic motor vehicle inspections. Accident Research Centre. Report nº 164, 2000. ACKNOWLEDGEMENTS [14] Sakai H.: Theoretical and Experimental Studies The authors gratefully acknowledge the funds on the Dynamic Properties of Tyres III. provided by the Spanish Government through the Calculations of the Six Components of Force CICYT Project TRA2004-03976/AUT. and Moment of a Tyre, International Journal of Vehicle Design, 2(3):182-226, (1981). REFERENCES [15] Sayers M. W., Mousseau C. W., Gillespie T. D.: “Using simulation to learn about vehicle [1] Blab R:. Introducing Improved Loading dynamics”, International Journal of Vehicle Assumptions into Analytical Pavement Models Design, Vol. 29, Nos. 1 / 2, pp. 112-127,2002. Based on Measured Contact Stresses of Tires, [16] Gillespie T. D.: Fundamentals of vehicle International Conference on Accelerated dynamics, Society of Automotive Pavement Testing, Reno, NV, Paper Number: Engineers.Inc, (1992). CS5-3, (1999). [17] Van Schoor O., van Niekerk J. L., Grobbelaar [2] Díaz V., Ramírez M. Muñoz B.: A wheel model B., 2001. Mechanical failures as a contributing for the study of the wheel angle measurement cause to motor vehicle accidents- South Africa. in the periodic motor vehicle inspection. Accident Analysis and Prevention 33, 713-721, International Journal of Vehicle Design, vol.34, 2001. Nº3. pp.297-308 (2004). [18] Woltson B. R.: Wheel Slip Phenomena, [3] Dugoff H., Fancher P. S. and Segel L.: An International Journal of Vehicle Design, Analysis of Tire Traction Properties and Their 8(2):177-191, (1987). DIAGNOSTYKA’ 3(47)/2008 55 BARTELMUS, ZIMRÓZ, Problems And Solutions In Condition Monitoring And Diagnostic Of Open Cast …

PROBLEMS AND SOLUTIONS IN CONDITION MONITORING AND DIAGNOSTIC OF OPEN CAST MONSTER MACHINERY DRIVING SYSTEMS

Walter BARTELMUS, Radosáaw ZIMROZ

Wrocáaw University of Technology Diagnostics and Vibro-Acoustics Science Laboratory Pl. Teatralny 2, Wroclaw, 50-051, Poland Tel. (+48 71) 320 68 49, Telefax (+48 71) 344 45 12 E-mail: {walter.bartelmus, radoslaw.zimroz}@pwr.wroc.pl

Summary In the paper is given discussion on demands connected with vibration signal analysis for condition monitoring of driving systems of monster machinery used in open cast mines. Because of external load variation which cause that the vibration signal generated by driving systems is no-stationary there is a need to use special techniques for signal analysis. The description of proper signal analysis techniques is given together with machinery description and external load characteristic. The description leads to design, production technology, operation, change of condition factors analysis [1].

Keywords: condition monitoring, mining machines, open cast mines, external load variation.

PROBLEMY I ROZWIĄZANIA W DIAGNOSTYCE UKàADÓW NAPĉDOWYCH MASZYNOWYCH GIGANTÓW PODSTAWOWYCH GÓRNICTWA ODKRYWKOWEGO

Streszczenie W pracy przedstawiona jest dyskusja o wymaganiach związanych z analizą drganiowego sygnaáu dla monitorowania ukáadów napĊdowych maszynowych gigantów uĪywanych w górnictwie odkrywkowym. PoniewaĪ drganiowe sygnaáy generowane przez ukáady napĊdowe na skutek zmiennoĞci obciąĪenia generują niestacjonarny sygnaá istnieje potrzeba wykorzystania specjalnych technik do analizy sygnaáu. Przedstawiono opis niektórych maszynowych gigantów wraz z technikami analizy sygnaáów i charakterystyki zmiennoĞci obciąĪenia. Opis prowadzi do analizy czynników konstrukcyjnych, technologicznych, eksploatacyjnych i zmiany stanu [1].

Sáowa kluczowe: monitorowanie stanu, maszyny górnicze, górnictwo odkrywkowe, zmienne obciąĪenia.

1. INTRODUCTION 2. SOME MONSTER MACHINERY DRIVING SYSTEMS For driving of open cast machinery there are used systems which consists of electric motor, Fig. 1 shows a driving station for belt conveyor hydraulic coupling or damping mechanic coupling systems with the scheme of a driving subsystem. (in some cases both) and different type of gearboxes. The driving station incorporates three or four Taking into consideration type of the external load, subsystems. Fig. 2 shows a bucket wheel during the which may be classified as: varying slowly with operation and the scheme of bucket wheel drive with random cycles of many minutes or a few minutes, three subsystem positions. The discussed bucket varying quickly with cycle from less than a second wheel drive has three independent driving systems to a few seconds. For driving systems of belt driven by three electric motors. The subsystem conveyors the length of cycle depends mainly of the consists of planetary gearbox with a fixed rim z3 and length of a conveyor and from the current output of three cylindrical stages of gearboxes z4 – z9. One of bucket wheel or chain excavators. Taking into alternative solutions for a compact drive used for consideration driving systems a bucket wheel load driving bucket wheels is given in Fig. 3. The system period variation depends on a bucket frequency. consists of a bevel gear stage and a special solution Another period for bucket wheel excavators is gearbox which gives possibility of driving power connected with the slewing of a bucket wheel. The distribution into two ways. This possibility is given external load of driving systems causes non- by planetary gearbox in which the sun z3 and rim z5 stationary vibration which is the signal for the are rotated. driving system condition. This situation demands special treatment used for vibration signal analysis. 56 DIAGNOSTYKA’ 3(47)/2008 BARTELMUS, ZIMRÓZ, Problems And Solutions In Condition Monitoring And Diagnostic Of Open Cast … a) 2. EXTERNAL LOAD DESCRIPTION

The bucket wheel design which comes from digging principle is the reason of periodic variation of external load. Much longer period of load variation is connected with the boom slewing. The other operation factors influencing the load variability are connected with digging ground properties. These design and operation factors leads to external load variations and closely connected with generated vibration represented by time courses b) given in Fig. 4. In Fig. 4a) is given electric current consumption variation which represents directly load variation. The vibration signal variation is given in Fig. 4c). Fig. 4a) and c) show positive correlation between load and vibration. Fig. 4b) and c) shows negative correlation between load and rotation speed or between vibration and rotation speed. Detailed signal analysis can show load variations connected with bucket frequency. The vibration signal showing the value proportional to load variation is given in Fig. 1. a) General view of belt conveyor driving Fig. 10. The vibration signal procedures for load system, b) scheme of one subsystem variation extraction is given in [2-4]. a) a) b)

c) b) c)

Z 1 Z3 Z2

Z5 Z4 Z7

Z9 Z6

Z8 Fig. 2. a) Bucket wheel during operation Fig. 4. a) Electric current consumption variation b) scheme of subsystem bucket wheel drive b) input shaft rotation speed RPM variation c) subsystem positions c) vibration signal variation Z6 Z7

Z4 Z3 4. FACTORS INFLUENCING DIAGNOSTIC

Z5 SIGNALS nj n3 nj

Z8 Z9 Z8 Fig. 5 and 6 show general and detailed division of factors having influence the vibration signal. Fig. 5 shows that primary and secondary factors having influence generated vibration signal can be divided n 2 Z1 into four groups of factors. More details for further

Z2 n1 division of factors is given in Fig. 6. The detailed description of factors having influence to vibration Fig. 3. Bucket wheel alternative driving system signals is given in [1] it also well to look into publications [5  7] where many examples are given where different factors influencing vibration signals are considered. DIAGNOSTYKA’ 3(47)/2008 57 BARTELMUS, ZIMRÓZ, Problems And Solutions In Condition Monitoring And Diagnostic Of Open Cast …

Factors division

primary secondery

design production technology operation change of condition

Fig. 5. General division of the factors influencing diagnostic signals

Factors division

primary secondary

design production technology operation change of condition

load geometric material rotational speed wear environment

seizing macrogeometrical microgeometrical moduli of elasticity pitting other other

structural form admissible tolerances condition shape errors of surface other

Fig. 6. More detailed division of factors influencing diagnostic signals

5. VIBRATION SIGNALS ANALYSIS a) 4000

The choice of a vibration signal analysis depends 3500 of an external load variation. If the signal is constant 3000 or with the slow variation the first step for signal 2500 y c analysis is a spectrum analysis [69]. It is used for n e 2000 qu the object presented in Fig. 1. For deeper analysis er are used also other techniques for signal analysis F 1500 like cepstrum given in Fig. 7 and time-frequency 1000 spectrograms given in Fig. 8. 500

0 0 0.5 1 1.5 2 Time b)

y4 gear5 0.1

0.09

0.08

0.07 0.06 Fig. 8. Time-frequency [s]-[Hz] spectrograms 0.05 a) spectrogram of signal without regular peaks 0.04 b) spectrogram of signal with marks equivalent 0.03 regular peaks 0.02 0.01 Details for these techniques of signal analysis are 0 0 0.2 0.4 0.6 0.8 1 1.2 given in publications [6, 9]. Further analysis is connected with Fig. 9 where Fig. 7. Acceleration [m/s2] signal time [s] trace classification of gearing condition is given it is with series of peaks (peaks marked with arrows) based on publications [7, 8]. It should be noticed that and cepstrum for signal the inclination of the line separated the gear 58 DIAGNOSTYKA’ 3(47)/2008 BARTELMUS, ZIMRÓZ, Problems And Solutions In Condition Monitoring And Diagnostic Of Open Cast … condition classes are also the measures of a gear occurrence of distinct perturbations at the period condition. equivalent the second harmonic of the arm rotation frequency, the gear condition before replacement of the gearbox b) occurrence of distinct perturbations at the period of the arm rotation frequency Ta for an other gearbox in a bad condition. a)

Fig. 9. Effect of load on gear transmission condition symptom value, accelerations in band 100-3500 Hz: AyD – gear transmission classes; a d – gear transmission points, number of gear y b) transmissions [7, 8]

The obtained results of signal analysis lead to identification of distributed and local faults and it is also the measure of the increase of a backlash in rolling elements bearings. More details about increase of the backlash are given for considering the case when the cycle varying load occurs. Non-stationary vibration signal generated by some monster machinery needs special techniques for signal analysis. The signal analysis should give the background for condition evaluation and Fig. 10. Envelopes proportional to load variation: maintenance decisions. As the result of signal analysis one can expect detection of local and a) signal from bad gearbox before replacement  bad distributed faults and increase of elements play condition, b) signal from new replaced gearbox  caused by the increased bearing backlash as result of good condition harsh mine environment. For identification of the external load variation can be used signal a) demodulation techniques to obtain results as it is given in Fig. 10, [24] which envelopes are proportional to load variation; a) signal from gearbox before replacement b) signal from replaced gearbox. It can be seen that if the system is in bad condition it more yields under varying cyclic load as is given in Fig. 10a). Variability of some parameter, which characterized the digging process are given in Fig. 10 to 12. In Fig. 10 and 11 are compared results of b) signal analysis of two gearboxes in bad and good condition. Fig.10 gives envelopes proportional to load variation: a) signal from a bad gearbox before replacement b) signal from a new replaced gearbox. Fig. 11 gives time [s] – frequency [Hz] spectrograms a) signal from a gearbox before replacement b) signal from a replaced gearbox. Fig. 10 shows the periodic variation of the signal with the bucket digging period Tb. this period is also depicted in the time-frequency spectrogram given in Fig. 11. Additionally one can noticed weak a period Fig. 11. Time [s] – frequency [Hz] spectrograms: connected with a planetary gearbox arm/carrier a) signal from gearbox before replacement rotation Ta. Fig. 12 gives more clear evaluation of (bad condition) b) signal from replaced gearbox the period by Wigner-Ville distribution with change (good condition) in condition of the arm of the planetary gearbox: a) DIAGNOSTYKA’ 3(47)/2008 59 BARTELMUS, ZIMRÓZ, Problems And Solutions In Condition Monitoring And Diagnostic Of Open Cast … a) 2.5

2

1.5

1 Acc [m/s2]Acc 0.5

0 0 500 1000 1500 2000 [Hz] 2.5

2 b) 1.5

1 Acc [m/s2] Acc 0.5

0 0 500 1000 1500 2000 [Hz] Fig. 13. Examples of short time segment frequency analysis a) 8

7 Fig. 12. Evaluation of Wigner-Ville distribution with 6 change in condition of the arm of the planetary 5 gearbox a) occurrence of distinct perturbations at 4

second harmonic of the arm rotation frequency, 3

condition before replacement of the gearbox 2 b) occurrence of distinct perturbations at the arm 1 rotation frequency for an other gearbox 0 in a bad condition 400 410 420 430 440 450 460 470 480 490 500 b) Fig. 3 shows the scheme of a driving system which condition will be evaluated. It is a compact 9 system which consists of an electric motor, one stage 8 bevel gear (gears z1, z2), planetary gearbox with 7 rotating sun z3 and rotating rim z5 and a planet z4. 6 The power in the compact system is transmitted to 5 the main gear z9 through two pinions z8 to 4 accomplish it two gear are added z6 and z7. The 3 variation of different parameters characterizing 2 phenomenon of bucket wheel operation are given in 1 0 Fig. 4. Fig. 4 shows in c) an input shaft rotation 400 410 420 430 440 450 460 470 480 490 500 speed variation in RPM a) electric current Fig.14. a) Smeared spectrum caused by frequency consumption variation b) vibration signal variation. modulation b) non smeared spectrum Fig. 9 shows positive correlation between stopien planetarny sumy investigated values; Fig. 15 shows negative 35 correlation for investigated values. The linear 30 characteristics can be expected only for some range of an increased backlash in rolling elements 25 bearings. 20 To obtain vibration parameters the signals were 15 processed to get vibration spectrums as is given in Fig. 13. It should be stressed that the spectrums 10 should be done for short time intervals to obtain no 5 smeared spectrums. The intervals are equivalent to

0 a bucket digging period Tb. The examples for the 945 950 955 960 965 970 975 980 985 smeared and non smeared frequency component are given in Fig. 14. Fig. 15. Vibration signal RMS component sum versus rotational speed for planetary stage 60 DIAGNOSTYKA’ 3(47)/2008 BARTELMUS, ZIMRÓZ, Problems And Solutions In Condition Monitoring And Diagnostic Of Open Cast …

Planning and equipment selection Torino Italy, 20-22 September 2006 pp.71-76. [4] Bartelmus W. Zimroz R.: Vibration condition monitoring of planetary gearbox under varying external load, Mechanical Systems and Signal Processing (accepted for publication). [5] Bartelmus W. Mathematical Modelling and Computer Simulations as an Aid to Gearbox Diagnostics. Mechanical Systems and Signal Processing 2001 Vol.15, no. 5, pp. 855-871. Fig. 16. Graphic interface for data analysis system [6] Bartelmus W.: Computer-aided multistage gearbox diagnostic inference by computer Fig. 16 shows a graphic interface for a data simulation. Scientific Papers of the Institute of analysis system. In Fig. 16 one can see the course of Mining of Wroclaw University of Technology. rotation frequency variation in [Hz] as a function of No.100, 2002 (available on Internet at time [s] (at the gearbox side of hydraulic coupling), Dolnoslaska Digital Library). distribution of the rotation frequency [Hz], variation [7] Bartelmus W.: Condition monitoring of open of acceleration signal [m/s2] as a function of time, cast mining machinery. Oficyna Wydawnicza the plot of vibration parameters as a function of Politechniki Wrocáawskiej Wrocáaw 2006 [RPM]. The plot has been obtained as the results of (available on the Internet at Dolnoslaska Digital several vibration signals as are given in Fig. 4b) The Library). values of parameters equivalent 60[s] period for [8] Bartelmus W.: Vibration condition monitoring of a gearbox in good condition is given by green dots. gearboxes. Machine Vibration 1992 nr1 s.178- From the vibration signal given in Fig. 4 part of the 189. results has been automatically rejected. Into [9] Bartelmus W.: Diagnostic information on consideration are taken only spectrums, which have gearbox condition for mechatronic systems. been obtained for a proper loaded gearbox for the Transaction of the Institute of Measurement and distribution of the rotation frequency similar to Control 25,5 (2003), pp. 450-464. distribution given in Fig. 16. For the data presented in the plot given in Fig. 15, which is also given in Prof. dr hab. inĪ. Walter Fig. 16 the linear regression analysis has been done BARTELMUS. Specjalista and obtained the equation of the external load z zakresu eksploatacji, dynamiki yielding statistical characteristics in the form i diagnostyki maszyn. Autor y = ax + b for good and bad planetary gearbox. okoáo 150 publikacji w tym jednej ksiąĪki oraz wspóáautor 6. CONCLUSIONS trzech monografii. Pracownik Politechniki ĝląskiej (1967-1980) The Paper shows the different ways of signal Instytutu Górnictwa Odkrywkowego, Wrocáaw analysis, which have been used for condition (1980-1993), Politechniki Wrocáawskiej (od 1994). monitoring and diagnostic for monster machines Stypendia i wykáady: Polytechnic of Central used in open cast mines. Some of the ways are London, Swansea University College, Southampton connected with traditional ways of the signal Institute Systems Engineering Research Center, analysis; these ways can be used when external load Gunma University, Laboratoire d’Analyse des is constant or almost constant. There have been also Signaux & des Processus Industrieis,. Wykáady na shown new ways of signal analysis and its zaproszenia i w ramach programu Sokrates- interpretations. These ways are suitable for the Erasmus: Vaxjo University, Lulea University, varying external load. Compiegne UTC, Cranfield University.

LITERATURE Radosáaw ZIMROZ, PhD. MSc from Faculty of Electronics [1] Bartelmus W.: Root cause analysis of vibration (speciality -Acoustics) Wroclaw signals for gearbox condition monitoring. University of Technology, 1998, BINDT Inside Journal Vol 50 No 4 April 2008. PhD (2002) at Mining Faculty, [2] Bartelmus W., Zimroz R.: Influence of random WUT (with honours). varying load to vibration generated by planetary Until now with Machinery gearbox driving bucked wheel in excavators. Systems Group at Institute of International Conference on COMADEM Lulea Mining Engineering, WUT. 9 months postdoc Sweden June 2006. (2004) at Cranfield University (SOE/PASE/AMAC), [3] Bartelmus W., Zimroz R.: Bucket wheel UK. variability identification on vibration analysis. Proceeding of the International Symposium Mine DIAGNOSTYKA’ 3(47)/2008 61 YFANTIS, KAPASAKIS, Marine Gas Turbine Performance Diagnostics: A Case Study

MARINE GAS TURBINE PERFORMANCE DIAGNOSTICS: A CASE STUDY

Elias A. YFANTIS(1) and Panaghiotis KAPASAKIS(2) (1) Professor and Dean (2) Lieutenant Commander HN, Department of Naval Sciences Hellenic Naval Academy, Hatzikiriakio - 18539 Piraeus - Greece tel +302104581644 fax +302104181768 e-mail: [email protected]

Summary The paper discusses how performance models can be used for marine gas turbines. A particular performance model, built for on board training purposes is employed to demonstrate the different aspects of this process. The model allows the presentation of basic rules of gas turbine engine behavior and helps understanding different aspects of its operation. A smart designed graphics user interface is used to present engine operation in different ways: operating line, operating points of the components, interrelation between performance variables and parameters etc. The perception of fault signatures on monitoring parameters is clearly demonstrated. Diagnostics capabilities of the existing performance model installed at the Hellenic Naval Academy Gas Turbines Virtual Lab are examined and tested using the measurement data from real marine gas turbines. A case study is discussed in this paper where the model results and the real engine conditions are compared for diagnostic purposes.

Keywords: gas turbines, performance, diagnostics.

1. INTRODUCTION combustor with mechanically independent high pressure and low pressure rotating systems. Vessel’s The propulsion system of a specific type of propeller shaft can be driven by either one or both frigates consists of a conventional two shaft gas turbines at any given time [3]. CODOG configuration with two MTU 956 20V TB A computer model produces very easily a lot of 82 cruise diesel engines; two boost GE LM 2500 gas information that would be difficult, expensive and turbines, Renk Tacke reduction gearboxes and two some times even impossible to obtain on an actual Escher Wyss controllable pitch propellers. engine. Engine behavior can be studied at all Gas turbines are complex engineering systems possible permissible operating conditions, while and performance monitoring and fault diagnosis is even physically non-permissible conditions can be a subject that has to be addressed in gas turbine examined, if sufficiently deep modeling is involved. related –onboard and off board- training courses for Any physical quantity can be observed, without the all level of personnel involved with gas turbine need of expensive instrumentation, which must be operation and maintenance. Time and means for used on an actual engine. Even quantities that would education and training are limited and therefore they be impossible to obtain due to geometrical or must be used in an optimized manner. Computer operating restrictions can be obtained (for example, assisted training comes to provide tools to increase turbine entry temperature, interstage pressure for training efficiency and fulfil such requirements [1]. a multistage compressor or turbine etc). In addition, The present paper discusses how marine gas turbines for given ambient conditions and turbine inlet condition monitoring and fault diagnosis software is temperature, the change in the operating parameters used for practical purposes. (mass flow rate, compressor discharge pressure, Data from a GE LM 2500 marine gas turbine is compressor discharge temperature, fuel mass flow used for the examination and testing of the existing rate, which is directly related to specific fuel performance model installed at the Hellenic Naval consumption and thermal efficiency, and exhaust gas Academy Gas Turbines Virtual Lab [2]. A case temperature), when the compressor is fouled, can be study is discussed where the report from the practice directly and accurately evaluated. and the software generated malfunctions are compared. The compressor efficiency is the 3. UNDERSTANDING THE EFFECTS independent variable affecting a series of operating OF MALFUNCTIONS parameters. The error between the calculated and the actual values is finally obtained. The advantages of computer models implementation become even more pronounced 2. PROPULSION PLANT PERFORMANCE when operation under abnormal conditions, namely deteriorated engines or engines with faults, are LM 2500 is an aero-derivative robust engine of considered. If experience is to be gained by increased power output and high efficiency observing actual engines this will have to happen developed and manufactured by General Electric. either by (a) studying cases where faults have Overall engine design includes an annular occurred on an operating gas turbine or (b) by 62 DIAGNOSTYKA’ 3(47)/2008 YFANTIS, KAPASAKIS, Marine Gas Turbine Performance Diagnostics: A Case Study setting up tests in which faults have been artificially introduced. The understanding of the effects of malfunctions can be achieved through the simulation of component faults. Such faults are simulated by modification of the performance characteristics of the components. The modified characteristics are then introduced into the model and the deviations of cycle or performance parameters are observed. In this way one can, for example, demonstrate very easily the performance drop due to compressor fouling or exhaust gas temperature variations due to turbine nozzle erosion [1]. Map modifications are effected by using scalars, multiplying the component performance parameters. Such scalars have been, for example, introduced in the past under the term ‘‘modification factors’’ MF Fig. 1. GE LM 2500 performance layout defined as MF=X/Xref, where X is the current value of a parameter and Xref its value for a component in 4. GAS TURBINE PERFORMANCE AND intact condition [4]. ASSUMPTIONS Introduction of malfunctions gives several possibilities for demonstrating their effect on engine The information provided for the GE LM 2500 is parameters. First of all, by performing simulations based on engines having clocked 2500 working for healthy and faulty engines the impact of faults hours and on PLA actuator (torque motor located on can be directly assessed. The important notion of the GT main fuel pump) value of 72. PLA value ”fault signature” can be very easily introduced, when corresponds to 72 % propulsion plant control levers the above-mentioned possibility exists. Values of – engine loading conditions, 2500 RPM power measured quantities can be calculated for both turbine rotating speed and 22 knots vessel’s speed. healthy and faulty operation and their differences are Controlled pitch propeller variations were negligible calculated to provide the signatures. The model used and the ratio P/D attained a constant value of 1.45 is the one installed in the Hellenic Naval Academy [5, 6]. Gas Turbines Virtual Lab and it is equipped with the The following assumptions were made regarding capability of directly evaluating such a signature, the real gas turbine operation and the software whenever a fault is simulated. This particular simulation: application was developed by the National Technical University of Athens/ Laboratory of Thermal 4.1. No inlet or exhaust losses Turbomachines in cooperation with the Hellenic The pressure drop through the inlet air barrier Naval Academy/Naval Engines Laboratory and screen is estimated to be about 4 in H20 at maximum features a powerful, user friendly, functional, flow rate. The maximum total pressure loss is training tool which comes up with fast and safe estimated not to exceed 12 in H20 measured at the conclusions about engine healthy behaviour and inlet bell mouth, while the back pressure is estimated operating trend both in moderate and extreme to be about 6 in H20 and not to exceed 20 in H20, conditions. The detailed description of the specific static pressure, measured at the exhaust extension capabilities of the Gas Turbines Virtual Lab and its outlet [5, 6]. characteristics (performance simulation and monitoring) has already been demonstrated [1, 2]. 4.2. Constant atmospheric pressure In this paper special attention is paid on the Atmospheric pressure at sea level is considered software diagnostic capabilities. to be constant getting a default value of 1013 mbars. A layout of the screen related to the GE LM 2500 performance is given in the following Fig. 1. 4.3. Fuel heating value The fuel used has a lower heating value (LHV) of 42800 KJ/Kg which refers to F-76 fuel properties, regularly feeding marine gas turbines.

5. RESULTS AND DISCUSSION

A series of runs was performed and the results are compared to the actual values (measurements) available from the practice. To simulate the existence of a malfunction, modification factors with values different from unity are introduced. DIAGNOSTYKA’ 3(47)/2008 63 YFANTIS, KAPASAKIS, Marine Gas Turbine Performance Diagnostics: A Case Study

In the case study examined in this paper the compressor has ‘‘experienced’’ a change in efficiency as a percentage of the reference value, namely -3%, -1%, +1%, +3%. The influence of the four different efficiency levels on five different parameters is examined, i.e. mass flow rate (W), compressor discharge pressure (CDP), compressor discharge temperature (CDT), fuel mass flow rate (Wf), which is directly related to specific fuel consumption (SFC) and thermal efficiency (Șth), and exhaust gas temperature (EGT). Load is the sixth parameter demonstrated in the fault signature, although load assumed to remain constant throughout the entire case study. Fig. 4. 101% of the reference compressor efficiency In the following fig. 2 the effect of a 3% reduction in the compressor efficiency on the gas In the following fig. 5 the effect of a 3% turbine performance is presented. improvement in the compressor efficiency on the gas turbine performance is presented.

Fig. 2. 97% of the reference compressor efficiency Fig. 5. 103% of the reference compressor efficiency

In the following fig. 3 the effect of a 1% According to the previous figures when the reduction in the compressor efficiency on the gas compressor efficiency changes from 97% to 103% turbine performance is presented. of the reference value, the following results are obtained. Mass flow rate (W) changes from 97% to 103% of the reference value. There is a very slight deviation on the compressor discharge pressure (CDP) related to the reference value. Compressor discharge temperature (CDT) changes from 104% to 96% of the reference value. Fuel mass flow rate (Wf) changes from 103.5% to 96.5% of the reference value. Finally, exhaust gas temperature (EGT) changes from 106.5% to 93.5% of the reference value. In the following table data from the practice (real gas turbine behavior) and simulation program results (prediction of the virtual gas turbine behavior) are being presented and compared. The error is derived Fig. 3. 99% of the reference compressor efficiency through the following equation: calculated value  actual value %error u100 In the following fig. 4 the effect of a 1% actual value improvement in the compressor efficiency on the gas The calculated and actual values used for the turbine performance is presented. estimation of the error are related to the specific case when a reduction of 3% on the compressor efficiency is observed. 64 DIAGNOSTYKA’ 3(47)/2008 YFANTIS, KAPASAKIS, Marine Gas Turbine Performance Diagnostics: A Case Study

Table 1. Error between calculated and actual values [5] LM 2500-30 Marine Gas Turbine Performance OPERATING ERROR Data, MID-TD-2500-8, Revised July 1981. PARAMETER % [6] Training Manual, LM 2500 Marine Gas Turbine, Mass flow rate (W) -1.8 System Description (29/1/1993). Compressor discharge -2.5 pressure (CDP) Elias A. YFANTIS is Compressor discharge a Professor of Naval +3.9 temperature (CDT) Engines, Director of Fuel mass flow rate (Wf ) +2.7 the Naval Engines Exhaust gas temperature Laboratory, Academic +4.2 (EGT) Dean and member of the Academy Council The error obtained shows that the simulation of the Hellenic Naval software underestimates W and CDP and Academy. He has overestimates CDT, Wf and EGT. a Diploma and a PhD Degree in Mechanical 6. CONCLUSIONS Engineering. He is a member of the ASME, the SAE and the New York The Hellenic Naval Academy Gas Turbines Academy of Science. He is a member of the Applied Virtual Lab is used in order to act as a malfunction Vehicle Technology Panel of the RTO/NATO and generator and a diagnostic tool. Data from the member of the “International Advisory Board of the practice is used in order to evaluate the accuracy of Centre for Gas Turbine Diagnostics and Life Cycle the simulation software calculated values. As a case Cost”, Cranfield University. His research interests study the change in the compressor efficiency is are Thermodynamics and Combustion, Engine considered. Condition Monitoring, Simulation and Diagnostics Particular aspects, which can benefit from the and Virtual Laboratories. use of computer models, have been discussed. Specific simulation software designed and developed for training and – potentially- diagnostic purposes has been used, examined and evaluated. The evaluation results presented a significant accuracy when the software is used as a malfunction generator and diagnostic tool, since the error obtained is less than 4.5% for the parameters chosen to be under observation.

REFERENCES

[1] Mathioudakis, K., Aretakis, N., Kotsiopoulos, P. N., and Yfantis E. A. A Virtual Laboratory for Education on Gas Turbines Principles and Operation. ASME Turbo Expo 2006: Power for Land, Sea and Air, GT2006-90357, 8-11 May 2006, Barcelona, Spain. [2] Mathioudakis, K., Aretakis, N., and Yfantis E. A. A Possibility for On-Board Training for Marine Gas Turbine Performance, Monitoring and Diagnostics. MECON 2006, 29 August –1 September 2006, Hamburg, Germany. Proc. pp. 128-132. [3] Technical Manual, Propulsion Gas Turbine System 7LM2500-SA-MLG22 (MEKO-G), Operating Instructions, GEK 50500-12. [4] Stamatis A., Mathioudakis K., Papailiou K. D., 1990a, Adaptive Simulation of Gas Turbine Performance. ASME Paper 89-GT-205, 34th ASME TURBO EXPO 1989, June 4-8, Toronto, Canada, also, Journal of Engineering for Gas Turbine and Power, ASME, vol.112, No 2, April 1990, pp.168-175. DIAGNOSTYKA’ 3(47)/2008 65 KORCZEWSKI: Endoscopic Diagnostics Of Marine Engines

ENDOSCOPIC DIAGNOSTICS OF MARINE ENGINES

Zbigniew KORCZEWSKI

Instytut Konstrukcji i Eksploatacji OkrĊtów, Akademia Marynarki Wojennej 81-103 Gdynia ul. ĝmidowicza 69 fax: (058) 626-29-08, e-mail: [email protected]

Summary There have been presented within the paper optic (qualitative) and digital (quantitative) methods of endoscoping internal spaces with industrial endoscopes and videoscopes presented within the paper which are applied willingly during a technical shape evaluation of the complex technical objects. They stand for a very effective diagnostic tool and they are very often reached by marine engines operators of both the fleets: naval and merchant. There have been also demonstrated selected metrological techniques that could be applied during diagnostic examination of the working spaces within marine diesel and gas turbine engines. An evolution of contemporary endoscopies as well as theoretical bases of optic and digital endoscopy has been brought forward. There have been also shown the methods enabling measurements of surface defects on the chosen inner parts of marine engines’ construction structure. The are based on digital recording of endoscopy imagines: “Stereo Probe”, “Shadow probe”, laser method as well as RGB method.

Keywords: marine diesel and gas turbine engines, technical diagnostics, industrial endoscopy, measurements methods.

DIAGNOSTYKA ENDOSKOPOWA SILNIKÓW OKRĉTOWYCH

Streszczenie W artykule przedstawiono optyczne (jakoĞciowe) i cyfrowe (iloĞciowe) metody wziernikowania przestrzeni wewnĊtrznych z wykorzystaniem endoskopów i videoskopów przemysáowych, które stanowią bardzo efektywne narzĊdzie diagnostyczne, po które równieĪ coraz czĊĞciej siĊgają eksploatatorzy silników okrĊtowych zarówno flot wojennych jak i wielu armatorów cywilnych jednostek páywających. Zaprezentowane zostaáy wybrane techniki metrologiczne, które mogą znaleĨü zastosowanie w diagnostycznych badaniach przestrzeni roboczych okrĊtowych táokowych i turbinowych silników spalinowych. PrzybliĪono ewolucjĊ wspóáczesnych endoskopów oraz podstawy teoretyczne endoskopii optycznej i cyfrowej. Zaprezentowano metody pomiaru defektów powierzchniowych wybranych elementów struktury konstrukcyjnej silników okrĊtowych, które bazują na cyfrowym zapisie obrazu endoskopowego: metodĊ „Stereo” („Stereo Probe”), metodĊ „Cienia” („Shadow Probe”), metodĊ laserową, metodĊ RGB.

Sáowa kluczowe: okrĊtowe táokowe i turbinowe silniki spalinowe, diagnostyka techniczna, endoskopia przemysáowa, metody pomiarowe.

1. INTRODUCTION (lenticular) borescopes and elastic (optical) fiberoscopes as well as the dynamically developing The visual investigation of surfaces creating digital endoscopy, with use the more and more internal spaces of the high - and medium-speed perfect videoscopes, equipped with miniature digital marine engines with use of specialist view-finders so cameras at high resolution [6, 10]. called endoscopes is at present a basic method of technical diagnostics. The superficial structure of 2. MEASUREMENT METHODS APPLIED constructional material is visible during WITHIN DIGITAL ENDOSCOPY investigations like through magnifies glass, usually from sure increase, which makes possible the During visual inspection through an optical detection, recognition and the possible quantitative endoscope of surface restricting internal spaces of opinion of stepping out failures and the material engine it often lacks the reference patterns, which defects, and in result - the opinion of degree of waste can be used for qualification of dimensions of and the dirt of studied constructional elements. detected defects. The observed dimension is the In dependence on applied method of observation function of not only the real dimension of the defect, and the processing of visional painting of studied but also a distance between the speculum lens and surface it distinguishes oneself the classic optical studied surface. Because the manufacturers of endoscopy, from utilization to this aim stiff engines pass the admissible values of superficial 66 DIAGNOSTYKA’ 3(47)/2008 KORCZEWSKI: Endoscopic Diagnostics Of Marine Engines defects of the most vulnerable constructional manufacturers of endoscopy equipment i.e. elements, the evaluation of real dimensions of defect OLYMPUS and EVEREST [7, 12, 14]: represents the key diagnostic question. x length, In traditional, optical approach to this question there x multisegment length, length broken (circuit), is applied the comparative method, using with this x distance from point to base straight line aim the calibrated measuring profiles put on the x depth (salience), ending of fiberoscope's speculum lens - [10]. x area of surface ( the area). Additionally, a suitable objective head should be In every metrological option the measurement chosen, which will assure, in direction of exactness is defined. When the operator possesses observation, the optimum angular limitation of field high skillfulness it reaches even 95-98% [12, 14]. of sight straight on, alternatively - in direction of observation side. In dependence on a distance 2.2. „Shadow” method between the lens and studied surface there is Different area for the application of the matched such a width of sight field which will make triangulation theory, in order to mark the dimensions possible maximum increasing the studied surface at of surface defects takes the "Shadow" method being being kept on the required depth of sharpness (the applied in digital endoscopy. The ending of expressiveness of painting). Digital endoscopy videoscope's speculum head is equipped with the brings in this regard completely new possibilities. special optics generating the shadow about The digital analyzers of painting, co-operating with characteristic shape (the most often the line of measuring heads of "Stereo”, "Shadow” or "Laser” 1 straight line).inside the light stream (like the type, basing on theory of triangulation , are able to projector) on surface of the studied element. The qualify exactly a distance from the speculum lens to projection of the shadow holds at known angle of observed surface, and consequently to mark the the speculum head position, in relation to the dimensions of the detected surface defects. The observed surface and known angle of the measuring heads give the possibility of digital observation sector. The shadow generated in the processing the stereoscopic effects, which enables vicinity of detected defect is then located and measuring the seen paintings in such way, to give recorded by camera CCD placed inside an assembly the quasi three - dimensionality impression - with its head. The nearer to the observed surface is the depth, the massiveness and the mutual distribution. speculum head the nearer form the left side of monitor screen is the line of shadow. Because there 2.1. „Stereo” method is well-known the position of shadow generating the The basis of method "Stereo” is the suitable painting on the matrix of LCD monitor screen, utilization of the prism proprieties splitting the a magnification of the painting can be simply painting, what makes possible for digital camera its enumerated. Moreover, the linear dimension of registration from two points limited with span of distance among individual pixels, and then the real parallactical lenses, similarly to human brain dimensions of detected surface defects can be receiving information about the surrounding world evaluated. from the pair of eyes placed each other in a distance Within the "Shadow" method the same of several centimeters ( it is so called an eye optical measurement options like in "Stereo" method are base making about 65 mm) [3]. Paintings accessible. transmitted from parallactical lenses differ each The possibility of taking the immediate decision other, and assignment of the high resolution, in case of doubts regarding proper interpretation resultant, digital painting on monitor LCD represents (unambiguous distinction) of detected surface the effect of realization of a computational defects being effective with decrease or algorithm. A distance from the lens to the observed accumulation of material is the very essential surface is marked by counting pixels in horizontal advantage of the "Shadow" method. Such diagnostic plane of computer's monitor between analogous problems step out during the evaluation process of points of the left and right view of the observed technical state in the working space of combustion element. The larger distance to the surface is engines: piston or turbine. observed the larger distance is between cursors of Often, because of optical and light effects the the left and right view on monitor screen. In the next usual dirt on surfaces of the air and exhaust stage of measurement technology the appropriate passages, in figure of mineral settlings or the measuring technique should be chosen. The products of burning the fuel (the carbon deposit), is following metrological options are well-known for interpreted as the corrosive or erosive decrement of the realization of digital measurement within the the constructional material. The depression of "STEREO" method offered by significant surface (its larger distance from the speculum head) is associated with the shadow line breaking and 1 W. Snellius was the creator of triangulation theory shifting towards the right side of the screen, and its (1615). The measurement method consists in division of salience (its larger approaching to speculum head) - the measuring area into adjacent rectangular triangles the shadow line breaking and shifting towards the and marks on the plane the co-ordinates of points by left side of the screen. means of utilization of the trigonometrical functions. DIAGNOSTYKA’ 3(47)/2008 67 KORCZEWSKI: Endoscopic Diagnostics Of Marine Engines

The confirmed in investigations of diagnostic usage of cylindrical optics in a laser bundle of the shipping engines utilitarian values characterize the scanner enlarges superbly its application values (the method of ”Shadow”, and also high exactness laser ray is distracted into continuous circuit line) which, at keeping on required conditions of the giving the possibility of technical state opinion about measurement, might achieve even 95% [14]. The smaller surfaces e.g. internal spaces of combustion maximum approach to studied surface of the engines, especially intracylindrical spaces of piston speculum head represents the most essential factor engines. of high measurement exactness (the line of shadow moves towards the left side as the speculum head 2.4. RGB method gets closer to the surface) as well as maintaining A digital painting recorded in videoscope's perpendicular to this surface the position of computer consists of separate elements, so called speculum head (the line of shadow runs pixels. Each of them is specified as a group of value perpendicularly to the basis of a monitor screen). of colours: red, green and blue of the painting component in this point - RGB format. With them 2.3. Laser method suitable confusion can get all different colours. Each Laser method is the youngest measuring of the colour in RGB format characterizes the tint, technology applied in digital endoscopy. In technical degree of clearness (or the brightness), the degree of diagnostics two ways of laser rays utilization for the saturation and the cleanness. It worth pointing out in measuring surface defects are well-known. The first this place that the colour, as a special feature of one is so called marker (multipoint) laser method, material objects, optically favoured, is dependent on patented by German concern KARL STORZ Gmbh the objects' physical-chemical structure, the way of & Co. KG - [13]. The creature of the multipoint absorbing and reflecting the light rays by the measurement is the estimation of distance from the observed surface, the character of only light, videoscope’s speculum head to studied surface (the properties of the air and peculiarity of surroundings increase of the real painting) on the basis of a base (the light reflexes). Taking into consideration plane created by at least 3 the best well-fitting (from characteristic colours a simple spectral analysis of among 49) laser points of the marker matrix, the colour composition reflecting the registered throwed on the observed element by means of surface of construction element's painting can be distracting optical arrangement of the measuring carried out and this way its structure might be probe [10]. The smaller distance is to studied surface estimated. Because certain surface defects, as the the larger is the shift of laser gauges towards left alterations of physical properties of constructional side of the monitor screen. The painting of surface, material, characterize themselves with a definite in the next stage of measurement realization, is pattern of colours spectrum it is possible to recorded (stop-frame), and then by the utilization of recognise closely was can for mediation of triangulation theory of triangulation dimensions of comparative analysis of ghostly paintings recognize detected defect are marked, in the following options early development stages of the unfavorable (similarly to measuring systems of OLYMPUS and structure. The additional diagnostic information can EVEREST company): the length, depth (the be gathered by the analysis of digital record features salience), the area, distance from the point to the on the borders of individual colours: their sharpness, base straight line. brightness, surface density etc. A scanning method is the second way of Within the range of conducted research works on application of laser rays to measure surface defects. diagnosing marine diesel engines worked out by the The method has been patented by Australian paper author's scientific team, there has been also company REMOTE VISION SOLUTIONS Pty Ltd undertaken trials to implement RGB method for the [8]. The method is applied by the service groups of estimation of carbon deposits (degree of dirt) on the Boeing to search surface defects on a airplane hulk. surfaces restricted the combustion chamber of The optical arrangement of laser scanner makes cylindrical sets. the transformation (the dispersion) of the laser ray in The registered results of routine endoscopic the system of whirling mirrors. Dispersed laser investigations of an engine being in current bundle raining on studied surface undergoes operation were used in this aim. IPEG format was reflection, which is directed on assembling lens and applied in order to record the pictures. In spite of measuring detector. The quantity of reflected laser pilotage character (only) of undertaken light is dependent on the state of studied surface. If investigations, during acquisition there has been the scanning laser ray shifts over the surface defect paid special attention on keeping constant an absorption will occur, resulting from the smaller comparable intensity of lighting incidencing onto quantity of reflected light. As a consequence, the observed surfaces of the piston bottom, by making intensity of reflected light (density of light stream on records in the same piston position in relation to assembling lens) is smaller than in the case surface Inner Dead Centre (IDC). Because of the character being free from defects. The dimension evaluation of making estimation (only intensity of the surface of detected surface defects is then made on the basis dirt) there has been given up full analysis of the of spectral analysis of distracted laser bundle in colours distribution's intensity, but intensity of computer analyzing programme of the scanner. The 68 DIAGNOSTYKA’ 3(47)/2008 KORCZEWSKI: Endoscopic Diagnostics Of Marine Engines

a) Depth measurement of dent on the blade’s trailing b) Depth measurement of dent on the blade’s trailing edge by means of „Shadow” method in the option edge by means of „Stereo” method in the option of of “distance from point to base straight line” - distance from point to base straight line - 0,98 mm, 0,90 mm, at measurement accuracy index 4,8, at measurement accuracy index 3,6, representing 0,5 representing 0,25 mm mm

c) Convexity measurement of surface „bulge” on the d) Convexity measurement of surface „bulge” on the blade-bed by means of „Shadow” method - 0,20 blade-bed by means of „Stereo” method - 0,29 mm, mm, at measurement accuracy index 12,0, at measurement accuracy index 10,3, representing representing 0,15 mm 0,1 mm

e) Surface „bulge” area measurement (closed broken f) Surface „bulge” area measurement (closed broken line) on the blade-bed by means of „Shadow” line) on the blade-bed by means of „Stereo” method method - 0,55 mm2, at measurement accuracy index - 0,45 mm2, at measurement accuracy index 10,8, 11,5, representing 0,1 mm2 representing 0,1 mm2

Fig. 1. The results of measurement of surface defects on blades of naval gas turbine engines with the utilization of "Stereo” and "Shadow” method DIAGNOSTYKA’ 3(47)/2008 69 KORCZEWSKI: Endoscopic Diagnostics Of Marine Engines greyness tints schedule of monochromatic paintings [3] BrzeĞciĔska W.: Fotogrametria. Wydawnictwo has been taken into consideration. The introduced Szkolne i Pedagogiczne. Kraków 1998. method which bases on usable Able of Image the [4] Doglietto F., Prevedello D. M., Jane J. A., Han. Analyser v3.6 programme analysing the paintings in J., Laws E. R.: A brief history of endoscopic relation to intensity of pixels occurrence of given transsphenoidal surgery - from Philipp Bozzini colour or the pixels of greyness tints was described of the first World Congress of Endoscopic Skull in publication in detail [10]. Base Surgery. Neurosurgery Focus. Vol. The registered results of performed by the author 19/2005. endoscopic investigations of the naval gas turbine [5] Gardner A. P.: How to excel in endoscopic engine's rotor blades are presented in fig. 1. It order photography. British Journal of Urology. No. to estimate dimensions of detected surface defects 81/1998. "Stereo” and "Shadow” method have been applied [6] Hlebowicz J.: Endoskopia przemysáowa. Biuro as well as "EVEREST” videoscope XL PROTM Gamma. Warszawa 2000. type has been put into use (thanks to politeness of [7] Industrial endoscopy system guide. Version 2. company representatives in Poland). Oferta OLYMPUS Industrial. [8] InspectCam SDMS laser measurement system. 3. CONCLUSION Fast and Accurate. Oferta REMOTE VISION SOLUTIONS Pty Ltd. The detecting the material defects in the internal [9] Korczewski Z.: Endoscopic examination of spaces of machines and industrial devices by means naval gas turbines. Polish Maritime Research, of endoscopes utilization represents one of the No. 4/1998, p. 18-21. youngest methods of technical diagnostics, in [10] Korczewski Z.: Endoskopia silników contrary to medical diagnostics, where it has been okrĊtowych. AMW. Gdynia 2008. acknowledged as the key method of searching [11] Rainer M. E.: Philipp Bozzini - the father of pathological states inside human body since ancient endoscopy. Journal of Endourology. times. No.17/2003. Introduced in the article optical (qualitative) and [12] Remote visual inspection. Product guide. Oferta digital (quantitative) method of examining visually OLYMPUS Industrial. internal spaces with the utilization of optical [13] Techno pack with multipoint measuring system. endoscopes and industrial videoscopes are more and Oferta KARL STORZ - Endoscope. Industrial more willingly applied method in the evaluation Group. process of technical state of the complex technical [14] XL PRO videoprobe measurement system. objects. They constitute the very effective diagnostic Oferta EVEREST VIT. tool and the operators of marine engines (diesel and gas turbine) apply them very willingly in both the Prof. Zbigniew fleets: war and merchant. Especially merchant KORCZEWSKI, works for shipowners are very interested in endoscopy the Institute of Shipbuilding application during engines servicing. and Mainte-nance at the Polish Naval Academy in LITERATURE Gdynia. He conducts lectures within the scope of [1] Báachno J., Bogdan M.: A non-destructive operation and diagnosing method to assess a degree of overheating of gas marine power plants. In the turbine blades. Journal of Polish CIMAC. frame of scientific activity he directs the diagnostic Diagnosis, Reliability and Safety. team dealing with energetic investigations of marine Vol. 2 No. 2. 2007. propulsion systems propelled by means of gas [2] Breen J., Stellingwerff M.: Application of turbine and diesel engines. He specializes in optical and digital endoscopy. Proceedings 2nd unconventional diagnostic methods of the marine EAEA-Conference, Vienna 1995. engines having limited supervisory susceptibility.

DIAGNOSTYKA’ 3(47)/2008 71 BOURAOU, ZAZHITSKIY: Decision Making About Aircraft Engine Blades Condition By Using ...

DECISION MAKING ABOUT AIRCRAFT ENGINE BLADES CONDITION BY USING NEURAL NETWORK AT THE STEADY-STATE AND NON-STEADY-STATE MODES

Nadiia BOURAOU, Olexandr ZAZHITSKIY

National Technical University of Ukraine "KPI" Ukraine, 03056 Kiev, Peremogy Prosp., 37, fax: +38 044 241 77 02, e-mail: [email protected]

Summary The work is devoted to problem solution of the gas-turbine engines (GTE) blades condition monitoring and diagnosis of the crack-like damages at the steady-state and non-steady-state modes of GTE. It is based on the development of theoretical basis of the vibroacoustical diagnosis methods, the application of the modern signal processing methods and new information technique for decision making. The application of the following signal processing methods: Wavelet- transformation and dimensionless characteristics of the vibroacoustical signals is proved. Neural networks are used for decision making about blades condition by the above mentioned features application. Classification of turbine blade condition was carried out using a two-layer Probability Neural Network (PNN).

Keywords: blade, crack-like damages, vibroacoustical condition monitoring, neural networks.

INTRODUCTION excitation of the compressor and turbine blades, and consequently, they define the methods and The problem of prolongation of aircraft turbine algorithms of signal processing which must be engine working life and increasing their reliability is chosen or developed. the issue of the day. This problem may be solved Initiation and increase of a fatigue crack in the using improved existing and new methods and blade lead the instantaneous change of its stiffness. diagnostic instruments. Despite the fact that the Usually the change of stiffness is modeled by the progress in development of existing methods and piecewise-linear characteristic of the restoring force instruments is considerable, the problem of [1, 2]. Non-linearity leads to variation of oscillation functional diagnostics of fatigue defects in parameters and to occurrence of local non-stationary compressor and turbines blades of an aircraft engine component in the measured signal. is not solved yet. We propose to solve this problem The dynamic model of gas-turbine engine as an by using the vibration and vibroacoustical diagnostic object for fatigue cracks diagnostics in turbine methods [1]. Since the most fatigue defects of the blades and compressors was created. This model is GTE rotor components directly connect with used for simulation and analysis of vibroacoustical vibration processes which take place in operating processes which occur at the steady-state and non- engine. On the other hand, vibration and steady-state modes of GTE at absence and presence vibroacoustical methods provide the possibility to of small fatigue cracks in one blade of the turbine diagnose and non-destructively evaluate defects stage (the relative rigidity changing at the crack without disassembling the engine. It may be done presence is considered -=0,03;0,05). The three using advances in computer-based technology and modes of GTE are simulated and investigated: m1 - information handling methods for recognition of the steady-state (constant value of the rotor rotation rotor component condition in the operating engine. frequency); m2 - non-steady-state (the fast increase Creation of the monitoring system is based on of the rotor rotation frequency); m3 - non-steady- application and further development of low- state (the decrease of the rotor rotation frequency). frequency vibroacoustical diagnostic methods which Simulated signals were processed using Wavelet- use vibrating and acoustical noise as diagnostic transformation and amplitude dimensionless information. characteristics of the vibroacoustical signals. The preliminary Wavelet decomposition of signals is 1. PROBLEM STATEMENT used for the sensitivity increasing of the amplitude dimensionless characteristics of the vibroacoustical Generally monitoring is a continuous process of signals as fault features [3]. The following amplitude information gaining about the object vibrating dimensionless characteristics are used: J3 - peak condition, transformation and analysis of it, and factor and J - factor of background. Thus, the fault making decision about object technical condition. 4 The stages of the mentioned informative process features are detected and the following feature & (mi) (mi) depend on the engine operation modes. These modes vectors are formed: X mi (J3 ;J 4 ) , i 1,3. define specific character of vibrating and acoustical 72 DIAGNOSTYKA’ 3(47)/2008 BOURAOU, ZAZHITSKIY: Decision Making About Aircraft Engine Blades Condition By Using ...

-=0.03 -=0.05

a)

b)

c) Fig. 1. The learning sets of the feature vectors for the three modes of GTE: a) m1, b) m2 and c) m3 (ɨ – without damage S0, ɯ – with damage S1).

The mentioned vectors are used for the decision -=0.03 and -=0.05, but at the m1and m3 – only for making about blades condition: the absence (S0) or -=0.05. the presence (S1) crack-like damage. The aim of this work is efficiency analysis By using the feature vectors the learning and recognition of aircraft engine blades condition at test sets are received for the above mentioned three the steady-state and non-steady-state modes of GTE modes of GTE: m1, m2 and m3. The learning sets by using neural networks. are shown in Fig. 1 for the relative rigidity changing at the crack presence -=0.03 and -=0.05. 2. RECOGNITION OF BLADES CONDITION Axes indicate the peak factor and factor of background. Classification of turbine blade condition was As it can be seen from a given plot, the linear carried out using a two-layer Probability Neural division into blades conditions S0 and S1 occurs for Network (PNN) [4]. The first layer consists of 100 the learning sets only at the m2 mode for both cases neurons. As a second layer it is used the so called competitive layer from 2 neurons. These neurons determine correct solution probability - the input DIAGNOSTYKA’ 3(47)/2008 73 BOURAOU, ZAZHITSKIY: Decision Making About Aircraft Engine Blades Condition By Using ... vector belongs to a faulty type or not. Such The results of classification are shown in Fig. 2 classification is based on Bayes methods and needs for the relative rigidity changing at the crack probability density estimate for a condition type. presence -=0,03 and -=0,05. For that, set of learning vectors are used. Every The effectiveness of turbine blades condition vector is described by Gauss function with a center classification by PNN was judged by the coefficient in the point corresponding to this vector. The sum K (this coefficient is a value of correct of named functions according to the whole classification probability). Relationships between available set of learning vectors is equal to the coefficient K and the influence parameter V for probability density of input vectors for each test sets of the feature vectors is shown in Fig. 3. As condition types. The value of the Gauss function it can be seen from a given plots, PNN recognizes mean-square deviation V specifies the width of the the blades condition with the following minimum neurons activation function and defines their values of the correct classification probability: influence on a probability density estimate sum. - m1 (steady-state mode) – K=0.93 for -=0.05 at This implies that the parameter V influences on the the V=0.1 (recognition for -=0.03 is not correct classification result, therefore its value is – K=0.3); determined mostly experimentally. -=0.03 -=0.05

a)

b)

c) Fig. 2. The results of classification for three modes of GTE: a) m1, b) m2 and c) m3 (ɨ – without damage S0, ɯ – with damage S1). 74 DIAGNOSTYKA’ 3(47)/2008 BOURAOU, ZAZHITSKIY: Decision Making About Aircraft Engine Blades Condition By Using ...

- m2 (non-steady-state mode) – K=0.96 for both and cracked blades allow to form the learning cases -=0.03 and -=0.05 at the V=0.1,…0.12; and test sets of the feature vectors. - m3 (non-steady-state mode) – K=0.94 for both 2. Application of a Probability Neural Network cases -=0.03 and -=0.05 at the V=0.1. provides turbine blades condition classification using the peak factor and factor of background K 1 of the results of Wavelet Decomposition of 0,8 vibroacoustical signals with the high values of the correct classification probability at the 0,6 steady-state and non-steady-state modes of GTE. 0,4 3. Received results are new and justify efficiency of turbine blades condition recognition at the 0,2 presence of small crack-like damages. These 0 results can be used to create a vibroacoustical 0 0,03 0,06 0,09 0,12 0,15 V monitoring system for aircraft engine rotor components. a) K 1 REFERENCES

0,8 1 Bouraou N. I., Marchuk P. I., Tyapchenko A. N.: Condition Monitoring Diagnosis Method of Aircraft Engine Rotating Details, in 0,6 Proceedings of the 15th World Conference on Non-Destructive Testing, 11pp. 0,4 2 Wong C. W., Zhang W. S., Lau S. L.: Periodic forced vibration of unsymmetrical piecewise- 0,2 linear systems by incremental harmonic 0 0,03 0,06 0,09 0,12 0,15 V balance method. J. of Sound and Vibration, p.91-105. b) 3 Bouraou N., Sopilka Yu., Protasov A.: The K 1 features Detection of the small rigidity faults of rotary machine elements. In Review of 0,9 Progress in QNDE, p.152. 4 Sankar K. Pal. and Sushmita M.: IEEE 0,8 Transaction on Neural Networks, 3, #5, 683- 696 (1992).

0,7 Nadiia BOURAOU, Doctor of Science, Prof., Head of the 0,6 Department of Orientation and V 0 0,05 0,1 0,15 0,2 0,25 0,3 Navigation Systems (National c) Technical University of Fig. 3. Dependencies of correct recognition Ukraine "KPI"). Areas of probability on the influence parameter V of a neural Expertise: oscillations, network for three modes of GTE: a) m1, b) m2 probability theory, reliability and c) m3 (continuous lines – -=0.05, dotted theory, signal processing, lines – -=0.03). recognition and decision making, non-destructive testing and evaluation, So, in spite of diagnostic features irregularity aviation acoustic. and little changes at turbine blade condition changing from defectless to faulty one, PNN Olexandr ZAZHITSKIY, provides classification of diagnostic object Master’s of Science, Assistant condition with the high values of the correct (Department of Orientation classification probability at the steady-state and and Navigation Systems, non-steady-state modes of GTE. National Technical University of Ukraine "KPI"). Areas of CONCLUSION Expertise: signal processing, non-destructive testing and 1. Simulation and analysis of vibroacoustical evaluation, recognition and signals radiated at steady-state and non-steady- decision making, neuron networks. state modes by an engine rotor with defectless DIAGNOSTYKA’ 3(47)/2008 75 YAVLENSKY, BELOUSOV, ROGOZINSKY, VOLKOV, Digital Cinema Diagnostic System Based …

DIGITAL CINEMA DIAGNOSTIC SYSTEM BASED ON SPECTRAL ANALYSIS AND ARTIFICIAL INTELLIGENCE METHODS

Alexander YAVLENSKY, Alexander BELOUSOV, Gleb ROGOZINSKY, Alexander VOLKOV

Department of Devices and Systems of Cinema and Television, Saint-Petersburg State University of Cinema and Television, Pravda st. 13, Saint-Petersburg, Russian Federation Tel: +7 (812) 9308450, e-mail: [email protected]

Summary In this paper digital cinema diagnostic complex is described. Its working algorithms are based on wavelet preprocessing, statistic analysis and neural network classification. Considered methods were practically incarnated using modern systems of computer modeling.

Keywords: diagnostics, digital cinema, wavelets, artificial intelligence.

Modern system of digital screening and film Statistic and spectral analyses of noise components production demands modern approach not only from should be performed, including threshold function the point of view of equipment developing and type, wavelet basis and decomposition level designing, but also from diagnostic one. Being selection. a complicated, self-optimizing system, digital Wavelet analysis is based on representation cinema complex should be observed as a segment of (2) s(n) ¦ Ak\ k (n) informational networks, in case of diagnostic view, k and imply an estimation of big amount of various where - basis functions, - decomposition \ k (n) Ak parameters. coefficients. Such systems consist of number of simple parts, Due to its time-frequency localization, wavelets connected to each other. This condition makes allow to analyze the non-regular signal structure. difficulties for usage of ordinary automated systems Time localization implies that wavelet’s energy is of diagnostic and quality control. The most concentrated in some finite time interval. Frequency complicated are description of allowable working localization means compactness of wavelet’s rate for hardware parts. Fourier-form, in other words it means its energy Diagnostic signals are complex-structured and localized inside of finite frequency interval. generally transient: Wavelets could be constructed from mother wavelet s(n) S(n)  V 1 e1 (n)  V 2e2 (n) (1) function \ k (n) by translation \ k (n) o where s(n) – measured signal, S(n) – normal n  b \ k (n  b) and dilatation \ k (n) o \ k ( ), functioning signal, e1(n) - noise, e2 (n) - a distortions due to systems defects, V and V - where parameter b called transition and a – scale. 1 2 Wavelet transform in its discrete version uses a and distortion levels, n – discrete or continuous variable. b with steps divisible by 2: Wavelet analysis possesses an ability of time- m m frequency localization of signals distortion. That is a 2 ,b k2 , k, m  Z . why it is most preferable basis for diagnostic Thus, wavelet function could be represented like m problems solution. For our purposes, the most m 2 (3) informative components of signal are the noise ones \ m,k (n) a0 \ (a0 nk)

- e1(n) and e2 (n) , since they carry an original where a0 >1 (here we take a0 = 2). imprint of system, which makes diagnostics possible Spectral transform coefficients A(m,k) are (figure 1). defined as convolution of signal s(n) and wavelet The solution of diagnostic task for noise characteristics is rather complicated, because of \ m,k . Farther, in the course of getting diagnostic number of problems supposed to be solved. One of information from object, there comes a necessity of the most important problem is noise separation. It incoming data clustering to classes depending on should be divided to local components and object’s characteristics. background. For each of these two components special analysis and thresholding methods are used. 76 DIAGNOSTYKA’ 3(47)/2008 YAVLENSKY, BELOUSOV, ROGOZINSKY, VOLKOV, Digital Cinema Diagnostic System Based …

Fig. 1. Diagnostic algorithm

Clustering is performed using neural network should be considered as function of i, that is algorithms. The present state of neuron is calculated ' M ', i , where ' - modified parameter. like i T (4) Frequency-dependent threshold ' leading is g A G i A necessarily for excluding of resonance influence to T where G G ,G ,...,G - weight array, T- informational signal’s analysis while mechanics 1 2 n diagnostic. In case of sound and video processing it transposition. allows to exclude non-informational components. Array Ais formed from coefficients A(m,k)for Function g(A) depends on threshold value ' and fixed frequency ranges . For Z 1,Z 2 ,...,Z n diagnosed parameter r: components e (n) ande (n) to be extracted ­ A ½ 1 2 g(A) G mi sgn§ A r § ·· 1 . (3) ¦ i ® ¨ mi M ¨',i ¸¸  ¾ according to [1], the present state of neuron is i ¯ 2 © © ¹¹ ¿ calculated like For classification of diagnostic data Probabilistic T (5) g A G i f (A) Neural Network (PNN) is used. It has a structure T based on radial base network architecture. PNN where G G ,G ,...,G - weight array, 1 2 n structure [1] for MATLAB modeling is shown on f (A) - array function. figure 2. Let us observe an example of f (A) calculation. Informational signal’s spectrum could be presented by array T . Now, we A m A m 1, A m 2 ,..., A m n have to transform array’s components [2, 3]: A f (A ) mi > sgn A  ' 1 @ (6) mi mi 2 m i k i i=1,2,...,n; Fig. 2. PNN structure [1]. ­ 1, x ! 0; sgn x ® 1, x d 0, Let us observe the PNN working algorithm [1]. ¯ Let the learning multitude is defined and consists of Q pairs of input arrays. Each array consists of K where - threshold value for i frequency;, ' k i elements, which are related to some class. Thus, we get matrix T with dimension K x Q, its rows k , k 1,2,... 'ki 'i correspond to classes and columns do to input 1,1 Diagnosed parameters r define the values of arrays. First layer weight matrix IW is formed spectral coefficients (4). Among them, received from input arrays, taken from learning multitude during statistic or fuzzy analysis of coefficients (4), (matrix P’). If new array multitude is entered, the let us select such frequency ranges ŒdistŒ block calculates proximity of new array to existed ones. Then, proximities are multiplied by , ,..., which coefficients are Z 1 Z 2 Z n Ami deviations and got to activation function input monotonous functions of r. Threshold value (radbas). The nearest learning multitude array to ' i DIAGNOSTYKA’ 3(47)/2008 77 YAVLENSKY, BELOUSOV, ROGOZINSKY, VOLKOV, Digital Cinema Diagnostic System Based … input one will be presented in output array a1 by Program is based on following VIs - WA Read number close to 1. From File.vi, WA Detrend.vi, WA Denoise.vi, TSA Matrix LW2,1 ( second layer weight matrix) Stationarity Test.vi. corresponds to matrix T (current learning multitude In present version, data should be contained in tie-up matrix). By multiplication T and a1, wav file, although LabView supports other formats correspond elements of array a1 will be defined. like AIFF, TXT, BMP, JPEG. It should be clear that Thus, function compet (competition activation analysis of 1D signals (vibration, sound) would be function of second layer) forms 1 if largest value of performed separately from 2D signal one (pictures) array n2 and 0 otherwise. So, reviewed network because of wavelet analysis algorithm differences, solves problem of input arrays classification by K but in some cases it is useful to estimate distortion classes. correlation between vibrations, sound and picture in Considered model could be incarnated as the same moments of time. So, on the technologic LabView virtual instrument (VI for short). For this point, to perform a diagnostic procedure, one just purpose Advanced Signal Processing Toolkit is used. have to start LabView and choose some options. If to It has wide collection of different means of wavelet work with recorded data only, considered version of transform. For example, let us observe the part of diagnostic instrument is enough, without diagnostic complex, responsible for noise estimation. supplementary hardware needed to gather On fig. 3 and 4 front panel and block diagram of information. program are shown. REFERENCES

[1] Medvedev V. S., Potemkin V. G.: Neironnie seti. MATLAB 6. – M.: DIALOG-MIFI, 2002. 496 p. [2] Yavlensky K. N., Yavlensky A. K.: Vibrodiagnostika i prognozirovanie katschestva mekhanitscheskih sistem. – L.: Mashinostroenie, 1983. – 239 p. [3] Yavlensky A. K. (with others): Mekhanitsheskie sistemi vakuumno-kosmitscheskih robotov i manipulyatorov: T.2 – Tomsk: MGP “RASKO”, 1998. – 378 p.

Fig. 3. Front panel of virtual instrument

Assignment of front panel elements: 1. Oscillogram of informational signal and noise component, extracted during wavelet preprocessing 2. Oscillogram of noise (diagnostic) component and noise trend detected 3. Local analysis results (quantity of peaks of certain width in signal) 4. Statistic analysis results (quantity of segments, mean, square values)

Fig. 4. Block diagram of virtual instrument

DIAGNOSTYKA’ 3(47)/2008 79 SHEVCHENKO, ZHILEVICH, KOROLKEVICH, Features Of Algorithms Of Diagnosing And Maintenance …

FEATURES OF ALGORITHMS OF DIAGNOSING AND MAINTENANCE OF RELIABILITY OF HYDRAULIC DRIVES OF MACHINES

Vasily SHEVCHENKO, Michael ZHILEVICH, Alexander KOROLKEVICH

The Incorporated Institute Of Mechanical Engineering The National Academy Of Sciences Of Belarus Street Academic, 12, Minsk, ph. + 375 17 284 24 46; a fax + 375 17 284 02 41 e-mail: [email protected]

Summary The paper discusses the possibilities of applying some algorithms of diagnosing and maintenance of reliability of hydraulic drives of machines. Some typical algorithms of diagnosing are considered. The conceptual positions of structural maintenance of reliability of hydraulic drives have been proposed.

Keywords: technical diagnosing, reliability, hydraulic drive.

1. INTRODUCTION control of parameters (constructive, regime, operational, etc.) One of the basic functions of the test system of a hydraulic drive is the establishment of conformity Technical of its technical condition to normative Methods requirements. The properties of hydraulic systems, Diagnosing defining their technical condition, are subject to the certain changes while in service machines. Functional Logic Experience is confirmed, that with a significant part Diagnosing Methods of technogenic failures are tie up with the mistakes Diagnosing admitted at designing and operation of machines, and that the price of such mistakes is exclusively Test Physical Methods great [1]. Therefore it is necessary to pay the big Diagnosing Diagnosing attention to working off of algorithms of technical diagnosing of hydraulic systems, increase of their Fig. 1. Methods of technical diagnosing efficiency and reliability. The basic lacks of the specified methods are 2. FEATURES OF TECHNICAL increased requirements to a memory size of the DIAGNOSING OF HYDRAULIC DRIVES monitoring system of parameters and low efficiency of such systems. Especially it concerns to cases of the For machines that are complex, power-intensive control of realization of signals or their statistical and potentially dangerous to the person, characteristics for all conditions of functioning. environment and economy of technical devices especially important requirement is maintenance of 3. REQUIREMENTS TO ALGORITHMS trouble-free operation. Admissible values of OF DETECTION OF MALFUNCTIONS parameters and attributes, preemergency and emergency conditions of system are defined with Maintenance service of hydraulic systems of the obligatory account of features inherent in machines demands participation of highly skilled actually hydraulic systems. First of all is experts on which arms there are the modern precision a centralization of functions and the mass diagnostic devices having an output on a computer, distribution of influences which are carried out by allowing to define methods of elimination of a uniform working body (liquid). malfunctions and maintenance of reliability of Depth of troubleshooting and reliability of diagnosed devices. Such method of diagnostics is results of diagnosing define a degree of safe modern, but, unfortunately, yet has not received operation of the machine, and also reliability of the enough wide circulation basically for the reason that forecast of the technical condition of the system in operation often there are machines manufactured during the future moments of time. long time ago and not equipped by such complex While in service hydraulic drives of machines diagnostic systems. However even in this case at the complex of methods of diagnosing (fig. 1) is presence of the skilled expert it is possible to define applied. quickly enough and authentically the reasons of At functional diagnosing of the condition of malfunction of hydraulic system, using a so-called a hydraulic drive is defined by results of the current logic method [2]. 80 DIAGNOSTYKA’ 3(47)/2008 SHEVCHENKO, ZHILEVICH, KOROLKEVICH, Features Of Algorithms Of Diagnosing And Maintenance …

All malfunctions of the hydraulic system can be general algorithm of search of malfunctions is carried divided into two greater groups: out by means of connection of base algorithms - The malfunctions which at present are not according to graphic model. If necessary the influencing functioning of the machine (the algorithm can be supplemented with heuristic outflow, the raised noise, the raised temperature, algorithms of search of the possible defects caused by etc.); features of connection of elements in a uniform - Functional malfunctions (jamming, decline of system (structure of the system). productivity, ignitions, etc.); The mathematical model is developed for the Searching of malfunctions is carried out by formalized modeling hydraulic drives of any different algorithms. Experience of the experts and structure in an automatic mode their practical skills, as well as common sense, thus 2 2 ­d x 1 § dx · dx i [p p A2 B2 i sgn i have crucial importance. The operational ° 2 i1  i  i  i ¨ ¸  ° dt A1i © dt ¹ dt documentation, as a rule, contains a lot of the ° valuable information on background of the arisen dxi °  A3i ], i 1...NU, i z iɫ; problem (whether there were earlier the similar ° dt malfunctions, what works were spent on °d 2x 1 P  P  P ° i [p  Fp  z 0 T  maintenance service and adjustment of those or 2 i1 i ° dt A1i fi other units and systems). The logical analysis of ® 2 ° § dxi · dxi dxi such information allows to save a lot of time at - A2  B2 ¨ ¸ sgn  A3 ], i i ; ° i i dt dt i dt c diagnosing and prevents failures in due time. ° © ¹ Revealing the elementary malfunctions of type °dp Ea0 ap pi § dx dx · i f i f i1 , i 1...NU, i i ; of outflow, foaming or overheat of the working ° ¨ i  i1 ¸ z ɪ ° dt Vi © dt dt ¹ liquid, insufficient speed of the agencies, the raised ° E a p dpi a0  p i § dxi dxi1 dxik · noise in separate devices, etc. by means of sense ° ¨ fi  fi1  fik ¸, i iɪ. organs and on the basis of experience allows to ¯° dt Vi © dt dt dt ¹ avoid excessive complications of test systems. It Here: NU – quantity of sites; i – number of a site; ɪi – promotes increase of their own reliability and pressure; ɯi – moving; Vi – volume of a liquid on i-th certain concentration of functionalities on search site; ȿɚ – the module of elasticity of a liquid; ɚɊȺ – the and the analysis of more complex malfunctions. factor considering influence of pressure on ȿɚ; fi – the Experience of diagnosing and good knowledge area of section i-th site; ɊɌ – force of friction; Ɋ0, ɊZ – of the device of hydraulic system allow to define constant and item loadings; iɫ – a site with the the priority procedure for test of units. If at once at hydraulic cylinder; iɪ. – unit of a branching; Ⱥ1i, A2i, preliminary check the faulty unit is not found, A3i,B2i – the factors considering various kinds of spend deeper check of each unit by means of losses of pressure. special control devices and stands. After finding of the faulty unit the reasons and Adequacy of the mathematical model and possible consequences of malfunction (clearing or algorithm is confirmed by computer experiment. In replacement of a working liquid, adjustment of fig. 2 results of dynamic calculation of a brake drive safety and conditioning devices, etc.) are defined and a drive of the elevating mechanism of the car- and eliminated. dumper are presented. Algorithms of preliminary logic diagnosing of Transients at functioning hydraulic drives are hydraulic systems usually include a number of speedy and are accompanied by high-frequency consecutive checks of conformity of marks and pulsations of pressure and significant control over, installations of units, correctness of their and instant value of pressure exceeds a threshold of adjustments and signals of management [2]. adjustment of a safety valve. If not to consider With the purpose of increasing of efficiency of dynamics of internal processes in a hydraulic drive it built-in diagnosing systems of mobile machines can lead to the false conclusion about a technical base diagnostic models and schemes (base condition of a hydraulic drive. Not to admit possible algorithms) automatically in search of possible wrong of diagnoses, transfer of characteristics of malfunctions of hydraulic drives [3] are developed. quality of transient to the system of automatic search Thus in schemes of recognition transients in of defects for updating of actions in case of hydraulic system are considered. discrepancy of entrance and target signals is Adequate work of system of diagnosing is stipulated. possible only under condition of the coordination of character and parameters of transients in a hydraulic drive with parameters of the measuring equipment. For formation of a diagnostic model of the hydraulic drive the special technique in which basis the analysis of the basic scheme of a hydraulic drive and an establishment of communications between elements is put is offered. Synthesis of the DIAGNOSTYKA’ 3(47)/2008 81 SHEVCHENKO, ZHILEVICH, KOROLKEVICH, Features Of Algorithms Of Diagnosing And Maintenance …

a) subsystems of maintenance of reliability, entered into 18 structure of a hydraulic drive, carry out active 16 influence on the parameters defining reliability. 14 p3 The protection devices warning of possible p1 12 failures are known. More often they switch off the 10 system before elimination of the reasons of possible 8 refusals. Such a way is the most comprehensive at p, MPa p, 6 operation concerning simple systems. 4 p2 Generally management of the processes defining 2 reliability of a hydraulic drive of machines and their 0 elements can be realized on principles of 0 0,2 0,4 0,6 0,8 1 1,2 indemnification of indignations or deviations of the b) t, c parameters defining reliability. For such management 20 0,005 it is necessary to have an opportunity of registration p 16 0,004 of deviations, the nobility of the characteristic of indignations and to compensate them (fig. 3). 12 0,003

ɦ V z, z,

p, MPa 8 0,002 z x 4 0,001 Hydraulic drive

0 0 0 0,02 0,04 0,06 0,08 0,1 t, c D1 u Fig. 2. Results of dynamic calculation: FD CPU a – a brake drive (p1,p2,p3 – pressure accordingly on input of a contour, in a branching and in the y actuating cylinders),b – the elevating mechanism EC D2 (p – pressure ,z–moving platforms, t– time). Fig. 3. The scheme of management of the processes defining reliability of a hydraulic drive 4. DIAGNOSING AND MAINTENANCE OF RELIABILITY OF HYDRAULIC DRIVES Indignations V influencing on a hydraulic drive are registered by sensitive element D1 and act in Diagnosing of a hydraulic drive assumes formation device FD. The signal of a mismatch from obligatory check of conformity of pressure and of an element of comparison EC, formed as a result of liquid consumption to demanded values for comparison of a deviation of adjustable size ɯ (it is corresponding modes. Thus adjusting installations registered by sensitive element D2) and the set of units of system, streams of a working liquid coordinate y also come in FD. In FD operating through a safety valve and drainage system, and influence u, getting in a power drive of control also vacuum an input of the pump are supervised. system CPU is developed. According to the acted Well-known, that results of diagnosing, the signals the drive of a control system influences the information received at this stage, should be used in certain element of a hydraulic drive, providing its the further for maintenance of working capacity and normal functioning. Thus, in the given general reliability of the system of the hydraulic drive. scheme the complex of devices D1,D2, EC, FD, CPU Experience has shown, that traditional use of represents providing subsystem which changes progressive technological processes and high- character of influence of indignations in the quality materials by manufacturing the hydraulic necessary direction and influences a course of devices, already given a new push to development processes in a hydraulic drive as a result of change of in this area, has allowed at the same time to reveal cycling of loadings, a mode of greasing, air- the certain restrictions which cannot be overcome conditioning, inclusion of correcting parts, etc. by only one improvement of technology and Antifriction functions of subsystems of material resources. Prospects of overcoming of maintenance of reliability are characteristic for all difficulties and the further development of types of hydraulic drives. The cores from them – hydraulic drives of machines are connected with the exception of hit of aggressive impurity in system of application of new approaches of structural a hydraulic drive, creation of a greasing layer maintenance of their reliability. between cooperating elements, indemnification of Structural maintenance of reliability of deterioration and deformation, maintenance of hydraulic drives of machines is carried out due to demanded properties of a greasing liquid, reduction purposeful development or change of the block of specific pressure and speeds of influence. diagram of a hydraulic drive at a stage of its For severe loading hydraulic drives the most designing [4]. Its basic essence says that, so-called typical functions of subsystems of maintenance of 82 DIAGNOSTYKA’ 3(47)/2008 SHEVCHENKO, ZHILEVICH, KOROLKEVICH, Features Of Algorithms Of Diagnosing And Maintenance … reliability are unloading (indemnification of D. Sc., professor Vasily deformations, decrease in specific pressure, SHEVCHENKO works as the reduction of vibrations, etc.), and also maintenance main scientific employee of tightness of systems. managing sector of hydraulic drives of machines of THE CONCLUSION Incorporated institute of mechanical engineering of the Technique and the algorithms of simulation of National Academy of sciences of a hydraulic drive are applied as the software Belarus. Area of scientific product for engineering calculations on mechanical interests: mechanics, systems of engineering firms. drives, reliability of machines. The winner of the The offered conceptual positions and schemes State premium in the field of a science and technic. of structural maintenance of reliability of hydraulic Are the author more than 200 scientific works and drives of machines are based on characteristic inventions, including 7 monographies and manuals. properties of hydraulic systems, results of diagnosing of their conditions and introduction of Prof., senior lecturer Michael subsystems of active maintenance of reliability. THILEVICH works as the Distinctive property of the specified subsystems is leading scientific employee of their ability automatically to support and if the Incorporated institute of necessary to restore working functions of hydraulic mechanical engineering of the drives during their long operation. National Academy of sciences of Belarus. Area of scientific THE LITERATURE interests: diagnostics of hydraulic drives of machines, [1] Reliability of technical systems and technogenic simulation of dynamic risk. The electronic manual http: processes of hydraulic drives. He is the author more //www.obzh.ru/nad/10-4.html. than 70 proceedings and inventions. [2] Kornjushenko S. I. Diagnos and treatment. “Building technics and technologies“ 2006. ʋ1 Prof., senior lecturer (41). P. 32 – 37. Alexander KOROLKEVICH [3] Thilevich M. I.: Substantiation of parameters of works as the leading scientific the built in system of diagnosing of the employee of the Incorporated incorporated hydraulic drive for increase of institute of mechanical safety and efficiency of career dumpers. The engineering of the National Dissertation on competition of a scientific Academy of sciences of degree., Minsk, 2006. 249ɫ. Belarus. Area of scientific [4] Bogdan N. V., Kishkevich P. N., Shevchenko interests: methods of designing V. S.: Hydraulic and pneumatic automatics and a of hydraulic drives of machines. He is the author hydraulic drive of mobile machines. Operation more than 100 proceedings and inventions. and reliability ɝɢɞɪɨ-and pneumatic systems: the Manual. – Minsk: Uradzai, 2001.396 p. DIAGNOSTYKA’ 3(47)/2008 83 VOLKOVAS, GULBINAS, SLAVICKAS, Evaluation Of Quality Of Heterogeneous Mechanical Systems Using …

EVALUATION OF QUALITY OF HETEROGENEOUS MECHANICAL SYSTEMS USING IMPEDANCE METHOD

Vitalijus VOLKOVAS, Ramunas GULBINAS, Edmundas Saulius SLAVICKAS

Technological Systems Diagnostics Institute, Kaunas University of Technology, Lithuania. Kestucio str. 27-b, Kaunas, LT- 44312. Fax: +370 (37) 32 37 20, E-mail: [email protected]

Summary The paper analyzes methodology for investigation of dynamic characteristics of heterogeneous systems: honeycomb optical tables and pipelines with sediments, applying mechanical impedance. The developed methodology may be used to assess the quality of plate or cylindrical heterogeneous structures, according to the dynamic characteristics and parameters established in the methodology. It was shown that impedance characteristics are informative to determine some parameters of the quality of honeycomb optical tables and are correlated with the thickness of the pipe’s inner sediment layer. Therefore it is possible to choose typical resonance frequencies and according to their changes, the decision about inner layer presence and its value can be made.

Keywords: heterogeneous mechanical systems, quality, impedance, diagnostics.

1. INTRODUCTION The characteristic features of the tables of honeycomb construction are the following: The composite structures are not just used for resistance to impact of static and dynamic forces and creation of various constructions in the technology, optimal ratio of mass and rigidity. but they also could be formed as the by-product The features of honeycomb are determined by while implementing some technological processes. core density, size of cells, material, the way of For example, for precise measurements various joining (pasting, clinching, welding), thickness and light honeycomb constructions, which are form of walls, fillings, etc. When the core density characterized by optimal ratios of mass, compression increases (i.e. the size of cell diminishes), the strength and bending strength are often used. In this rigidity, shear module and mechanical connection case the quality of composite structure (e.g. the with upper and lower layers get bigger, and in such honeycomb optical tables’ tabletops (Fig. 1) have a way they improve the characteristics of table. been more and more widely applied in laser These layers are made from steel, aluminium, plastic technologies and nanotechnologies) is related with and other materials. good dynamic characteristics and properties of The main features of tabletop are the following: vibration damping in vertical and other directions. local flatness, general flatness, static/dynamic rigidity and maximum relative tabletop motion.

Fig. 1. Examples of Honeycomb Plates’ Constructions 84 DIAGNOSTYKA’ 3(47)/2008 VOLKOVAS, GULBINAS, SLAVICKAS, Evaluation Of Quality Of Heterogeneous Mechanical Systems Using …

All these items describe the main values of characteristics of isolation system and concrete quality parameters of table. They are environmental vibrations, as well. introduced in the producers’ catalogues and the 4. Maximum Relative Tabletop Motion is consumers may perform the comparative analysis determined after the transmissibility of isolator, and choose the product suitable for the needs. factor of compliance strengthening Q and power’s The by-product may be the formation of the spectral density are evaluated. heterogeneous sediment inside the pipes while There were experimental tests made for optical implementing the technological processes in the plates of honeycomb construction 1HT 12-24-20 [4]. industrial companies, for example while processing The transmission characteristics were determined by the oil inside the pipe the layer of sediments (such as applying mode analysis. The reaction of the object to combustion char and like it) is formed. In this case the strike was fixed by stimulating vibrations in the quality of pipes is related to the thickness of certain points with the help of special strike gauge sediment’s layers, which affect the dynamic MODALHAMMER mod. 2302-10, and characteristics and efficiency of pipes. The control accelerometer that was fastened near the strike place of the sediment layer is characterized by the fact that and connected to analyzer PULSE 3560 (Fig. 2). the measured dimension is integral in certain distance, which is much bigger than the wall’s thickness, e.g., in pipe’s section or area, while the volume of measurements, i.e. lengths of pipes, are very big. In both cases we deal with heterogeneous mechanical systems with special features. In such a way the technology encounters plate and cylinder composite constructions, while it is important for the technological development and safe maintenance to control simply and reliably the quality characteristics and parameters of such constructions.

2. DETERMINATION OF DYNAMIC CHARACTERISTICS OF OPTICAL PLATE OF HONEYCOMB CONSTRUCTION Fig. 2. Determination of Dynamic Characteristics of Optical Plate of Honeycomb Construction While assessing the quality of optical tables, it is 1HT 12-24-20 necessary to prepare and approbate the methodology for determination of dynamic characteristics of Due to the created methodology the following honeycomb optical tables. items were determined: The features of optical table are characterized by - Dynamic compliance (10…1000) Hz in static and dynamic rigidity. The maximum frequency range; deflection under the impact of static load is - First resonant frequencies; considered to be the measure of tabletop’s rigidity. - Dynamic deflection coefficient – DDC; Dynamic rigidity describes the resistance of table - Maximum relative tabletop motion – MRTM. plate to the impact of vibrations. The dynamic For this purpose the received dynamic characteristics of the table can be measured the best compliance curve (Fig. 3) of analyzed plate was by dynamic compliance – the dimension that is used, where two quite high resonant peaks of 199 Hz inverse to the dimension of dynamic rigidity. The and 220 Hz may be observed. dynamic compliance experimental curve may be The table until 80 Hz is presumed to be determined by impedance method. Then the a perfectly solid body, where compliance diminishes following things may be calculated using inversely to square of frequency (Ȧ = 2ʌf). The experimental data [1]: compliance curve helps the frequencies that are 1. Dynamic Deflection Coefficient – DDC that higher than 80 Hz to diverge from the straight line of assesses the relative dynamic characteristics of perfectly solid body. The table cannot be considered tabletop. It is determined in compliance curve. to be a perfectly solid body, it starts deforming 2. Transmissibility of Isolator - T – frequentative because of the vibration’s impact. function of ratio of tabletop’s motion and reaction Dynamic Deflection Coefficient (DDC) and of floor (base) that is expressed in decibels (dB). Maximum Relative Tabletop Motion (MRTM) were 3. Relative Tabletop Motion, i.e. relative movement calculated according to the methodologies of between two points of tabletop. The quicker the Newport and Melles Griot [2, 3]. motion is, the less possible justiration of the The factor of compliance strengthening Q of components mounted on the table surface table’s point 1 in the resonant frequency fn is becomes. The relative motion depends not only on calculated by drawing the characteristics of perfectly the dynamic characteristics of tabletop, but on the DIAGNOSTYKA’ 3(47)/2008 85 VOLKOVAS, GULBINAS, SLAVICKAS, Evaluation Of Quality Of Heterogeneous Mechanical Systems Using …

Fig. 3. Dynamic Compliance Curve and Resonant Frequencies (in point 1) of Optical Plate 1HT 12-24-20. solid body, i.e. the straight line, by the compliance Consequently, the performed calculations are curve (Fig. 3). Starting from the lowest frequency, correct and corresponding to the dynamic the compliance curve slopes down without characteristics presented in the catalogues [2-4]. discontinuities until the optical table is rigid and no relative motion is observed on the surface. 3. EXPERIMENTAL TEST OF INNER The ratio calculated according to the curve of SEDIMENT LAYER IN PIPES Fig. 3 in the zone of the first resonance (fn = 199 Hz) is the following: The known measurement methods of thickness Q = A/B = 71.7 [4, 5] are not effective while measuring the thickness Dynamic Deflection Coefficient (DDC) is of heterogeneous sediment. Firstly it is explained by calculated according to the formula: a very large number of measurements done in the 3 -3 measurement point in order to receive an integral DDC = Q f n = 0,00302 = 3,02·10 value in certain range. The sediment layer is Inverse dimension of dynamic deflection honeycomb and it is quite difficult to receive a better coefficient is DDC-1 = 331. reflection from the inner sediment layer that is Maximum relative tabletop motion is calculated necessary for resonant or impulse measurement of according to the formula: thickness. According to the tests, the wave interference method [6] that is now used for the MRTM = CT (Q f n x PSD) , 3 control of sediment layer has a number of where: T- transmissibility of isolator, PSD – power’s advantages, but it also has several disadvantages. spectral density, C – constant that determines the These could be the fact that the measurements are acceleration units and assesses the worst case done when the acoustic transformers are in good between any two points. In the case being analysed contact with the exterior surface of the pipes. This C = 0,623 m/s2; PSD = 10-9 g2/Hz (in the surface is often corrosive, mechanically damaged or environment close to busy roads); T = 0,01 (accepted difficult to access. In order to overcome these according to [3] Melles Griot Super Damp TM difficulties, the possibility to use the impedance qualitative isolators). method was analyzed experimentally by determining 71,68 the existence and thickness of the inner layer of MRTM = 0,623·0,01· 10 9 = 0,59·10-9 3 ˜ multilayered cylinder system. 199 The experimental test was done with the steel 2 [mm/s ]. fragments of the pipes (~1 m length), which have the Such a dimension may be explained by the fact calibrated thickness of inner sediment layer of that relative motion of tabletop’s point is measured 0 mm, 10 mm and 20 mm, the real (formed while 2 under the impact of 1 m/s acceleration, therefore the exploiting) inner sediment layer from 10 mm to 25 value of the maximum relative motion in the point 1 mm, or which do not have any inner sediment layer of probative table 1HT 12-24-20 is 0,59 nm. at all. The external diameter of the fragments of The calculated maximum relative plate motion is analyzed pipes is 150 mm, inner – 134 mm, their the following: layer of sediment – slash – is formed while using the -7 2 -9 MRTM = 0,00176·10 m m/s = 0.176·10 m. pipes in oil processing (process of petrol making). Relative Tabletop Motion (max) in the catalogue The tests used the method of mechanical of MELLES GRIOT [3] is the following: < 0,18 nm impedance in order to compare its possibilities and -9 (7·10 in.) results with the already approved and analyzed method of interference [6, 7]. The methodology of the mechanical impedance was implemented with 86 DIAGNOSTYKA’ 3(47)/2008 VOLKOVAS, GULBINAS, SLAVICKAS, Evaluation Of Quality Of Heterogeneous Mechanical Systems Using … the help of the specialized equipment of the While analyzing the impedance characteristics, company Brüel & Kjær GmbH, PulseTM 360, with the attention was paid to the changes of their the stroke hammer of the 8202 type and software BZ parameters (resonant frequencies, forms, etc.) and 7760. Such a set of equipment allowed recording their correlation with the condition of pipe’s inner a number of impedance curves, getting their average layer in the measurement place. and analyzing them. The characteristic shock’s In order to eliminate the pipe’s parry conditions, frequency range is presented in Fig. 4. The which affect the experiment, we chose the range of registration of the impedance characteristics was frequencies from 500 Hz to 5000 Hz for the test. done in various points of pipes’ fragments in the The results of the experimental test are shown in radial direction and along the pipe. Figs. 67.

[N] Autospectrum(Jega) - Input Working : Input : Input : FFT Analyzer 1

300m

100m

30m

10m

3m

1m

300u

100u 200 500 1k 2k 5k 10k [Hz] Fig. 4. Mechanical shock’s range

[N/(m/s)] Laipt. Neslif. 90 laip (Jega, Greitis) - File (Magnitude) Working : Input : Input : FFT Analyzer 1M

300k

100k

30k

10k

3k

1k

300

100 100 200 500 1k 2k 5k 10k [Hz] Fig. 5. Impedance of the calibrated pipe with the thickness of slash layer of 0 mm: continuous curve – excitation (stroke by hammer) in the pipe’s polished place; spot curve – in the unpolished pipe’s place DIAGNOSTYKA’ 3(47)/2008 87 VOLKOVAS, GULBINAS, SLAVICKAS, Evaluation Of Quality Of Heterogeneous Mechanical Systems Using …

[N/(m/s)] Laipt_2cm_(Jega, Greitis) - File (Magnitude) Working : Input : Input : FFT Analyzer

1M

300k

100k

30k

10k

3k

1k

300

200 500 1k 2k 5k [Hz] Fig. 6. Impedances of the calibrated pipe: continuous curve – 0 cm of slash (cursor’s frequency 1036 Hz); spot curve – 1 cm of slash (cursor’s frequency 1040 Hz); stroked curve – 2 cm of slash (resonance frequency 1224 Hz)

[N/(m/s)] Nr2 30cm (Jega,Greitis) - Input (Magnitude) Working : Input : Input : FFT Analyzer 3M

1M

300k

100k

30k

10k

3k

1k

200 500 1k 2k 5k 10k [Hz] Fig. 7. Impedance of the control pipe with real layer of slash (ca. 25 mm) (cursor’s frequency 1296 Hz).

According to the results averaged on the basis of slash. It can be explained by the fact that the identical measurements, the method of impedance is casing’s resonance in case of n=3 and m=2 indifferent to the condition of surface affected by the (parameters of frequency equation) is about 1020 Hz impulse power in the contact area (see Fig. 5). The without slash, while the layer of slash affects more impedance characteristics essentially do not differ the characteristics of system’s rigidity than mass, from 1000 Hz, independently whether the excitation that is why the resonance (see Fig. 6) has shifted. is done in the polished or unpolished part of the The impedance of the control pipe (the other pipe pipe. Besides, it has been noticed that the that has a 20-25 mm slash layer) in the frequency characteristic of the impedance in the pipe with real range of 1000÷1300 Hz showed a similar change. slash does not depend on the excitation place in This result is correlated with the impedance terms of average in the analyzed ranges of characteristics of the stepped (calibrated) pipe. frequencies. According to the presented results, the resonance The results presented in Fig. 6 show the general frequency changes the most in case of thicker slash changing tendency of resonance in pipes without layer. This shows that thinner layers change the slash and in real pipes with the 10÷20 mm layer of dynamic characteristics of the mechanical system 88 DIAGNOSTYKA’ 3(47)/2008 VOLKOVAS, GULBINAS, SLAVICKAS, Evaluation Of Quality Of Heterogeneous Mechanical Systems Using … a little. However after the complete modal analysis [6] Patent of Lithuanian Republic (V. Volkovas and is done, it is possible to expect other n, m V. Sukackas), Nr. 3304, 1996. combinations and relevant frequencies, in case of [7] Sukackas V., Volkovas V.: Investigation of the which this shift may get clearer. Lamb Wave Inference in a Pipeline with Sediments on the Inner Surface // Russian 4. CONCLUSIONS Journal of Nondestructive Testing. ISSN 1061- 8309. Kluwer Academic/Plenum Publishers, The done tests showed that the measurements of Vol. 39, No 6, 2003, p.p. 445-452. mechanical impedance are informative enough to identify the dynamic characteristics of Professor, habil. Dr. Vitalijus heterogeneous mechanical systems. In general VOLKOVAS  Head of meaning, the quality of these systems can be defined Department of Solid Mechanins by certain quantitative parameters, determined from of Kaunas Universtiy of the impedance curves. Technology, Director of the The following parameters are determined in case Institute of Technological of plane heterogeneous mechanical systems: Systems Diagnostics and Head dynamic compliance, first resonant frequency, of the Laboratory of dynamic deflection coefficient, maximum relative Vibrodiagnostics. and moni- top motion. toring. Member of IFToMM national Committee, With regard to the cylinder non-homogeneous member of Lithuanian Society of Scientists and mechanical systems it was determined that: Executive Committee of the Lithuanian Associotion  the impedance characteristics are correlated with on NDT&TD. Member of editorial boards of the thickness of heterogeneous layers of inner Journals „Diagnostyka“ (Poland), “Mechanika” and cylinder systems; „Vibroengineering“ (Lithuania), subst. of editor of  the information on the dynamic behavior of Journal “Measurement” (Lithuania). cylinder system is received from the impedance characteristics and it allows selecting the Dr. Ramunas GULBINAS  Head of the frequencies, which are informative for the Laboratory of Machinery Vibrations and Acoustic evaluation of inner sediment layer’s thickness; Noise Level Testing of the Institute of Technological  the shift of pipe’s resonance frequency may be Systems Diagnostics. Member of Lithuanian Society used to create the simple indicator of critical of Scientists. thickness of sediment layer;  contrary to the Lamb Wave Inference method Dipl. Ing. Edmundas Saulius SLAVICKAS is [6], the impedance method does not need special specializing in the field of measurements of conditions of acoustic contact. vibration of mechanical systems. The developed methodology may be used to assess the manufacturing quality of optical tables, according to the dynamic characteristics and parameters established in the methodology, and to solve the identification task of the technical condition of heterogeneous systems.

REFERENCES

[1] Gulbinas R., Slavickas E. S., Volkovas V., Stankus R.: Investigation of Dynamics of Optical Tables. Proc. of the 5-th Int. Conf. “Vibroengineering 2004”, Kaunas Technological University, p.p. 137-139. [2] Data about Optical Tables and Vibro-Isolators of Firm Newport www.newport.com. [3] Data about Optical Tables and Vibro-Isolators of Firm Melles Griot www.mellesgriot.com. [4] Alleyne D., Cawley P. A Two-Dimensional Fourier Transform Method for the Measurement of Propagating Multimode Signals’. J. Acoust. Soc. Am. 89/3/, 1991, p. p. 1159- 1198. [5] Devices and Equipment for NDT of Materials. A Review of Production Program of Krantkramer Firm. www.krantkramer.com. DIAGNOSTYKA’ 3(47)/2008 89 DRUZHININ, SAPOZHNIKOVA, TAYMANOV, Diagnostics Of Control Rod Drive. Possibilities To Extend ...

DIAGNOSTICS OF CONTROL ROD DRIVE. POSSIBILITIES TO EXTEND ITS LIFETIME AT NPP’S

Igor DRUZHININ, Ksenia SAPOZHNIKOVA, Roald TAYMANOV

D. I. Mendeleyev Institute for Metrology (VNIIM) 19 Moskovsky pr., St. Petersburg, Russia, 190005, fax: (+7-812) 251-99-40, e-mail: [email protected]

Summary According to specifications, lifetime of a linear stepping electromagnetic drive in WWER- 1000 nuclear reactors is limited by the pre-assigned lifetimes of the drive components such as electromagnets, a latch assembly, drive rack, and control rod position sensor. The lifetime of the drive components is from 7 to 20 years. On the basis of diagnostics of these components under real conditions, actual lifetimes of the components at nuclear power plant (NPP) were forecasted. It is shown that in order to extend the lifetime of the drive components up to the reactor lifetime (60 years), it is necessary to replace the sensor of the DPL-type by the new one (of the DPL-KV type) and to decrease the electromagnet temperature. The Measuring and Diagnostic System with the DPL-KV sensor enables diagnosing the sensor, electronic processing unit and drive. Economic effect due to the cumulative technical decisions is estimated to be 7 mln.euro per NPP unit.

Keywords: linear stepping electromagnetic drive, control rod position sensor, lifetime, nuclear reactor.

DIAGNOSTYKA NAPĉDU PRĉTA STERUJĄCEGO. MOĩLIWOĝCI ZWIĉKSZENIA CZASU EKSPLOATACJI NAPĉDÓW W ELEKTROWNIACH ATOMOWYCH

Streszczenie Zgodnie z dokumentacją czas eksploatacji elektromagnetycznych silników krokowych prĊtów sterujących w reaktorach jądrowych WWER-1000 jest ograniczony ustalonymi czasami eksploatacji ich skáadników (blok przesuniĊcia, prĊt, elektromagnesy, czujnik) – od 7 do 20 lat. Na podstawie wyników diagnostyki tych skáadników w warunkach realnych zostaáa zrobiona prognoza ich rzeczywistej ĪywotnoĞci w elektrowniach atomowych. Wykazano, Īe w celu wydáuĪenia czasu eksploatacji skáadników do czasu Īycia reaktora (60 lat) trzeba wymieniü czujnik DPL na inny, bardziej niezawodny – DPL-KV i znacznie obniĪyü temperaturĊ elektromagnesów. Ukáad pomiarowo-diagnostyczny wyposaĪony w czujnik DPL-KV pozwala przeprowadzaü diagnostykĊ czujnika, osprzĊtu elektronicznego i silnika. Caákowity efekt ekonomiczny wprowadzenia zaproponowanych rozwiązaĔ technicznych jest szacowany na 7 milionów Euro na blok energetyczny.

Sáowa kluczowe: elektromagnetyczny liniowy silnik krokowy, czujnik poáoĪenia prĊta sterującego, czas eksploatacji, reaktor jądrowy.

1. INTRODUCTION An internal drive part (the motion unit, rack, and a lower part of the sensor) is in the primary coolant At present, more than twenty power units with (the temperature is up to 350 °C and pressure is up nuclear reactors of the WWER-1000 type are in to 18 MPa), whereas an external part (the operation in Russia, Ukraine, Bulgaria, and Czechia. electromagnet assembly and a higher part of the The expected lifetime of such reactors will be about sensor) is outside the coolant. 60 years (the lifetime of 30 years, which was The housing provides hermeticity of the internal specified initially, will be prolonged). part. Under a reactor control mode, the In most WWER-1000 power units, the control electromagnet assembly is moving the rack with the and protection system applies linear stepping drives control rod by interaction with the motion unit which move control rods (CR’s). through the housing. The sensor function is Each drive comes in the form of a high cylinder measuring the control rod position. Under the and comprises mechanical unit (a housing, motion emergency shutdown mode, the electromagnet unit, and rack) and electrical unit (an electromagnet assembly is de-energized. The rack drops together assembly and sensor). with the control rod on an arresting device. 90 DIAGNOSTYKA’ 3(47)/2008 DRUZHININ, SAPOZHNIKOVA, TAYMANOV, Diagnostics Of Control Rod Drive. Possibilities To Extend ...

The specified lifetime of these drives is not deterioration. Falling loads cause distortion of the longer than 20-30 years. Accordingly, during the rack and wear of a rack damper. expected reactor lifetime, all the drives should be The most period of time during the fuel cycle, replaced at least once, that is not economic. the drives of the emergency shutdown groups are at This problem is particularly important for the the top end switch and do not move. Meanwhile, the drives of the SHEM-2 type, which are in operation drives in the control group move only in the higher at the most WWER-1000 Nuclear Power Plants part of the displacement range. Correspondingly, (NPP’s). The specified lifetime of mechanical unit only a part of the drive teeth is wearing in of this drive is 20 years. The lifetime of electrical interaction with the rack latches. Therefore, unit of the drive is 10 years or less. In the according to the criterion of rack wear, the rack specification, the value of rated life is 6000 double resource depends on the maximum load per rack strokes and the number of rated drops from any tooth. height in response to scram signal (the function of The MS’s were recording all the steps of the emergency shutdown) is 200. drives during each fuel campaign. Operation experience of the SHEM-2 drives as Fig. 1 shows a typical diagram of distribution of well as theoretical and experimental investigations the number of steps during the campaign. In order to show that at the operating NPP the lifetime of such identify the position of the control rod, a two- digit a drive can be enhanced up to the reactor lifetime. decimal scale is applied. After matching the control The prolongation can be achieved by drive rod positions with the number of the teeth, it follows modernization at a reasonable cost. from this diagram that the maximum load per tooth during a campaign is 300 times. 2. LIFETIME OF MECHANICAL UNIT

In the first place, the validity of correlation between the specified drive lifetime and specified resource as well as that of the specified resource should be analyzed taking into account operation experience. The main components of the drive mechanical unit are the motion unit and rack. Dynamic mechanical loads (vibration and shocks) and friction Fig. 1. Typical diagram of distribution of the influence the durability of these components number of steps during the campaign significantly. Correspondingly, the lifetime of the motion unit and rack must depend on the time of The average duration of the fuel cycle was 10 using of the resource. months or so. Therefore, the resource of the rack The estimate of the average lifetime of the drives teeth defined on the basis of maximum load per rack can be got on the basis of information about tooth is approximately 80 campaigns. Taking into operation of the SHEM-2 drives during a fuel cycle. account maintenance works (2 months), that is equal Such information was obtained in the Measuring to 80 years. If the duration of the campaign and Diagnostic Systems with the DPL-KV sensor increases up to 18 months, the above estimate does (MS’s) [1, 2], which were in operation at the 2nd not change significantly and exceeds 60 year if we power unit of the Kalinin NPP in Russia. According use any method of assessment. In reality, the to the information obtained in 9 fuel campaigns, in resource will be much higher because of the control group, each drive made not more than transposition of the drives from the control group to 217 double strokes, while in the emergency the emergency shutdown group. shutdown group the drives made only 100 double According to the information obtained in the strokes. MS’s, if the equipment is faultless (there is no If the operation intensity is the same, the unscheduled outage) the average number of the resource of the motion unit for the control group drops is not less than 4 times a year. Consequently, drives will be equal to approximately 250 years. The the estimate of the forecasted drop rack resource is drives are usually operating in the control group 50 years. In case of 18-months fuel cycle and the only during several years. After that, they are moved same number of drops during the campaign, the to the emergency shutdown group. So, during 60 forecasted drop rack resource is more than 80 years. years of operation, the SHEM-2 motion unit will use Taking into account that after the life test only including 200 drops, the condition of the rack is a small part of its resource. satisfactory, we can consider the real rack resource The lifetime of the rack depends on both the to be significantly higher than the specified one. intensity of control mode (step movement) and the The diagnostic information obtained in the 9 number of drops. The interaction of rack teeth with campaigns of the MS’s operation leaves no doubt latches of the motion unit leads to rack teeth DIAGNOSTYKA’ 3(47)/2008 91 DRUZHININ, SAPOZHNIKOVA, TAYMANOV, Diagnostics Of Control Rod Drive. Possibilities To Extend ... that the drop time will not exceed the maximum The simulation results made it possible to determine permissible value of 4 c during 60 years. the optimum parameters of the screen, which can provide decrease in temperature not less than 10 °ɋ. Such decrease corresponds to the forecasted lifetime of 60 years. Tests of the electromagnet with the screen under actual environment conditions confirmed the correctness of calculations. The least durable component of the electrical unit is the DPL sensor. Its specified lifetime is 5-10 years. The reason for rejection of the DPL sensors during operation is mainly concerned with oxidation of a wire and increasing the sensor coil resistance. Fig. 2. Dependence of drop time of the rack on time As a result, actually, the sensor lifetime is 3-7 years. for three MS’s (extrapolation to 60 years) In order to avoid the multiple replacing of the DPL sensor during the power unit operation, the An additional argument that the lifetime of the new sensor of the DPL-KV type that is much more racks is longer than 60 years is the positive result of longevous should be applied. At the same time, an tests of similar racks used for SHEM-I and SHEM-3 opportunity appears to eliminate the other defects drives. These tests included 400 and 470 drops, which are inherent for the SHEM-2 drive and correspondingly. electronic unit operating with the DPL sensor. They Thus, the forecasted lifetime of the main are: significant error of position measurement, components of the SHEM-2 mechanical unit unreliability, very small capability to organize exceeds the lifetime of the reactor. diagnostics, etc. The electrical components of the DPL-KV sensor are characterized by the lifetime 3. LIFETIME OF ELECTRICAL UNIT longer than 60-80 years. The validity of these values is confirmed by diagnostic information obtained in Estimate of a forecasted lifetime of the electrical 9 operation campaigns. The sensor signals (at the unit of the SHEM-2 (electromagnet assembly and same positions and under the same conditions) has DPL sensor) is of great importance, because these changed insignificantly. Resistance of the components are very expensive whereas their inductance coils tended to stabilization (Fig. 3). specified lifetime is too short. The drive of the SHEM-2 type comprises three 1,035 electromagnets identical in design. Depending on 1,030 their designation, they perform functions of pulling, 1,025 locking, and fixing electromagnet. According to [3], 1,020 life time of these electromagnets depends on the coil ʋ2 1,015 temperature when the drive is not moving. The Resistance ʋ4 1,010 lifetime of the fixing electromagnet is 25-30 years, ʋ5 for pulling and locking electromagnets it is much 1,005 longer than the lifetime of the reactor. 1,000 Possible methods for decreasing the 0 10203040506070 electromagnet temperature were analyzed. They are: t, years x increasing the heat transfer from the Fig. 3. Dependence of sensor resistance (relative electromagnet to air; value) on time (extrapolation up to 60 years) x decreasing the power of inner heat generation in the electromagnet coil by reducing the value of The forecasted variations of the average value of current; the DPL-KV coils resistance are not more than x decreasing the radiant heat exchange between the 3.5%. In practice, such variations do not influence housing and electromagnet. the MS operation. All the more, its algorithm Decreasing the radiant heat exchange can be includes adaptive procedures which correct the achieved by application of a thin heat-reflective signals under changing environment. screen with a high reflectivity that does not change The specified lifetime of a sensor housing is not significantly during long term operation. The screen less than 30 years. It can be prolonged up to 40-60 should be inserted at the inner surface of the fixing years on the basis of diagnostics results. electromagnet [3, 4]. It follows from the above-said that the stability Simulation of the electromagnet operation has of the DPL-KV parameters is sufficient to provide shown that application of the screen is the most its reliable functioning during the expected reactor effective method for electromagnet temperature lifetime. In order to apply the DPL-KV sensor, it is decreasing. Insertion of the screen can be carried out necessary to replace the standard rack by the similar at the operating NPP during maintenance works. rack that includes a multi-component shunt. Since 92 DIAGNOSTYKA’ 3(47)/2008 DRUZHININ, SAPOZHNIKOVA, TAYMANOV, Diagnostics Of Control Rod Drive. Possibilities To Extend ... the racks which have partly used their drop resource allows to diagnose the condition of a rack (that is the most critical parameter) will be also damper); replaced, such a replacement will provide the x checking whether the CR has fallen down on the sufficient resource of the new racks which will be arresting device. capable to operate up to the end of the reactor The high diagnostic sensitivity allows to reveal lifetime. incipient defects (even before appearance of Thus, the replacement of the DPL sensor and a significant failure). Diagnostic capabilities can be standard rack with the DPL-KV sensor and the new further enhanced in the future. rack, as well as insertion of the heat-reflective Information about all CR moves, control screen inside the fixing electromagnet enable commands, operation modes, malfunctions or prolonging the drive lifetime up to the lifetime of the failures as well as operator’s actions are logged in reactor. a “black box” recorder. At the same time, the MS estimates the drive operational conditions by 4. DIAGNOSTIC CAPABILITIES OF MS accumulating parameters like the number of drops, steps made, input control signals, etc. It should be noticed that the lifetime of each The real time CR position is displayed on a front drive component set in the documentation was panel. Complete set of information can be calculated and confirmed by the life test of several transferred to a special palm computer and shown on samples. The actual resource of each component its display. If necessary, data can be transferred to depends, to a great extent, on the quality of a PC for archiving and analyzing. Each MS can be manufacturing, which is checked ineffectively in connected to a local network. In this case, the MS many cases [5]. Removal of a failue arisen in the can perform cross system diagnostics. This improves drive during the reactor operation leads to the MS fault-tolerance. For instance, the local significant expenses. In order to minimize the network gives an opportunity to inform operators probability of such a failure, it is necessary to about the wrong positions of CR, including the case diagnose the drive directly during the fuel cycle. of CR position mismatch in the control group as Such diagnostic possibilities are realized in the MS, well as of any CR slipping down from the end which provide: switch. x CR position measurement (accuracy is within Based on diagnostic information obtained during 0.3% under all possible conditions); system operation, an individual “registration x metrological diagnostic check of the reliability of certificate” is automatically issued for each drive. CR position measurements (it is not necessary to This certificate contains an assessment of drive calibrate the sensor during operation); condition as well as recommendations for operators x fault tolerance (maintenance of operating how to carry out preventive maintenance. integrity if any signal wire breaks or any coil Fig. 4 illustrates the diagnostic capabilities of the fails); MS. The diagnostics applies the diagrams of x automatic correction of a sensor, shunt and displacement. The diagram enables: processing unit parameters (this eliminates the x defining the actuation time of the motion unit influence of temperature variations, defects of latches, joints, material and component aging); x checking the correctness of response to x filtration of various noises; a cyclogram of the electromagnet current, x self-diagnostics of whole MS with failure x checking the control rod and rack coupling. localization; The ability to obtain such diagrams is determined x generation of textual recommendations for by both the high displacement sensitivity of the malfunction elimination; sensor and the fact that the time interval between x assessment of the condition of main drive two consecutive control rod position measurements components (rack teeth, latches of the motion is very short. In case of the drive fault, the form of unit and, partly, electromagnets); the diagram is changing. This makes possible to find x drive condition diagnostics (step missing or out the origin of the fault or to reveal the incipient delay, as well as teeth slippage); fault (even before appearance of a significant x checking of the CR and rack coupling; failure). x sensor to processing unit connection diagnostics; x control connection diagnostics; x measurement and recording of CR drop time diagram (this allows to diagnose the condition of a guide sheath and rack curvature in case of CR emergency shutdown or spontaneous drop); x determination of the top and bottom oscillation points during the CR damping process (this DIAGNOSTYKA’ 3(47)/2008 93 DRUZHININ, SAPOZHNIKOVA, TAYMANOV, Diagnostics Of Control Rod Drive. Possibilities To Extend ...

[5] Taymanov R. Ye., Sapozhnikova K. V., 25 Lukashev A. P, Pronin A. N.: About Estimation 20 of the Quality of Manufacturing of Sensors with

15 the Long-term Calibration Interval, Datchiki I Systemi, 2006, 9, p. 67 - 80. 10 5 Dr. Igor DRUZHININ. Dr. Displacement,mm 0 Druzhinin is currently principal

-5 engineer at the VNIIM. He is the 0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1 author of more than 20 Time, s publications. Major research activity during recent years: Fig. 4. Diagram of drive rack move: a step up measurements of various non- electric quantities under harsh 5. CONCLUSION environment, increasing the nuclear power plant equipment lifetime. In order to increase the lifetime of the control rod drive up to the expected lifetime of the nuclear Dr. Ksenia SAPOZHNIKOVA. reactor (60 years), it is necessary to replace the Dr. Sapozhnikova is deputy head standard sensor and standard rack with the DPL-KV of laboratory at the VNIIM, sensor and the new rack, as well as to decrease lecturer at the Courses for Power noticeably the electromagnet temperature, using the Plants Chief-metrologists. Main screen. research area: measurements The MS with the DPL-KV sensor excels the under harsh environment and analogues at reliability, diagnostic capabilities, fault- metrological diagnostic check of tolerance, and accuracy significantly. Diagnostic sensors. Dr. Sapozhnikova is the author of more capabilities of the MS make it possible to switch than 80 publications. from pre-assigned terms to forecasting equipment condition during the future campaign and to Dr. Roald TAYMANOV. repairing the drives depending on their technical Dr. Taymanov is head of condition. laboratory at the VNIIM, The economical effect of the totality of the academician of the Russian suggested technical decisions is about 7 million euro Metrological Academy. He has per power unit. These decisions can be also applied been the scientific leader of to the other types of the stepping drives, for example many various projects on SHEM-3, used at the WWER-1000 nuclear reactor. signals transformation and measure-ments. Dr. Taymanov REFERENCES is the author of more than 160 publications. Listed [1] Sapozhnikova K., Taymanov R., Druzhinin I.: in “Marquis Who’s Who in Sci&Eng 2006-2007” “Who’s Who in the World 2008”, and “Outstanding About the Effective Approach to the st Modernization of the NPP Control and Scientists of the 21 Century”. Emergency Shutdown System, IAEA Technical Meeting 13-16 September 2005, Chatou, France, http://entrac.iaea.org/I-and- C/TM_EDF_2005_09/Session%2 4%20Wedn%20PM/Sapozhnikova%20paper%2 0updated.pdf. [2] Taymanov R., Sapozhnikova K., Druzhinin I.: Measuring Control Rod Position, Nuclear Plant Journal, 2007, 2 (March-April), USA, p. 45-47. [3] Druzhinin I. I., Gerasimov N. P., Taymanov R. Ye.: Possibility for Enhancing the Lifetime of Electromagnets of the Drives in the Control and Protection System in a Nuclear Reactor, Energoma-shinostroenie, 2006, 2-3, p. 18 - 24. [4] Patent RU 44867 U1, ɆɉɄ7 H01F7/16, G21C7/12. Electromagnet of the Drive moving a Control Rod in a Nuclear Reactor, Gerasimov, N. P., Druzhinin, I.I., Taymanov, R. Ye., 27.03.2005, Bulletin, 2005, 9.

DIAGNOSTYKA’ 3(47)/2008 95 MICHALSKI: Diagnostics In Car Maintenance

DIAGNOSTICS IN CAR MAINTENANCE

Ryszard MICHALSKI

Katedra Budowy, Eksploatacji Pojazdów i Maszyn, Wydziaá Nauk Technicznych UWM w Olsztynie, [email protected]

Summary The paper proposes an extension of the currently applied vehicle inspection, based on the OBD II system, and periodical inspection, by vehicle dynamic load testing. The method involves generating faults based on an analytic model of traffic. The method takes into account the static and dynamic properties of a vehicle within the whole range of its operation, taking into account the variable traffic intensity, and enables early detection of defects which accumulate at variable random loads. The highest precision of computation is ensured by traffic models with elastic and damping constraints, with variable rigidity and damping.

Keywords: diagnostics, vehicle maintenance, dynamic load.

DIAGNOSTYKA W UTRZYMANIU POJAZDÓW SAMOCHODOWYCH

Streszczenie W pracy zaproponowano rozszerzone podejĞcie do obecnie stosowanej kontroli stanu pojazdów, opartej na OBD II i kontroli okresowej, o pomiary obciąĪeĔ dynamicznych pojazdu. Zaproponowano metodĊ generacji niezdatnoĞci pojazdu na podstawie modelu analitycznego ruchu drogowego. Metoda ta uwzglĊdnia wáasnoĞci statyczne i dynamiczne pojazdu w caáym zakresie jego pracy z uwzglĊdnieniem zmiennej intensywnoĞci ruchu i pozwala na wczesne wykrycie rozwoju uszkodzeĔ kumulujących siĊ przy zmiennych obciąĪeniach losowych. NajwiĊkszą dokáadnoĞü obliczeniową mogą zapewniü modele ruchu z wiĊzami sprĊĪystymi i táumiącymi o zmiennej sztywnoĞci i táumieniu.

Sáowa kluczowe: diagnostyka, utrzymanie pojazdu, obciąĪenie dynamiczne.

1. INTRODUCTION - a driver’s errors made in the “man – car – road – service” system, The basic tasks of technical diagnostics include - construction errors made in the process of detection and identification of failures and faults in designing the system of a vehicle operation, cars, which – directly or indirectly – result in their - manufacturing errors, made in the production unreliability. The elements of faults in which process, various factors result in disturbing the balance of - maintenance errors, made in the condition forces, moments or balance in the continuity of inspection and during the process of energy and information flow, which may result not maintenance. only in a vehicle fault, but also in its failure or even The problem of diagnosing the condition of a car accident. Another task is the choice and a vehicle during its operation can be reduced to implementation of a strategy of car maintenance, checking the correctness of parameter changes in whose aim is to slow down the process of its relation to the operation time t and determination of degradation and to restore its fitness for use [6]. status control cycles as well as locating the faults in The theoretical foundations for developing a vehicles. various diagnostic methods are algorithms of car Diagnosing a vehicle may be reduced to inspection on various levels of its complexity comparing actual characteristics u with assumed (decomposition). The knowledge of the dynamics of ones uo, according to the following relationship: a vehicle steering provides more in-depth and comprehensive information about the vehicle tk U e u e ,t  u e,t k dt min condition and static characteristics. The degrading ³ >@0 0 effect of dynamic load on kinematic nodes is greater t0 than that of static load or natural wear. The causes where: of failures and faults in the system of vehicle U e - function of weight, U >@0,1 ; operation system are the following [6]: e0 and e - a set of assumed and actual diagnostic - dynamic input function in the course of vehicle parameters; operation, t - variable time of vehicle condition inspection; 96 DIAGNOSTYKA’ 3(47)/2008 MICHALSKI: Diagnostics In Car Maintenance k 1,2 - exponent of integrand. by new diagnostic signals, taking into account the The essence of the currently applied diagnostic intensity of the vehicle work and the dynamic load systems can be reduced to the action of a summing in the process of its use. node e e e 3. IDENTIFICATION OF DYNAMIC LOAD ¦ i  0i ' i i AND VIBRATIONS IN VEHICLE MOTION where the programmed values e0 are compared with the set e which describes the current status, taking A vehicle and its elements are adapted for into account the defects and faults of a vehicle. transferring specific static and dynamic load. The load taken into account in the process of a vehicle 2. STANDARDS OF ON-BOARD design, in terms of the course as well as the values DIAGNOSTICS (OBD) and time, correspond to the average operating conditions. Such conditions can be referred to as the The current standards adopted for the assessment average vehicle load with cargo, good road surface of the technical condition of cars as OBD I in the condition, moderate utilisation of engine power and last decade of the 20th century have a number of driving speed [2]. restrictions. Since 1996, the OBD II standard has The level and nature of the dynamic load been applied for cars. Technical diagnostics in this originating in a vehicle depends on its structure and approach consists in identification and location of technical condition and on the traffic conditions. a failure where it originated, usually defects in the Dynamic load is a result of vibrations produced by power transmission system, assemblies and the a vehicle motion. The vibration intensity and, vehicle functional systems from the point of view of consequently, the level of dynamic load increases road safety and environment protection [4]. with: The principles of on-board diagnostics were x increase in the driving speed and vehicle weight; determined by SAE (Society of Automobile x increase in the power transmitted by the power Engineers) in its standard SAE J 1830, while the transmission system; requirements of on-board diagnostics by European x deterioration of the road surface condition; car manufacturers in the standard ISO 9141-2. These x driving style, including frequent changes in the standards constitute a set of technical and legal engine rotational speed, engaging the clutch and requirements which define the on-board diagnostic gears; system as diagnostic procedures, installing x increasing clearance in connections, which cause diagnostic sockets, processing the diagnostic signal, parts to hit one another. monitoring the basic parameters of the power The most dangerous values of high dynamic load transmission system, including exhaust emission and the resulting extreme tensions in the vehicle parameters. parts may be caused by: The criterion of assessment of whether an x driving at high speed on a bumpy road; assembly and its functional elements work properly x vehicle overloading; is the exceeding of the adopted threshold of 50% of x construction changes which are at variance with the diagnostic parameter, which is signalled and the documentation provided by the vehicle recorded with relevant error codes [5]. The OBD II manufacturer, e.g. increasing the engine power, system is able to detect faults which are the main wheel track; elements of a power transmission system, i.e. x resonance in the power transmission system or increased waste emission above the accepted suspension. threshold. They are mainly related to signalling of Additionally, all the rotational elements with faulty operation of electronic systems, sensors in the clearance in connections, imperfect shape or engine power supply system, leaks in the fuel balance, may result in periodical dynamic system, etc. Ultimately, the existing OBD II will interactions. They increase with increasing provide a basis for developing rotational speed, and the coincidence of the a comprehensive diagnostic system for an entire rotational speed with the natural frequency of an vehicle, with the possibility of connecting to element or assembly results in resonance and rapidly a remote service for locating any defects and failures increasing dynamic load. of the vehicle with the use of the GPS system. The actual load changes are random. Despite the The adopted system has a number of restrictions complex character of load which corresponds to which are a consequence of reducing the status particular stages of a vehicle drive, they can be control only to selected test points and the location approximately substituted with blocks of repeatable of errors at the assembly and test element level. The cycles with predetermined amplitude and frequency. difficulties encountered here include diagnostic The results of such tests can be shown, e.g. as inference, identification of diagnostic relations and a Wöhler’s diagram. The diagram was used to locating the types of failures. Therefore, this paper present the relationship between the variable load ıA proposes to extend the scope of a vehicle diagnostics and the number of load cycles N before the vehicle DIAGNOSTYKA’ 3(47)/2008 97 MICHALSKI: Diagnostics In Car Maintenance damage (destruction). If, while driving, a vehicle ni element is subjected to load with an amplitude of D i N ıAi, it can withstand without defect the number Ni of i load change cycle, which may correspond to the where: number of kilometres of driving (Fig. 1). ni - number of load cycles with the amplitude ıAi; Further considerations involved Palmgren- Miner’s hypothesis of fatigue life, which describes N i - the number of load cycles ıAi, after which the the process of linear accumulation of fatigue failures element becomes damaged. of machine elements. The greater the load changes The number ni is determined from the load amplitude, the more intensive the microdamage change course in the analysed element, and Ni is development. If the load cycles are repeated, determined from the Wöhler’s diagram. An element microdamages accumulate (add up) in the element damage, i.e. exceeding its durability, takes place and the element becomes damaged (e.g. cracks) after after the sum of load cycles is equal to: a certain number of load cycles. D ¦ Di 1 In order to describe the process of damage i accumulation, the coefficient (D) is introduced: These considerations will be used in a computational example. An example changes course of a bearing load is shown in Fig. 2.

Zmienne obciąĪenie dynamiczne o

amplitudzie ıAi

Fig. 1. Idealised load changes during the tests of the elements durability

Fig. 2. An example change course of a drive wheel bearing dynamic load 98 DIAGNOSTYKA’ 3(47)/2008 MICHALSKI: Diagnostics In Car Maintenance

It has been chosen to show the effect of the 4. A FAILURE – DIAGNOSTIC - ORIENTED manner of a vehicle use and the dynamic load MODEL OF DYNAMIC LOAD IN which results from it on the vehicle condition. A POWER TRANSMISSION SYSTEM Table 1 shows example results of a bearing fatigue test, performed by the manufacturer. These are the Considerable dynamic load may appear in the numbers Ni of load cycles with the amplitude of ıAi power transmission system, due to: until the bearing is destroyed. They were x variable nature of the action of resistance forces juxtaposed with the results of bearing load test while driving; results during the drive over the distance of 10 km. x uneven work of a combustion engine; Static load have been omitted in Fig. 2. x improper action of the driver on the clutch, The random changes of the operational load was brake system and when changing gears; the basis of determination of the load spectrum. To x lack of balance and kinematic compatibility of do this, the distribution of peak loads, i.e. local the drive shaft and axle-shafts; extremes, were found. After the extreme values x imprecise workmanship, wear and clearance in were counted on particular levels of load, the assemblies of the drive transmission system. numbers of cycles ni of load with amplitude of ıAi These factors produce primarily torsional were determined and shown in Table 1. vibrations as well as vibrations and noise in the Adding up the drive wheel bearing load cycles power transmission system. over a distance of 10 km: A vehicle model is a set of interconnected n 1 2 2 4 19 D i   0    partial models: ¦ N 1000 10000 200000 500000 1000000 i x a driver model; D 0,001237 x a steering system model; Using the linear hypothesis of damage x a power transmission system model; accumulation, the bearing forecast has been x a model of wheel cooperation with the ground calculated, expressed by the mileage done before (a wheel model in combination with the ground a failure (L) for D=1. model); x a chassis model; 1˜10 L # 8084km x a work environment model. à 0,001237 The driver model describes the forcing actions Mileage 10 km - failure coefficient D = 0.002476 performed by the driver (the force on the steering X km -failure coefficient D = 1. wheel, the position of the clutch and brake levers, Therefore, the mileage before the damage batching fuel, etc.). It is related closely to the calculated for the second case is equal to: function of senses, mental processes in the brain, 1˜10 reflexes, the level of manual skills. The steering L 4039km 0,002476 system model describes the geometrical and dynamic relationships in the steering mechanism. It The computational example clearly confirms helps to examine clearances and friction. It helps that the higher the values of dynamic load during determine the position of turning wheels and the drive, the shorter the time of the bearing work moments on the stub axles depending on the before a failure. Similar physical phenomena occur steering wheel position and the force applied to it, with other structural elements of a vehicle. as well as the shift of the suspension elements. Therefore, assuming a 20% surplus of an element The basic tool applied in the process of the durability, the technical condition of a bearing vehicle model construction, are automatic system of should be checked in the first case after the mileage generating the equations of motion, used to analyse of 6,500 km and in the second case – after that of multi body systems (MBS - Multi Body System 3,400 km. In the adopted standards of on-board Formalizm), which include: [8]: diagnostics, the issues of vehicle dynamics have been practically omitted. Vibrations during the vehicle work not only create dynamic load of elements and assemblies, contributing to their lower durability, but they also significantly affect the effectiveness and efficiency of the driver’s actions and those of people travelling in the vehicle.

* - Failure – loss of a vehicle’s ability to perform the required functions. Critical failure – one which poses a threat to humans, results in considerable material damage or other unacceptable outcome [PN-93/N-50191]. **- Fault – a condition of a vehicle in which it is unable to perform the required functions [PN-93/N-50191]. DIAGNOSTYKA’ 3(47)/2008 99 MICHALSKI: Diagnostics In Car Maintenance

Table 1. Number of load cycles ni, isolated from dynamic load changes of a drive wheel bearing, assumed during the drive over the distance of 10 km Amplitudes of load cycles Number of load cycles ıA1 ıA2 ıA3 ıA4 ıA5 ıA6 600MPa 500 MPa 400 MPa 300 MPa 200 MPa 100 MPa Ni 1000 10 000 50 000 200 000 500 000 1 000 000 Number of load cycles ni, recorded during the drive in nominal 1 2 0 2 4 19 conditions Number of load cycles during 2 3 6 5 10 11 extreme driving x ADAMS (Automated Dynamic Analysis of Assuming the forces of gravity, elasticity and Mechanical Systems); attenuation as external forces, the equation has the x SIMPACK (SIMulation PACKage for following form: multibody systems); x MEDYNA (MEhrkörper DYNAmik); d § wE · wE ¨ k ¸  k Q , i 1,2,..., s . x DADS (Dynamic Analysis and Design System); ¨ ¸ qi dt © wqi ¹ wqi x SD/FAST (Symbolic Dynamics/FAST); x MADYMO (MAthematical DYnamical An attempt to link dynamic stresses (ıd) (load) MOdel); of any vibrating element in a vehicle with the speed According to Schiehlena W., the most common of its vibration can be presented as: and advanced MBS systems include: ADAMS and V V ˜ U ˜ c ˜ k for ım=0 DADS. d p d DADS helps analyse dynamics in the time where: domain of rigid body systems (the latest version of ıd – dynamic stress; the program enables analysing pliable bodies). Vp – peak amplitude of vibration speed, measured Such bodies may be interconnected by kinematic at the site of maximal dynamic deformations; and pliable nodes. ȡ – density of the material mass; The model notation and analysis employing it c – velocity of speed propagation in the material; are made possible by the following program blocks: kd – the dynamic coefficient depending on the x DADS Pre-processor – makes it possible to energy concentration; enter model data; ım – working stress. These equations can be used to identify failures x DADS Analysis – generates motion and node equations which are used to determine: position, in specific functional circuits of a vehicle. speeds, accelerations and reactions. Diagnostic inference employs models of R relationship definite on the cartesian product of the x DADS Postprocessor – generates computation sets of a vehicle failures and faults results in the form of time series; x DADS Graphic Environment – generates F ^f i ;i 1, I` graphic interpretation of computation results and diagnostic signals (vibrations, power, moment, (animation). efficiency, clearances, etc.) The DADS environment has been used to : generate the equations of motion of a vehicle. S ^s j ; j 1, I` RF / S  F u S A system of difference equations is generated The expression f i Rsj means that the diagnostic automatically based on Lagrange’s equation of the second type in it input form [8]: signal sj identifies the fault fi. For example, the occurrence of the failure fi (clutch clearance) results in power drop and is a diagnostic signal with the E d § wEk · wEk w p value of 1. Then the matrix of relationships R can ¨ ¸   Q , F/S dt ¨ wq ¸ wq wq qi be shown as a binary diagnostic matrix, defined as © i ¹ i i follows: i 1,2,..., s ; °­0 œ f i , S j  RF / S where: r f i , S j ® 1 œ f , S  R Ek – the system kinetic energy; ¯° i j F / S Ep – the system potential energy; The relationships RF/S may be defined by qi – i-th generalised coordinate; assigning to each diagnostic signal a subset of Qqi – generalised force corresponding to the failures F(sj), detected by such a signal: i-th generalised coordinate; F s j ^f i  F : f i Rs j ` s – the number of the system’s degrees of freedom. 100 DIAGNOSTYKA’ 3(47)/2008 MICHALSKI: Diagnostics In Car Maintenance

Binary diagnostic matrices can be determined 4. BocheĔski C.: Badania kontrolne samochodów. by conducting simulation tests on a physical model WKià. Warszawa 2005. of 5. Merkisz J., Mazurek St.: Pokáadowe systemy a vehicle or station tests with programmed dynamic diagnostyczne pojazdów samochodowych. loads, and based on expert knowledge. WKià. Warszawa 2002. 6. NiziĔski S., Michalski R.: Utrzymanie pojazdów 5. SUMMARY i maszyn. Wydawnictwo Technologii Eksploatacji-PIB. Radom-Olsztyn 2007. With the development of car constructions, the 7. Korbicz J., KoĞcielny J., Kowalczuk Zdz., dynamics of their work increase, which creates the Cholewa W.: Diagnostyka procesów, Modele, demand for new diagnostic models based on Metody sztucznej inteligencji, zastosowania. failure-oriented physical models of vehicle motion. WNT 2002. Such diagnostic models help improve the 8. Szczyglak P.: Wpáyw wáasnoĞci dynamicznych effectiveness of vehicle maintenance by adapting agregatu maszynowego na jego statecznoĞü. the vehicle inspections to operational input Rozprawa doktorska. Politechnika Warszawska functions. Thanks to early detection of a fault in 2005. a vehicle it is possible to prevent a failure and to improve traffic safety. Prof. dr hab. inĪ. Ryszard This paper proposes an extended approach to MICHALSKI jest the currently applied vehicle condition inspection, kierownikiem Katedry based on OBD II, and periodical inspection, by Budowy, Eksploatacji a vehicle dynamic load measurement. It proposes Pojazdów i Maszyn na a method of generating a vehicle fault based on an Wydziale Nauk analytic model of traffic. The method takes into Technicznych UWM account the static and dynamic properties of w Olsztynie. a vehicle throughout the period of its operation, W dziaáalnoĞci with variable motion intensity, and enables early naukowej zajmuje siĊ detection of failures which accumulate at variable diagnostyką techniczną, niezawodnoĞcią, random loads. The highest precision of technologią napraw i analizą systemową computation is ensured by traffic models with eksploatacji pojazdów i maszyn roboczych. Posiada elastic and damping constraints, with variable w swoim dorobku naukowym ponad 260 publikacji rigidity and damping. naukowych i naukowo-technicznych oraz However, there are some obstacles in this 6 patentów. Jest autorem lub wspóáautorem approach which are related to the structure of the opracowaĔ zwartych w tym: Pokáadowe systemy diagnostic matrix, the absence of clear diagnostic nadzoru maszyn ze sztuczną inteligencją (1997), relations, the effect of non-linearity of models Metody oceny stanu technicznego, wyceny which determine the failure development. As pojazdów i maszyn (1999), Diagnostyka obiektów a consequence, it may result in missing the failures technicznych (2002), Diagnostyka maszyn to which the adopted diagnostic signals are roboczych (2004), Utrzymanie pojazdów i maszyn sensitive. (2007). Jest czáonkiem Zarządu Gáównego Polskiego Towarzystwa Diagnostyki Technicznej, REFERENCE Redaktorem Naczelnym czasopisma „Diagnostyka”, czáonkiem zarządu Polskiego 1. PN-93/N-50191 NiezawodnoĞü, jakoĞü usáugi. Naukowo-Technicznego Towarzystwa 2. Prochowski L.: Mechanika ruchu, pojazdy Eksploatacyjnego, czáonkiem zespoáu samochodowe. WKià. Warszawa 2005. Ğrodowiskowego Podstaw Eksploatacji Komitetu 3. Michalski R, Wierzbicki S.: Diagnostyka Budowy Maszyn Polskiej Akademii Nauk, pojazdów samochodowych. Wybrane Komitetu Motoryzacji i Energetyki Rolnictwa PAN zagadnienia transportu samochodowego. Oddziaá w Lublinie. PNTTE. Warszawa 2005. DIAGNOSTYKA’ 3(47)/2008 101 ĩÓàTOWSKI, Technical Diagnostics Of Folded Objects. Directions Of Development

TECHNICAL DIAGNOSTICS OF FOLDED OBJECTS. DIRECTIONS OF DEVELOPMENT

Bogdan ĩÓàTOWSKI

University of Technology and Life Sciences 85-796 Bydgoszcz, ul. Prof. S. Kaliskiego 7, e-mail: [email protected]

Summary The destruction processes of technical systems extort need the changes of supervising their technical state. Modern technical diagnostics methods are the tool of diagnosing their technical state what is the basis undertaken the decision. The technical diagnostics, he is one from basic sciences about the rational exploitation of objects next to tribology, reliability, safety and exploitation theory. The acquaintance of physical phenomena drawing ahead while functioning machine engine makes possible the qualification of qualitative relationships between destructive processes and the machine engine condition. The great dispersion of the initial properties of machine, how and undeterminancy and the continuity of aging and the cells processes and the tasks of the diagnostics machine which has to work out the specific gathering of diagnosing clearly use up outline methods. The important problems of the article enclose: modelling the machine engines dynamics in the aspect of diagnostics, multidimentions and her the various ways of solving SVD, PCA, Date Fusion, diagnostics integrated with the object, artificial intelligence, diagnostic agent and system solutions of folded objects diagnostics.

Keywords: technical state, dynamics, destruction, symptoms, fault diagnostic system, development.

DIAGNOSTYKA ZàOĩONYCH OBIEKTÓW TECHNICZNYCH. KIERUNKI ROZWOJU

Streszczenie Procesy destrukcji systemów technicznych wymuszają potrzebĊ nadzorowania zmian ich stanu technicznego. Metody diagnostyki technicznej są narzĊdziem diagnozowania ich stanu technicznego, co jest podstawą podejmowanych decyzji. Diagnostyka techniczna, obok tribologii, niezawodnoĞci, teorii bezpieczeĔstwa i teorii eksploatacji jest jedną z podstawowych nauk o racjonalnej eksploatacji obiektów. Poznanie zjawisk fizycznych zachodzących w czasie funkcjonowania maszyny umoĪliwia okreĞlenie związków jakoĞciowych miĊdzy zachodzącymi procesami destrukcyjnymi a stanem maszyny. DuĪy rozrzut wáasnoĞci początkowych maszyny, jak i nieoznaczonoĞü i ciągáoĞü procesów starzenia i zuĪyü wyraĨnie zakreĞlają cele i zadania diagnostyki maszyn, która musi wypracowaü sobie specyficzny zbiór metod i Ğrodków diagnozowania. WaĪne problemy tego referatu obejmują: modelowanie dynamiki maszyn w aspekcie diagnostyki, wielowymiarowoĞü i jej róĪne sposoby rozwiązywania, SVD, PCA, Data Fusion, diagnostyka zintegrowana z obiektem, sztuczna inteligencja, agent diagnostyczny oraz systemowe rozwiązania diagnostyki obiektów záoĪonych.

Sáowa kluczowe: stan techniczny, dynamika, destrukcja, symptomy, system diagnostyczny, rozwój.

1. INTRODUCTION The search of ways and manners to describe the energetic wear and tear of machines and their The presented considerations concern modern elements already has some methodological premises approach towards the dynamic state modeling and first applications. However, these are still objects with the use of descriptions and researches theoretical considerations, here and there supported within the scope of identification, distinguishing the by practical researches approximate to the sought- possibilities modern IT technologies and issues after energetic mappings, not always, however, with directly supporting different methods of machine a simple physical interpretation. The modeling of dynamics evaluation and forming. Emerging dynamic state changes and object functioning evolutionary dynamic models improve the capability, taking into consideration variable load as methodology and inference in dynamic state well as individual approach towards the state evaluation which is increasingly often used for the changes of each element, is merely the beginning optimization of constructions and which supports within the scope of evolutionary models application. exploitation decisions. Such models require an analytic base while at the 102 DIAGNOSTYKA’ 3(47)/2008 ĩÓàTOWSKI, Technical Diagnostics Of Folded Objects. Directions Of Development same time appropriately reflect construction and Dynamics analysis of a system consists of the exploitation changes taking place during the time of following stages [15, 18]: machine’s life. -stage I – accurate determination of the system, its The knowledge of dynamic state and system crucial characteristics and building a physical structure allows to describe its behavior, as well as model whose dynamic characteristics will be, to an allows to create forecast models of system behavior adequate degree, correspondent to characteristics in the function of dynamic evolution time, based on of the real object; the model of technical state symptoms increase. -stage II – analytical description of dynamic Most frequently, however, equations describing phenomena reflected by a physical model, i.e. system behavior in the function of dynamic finding a mathematical model, differentia evolution time are not known, which justifies the equations describing the motion of the physical need of implementing new tools for dynamic state model; research. There is, therefore, the requirement of -stage III – studying dynamic characteristics of experimental verification for analytic models of a mathematical model on the basis of solving technical objects, for a proper model is one which differentia equations of the motion, determining verifies itself in practice. Therefore, an experiment is the predicted motion of the system; only an inspiration for further research leading to -stage IV – making design decisions, i.e. accepting construction optimization. physical parameters of the system, with In the procedure of dynamic state experimental modernization adjusted to expectations. The identification, we have distinguished: the synthesis and optimization leading to obtaining methodology of information acquisition; the required dynamic characteristics of the methodology of vibration estimators making; the construction. methodology of information processing; the The presented procedure is based on the methodology of cause-and-effect concluding as well knowledge of the system’s model, and conclusions as the methodology of results static preparation. drawn form actions on models depend on their Within these ranges, many new programs have been quality. offered within the framework of specified state identification procedures, verifying their usefulness 3. TECHNICAL SYSTEM STATE ANALYSIS in partial researches of the critical machines dynamic state. The technical systems state analysis is compound Connection of the identification of the dynamic for a set of mathematical procedures that they can be condition of machine with the technical diagnostics related to each other to develop analysis of superior gives many new problems and shows on roads order and to find relationships between procedures developmental fields. and states in different systems. The procedures can be classified according to the analysis stage in that 2. DYNAMIC STATE IDENTIFI-CATION they are executed: pre-processing, processing, post- METHODS processing. There are many relationships that can be possible Researches of dynamic characteristics and loads with the procedures, in this work is proposed only of machines are performed directly on objects and a few relations (see fig. 1), it is possible to formulate with the use their physical and mathematical models. other relations of procedures to do another kind of Direct object examinations are very costly and time- methodologies or analysis. consuming: they require a technically apt object and PRE-PROCESSING they often lead to its damage or destruction. In order Import *.UNV format to avoid many of these difficulties, in the process of The Universal file format (.UNV) is a set of creating new constructions or modernizing existing ASCII file formats widely used to exchange analysis ones, more commonly and widely simulation and test data. These files are text plane archives of methods and model examining techniques are data set from experimental tests and they are implemented in place of researches on objects. available to the public domain. The data set can Dynamic is a science on how things change in come from data acquisition systems with different time, and forces which are the cause of these sources as acceleration, displacements, temperature, changes [14, 15, 18]. The aim of the system dynamic noise, etc. the data are in the dynamic time domain, study is understanding the rules of functioning, the t. Starting from .UNV files, the Measures Matrix, changes of dynamic loads state, and foreseeing the Mmtx, could proper behavior of the system. The necessity to know the system’s dynamics stems from growing requirements set for machines. Along with the increase of their motion velocity and load values, the increase of requirements concerning their durability and reliability, as well as the necessity of automatic steering, the importance of the construction dynamics analysis increases. DIAGNOSTYKA’ 3(47)/2008 103 ĩÓàTOWSKI, Technical Diagnostics Of Folded Objects. Directions Of Development

External be made with the q observation vectors; the Data acquisition devices DIFA system whit *.xls observation vectors have p observations, that means: format Measures  

ªºm11 m 12 m 1q o Observation 1 Pre-processing «» m21 m 22 m 2q o Observation 2 To read *.UNV To build the Mmtx «» plane text format symptoms matrix «»  exporting the from the signals «» m m m Observation measures to *.xls measurements ¬¼«»pp12 pqo p

Processin To build symptom matrix Input/Output SVD procedure The symptoms result from a Measures Matrix, Join relationship Mmtx, and are separated in two classes according to functions g Join Optimum procedure the related between the signals: single signals analysis, related signal analysis. Neural Network Spline procedure These distinctions are due to the mathematical differences procedures, and generate a classification of symptoms (see Table 1). Post-processing MAC procedure

Fig. 1. Flow diagram of the technical systems state analysis

Table 1. Classification of symptoms Indicators Single symptoms Indicators Related symptoms Average RMS time domain Time in highest correlation RMS frequency domain RMS power Pick Kurtosis Amplitude Frequency in highest correlation Skewnees Standard deviation Cross-correlation Shape factor

Crest factor Amplitude in highest correlation Impulse factor Looseness factor Velocity of rice frequency Frequency Displacement of Rice frequency Covariance in highest correlation Harmonic index Correlation time Correlation frequency Auto-correlation Correlation Amplitude Frequency Correlation covariance Noise level [%] Distribution < 2(std) Probability Density [0,2std] Coherence Frequency of maximum coherence Spectrum (Fast Fourier Transformation)

Frequency of Power spectrum High order Bi-coherence Tri-spectrum Amplitude of maximum coherence Wigner Ville Wavelet 104 DIAGNOSTYKA’ 3(47)/2008 ĩÓàTOWSKI, Technical Diagnostics Of Folded Objects. Directions Of Development

The symptom matrix is develop starting from the Smtx choose symptoms Smtx { S , S ,..., S ,..., S } , where 12 j n Center & Normalized Ss represents the j-th symptom vector in the life relative to the initial value jj1 Sj S j of Symptom Vector s1j time of the system T , that means: Smtx1 S j Singular Value p T Decomposition Smtx1 U˜/˜ V sssState ªº11 1jn 1 o T1 Singular value / OO O «» 12 n    &&& «» Singular vectors V ^`III1,...,jn ,..., Smtx «»sssiijin1  oState Ti «»   First generalized & T «» SG OI˜¦˜ U «»sss damage 11 11 ¬¼mmjmn1  oState Tm For r = 1tomdo

PROCESSING T Smtx1 TT1 ,...,rrrr U1...˜/˜ 1... V 1...

Evol1(r) O1 SVD procedure Evolution of DS(r) ¦Oj Given a set of n data, damage AF(r) Smtx1TT ,..., , where symptom 1 r Smtx ^ S12, S ,..., Sjnn ,..., S 1 , S ` DA(r) 3Oj n vector Sj  a multidimensional space, and contain m stages in the time domain T . The approach is focused to realize a linear analysis of the Sort Smtx1 relative to Yes symptomatic space. A singular value and states values T of the Sort Smtx1(Sk) Symptom Vector Sk Ss s s corresponding singular vectors of the matrix Smtx kk 12 k mk are a scalar V and a pair of vectors u and v that No satisfy Smtx˜ v V ˜ u , End SmtxT ˜ u V ˜ v . (1) Fig. 2. Flow diagram of SVD With the singular values on the diagonal of the OPTIMUM establish the variation coefficient as matrix / and the corresponding singular vectors forming the columns of two orthogonal matrices U the parameter f 1j , and V, it is obtain: V j , (3) , and f 1j Smtx˜ V U ˜/ S T j Smtx˜ U V ˜/, (2) where V is the standard deviation and S is the Since U and V are orthogonal, this becomes the j j singular value decomposition. average, of the j-th symptomatic vector.

The SVD of an m-by-n Smtx matrix involves an The parameter f 2 j can be calculate to three m-by-mU, and an n-by-nV. In other words, U and V different ways: are both square. If Smtx is square, symmetric, and - correlation coefficient, positive definite, then its eigenvalues and singular - sensibility of symptoms, value decompositions are the same. But, as Smtx - the Modal Assurance Criterion (MAC). departs from symmetry and positive definiteness, the A time vector is created to computer the f 2 in difference between the two decompositions j increases. the correlation coefficient. The time frame 4 The singular value decomposition is the represents the existent connectivity between the appropriate tool for analyzing a mapping from one states T . 4 is vector space into another vector space, possibly with a linearly spaced and increase vector, it is mean a different dimension. The procedure calculates the a degradation linearly progressive between the First Generalized Damage, GS1, and Evolution of states is assumed; when is Damage (see Fig. 2). ^TTT1,...,jm ,..., ` T1 OPTIMUM method the first stage of fault, T is the second stage of fault Given a set of n data, 2 and consecutively. Smtx S, S ,..., S ,..., S , S , where symptom ^ 12jnn 1 ` Then the correlation coefficient is defined as: vector S  n , a multidimensional space, and j 1 CS 4, j , (4) f 2 contain m stages in the time domain T , time of j CCSS 44,, ˜ j j system life. The approach is focused to realize an analysis of the symptomatic space with statistical C is the variance of the vector elements and its criteria. expression is: DIAGNOSTYKA’ 3(47)/2008 105 ĩÓàTOWSKI, Technical Diagnostics Of Folded Objects. Directions Of Development

ªº, (5) Smtx CS4, jjS E 4PP4 S czas ¬¼ j where is the mathematical expectation and V j E f 1 j S P ES . Variation j S j j coefficient * f 1j f 1 j The calculation of the sensibility of symptoms 1 max f j has similar expression to the variation coefficient in the parameter f 1j , therefore exist a relationship Correlation 1 CS 4, j Correlation Yes 2 strongly linear. The expression of sensibility of coefficient f j coeff. CCSS44,, ˜ symptoms is jj No SSjjmax min . (6) f 2 j S Sensibility Yes 2 SSjjmax min j Sensibility of f 2 sympt. j S MAC procedure uses the time frame 4 defined symptoms j at correlation coefficient to realize the calculus, then No T 2 MAC assumes equal considerations in the stages 3 4˜S j MAC f 2 j SSTT characteristics T . MAC expression has the form jj˜4˜4 T S2 4˜j . (7) f 2 j TT * 4˜4˜ SSj ˜j f 2 * j f 2 j 2 To carry out the maximization of the max f j parameters fi , a normalization of each element of them relative to the maximum value is executed: **22 Lf1112 f jj   j fi * j ; where i = 1, 2. (8) fi 1 Lj j w max fi j 1 L j ¦ j * fi represents the statistical behavior of the each symptom performance, which later on will permit to Sort L LL  L mark the coordinates of ideal point. 12 n It is possible to get an index to quantify the End relationship between each symptom performance Fig. 3. Flow diagram of OPTIMUM * fi and the ideal performance. This relationship is established through trigonometric focus, the calculus INPUT/OUTPUT RELATIONSHIP of the existing norm between the statistical FUNCTIONS Lj ** symptoms performance points 12 to xf j , yf j Spectrum diagram the ideal point xy 1, 1 The spectrum diagram is the Discrete Fourier Transform (DFT) of the symptomatic vector , 22 S j ** Lf 1112  f. (9) j jj computed with a Fast Fourier Transform (FFT) algorithm: Then, L are converted weight expressions w i j N (coefficients) to have better interpretation of results: FFT() k S (1)(1)ik  , (11) ¦ jNZ i 1 1 L j ; (10) w (2 S i ) j n N 1 L where, ZN e is a N root of unity. ¦ j j 1 it is requisite w 1. It is compute with a Hanning window: ¦ j The global algorithm OPTIMUM is described in §·§·k , (12) YS>@k 10.51cos2 ¨¸  ¨¸ the Fig. 3. ©¹©¹n 1 and kn 0, ..., 1. Transfer function Given input signal vector x and output signal vector y of de measures matrix Mmtx, the relationship between the input x and output y is modeled by the linear and time-invariant transfer function . The transfer function is the quotient TFxy () f of the cross power spectral density of x and y, Pfxy ()

and the power spectral density Pfxx () of x: 106 DIAGNOSTYKA’ 3(47)/2008 ĩÓàTOWSKI, Technical Diagnostics Of Folded Objects. Directions Of Development

Pf() Polynomials are the approximating functions of TF() f xy . (13) xy Pf() choice when a smooth function is to be xx approximated locally. For example, the truncated uses a periodic Hamming window and TFxy Taylor series: n estimates the transfer function at positive i x aDfaii ! (15) frequencies only; in this case, the output is ¦  TFxy i 0 a function whit (p+1)/2, p is the quantity of provides a satisfactory approximation for fx if f measures. If x or y is vector whit complex elements, is sufficiently smooth and x is sufficiently close to TF estimates the transfer function for both positive xy a . But if a function is to be approximated on and negative frequencies. a larger interval, the degree, n, of the approximating The cross power spectral density of the Pxy polynomial may have to be chosen unacceptably discrete-time signals x and y using the Welch's large. The alternative is to subdivide the interval averaged, modified histograme method of spectral >a, ..., b@ of approximation into sufficiently small estimation. The cross power spectral density is the intervals ªº[[,..., , with: distribution of power per unit frequency whit 50% ¬¼jj1 overlap. ab [ jj  [ 1 , (16) Coherence so that, on each such interval, a polynomial pj of Cohe() f finds the magnitude squared coherence xy relatively low degree can provide a good estimate of the input signals x and y using Welch's approximation to f. This can even be done in such averaged, modified periodogram method. The a way that the polynomial pieces blend smoothly. magnitude squared coherence estimate is a function Any such smooth piecewise polynomial function is of frequency domain with values a range of values called [0,1] that indicates how well x corresponds to y at a spline. I. J. Schoenberg coined this term since each frequency. The coherence is a function of the a twice continuously differentiable cubic spline with power spectral density , of x and y and the Pxx Pyy sufficiently small first derivative approximates. Spline interpolation cross power spectral density Pxy , 2 From the values of the underlying measure Pf xy () , (14) vector at the points in the time vector Cohe() f M j xy PfPf()˜ () xx yy time ,..., ,..., uses a cubic spline is calculated whit a periodic Hamming ^TTT1 jm` Cohexy () f interpolation to find * , window of length to obtain eight equal sections of x Mmjj 1 ,..., m ijrj ,..., m and y, and the value to obtain and 50% overlap. with r 100. It means, to involve the construction Cross correlation and subsequent use of a piecewise-polynomial It is applied the cross correlation function XCF approximation. between two unvaried and stochastic time series. In the observation j-th, , the ordered The sample cross correlation function between x and 4 T j y vectors is a vector of length 2(nLags)+1, which pair is and the spline interpolation M jj,T corresponds to lags 0,rr 1, 2,...,nlags . The ^ ` calculate the piecewise-polynomial function f that central element of XCF contains the 0-th lag cross satisfies correlation. fMallj, . (17) Optimization T jj The optimization procedure is realized following There are two commonly used ways to represent three basics steps: a polynomial spline, the PP-form and the B-form. - fitting the points to a polynomial expression, A spline in PP-form is often referred to as a - calculate the differential function of the piecewise polynomial, while a piecewise polynomial polynomial expression, in B-form is often referred to as a spline. This - to find the roots of the differential function. reflects the fact that piecewise polynomials and Next, it will be develop each of the steps. (polynomial) splines are just two different views of Fitting the points to a polynomial expression the same thing. Given a set of q data, The PP-form of a polynomial spline of order , where the k provides a description in terms of its breaks Mmtx ^ M12, M ,..., Mjqq ,..., M 1 , M ` measure vector have a discrete [ j , ...,[ j 1 and the local polynomial coefficients Mmjj 1 ,..., m ijpj ,..., m number of observations p, the task is calculate a new c ji of its l pieces. set of observations * , which k 1 Mmjj 1 ,..., m ijrj ,..., m k Mc* TT[  , (18) , and * represent a polynomial function. ¦ j ji rp!! M j i 1 where j 1,2, ...,l . DIAGNOSTYKA’ 3(47)/2008 107 ĩÓàTOWSKI, Technical Diagnostics Of Folded Objects. Directions Of Development

A cubic spline is of order four corresponds to the Mmmm* ªº,...., ,..., ' jjijrj ¬¼1 fact that it requires four coefficients to specify a czas >@TTT0,....,ir ,..., cubic polynomial. i 1 Lagrange Interpolation c 1 gmm ˜ Calculate the Lagrange polynomial ij i1 j interpolation of p-1 order from the measure vector Mm,..., m ,..., m jj 1 ijpj No p (TTTT )( )...( TT  )( TT  )...( TT  ) g d 0 Mm*() 01ii 1 1 p jijT ¦ ˜ i 1 (TTTTi01 )( i )...( TT ii  1 )( TT ii  1 )...( TT ip  ) Yes (19) Approximation TT Root ii1 The approximation finds the coefficients of cj 2 mm a polynomial curve fitting of 2, 3, 4 and 5 Amp ij i1 j p T cj 2 order that fits the data M j , in a least squares sense.

The polynomial has the expression cc 1 ii 1 nn1 . (20) p TTT pp12  ppnn T 1 Differential function of the polynomial No expressions i > r-1 A procedure of numeric differentiation of the fitting function is realized Yes ** * dMji TT M ji 1  M ji T , (21) End d TTTii1  Fig. 4. Flow diagram of Incremental Quest this procedure is a very simpler way to calculate the differential function, besides the computational There, the network is adjusted, based on calculus is really fast, and it is important when there a comparison of the output and the target, until the are a lot of data. The exactitude is reasonable due to network output matches the target. Typically many the high density of data; the function such input/target pairs are used, in this supervised * result from the fitting learning, to train a network. Mmjj 1 ,..., m ijrj ,..., m Target process, then the number of observations is assured, r 100 . Neural Network Differential functions roots Including connections Compare To determinate the local maximum and Input (called weights) Output minimum points of the function * is necessary between neurons M j solve the expression: Adjust * weigthts dMj T 0 . (22) dT Fig. 5. Neural arquitecture The Incremental Quest method is described in Fig. 4. Batch training of a network proceeds by making Neural Network weight and bias changes based on an entire set Neural networks are composed of simple (batch) of input vectors. Incremental training elements operating in parallel. These elements are changes the weights and biases of a network as inspired by biological nervous systems. As in nature, needed after presentation of each individual input the network function is determined largely by the vector. Incremental training is sometimes refered to connections between elements. We can train a neural as adaptive training. network to perform a particular function by Each element of the input vector X=[x1,x2,…, adjusting the values of the connections (weights) xp] is connected to each neuron input (see Fig. 6). between elements. Commonly neural networks are The i-th neuron has a summer that gathers its adjusted, or trained, so that a particular input leads to weighted inputs and bias to form its own scalar n a specific target output (see Fig. 5). output j. The various ni taken together form an R-element net input vector n. Finally, the neuron layer outputs form a column vector a. It is common for the number of elements in the input vector to a layer to be different from the number of neurons in the layer ( pRz ). A layer is not constrained to have the number of its inputs 108 DIAGNOSTYKA’ 3(47)/2008 ĩÓàTOWSKI, Technical Diagnostics Of Folded Objects. Directions Of Development equal to the number of its neurons. The developed - Finite element models are used for test/analysis neural network has two layers: correlation. In most industrial applications (see - in the first layer, the number of neurons is R=2p, Fig. c), test and FEM nodes are not coincident so - in the second one, R=1. that special care must be taken when predicting layer 1 layer 2 layer l       FEM motion at test nodes/sensors (shape 1 1 2 2 l l n1 a n1 a n1 a observation) or estimating test motion at FEM ¦ f ¦ f ¦ f 1 1  1 DOFs (shape expansion). x 1 2 l 1 b b b 1 1 1 1 1 1 1 1 2 2 l l n2 a n1 a n1 a ¦ f1 ¦ f1  ¦ f1 x l 2 b 1 b 2 b 1 2 1 2 1 2        l xp 1 1 2 2 l nR a nR a nR a ¦ f1 ¦ f1  ¦ f1 b 1 b 2 b l 1 R 1 R 1 R Fig. 6. Arquitecture of neural network The sum of the weighted inputs and the bias forms the input to the transfer function f.

The training process requires a set of examples of proper network behavior - network inputs x and a. Input/Output b. Wire frame c. target outputs y. During training the weights and Finite elements biases of the network are iteratively adjusted to Fig. 7. Representations of the measured motion minimize the network performance function. The performance function used for feedforward networks Correlation criteria seek to analyze the similarity is the Sum Square Error (SSE), it means, sum and differences between two sets of results. Usual squared error between the network outputs, a, and applications are the correlation of test and analysis the target results and the comparison of various analysis outputs t. results. Ideally, correlation criteria should quantify In the present work, the batch steepest descent the ability of two models to make the same training function is used. The weights and biases are predictions. updated in the direction of the negative gradient of Since, the predictions of interest for a particular the performance function. There is only one training model can rarely be pinpointed precisely, one has to function associated with a given network. use general qualities and select, from a list of The larger the learning rate, the bigger the step. possible criterion, the ones that can be computed and If the learning rate is made too large, the algorithm do a good enough job for the intended purpose. becomes unstable. If the learning rate is set too The Modal Assurance Criterion (MAC) and small, the algorithm takes a long time to converge. Pseudo Orthogonally Checks (POC) are very POST-PROCESSING popular and useful criteria. Other criteria should be MAC procedure used to get more insight when you don’t have the Modal testing differs from system identification desired answer or to make sure that your answer is in the fact that responses are measured at a number really foolproof. The following table gives a list of of sensors which have a spatial distribution which criteria. allows the visualization of the measured motion. Shape correlation tools can also be used to Visualization is key for a proper assessment of the compare frequency responses. Thus the MAC quality of an experimental result. One typically applied to FRFs is sometimes called FRAC. considers three levels of models: MAC is the most widely used criterion for - Input/output models are defined at sensors. vectors correlation (mainly because of its In the fig. 7a, one represents these sensors as simplicity). The MAC is the correlation coefficient arrows corresponding to the line of sight of vector pairs in two vectors sets, RM and EM. RM measurements of a laser vibrometer. Input/output is the reference matrix or the observation of models are the direct result of the identification analytical analysis and it is defined as procedure. RM R,.., R ,..., R where - Wire frame models are used to visualize test ^ 1 j n ` results. They are an essential verification tool for T , and EM is the estimate Rrjjijmn ªº1 ,..., rr ,..., the experimentalist. Designing a test well, includes ¬¼ matrix or experimental analysis and it is defined as making sure that the wire frame representation is T EM E,.., E ,..., E where Eeeeªº,..., ,..., , sufficiently detailed to give the experimentalist ^ 1 jn` jjijmn ¬¼1 a good understanding of the measured motion (see then MAC is given by: fig. 7b). With non-triaxial measurements, a significant difficulty is to handle the perception of motion assumed to be zero. DIAGNOSTYKA’ 3(47)/2008 109 ĩÓàTOWSKI, Technical Diagnostics Of Folded Objects. Directions Of Development

T 2 limits. The MAC compute vector pairs for all ^`REij^` (23) MAC vectors ij T T ^`^`EERRj ji^`^` i M AC analysis Relationship Table 2. Different correlation criteria High (MAC > 0.9) Y1 Low (0.9 < M AC < 0.05 Criterion Description Advantage Limitation No (MAC < 0.05) Modal Y2 Assurance Criterion. The It can be e most popular used in all Y3 c en criterion for cases. A er ef correlating MAC It can give very Y4 R vectors. criterion misleading MAC Insensitive to applied to results without Y5 vector scaling. frequency warning. Sensitive to responses is Y1 Y2 Y3 Y4 Y5 sensor selection called Est i m at i o n and level of FRAC. a. Two dimension representation response at each MAC analysis sensor. Relationship Pseudo High (MAC > 0.9) Orthogonally 1 Low (0.9 < MAC < 0.05) Checks. No (MAC < 0.05) Required in It gives a It requires the some industries 0.5 much more definition of a for model reliable mass validation. This POC indication of associated with criterion is only 0 correlation the known defined for than the modeshape modes since MAC. components. Y1 other shapes do verify Y2 orthogonality Y3 conditions. Y5 Modeshape Y4 Y4 Y3 pairing (based Y5 Y2 on the MAC) The same of The same of Ref er en ce Y1 Error and relative MAC. MAC. Estimation frequency error and MAC b. Three dimension representation correlation. Fig. 8. MAC diagram Relative error. It does not tell Insensitive to Extremely much when Rel scale when using accurate in the two sets, it is means, one way to represent the correlation the modal scale criterion. analysis is, each vector from data set RM is poor. factor. compared with each vector from data set EM. Coordinate A very fast The product of a couple vectors is a scalar Modal tool giving quantity. If and are identical the numerator Assurance It does not R j E j more insight Criteria compare systematically COMAC into the and denominator are equal, giving a MAC value of sets of vectors to give good reasons of analyze which indications. 1. If the two vectors are orthogonal to one another poor sensors lead the numerator is 0 and hence the MAC value is zero. correlation. poor correlation. In the Fig. is shown a two identical set of vectors What if analysis, non-orthogonals. where It is more coordinates are The MAC plot is an n x n grid (n is the number MACCO precise than It is slower. sequentially of vectors being compared) enabling comparison of COMAC. eliminated from all vectors from data set RM with all modes from the MAC. data set EM. The leading diagonal represents the correlation between the vectors of the same number For two vectors that are proportional MACij 1 from data sets RM and EM. If the ordering of the vectors in each set is the same then the correlation (perfect correlation). If MAC t 0.9 is generally ij here should be high. considered as much correlated. If 0.9 dMACij 0.6 should be considered with much caution (they may or may not indicate correlation). Next is showed - fig. 8 - the 2D and 3D representation of MAC. The height of the patches associated to each vector pair are proportional to MAC value, the colour is relative to high and low 110 DIAGNOSTYKA’ 3(47)/2008 ĩÓàTOWSKI, Technical Diagnostics Of Folded Objects. Directions Of Development

MAC analysis Relationship construction. The proposed methods of PCA, SVD, High (MAC > 0.9) OPTIMUM permits on choice of the best measures Low (0.9 < MAC < 0.05) by defined the criterions of quality usefully. 1 No (MAC < 0.05) LITERATURE 0.5 1. AmeliaĔczyk A.: Optimization multicriterion in 0 problems steering and management. Ossolineum, Wrocáaw 1984. Y1 2. Bendat J. S., Piersol A. G.: Method of analysis and measurement of random signals. PWN, Y2 Warszawa 1996. Y3 3. Bishop R. D., Gladwell G. M.: Matrice analysis Y5 Y4 Y4 of tremblings. PWN, Warszawa 1972. Y3 Y5 Y2 4. Broch J. T.: Mechanical Vibration and Shock Ref er en ce Y1 Measurements. Brüel & Kjaer, 1980. Estimation 5. Cempel C.: Practical of vibroacoustic. PWN, Fig. 9. MAC with identical matrix Warszawa 1989. 6. Eykhoff P.: Identification in dynamic 5. SUMMARY arrangements. BNInĪ., Warszawa 1980. 7. Giergiel J., Uhl T.: Identification of mechanical Initial qualities in case of machines are usually arrangements. PWN, Warszawa 1990. geometric features (e.g. clearance, element 8. Giergiel J.: Mechanical trembling. AGH, permanent set) and material characteristics (e.g. Kraków 2000. immediate resistance, fatigue strength) of their 9. Kaczmarek J.: Basis of theory of trembling and elements. The set of all the state features: geometric dynamics of machines. WSM, Szczecin 1993. features, material and machine’s functional 10. Moczulski W., Ciupke K. (Eds.): Knowledge parameters, necessary from the point of view acquisition for hybrid system sof risk assessment reliability, can be treated as and critical machinery diagnosis. Part V. ITE a multidimensional random process because many of Radom 2008 pp.295-356. the features change randomly in the machine’s 11. Morrison F.: Art of modeling of dynamic exploitation course. arrangements. WNT, Warszawa 1996. The course of changes in the technical state due 12. Paczkowski W. M.: Chosen problems of discreet to external reactions, depends not only on the level evolutionary optimization. Szczecin 1999. of those reactions but also on the above mentioned 13. Tylicki H.: Optimization of process of features at the initial moment t = 0, i.e. on the initial forecasting of technical state of motor vehicles. technical state. The object’s technical state at the ATR Bydgoszcz 1999. moment t depends, therefore, on time which elapsed 14. Uhl T., Lisowski W.: Practical problems of from the beginning of the exploitation, on the course analysis of modal construction. CCATIE, of external reactions within the whole time range Kraków 1996. 15. Uhl T.: By computer helped identification of from t0 to t, as well as on the initial technical state. The values of the accepted amplitude features are models of mechanical constructions. WNT, directly dependent on the machine’s technical state. Warszawa 1997. Amplitude features are qualities directly connected 16. ĩóátowski B., NiziĔski S.: Modelling process of with the state, useful for an easy theoretical mapping exploitation machines. Bydgoszcz 2002. of appearing (in time) changes of the object’s proper 17. ĩóátowski B.: Investigation of dynamic machine. functioning abilities. ISBN – 83-916198-3-4, Bydgoszcz 2002. Tools for the state evaluation are methods of 18. ĩóátowski B., Cempel C. (red.): Diagnostics dynamic state research, supported by modern IT engineering of machines. ITE, Radom 2004. technologies to which this work is dedicated. Technical reality is the result of the models analysis which describe it more or less properly. The process whose aim is to build the best operational model (mathematical or empirical) is called the identification process. It is composed of the problems: modeling, experiment, estimation and model verification. The problems of optimization are important question. This area of investigations, in this the reduction of information quantity and search the best symptoms of destruction state elements of DIAGNOSTYKA’ 3(47)/2008 111 WILK, àAZARZ, MADEJ, Gear Fault Detection Using Vibration Analysis

GEAR FAULT DETECTION USING VIBRATION ANALYSIS

Andrzej WILK, Bogusáaw àAZARZ, Henryk MADEJ

Faculty of Transport, Silesian University of Technology 40-019 Katowice, ul. KrasiĔskiego 8, fax (+48) 32 603 41 08, email: [email protected]

Summary The article includes results of the team’s research work on vibroacoustic diagnostic of gearbox components’ faults. A review of simulation and experimental researches that aimed at elaboration of methods which would enable early identification of teeth faults in the form of working surface pitting, spalling of tooth crest, crack at the tooth bottom as well as partial breaking of a tooth, is presented. Assessment of selected methods of processing the vibroacoustic signals during the detection of gear faults has been carried out while faults occur in gear bearings working under various conditions. The initially processed vibration signals analyzed within time and frequency domains constituted a basis for preparation of detection measures that were sensitive to early stages of damage. The measures obtained as a result of simulation and experimental tests were used to construct a set of neuron classifier models to diagnose the type and degree of toothed wheels faults with a validation error below 5%. The achieved qualitative and quantitative conformity of simulation and experimental research results has shown that application of an expanded and identified dynamic model of the gear in a power transmission system enables the acquisition of reliable diagnostic relations.

Keywords: gearbox, vibration, gear fault, diagnostics.

WYKRYWANIE USZKODZEē PRZEKàADNI NA PODSTAWIE ANALIZY DRGAē

Streszczenie W artykule zawarto wyniki prac zespoáu w zakresie diagnostyki wibroakustycznej uszkodzeĔ elementów przekáadni zĊbatych. Przedstawiono przegląd badaĔ symulacyjnych i doĞwiadczalnych, których celem byáo opracowanie metod pozwalających na wczesną identyfikacjĊ uszkodzeĔ zĊbów w postaci pittingu powierzchni roboczych, wykruszenia wierzchoáka, pĊkniĊcia u podstawy zĊba oraz czĊĞciowego wyáamania zĊba. Dokonano oceny efektywnoĞci wybranych metod przetwarzania sygnaáów wibroakustycznych w procesie wykrywania uszkodzeĔ kóá zĊbatych przy jednoczesnym wystĊpowaniu uszkodzeĔáoĪyskowania przekáadni pracujących w róĪnych warunkach. WstĊpnie przetworzone sygnaáy drganiowe analizowane w dziedzinie czasu i czĊstotliwoĞci stanowiáy podstawĊ do opracowania miar diagnostycznych wraĪliwych na wczesne stadia uszkodzeĔ. Miary otrzymane w wyniku symulacji oraz badan doĞwiadczalnych wykorzystano do budowy zestawu wzorców klasyfikatora neuronowego diagnozującego rodzaj i stopieĔ uszkodzenia kóá przekáadni z báĊdem walidacji poniĪej 5%. Uzyskana zgodnoĞü jakoĞciowa i iloĞciowa wyników badaĔ symulacyjnych i doĞwiadczalnych wykazaáa, Īe wykorzystanie rozbudowanego i zidentyfikowanego modelu dynamicznego przekáadni w ukáadzie napĊdowym umoĪliwia pozyskanie wiarygodnych relacji diagnostycznych.

Sáowa kluczowe: przekáadnie zĊbate, drgania, uszkodzenia kóá, diagnostyka.

1. INTRODUCTION their components in the initial phase of fault occurrence. Toothed gears are designed for cooperation with One of the most frequently applied methods is sources of propulsion of higher and higher power measurement of the vibroacoustic signal and on this and are exposed to high external dynamic loads. In basis, determination of measures sensitive to the design process, designers are trying to achieve different types of damage. the highest possible ratio of power transmitted The rate of propagation of vibroacoustic through wheels to their weight. A gear working disturbance caused by a changed condition of a gear under high load should be either sporadically or makes the vibroacoustic methods particularly useful constantly monitored to ensure safe operation. The in diagnosing early stages of faults. techniques of diagnosing the technical condition of Recently, techniques of contactless measurement gears are oriented towards identification of faults of of vibration have developed considerably. They enable measuring the vibration speed of rotating 112 DIAGNOSTYKA’ 3(47)/2008 WILK, àAZARZ, MADEJ, Gear Fault Detection Using Vibration Analysis bodies. Measurements of the vibration speed of residual signals are obtained, as well as a signal rotating shafts make it possible to eliminate the containing only bands of meshing frequency and its consequences of complex and variable in time harmonics. On the basis of the first two signals, transmittance of the bearing/gearbox system, which numerical estimators of amplitude and allows obtaining effective symptoms of faults. dimensionless discriminants are calculated. Analysis An essential issue in the diagnosing of gearboxes methods are used in time domain, frequency domain, is the ability to differentiate between various or in time and frequency domain, as well as phenomena influencing the vibroacoustic signal statistical moments of higher orders. Those connected with both, normal operation of the dimensionless discriminants which are based on gearbox and development of faults in its statistical moments of higher orders (FM4, M6A, components. M8A, NA4 ...) are most often determined using Toothed wheels and bearings are the gearbox differential and residual signals [1, 4, 10, 12]. components most susceptible to damage. The In case of simultaneous occurrence of faults in modern diagnosing methods of gearboxes are wheels and bearings, it is justified to apply comb oriented to the detection of early phases of fault filtration, thus enabling separation of vibration occurrence, e.g. spalling of tooth crest, crack at the signals generated by different faults [9]. tooth bottom, fatigue chipping of the upper layer, or Input signal (vibration) galling of the interacting surfaces. In the diagnosing of rolling bearings, detection of initial stages of Signal tacho Preprocessing damage to the bearing race or rolling elements is extremely important. Time synchronous SIGNAL PROCESSING averaging AND FEATURE EXTRAC TION The development of computer hardware and x statistical damage metrics signal processing methods enables using advanced xRMS , x, xˆ,kurtosis,SF, ER,... , signal analysis methods in the time-frequency plane. x FFT, short time Fourier transform SFFT, x envelope detection, The methods allow observation of non-stationary x demodulation, impulse disturbance induced by faults in their initial x cepstrum, x wavelet transform (DWT, CWT, WPT), x Wigner - Ville transform, stages. FFT x bispectrum, trispectrum, ... , Experimental research on gearboxes is difficult x correlation, coherence. to carry out, as well as costly and time-consuming, and in the case of gears produced as single items, Comb filtering most often impossible. In such cases, it is justified to use an identified dynamic model of a gear in a power transmission system [5]. Such model will Bandpass filter allow making a series of numerical experiments and Signal spectrum i*f ± ǻf analysis of the simulation results will enable z expanding the diagnostic knowledge and obtaining Differential Residual higher certainty of the diagnosis. spectrum spectrum IFFT For the monitoring of the condition of many IFFT IFFT power transmission units, expert systems are Signal containing only Differential Residual the mesh frequencies created, which use artificial intelligence signal signal and their harmonics components. A properly constructed and taught system can automatically recognize the existing faults. Neuron networks in the process of learning FM4, M6A, M8A, … NA4 (NA*4), ... NB4 FM0 acquire the ability of generalizing knowledge, which allows detection of faults in their early phases, often not noticed during diagnosing. A basic problem Fig. 1. Methods of vibroacoustic signal processing. while constructing such systems is to define a set of input data and acquire an appropriately large set of 3. MODEL OF TOOTHED GEAR WORKING training data [1]. IN A POWER TRANSMISSION SYSTEM

2. SIGNAL ANALYSIS METHODS In the simulation tests, a dynamic model was IN THE DIAGNOSING OF GEARBOXES used representing a toothed gear working in a power transmission system (Fig.2). The model was created In vibroacoustic diagnostic of gearboxes, in the Matlab–Simulink environment. It takes into a number of different signal analysis methods are account the characteristics of an electric driving used [2, 6, 7, 13, 14]. Figure 1 presents a general motor, single-stage gear, clutches and working classification of the existing signal processing machine. methods. The basis consists of a properly selected vibroacoustic signal (WA) which, in order to eliminate incidental disturbance, can be synchronously averaged and from which, by applying appropriate filtration, a differential and DIAGNOSTYKA’ 3(47)/2008 113 WILK, àAZARZ, MADEJ, Gear Fault Detection Using Vibration Analysis

research that the use of a contactless laser measurement of transverse vibration speed of rotating gear shafts, combined with advanced methods of signal processing, such as Wigner-Ville distribution or continuous wavelet transform, enables detecting such fault in its initial stage. This method of measurement eliminates the influence of complex transmittance of the bearing - gear casing system [9]. Fig. 3 shows the results of time and frequency analysis WV of differential signal. In the WV distribution, an increase of amplitude occurs within the pinion turn angle corresponding to Fig. 2. Dynamic model of toothed gear in a power the cooperation of the damaged tooth. transmission system.

The simulation model allowed taking account, in the calculations, of cyclic and random deviations which occurred in the mesh [3, 5, 11]. The utilization of a dynamic model of gear in a power transmission system was possible owing to very good identification and tuning of the model parameters. It gave very high qualitative and quantitative consistency of simulation results with the results obtained from tests of a real object [1, 5, 6, 9]. The gear model also enabled mapping of local faults consisting of a crack at the tooth bottom or chipping of tooth crest, and faults of rolling Fig. 3. WV Time/frequency distribution of bearings’ components. differential signal – measurement of vibration speed The chipping of tooth crest throughout its length of pinion shaft in the direction of the force acting was modeled as tooth contact section shortened by between the teeth – 1 mm chip of the pinion tooth. a value equal to a predetermined part of pitch. The effect of a changing tooth contact section length on For easier interpretation of the results obtained, the meshing time was taken into account as well. summation was performed of WV distribution Chipping of tooth crest in a pinion results in discrete values (formula 1) in accordance with the a premature finish of operation by a couple of teeth, equation: whereas chipping of the reference cone apex results B in a delayed start of cooperation between the couple S I WV l ,k (1) of teeth. WV ¦ WV WV k A A crack at the tooth bottom is accompanied by WV reduced rigidity of meshing. Therefore, a fault of WV lWV ,kWV WV t, f (2) this sort was mapped as a percentage reduction of where: rigidity of the cooperating couple of teeth in relation lWV, kWV – discrete values of time and frequency, to a couple without faults. respectively, Analysis of the effect of the crack depth in the A, B – discrete values corresponding, respectively, tooth root on a change in mesh rigidity was to limit frequencies of the summation interval fA, fB. presented in monograph [9]. Faults of working surfaces of cooperating In the presented in Fig. 4 sum of WV distribution, elements of rolling bearings were modeled in local maxima coming from the chipping of the tooth a similar way, by reducing the bearing rigidity while crest in the pinion are clearly visible, which the damaged piece of surface was in the load facilitates localization of the fault. transmission zone [9]. The sums SWV I of WV distributions, obtained from measurements (Fig. 4) and simulations (Fig. 5) 4. DETECTION OF TOOTH CREST show high consistency. CHIPPING

Initial phases of tooth crest chipping development in a toothed gear do not significantly influence the general level of vibration. Hence, detection of damages of this type in the early phase is very difficult. It appears from the previous 114 DIAGNOSTYKA’ 3(47)/2008 WILK, àAZARZ, MADEJ, Gear Fault Detection Using Vibration Analysis

5. NEURON CLASSIFIER OF TOOTHED WHEEL FAULT

The results of research connected with the structure of neuron classifiers, which were taught and verified on the basis of data obtained from a simulation model of a toothed gear working in a power transmission system, were presented in monograph [1]. For constructing the models, signals of transverse vibration speed of wheel shaft, analyzed by means of FFT and CWT, were used. Based on Fig. 4. The sum of time/frequency WV distribution in preliminary tests, an artificial neuron network of 0÷4500 [Hz] band, generated from a differential MLP type was chosen as the classifier. Sets of signal of pinion shaft vibration speed measured in models were built on the basis on vibration signals the direction of the force acting between the teeth – of a toothed gear working in the following experimental research result. conditions: - M 138 [Nm], n 900 [r.p.m.], - M 138 [Nm], n 1800 [r.p.m.], - M 206 [Nm], n 900 [r.p.m.], - M 206 [Nm], n 1800 [r.p.m.]. A neuron classifier was built, capable of recognizing the degree of fault in wheel teeth in the form of a crack at the tooth bottom or chipping of tooth crest in a gear working at different shaft speeds and different load torques. It was assumed that the following classes would be recognized: a crack at the tooth bottom in the form of percentage reduction of rigidity of a couple of teeth in case of such fault: - class 1 @ 0 y 9 %, Fig. 5. The sum of time/frequency WV distribution in - class 2 @ 10 y 19 %, 0÷4500 [Hz] band, generated from a differential - class 3 @ 20 y 29 %, signal of pinion shaft vibration speed recorded in the - class 4 @ 30 y 40 %, direction of the force acting between the teeth – chipping of tooth crest, in the form of per cent simulation result. length of pitch, by which the tooth contact section shortens as a result of such fault: Based on the research, it can be affirmed that - class 5 @ 0 y 9 %, processing of the signal of transverse vibration speed - class 6 @ 10 y 19 %, of gear shafts, measured in the direction of the force - class 7 @ 20 y 29 %, acting between the teeth, and the use of analyses in, - class 8 @ 30 y 40 %, simultaneously, time and frequency, or time and The training process and the testing validation scale domains (CWT), facilitate effective detection process are presented in Fig. 6. of chipping of a tooth crest. Using the sums of WV When using both, the models obtained from FFT distribution (Fig. 4, 5) or scalograms [9], measures analysis and CWT analysis, the authors managed to where built enabling the evaluation of the tooth chip build neuron classifiers which can diagnose the type depth. and degree of fault of a gear wheel tooth with Computer simulations of a toothed gear with a validation error below 5%. damaged components, made using its expanded and Irrespective of the model building method, the identified dynamic model, made it possible to verify testing error for data taken from a real gear was ca. the measures of the case of tooth crest chipping 60%. during the operation of gears of different In the successive stage, apart from data taken geometrical parameters of toothed wheels, at from a dynamic model of a gear, part of data coming different rotational speeds, loads or deviations in from tests of a real gear were added to the training wheel workmanship. set. The testing error value obtained in that case was ca. 20% in the case where for the construction, layers of both sigmoidal and tangensoidal hidden neurons were used [1]. DIAGNOSTYKA’ 3(47)/2008 115 WILK, àAZARZ, MADEJ, Gear Fault Detection Using Vibration Analysis

The research has shown that it is possible to in a power transmission system for simulating the build a neuron classifier of two fault types of wheel faults of its components enables the acquisition of teeth in different stages of advancement for a gear credible diagnostic relations. working at different rotational speeds of shaft and The simultaneous application of experimental with different load torques. methods and computer simulations has facilitated the creation of input data to the system diagnosing local damage of wheels, working based on artificial intelligence methods. The research presented in the monograph [1] shows that artificial neuron networks, taught using data obtained from the model and from a real gearbox, offer the highest correctness of classification of the type and degree of fault in gears.

REFERENCES

[1] Czech P., àazarz B., Wojnar G.: Wykrywanie lokalnych uszkodzeĔ zĊbów kóá przekáadni z wykorzystaniem sztucznych sieci neuronowych i algorytmów genetycznych. ITE Katowice - Radom 2007. [2] Dąbrowski Z.: RóĪne metody diagnozowania przekáadni zĊbatych na podstawie pomiaru drgaĔ i haáasu. XXII Ogólnopolskie Sympozjum Diagnostyka Maszyn, WĊgierska Górka marzec 1995 r., s. 35÷44. [3] Dąbrowski Z., Radkowski S., Wilk A.: Dynamika przekáadni zĊbatych – Badania i symulacja w projektowaniu eksploatacyjnie zorientowanym. Wydawnictwo i Zakáad Poligrafii Instytutu Technologii Eksploatacji w Radomiu Warszawa – Katowice – Radom 2000. [4] Lebold M., McClintic K., Campbell R., Byington C., Maynard K.: Review of Vibration Analysis Methods for Gearbox Diagnostics and Prognostics. 54th Meeting of Society for Machinery Failure Prevention Technology, Virginia Beach, VA, May 1-4, 2000, s. 623÷634. [5] àazarz B.: Zidentyfikowany model dynamiczny przekáadni zĊbatej w ukáadzie napĊdowym jako podstawa projektowania. Wydawnictwo i Zakáad Poligrafii Instytutu Technologii Fig. 6. Chart of the adopted methodology of working Eksploatacji w Radomiu, Katowice-Radom with neuron classifiers [1]. 2001. [6] àazarz B., Madej H., Wilk A., Figlus T., 6. CONCLUSIONS Wojnar G.: Diagnozowanie záoĪonych przypadków uszkodzeĔ przekáadni zĊbatych. As results from the research presented in the ITE Katowice - Radom 2006. paper, the methods applied to process the signal of [7] àazarz B., Wojnar G.: Detection of Early transverse vibration speed of shafts, measured in the Stages of Pinion Tooth Chipping in direction of the force acting between the teeth, and Transmission Gear. Machine Dynamics the use of analyses in, simultaneously, time and Problems 2003,vol. 27, No 3, s. 23÷34. frequency or time and scale domains (CWT), [8] àazarz B., Wojnar G.: Model dynamiczny facilitate effective detection of various faults of ukáadu napĊdowego z przekáadnią zĊbatą. XVII toothed wheels. Ogólnopolska Konferencja Przekáadnie ZĊbate, The achieved high qualitative and quantitative WĊgierska Górka 09÷11.10.2000r., s. 101÷108. conformity of simulation and experimental research [9] àazarz B., Wojnar G., Czech P.: Wibrometria corroborates the fact that application of an expanded laserowa i modelowanie  narzĊdzia and identified dynamic model of a gearbox working 116 DIAGNOSTYKA’ 3(47)/2008 WILK, àAZARZ, MADEJ, Gear Fault Detection Using Vibration Analysis

wspóáczesnej diagnostyki przekáadni zĊbatych. Andrzej WILK prof. Ph. D., ITE Katowice - Radom 2007. D.Sc. Eng. Head of Department [10] Madej H.: Wykorzystanie sygnaáu of Automotive Vehicle resztkowego drgaĔ w diagnostyce przekáadni Construction Silesian zĊbatych. Diagnostyka, vol. 26, 2002, s. 46÷52. University of Technology in [11] Müller L.: Przekáadnie zĊbate - dynamika. Katowice. Scientific interest: WNT, Warszawa 1986. diagnostics of powertrain with [12] Stewart R. M.: Some useful data analysis gearboxes, machine design. techniques for gearbox diagnostics. Report Member of Polish Society of MHM/R/10/77. Machine Health Monitoring Technical Diagnostics. Group. Institute of Sound and Vibration Research. University of Southampton 1977. Bogusáaw àAZARZ Ph. D., [13] Wilk A., àazarz B., Madej H.: Metody D. Sc. Eng. professor of The wczesnego wykrywania uszkodzeĔ Silesian University of w przekáadniach zĊbatych. Przegląd Technology. Scientific interest: Mechaniczny, nr 3, 2002, s. 14÷18. modelling of dynamic [14] Wilk A., àazarz B., Madej H., Wojnar G.: processes, diagnostics of tooth Analiza zmian wibroakustycznych symptomów gear, machine design and diagnostycznych w procesach zuĪyciowych kóá vibroacoustic signal proces- zĊbatych. Materiaáy XXX Jubileuszowego sing. Member of Machines Ogólnopolskiego Sympozjum Diagnostyka Construction Committee and Maszyn, WĊgierska Górka 03÷08.03.2003 r. Polish Society of Technical Diagnostics.

Henryk MADEJ, Ph. D., D. Sc. Eng. work as a professor in the Department of Automotive Vehicle Construction Silesian University of Technology in Katowice. He deals with IC engines and powertrains diagnostics with the application of vibroacustical signal analysis, also with minimization of machine vibroactivity. Member of Polish Society of Technical Diagnostics. DIAGNOSTYKA’ 3(47)/2008 117 LIŠýÁK, NAMEŠANSKÝ, Human And Computer Recognition Of Nighttime Pedestrians

HUMAN AND COMPUTER RECOGNITION OF NIGHTTIME PEDESTRIANS

Štefan LIŠýÁK, Pavol NAMEŠANSKÝ

Department of Road and City Transport, Faculty of Operation and Economics of Transport and Communication, University of Žilina, Univerzitná 1, 01026 Žilina, Slovakia E-mail: [email protected], [email protected]

Summary This study investigates recognition and detection abilities of nighttime pedestrians by observers and using a photographic method. The examination of the visibility of nighttime pedestrians is done and afterwards evaluated by two methods. In the first method observers were asked to detect a pedestrian from a slowly moving vehicle. In each attempt, detection distance was recorded. From the place of detection, a digital photography was taken from the still vehicle. Analysis of the digital photography in computer graphic software is the keystone of the second method. The article also shows potential effect of retroreflector positioning on recognition of nighttime pedestrians.

Keywords: pedestrian, recognition, detection, visibility, retroreflector.

1. INTRODUCTION 2. METHOD

Reduced visibility is the major contributor to 2.1. Task pedestrian accidents at night. The visible distance of Observers performed a recognition task while dark-clad pedestrians is typically less than one-third seated in the rear passenger’s seat of a vehicle, with of the stopping distance at normal highway head positioned above and between front seat backs. speed. [1] Statistics show, that number of driven Two vehicles were used, in each case, with low kilometres is minor during nighttime traffic, but beam lamps. Specifically, an observer’s task was to accidents, which happen in dark or in lower say stop right in the moment of recognition of visibility conditions are usually tragic. When scaled a silhouette of a pedestrian. The pedestrian was by the number of miles driven, pedestrian fatality standing on the right side of the road, ahead of the rate is three times higher at night. Part of the reason vehicle. In each attempt, the vehicle was stopped as is a bigger chance of driver drinking and fatigue, but soon as possible, always with the same, specially the critical factor is lower visibility due to reduced trained driver. The speed of the vehicle was being ambient illumination. [3] held approximately on the level of 17 km/h. This study is oriented on investigating and evaluating of visibility of still object – pedestrian. 2.2. Observers Measurements were made during night. In this 35 observers participated in this study. 31 article, there are included: 3 measurements, in observers were aged between 20 and 27 and 4 were 2 different sites, with 2 different vehicles, with aged between 45 and 60. There were 11 females and 2 different groups of observers. Measurements and 25 males in general. All observers were licensed evaluation were made using 2 methods. In the first drivers. method, let us say “human method”, we measured distances, when pedestrian was recognized by 2.3. Sites observer from the moving vehicle. In the second There are three measurements included in this method, let us say “computer method”, digital study from two different sites. The first photographs were evaluated using a computer measurement was done on 19th June 2007 in software. These digital photographs were taken from Dresden, Germany. The second and the third the still vehicle right from the place, where observer measurement was done on 8th December 2007 in had detected a pedestrian. It allowed us to compare Dolný Hriþov, Slovakia. these two methods. We could analyze applicability Since other vehicles, either preceding or and reliability of computer method. Furthermore, in oncoming, would influence the visibility of the 2 measurements there was investigated an effect of pedestrians, the measurements were conducted on retroreflector positioning on recognition of rural roadway section with no traffic (Dresden) and silhouettes of pedestrians. on the landing runaway of the airport Žilina in the time without operation (Dolný Hriþov). 118 DIAGNOSTYKA’ 3(47)/2008 LIŠýÁK, NAMEŠANSKÝ, Human And Computer Recognition Of Nighttime Pedestrians

2.4. Equipment There were two different vehicles with different headlights used in this study. In the first measurement, we used Mercedes Benz S 500 (made in 2007) with bi-xenon factory headlights. In the second and the third measurements, we used Škoda Octavia 1,6 GLX (made in 1999) with factory headlights, but with new bulbs OSRAM Bilux H4 12V, 60/55 W. For measuring distances, we used measuring Fig. 2. Sample photography – measurement 1 tape and white chalk. For taking digital photographs, Measurement 2 parameters: we used digital SLR camera Nikon D70s with lens Date and time: 8.12.2007, 19:40 – 21:00 AF-S DX Zoom-Nikkor 18-70mm f/3.5-4.5G IF Weather cond.: clear-somewhat cloudy, humid, ED. We used tripod. foggy, stronger wind occasionally, 4-6 C 2.5. Procedure q Shutter speed: 10s, Aperture: 5,6, The observers were seated in the middle of rear WB: 4000 K, Colour space: Adobe RGB seat. There were two people in the vehicle during Focal length: 40mm (60mm equal to 35mm film) one measurement – driver and observer. They were told that this study investigated how well drivers can recognize pedestrian at night. Particularly, the observers were instructed to say stop whenever they were sure, they recognized pedestrian as a person (silhouette was relevant). The observers had known, where the pedestrian was standing – on the right side of the road. So, they had expected, where the pedestrian were going to appear. However, they were instructed to direct their gaze primary not on the right side of the road. The pedestrian was Fig. 3. Sample photography – measurement 2 dressed in dark clothes, as shows fig. 1: black shoes, blue/grey denim jeans, dark matte jacket, and black Measurement 3 parameters: cap with white marking. Date and time: 18.12.2007, 19:20 – 21:00 Weather cond.: clear-somewhat cloudy, light wind, 1-2,5qC Shutter speed: 10s, Aperture: 5,6, WB: 4000 K, Colour space: Adobe RGB Focal length: 44mm (66mm equal to 35mm film)

Fig. 1. Photography of the pedestrian After stopping the vehicle, detecting distance was recorded and digital photography was taken. Fig. 3. Sample photography – measurement 2 Camera was mounted on a tripod. The tripod was Computer method mounted and fixed on the right front seat. Conditions and digital photographs parameters: 1. WB calibration and JPG creation. Common parameters: *.NEF file was recorded during the Sensibility ISO: 200, Resolution: 3008 x 2000, measurement. The reason was simple. It allowed us Colour space: Adobe RGB, File: *.NEF, after to make a white balance calibration. All WB calibration converted to *.JPG, photographs, in particular measurement, they were Long time noise reduction was activated. calibrated to the same WB value. NEF file works in Measurement 1 parameters: 12 bit depth for each of RGB channel and also Date and time: 19.6.2007, 23:00 – 2:00 enables specific photography settings: sharpness, Weather cond.: clear sky, light wind,15-17 C q tone compensation, colour mode, saturation. There it Shutter speed: 15s, Aperture: 9, was used the best configuration to distinguish pixels WB: 3800 K, Colour space: Adobe RGB with the similar level of luminosity: sharpness-high, Focal length: 18mm (27mm equal to 35mm film) tone comp.-low contrast, colour mode-Nikon Adobe DIAGNOSTYKA’ 3(47)/2008 119 LIŠýÁK, NAMEŠANSKÝ, Human And Computer Recognition Of Nighttime Pedestrians

RGB, saturation-moderate. After this process, NEF Measurement 1 120 Correlation was converted and saved as JPG with the highest 110 100 -0.595 quality. 90 80 70 2. JPG processing 60 50 Distances For JPG processing, it was used conventional 40 Differences 30 graphic software – Corel Photo Paint, which 20 10 Distances [m], Differences [-] Differences [m], Distances provides useful statistical information about 0 -10 1234567891011121314Observers a bitmap picture. A histogram is very good tool, how -20 to measure luminosity distribution in a picture or in -30 a selected area. The histogram represents a bar graph Measurement 2 70 Correlation of the total number of pixels that appear at different 60 -0.787 levels of luminosity. The horizontal axis represents 50 the luminosity level, while the vertical axis 40 30 Distances represents the number of pixels at each luminosity 20 Differences level found within the current image. The left side of 10 the horizontal axis represents the darkest tones 0 123456789101112Observers within the image, while the right side represents the Distances [m], Differences [-] -10 -20 lightest tones within the image. The histogram -30 provides these statistics: weighted arithmetic mean, Measurement 3 standard deviation, median, range, … JPG file Correlation 80 -0.664 distinguishes 256 levels (8 bit) of luminosity in one 60 Distances pixel. 40 Differences Drivers detect pedestrians by their contrast, the 20 0 difference in brightness between pedestrian and 1234567891011Observers background. The keystone of the computer method -20 -40

was to compare luminosity of the pedestrian and Distances [m], Differences [-] -60 luminosity of the background. There was used -80

“mask” tool, which provides possibility to cut area -100 of the pedestrian from the background. Histogram of the pedestrian was displayed. It was used “invert Fig. 4. Human – Computer recognition mask” tool, what allowed us to display histogram of Measurement 1 - Retroreflectors 120 the background. The difference of weighted 110 arithmetic means of the pedestrian and the 100 background is the measure of the contrast. 90 80 NONE 70 ANKLE SHOULDER 60 3. RESULTS Distances [m]

50 Primary task was to measure detecting distances 40 30 Obs erve rs and compare human to computer method. Secondary 20 task was to investigate an effect of retroreflector 1234567891011121314 Measurement 2 - Retroreflectors positioning on recognition of silhouettes of 120 pedestrians. Graphic charts show the results – figure 110 100

4 and 5. Correlation coefficient indicates the 90

80 strength and the direction of a linear relationship NONE 70 ANKLE between two variables – detection distances and SHOULDER 60 Distances [m] differences of weighted arithmetic means of 50 luminosity. The problem of visibility and 40 recognition is complex and there is a great 30 Obs erve rs 20 contribution from complicating factors. 123456789101112 Fig. 5. Effect of retroreflector positioning Correlatio Negative Positive Small -0,3 to -0,1 0,1 to 0,3 Medium -0,3 to -0,5 0,3 to 0,5 Large -0,5 to - 1,0 0,5 to 1,0

Fig. 6. Interpretation of a correlation coefficient 120 DIAGNOSTYKA’ 3(47)/2008 LIŠýÁK, NAMEŠANSKÝ, Human And Computer Recognition Of Nighttime Pedestrians

DISCUSSION necessary to point out, that we deal with static pedestrian and we tried to recognize the silhouette. In the primary task we investigated recognition We predicted, that the position of the retroreflector distances of the pedestrian in different conditions. on the ankle would be the best for recognition. It is Graphic charts – Fig. 4, shows distances, when the obvious from the Fig. 6., that the longest recognition pedestrian was recognized by an observer. Mean distances were in the case of shoulder position. recognition distances and standard deviation shows From the study of the photographs it is obvious and Fig. 7. visible, that the ankle position of the retroreflector causes glare of the observer and also driver. Mean Standard Consequently, the ability to recognize a silhouette is Measurement recognitio deviatio reduced. There is no doubt, that the application of n distance n the retroreflector, whether fixed to shoulder or 1 92,29 m 9,33 ankle, enabled visibility of the light spot from the 2 36,53 m 12,66 long distance (more than 200 m). 3 61,82 m 10,47 Generally, many researches confirmed that Fig. 7. Arithmetic mean of recognition distances pedestrians are visible at greater distances when they wear a reflective tag or vest. However, there are The longest distances were measured in the some drawbacks to reflective material. One is that measurement 1, when there was favourable weather reflective material sends light primarily in one and the measurement was done with Mercedes direction. If the headlights hit the material at the S with bi-xenon headlights. The shortest distances wrong angle, the reflected light goes in the wrong were measured in the measurement 2, when there direction and does not hit the driver’s eye, and the was foggy weather and the measurement was done reflector will appear dark. Further, if the reflective with Škoda Octavia. Thus, the influence of weather material covers a small part of the body, then the is evident. Foggy weather is one of the most driver may detect its light but may not recognize it dangerous weather conditions. It is also confirmed as being a person. Reflective material may also by the road accidents. When scaled by total number cause pedestrians to be overconfident. [3] of traffic accidents, fatality rate is the highest in foggy weather. REFERENCES The computer method, which we have tried, is not enough accurate. It is confirmed by a correlation [1] Luoma J., Schumann J., Traube E. C.: Effects coefficient. To improve accuracy, reliability and of retroreflector positioning on nighttime objectivity of the used method, it is necessary to recognition of pedestrians. The University of improve evaluating process of the photographs. Michigan Transportation Research Institute, There are some proposed improvements: MI, USA, 1995. a) An algorithm, which would be able to separate [2] McCarley J. S., Krebs W. K.: Visibility of road (automatically and reliable) pedestrian from hazards in thermal, visible, and sensor-fused background. night-time imagery. Operations Research b) In the evaluation of luminosity of background, to Department, Naval Postgraduate School, divide it to close and far background. Another Monterey, CA, USA, 1999. option would be to develop an integral method [3] Green M.: Seeing pedestrians at night. with stronger importance of the close www.visualexpert.com background. However, there is a question what is [4] Kasanicky G.: Kramer F., Nitsche K.: the close background. Zwischenergebnis zur Studie – Heutige c) The same or similar process with pedestrian as in Methoden und Verfahren zur Rekonstruktion the case of background. von dunkelheitsunfällen / optische The best solution though would be to develop Wahrnehmbarkeit, 2006. a method, which would be able to evaluate the contrast between pedestrian and background, Prof. Ing. Štefan LIŠýÁK, CSc. reliable and quick. The existence of the algorithm, [email protected] which would be able to recognize reliable the University of Žilina, pedestrian by an analysis of an image, in visible, IR Univerzitná 1, 010 26 Žilina, Slovakia spectrum or using fused image, is the condition for developing intelligent and maybe autonomous vehicle safety system. The biggest problem is with Ing. Pavol NAMEŠANSKÝ the reliability, because there are uncountable [email protected] different traffic situations in different weather University of Žilina, conditions. Univerzitná 1, 010 26 Žilina, Slovakia The secondary task was to investigate an effect of retroreflector positioning on recognition of nighttime pedestrians. Fig. 5. shows the results. It is DIAGNOSTYKA’ 3(47)/2008 121 SOKOLOVA, BALITSKY, On Some Critical Machinery Vibration Monitoring Algorithm And …

ON SOME CRITICAL MACHINERY VIBRATION MONITORING ALGORITHM AND ITS APPLICATION FOR INCIPIENT FAULT DETECTION AND LOCALIZATION

Asja SOKOLOVA, Felix BALITSKY Machinery Engineering Research Institute, Russian Academy of Science 4, M. Kharitonievskiy, Moscow, 101990 Russia, tel/fax: +7 495 333 51 39, [email protected]

Summary Further development of proposed earlier [1] critical machinery vibration monitoring algorithm intended for faults early detection, unlike standard monitoring techniques, is propounded. The algorithm is founded on nondimensional S-discriminants, calculated from current amplitude- clipped vibration signal parameters referred to the ones for the machine being in good (normal) condition. These parameters have an inherent high sensibility to amplitude spikes magnitude and amount growth, which takes place at vibration signal under the machine degradation, due to suppressing intrinsic machine vibration hash. The paper shows that really effective high speed machinery condition monitoring technique based on using casing vibration data should mandatory take into account the acceleration parameters calculated both in wide and narrow frequency bands.

Keywords: vibration condition monitoring, non-dimensional discriminant, machine defect.

1. INTRODUCTION applications and on early detection of gas turbine engine (GTE) rolling bearing developing defects at To assure technical, ecological and human safe a gas pipeline compressor station. machinery exploration one needs to use condition monitoring and diagnostics algorithms which enable 2. A NEW APPROACH TO THE VIBRATION to detect operational malfunctions at the very early MONITORING OF HIGHSPEED stage of their development, i.e. algorithms based on MACHINERY specific methods which detect even week changes of vibration signals originated from the incipient faults. As it is seen from practice, the more complicated It is the well-known fact that vibrations carry all is a machine unit, the more dispersion of measured information about machine dynamic behavior, parameters within general scope of similar machines including defects manifestation as well. The wider proves. Due to that it is important to use the frequency range we use for machine problem individual approach when vibration monitoring is analysis and the more diagnostics algorithms take applied. into consideration the main specific features of One feasible way for the early malfunctions machine dynamic model, the better are results. It is recognition is preliminary “passportization” of well-known also another fact, that generally most of vibration spectrums and posterior comparison of machinery monitoring and protection system measured current spectra with the passport spectral algorithms are based on the estimation of the RMS data to estimate changes derived from machine (root mean square) value of vibration velocity over condition degradation during its exploration. The the range of 10 to 1000 Hz, or amplitude divergence point is that the procedure is not effective enough (for wide or narrow frequency band) from base line because these changes are small and due to the meanings under good machinery condition. procedure is not automatic. The conventional vibration velocity range There was suggested another way with using (10- 1000 Hz) of high-speed-rotation machinery some non-dimensional vibration characteristics contains merely several first shaft rotation frequency (S-discriminants) which values are independent harmonics affected only by rough machinery either on machine type or vibration units, but only condition changes, for example due to unbalance, on type (kind) of defect and its severity. Such well- eccentricity, misalignment, part breakage, and so on. known discriminants as peak-factor , excess X p /V But technique of incipient fault detection (such as E P /V 4  3 (or kurtosis), indexes of amplitude or erosion, corrosion, pitting, scuffing, and so on) bases 4 on some other principles, because their symptoms frequency modulation, prof. Cempel dimensionless are situated in high frequency vibration range [1-3]. discriminants, and so on, often have been used for The paper gives evidences of the necessity of this aim [1]. However above-mentioned combined approach with incipient fault detection nondimensional parameters have an essential and diagnostics methods as its foundation. The paper demerit, namely – when a numerator growing along presents some investigation results on the subject of with defect development (due to the number and safe machinery operation ensuring in atomic amplitude of vibration overshoots enlargement), 122 DIAGNOSTYKA’ 3(47)/2008 SOKOLOVA, BALITSKY, On Some Critical Machinery Vibration Monitoring Algorithm And … a denominator behaves similarly and dependences of specified amplitude clipthreshold P (see for example these characteristics on machinery time operation are the red lines at Fig. 2 which are drawn here at nonlinear as a rule. It is known [1], that such amplitude values of ±2Vn). Thus it is possible to get sensitive vibration parameter to overshoots rid of the own machine vibration which is intrinsic appearance as excess coefficient, having high in absence of any malfunctions, and due to it sensitivity to any damages at an early stage with improve the parameters sensitivity. small amount of overshoots per time unit, lost its sensitivity to well-developed damage with overshoots quantity growing on.. Dependence of excess (and other standard dimensionless parameters) values from pulses number (which is modeling the machinery parts degradation degree along the operational time) is nonlinear function that means non-single-valued nature of these parameters, in contrast to S-discriminant Id (see Fig. 1).

Fig. 2. Examples of vibration acceleration time histories: a) for machine in good condition, b) for machine with injured elements (scuffing in contact zone) Fig. 1. Influence of pulse quantity m (within time realization N=1024) on excess E, peak-factor P-F, As a result there were formed the relative normalized amplitude deviation V t /V n and S- indexes from some statistical parameter of the discriminant Id values for meaning of relative pulse clipthreshold exceeding for current vibration signal divided by the same characteristic of reference amplitude A/V =5, where V ,V – are standard s s n signal, measured under machinery normal condition. amplitude deviations of random noise signal and One of them is dispersion index Id of threshold reference summary noise and pulses sequence exceeding, formulated as follows: § N 2 · All that is the foundation for suggestion such ¨ x P ¸ ¨ ¦>@i t K ¸ kind of monitoring and protective algorithms which I i 1 ˜ t ¨ d N 2 K ¸ would be more sensitive to early fault events and ¨ n ¸ x j P besides have monotonic dependence from machinery ¨ ¦>@ n ¸ © j 1 ¹ degradation state during operational time. To properly realize the “critical” machines condition Here (xi)(t) and (xi)(n) are values of vibration monitoring and incipient fault detection techniques amplitude components, calculated for current and was proposed the algorithm of estimation of some reference (i.e. normal, without any faults) machine dimensionless S-discriminant magnitude declining conditions; P , (O= 0.5 – 3.0) – amplitude from the value equal to an unit that is placed in OV n correspondence with machinery normal condition. clip-threshold, V 0 – standard deviation (RMS) of Analysis of machine vibration waveforms with vibration signal for normal (reference) machinery operational damage in progress shows that vibration condition; K(t) and K(n) – are numbers of overshoots becomes unstable and mandatory feature of an above the threshold P for current and normal incipient fault is an appearance of single or multiple vibration signals. signal overshoots (Fig. 2) deriving from interaction An amplitude discriminant significance equals to conjugate parts format changes due to erosion, 1, when the machine is under its good condition, but corrosion, pitting, contact surfaces welding and becomes >1 or >>1, if any kind of damage would scuffing, and so on. take place. Thus there were formed dimensionless To raise the sensitivity of the vibration features amplitude discriminants, featuring high sensitivity to to incipient faults and minimize the influence of instability, caused by machinery operational uninformative regular machine vibration (i.e., noise), imbalance, resulted from any fault, and noise there were suggested stochastic dimensionless immunity to internal machinery masking characteristics (amplitude discriminants) to be interference. Some practical (experimental) formed for vibration signals, clipped above the examples of S-discriminant successful application DIAGNOSTYKA’ 3(47)/2008 123 SOKOLOVA, BALITSKY, On Some Critical Machinery Vibration Monitoring Algorithm And … for incipient detection of multistage gearbox parts recognizable. The first significant discriminant defects and of gas turbine unit rolling bearing faults overshoot which corresponds to dramatic change are given below. condition of ball-bearing one could see at 25.06.04 point, i.e. three weeks before the train has been 2. VIBRATION ANALYSIS RESULTS OF GAS stopped. TURBINE ENGINE (GTE) AND GEARBOX

The example below is given to clearly prove that standard monitoring technique is not effective to evaluate changes of complex high speedy machinery and is not applicable to incipient faults detection and recognition (malfunctions of bearing supports, tooth gears, compressor and turbine blades, and so on) due to practically absence of information of these machine parts operability in vibration velocity signals. Really, plant’s limiting values of vibration velocity RMS within 10…1000 Hz frequency band (overall level) could be exceeded only under great Fig. 4. Narrow band (1.0-1.75 kHz) case vibration machine degradation that distorts of its shaft line various parameter behavior during ball bearing (bearing breakage, blade breakaway, and so on). damage development in LPC of gas turbine engine Accordingly with GTE failure statistics there DG-90 front support. about 80% breakdowns are the result of bearing malfunctions. Typical plot of lateral vibration Some practical examples of S-discriminant acceleration spectrum being acquired at one of navy application for multistage gearbox parts GTE type DG-90 casing measurement points is malfunctions are described below. Wide-band given in Fig. 3 for front support of low pressure vibration discriminant analysis permits to detect compressor (LPC) in vertical direction [2]. informative measurement point for consequent detailed narrow-band signal analyses. In Fig. 5 there are given the dependences of discriminants from operation time of research reactor IBR-2 movable reflector-modulator [3], that are calculated for narrow frequency band (5.0-6.3 kHz) casing gearbox vibration, measured in points #1-4.

Fig. 3. GTE casing vibration acceleration spectrum

It is clear, that frequencies of bearings generated vibration lay, as a rule, exceed 1000 Hz and as a matter of fact the deterioration of LPC front support ball bearing in question had not been detected timely with the conventional on-line monitoring system because of no one vibration velocity threshold had not been exceeded. Fig. 5. Narrow-band gearbox casing vibration In Figure 4 the several non-dimensional vibration discriminant trends for measurement points (#1-4) parameters plots are shown under degradation process in ball-bearing of front support of low In Fig. 6 are given the amplitude modulation pressure compressor (LPC) of navy gas turbine indexes, calculated for informative measuring point engine DG-90 during operation time 01.06.04 to #4 on the ball bearings defect passing frequencies. 19.07.04, when machinery was stopped due to metal In such a way it’s possible to realize the chips appearance in lubrication. procedure of machinery deterioration development All parameters were measured for narrow band detection generally, and after that – specific (1.0÷1.75 kHz) case vibration acceleration – around malfunction localization by means of frequency of the inner race frequency (BPFI) [2]. The identification. For objective estimation of operating machinery condition it is reasonably to estimate both discriminant Id actually has max value of _ 100, but Figure 4 scale (0 to 25) was chosen for purpose the wideband and narrowband vibration parameters other parameters (peak-factor P-F, normalized evolution during machinery service life such as S-discriminant. standard deviation V t /V n , excess E) – to be 124 DIAGNOSTYKA’ 3(47)/2008 SOKOLOVA, BALITSKY, On Some Critical Machinery Vibration Monitoring Algorithm And …

SOKOLOVA Asja ………. Grigorievna, Ph. D., Head of vibroacoustical machinery diagnostics laboratory, Machinery Engineering Research Institute of Russian Academy of Science, Moscow, Russia

BALITSKY Felix Janovich, Ph. D., Senior scientific researcher of vibroacoustical Fig. 6. Trends of ball bearings defect passing machinery diagnostics labora- frequencies amplitude modulation indexes tory, Machinery Engineering Research Institute of Russian Only combined utilization of a set different Academy of Science, Moscow, methods and algorithms for operating condition Russia monitoring of critical machinery, leaning on the information, obtained with incipient failure detection methods can give a reliable estimation of its condition.

REFERENCES

[1] Sokolova A. G., Balitsky F. Ja.: Sensitive and Noise-immune Vibration Discriminants for Instability Phenomena Detection caused by Incipient Machinery Deterioration, Proceedings from 1st International Symposium on Stability Control of Rotating Machinery “ISCORMA-1”, South Lake Tahoe, California, USA, 20-24 August 2001 (published in CD). [2] Sokolova A. G., Balitsky F. Ja., Pichugin K. A.: Dimensionless Machine Vibration Sdiscriminants as a Mean to Improve monitoring and Get Incipient Fault Detection, The 9th European Conference on Non-Destructive Testing (EC NDT), Conference Proceedings / Poster Presentation N 82, 8 pp, Berlin, September 25-29 2006. [3] Sokolova A. G., Balitsky F. Ja., Sizarev V. D.: Detection Technique for Faults Perturbing Rotor Stability, Illustrated by Means of an Example of the Movable Reflector-Modulator of the Pulsed Research Reactor. Proceedings of The Second International Symposium on Stability Control of Rotating Machinery-ISCORMA-2, pp 170-179, Gdansk, Poland, 4-8 August 2003. DIAGNOSTYKA’ 3(47)/2008 125 NIZIēSKI, KUPICZ, Method For Diagnosing Internal Combustion Engines With Automatic Ignition…

METHOD FOR DIAGNOSING INTERNAL COMBUSTION ENGINES WITH AUTOMATIC IGNITION ON THE BASIS OF TORQUE MEASUREMENT IN TRACTION CONDITIONS

Stanisáaw NIZIēSKI, Wáodzimierz KUPICZ

University of Warmia and Mazury Military Institute of Armour and Automotive Technology (WITPiS) 05-070 Sulejówek, 1 Okuniewska St., tel. +48 22 681 10 73, e-mail: [email protected]

Summary The paper presents a new method for diagnosing internal combustion engines with automatic ignition in traction conditions. Its substance involves determination of engine torque on the basis of recording of acceleration in road conditions. Extensive preliminary and primary experimental research has been conducted. Three variants of the internal combustion engine diagnostic model have been developed using a trivalent evaluation of states. Algorithms have been proposed to control a state and location of engine defects. The new method has been verified in traction conditions. A probability of a correct diagnosis of internal combustion engine is 0.85 ÷ 1.

Keywords: military vehicles, internal combustion piston engines, diagnostics.

METODA DIAGNOZOWANIA SILNIKÓW SPALINOWYCH O ZAPàONIE SAMOCZYNNYM NA PODSTAWIE POMIARU MOMENTU OBROTOWEGO, W WARUNKACH TRAKCYJNYCH

Streszczenie W pracy przedstawiono nową metodĊ diagnozowania silnika spalinowego o ZS w warunkach trakcyjnych. Istota polega na wyznaczeniu momentu obrotowego silnika na podstawie rejestracji przyspieszenia w warunkach drogowych. Wykonano obszerne wstĊpne i zasadnicze badania eksperymentalne. Opracowano trzy warianty modelu diagnostycznego silnika spalinowego, z wykorzystaniem trójwartoĞciowej oceny stanów. Zaproponowano algorytmy kontroli stanu i lokalizacji uszkodzeĔ silnika. Zweryfikowano nową metodĊ w warunkach trakcyjnych. PrawdopodobieĔstwo poprawnej diagnozy silnika spalinowego wynosi 0,85 ÷ 1.

Sáowa kluczowe: pojazdy wojskowe, táokowe silniki spalinowe, diagnostyka.

1. INTRODUCTION 4. so far there are no objective diagnostic methods for quantitative determination of a wearing Based on analysis and assessment of diagnostic degree of piston-crank system and camshaft of methods for internal combustion engines with internal combustion engines; automatic ignition, it should be stated that [3, 4, 6]: 5. a need arises to develop methods for finding 1. tool-free methods for research and assessment of a genesis and forecasting a state in an aspect of a state of internal combustion engines with possibilities for them to be used in practical automatic ignition, based on senses of eyesight, realization of the internal combustion engines hearing and touch may be used as supplementary servicing process; methods in mechanical vehicles diagnostic- 6. a significant number of existing methods stands servicing process, although they are not excluded for static and quasi-dynamic methods that are to be used as primary methods in practical useless for applications in board diagnostic realization of processes for diagnosing and systems; servicing internal combustion engines; 7. in our opinion, at present there is no sufficient set 2. a considerable number of diagnostic methods of diagnostic methods and means allowing for being used for internal combustion engines with full and effective control of a state and location automatic ignition is characterized with complex of damages in engine with automatic ignition diagnostic algorithms and a high labor both in stationary conditions and in particular consumption, which results in their low while driving; usefulness in diagnostics of those technical 8. there is a need to develop an effective method to objects; control a state and location of damages in 3. a large number of existing diagnostic methods internal combustion engine with automatic for internal combustion engines with automatic ignition for its further use in board diagnostic ignition has no fixed boundary values of systems in conditions while a vehicle is in diagnostic parameters; motion. 126 DIAGNOSTYKA’ 3(47)/2008 NIZIēSKI, KUPICZ, Method For Diagnosing Internal Combustion Engines With Automatic Ignition…

2. DIAGNOSTIC MODEL FOR VEHICLE is worthwhile stressing that a concept of ACCELERATION methodology, defined below, for determining 2.1. Gist of vehicle acceleration process a torque of internal combustion engine has not been In the vehicle acceleration process in a selected applied so far in diagnostic systems of mechanical gear, internal combustion engine’s torque from vehicles in traction conditions. crank-shaft is transferred to a clutch, gearbox and 2.4. Methodology for determining further through distribution box, final drive – onto a set of diagnostic parameters vehicle wheels. Driving force Fn, that is generated A problem with determining a set of diagnostic by the engine must be higher than total motion parameters of internal combustion engines in resistances force that consists of: inertia force Fb, traction conditions will be solved by the following rolling resistance force Ft, and air resistance force Fp steps: on a horizontal route. STEP I The higher driving velocity the higher total Measurement of vehicle linear velocity and motion resistances force and vehicle acceleration engine rotational speed keeps declining until stability of velocity at which STEP II driving force becomes balanced by the motion Determination of dynamic radius of a circle r resistances force, or when a maximum rotation d speed of the engine has been achieved. from dependence: 2.2. Equations of vehicle motion while v *i r c car acceleration d (2.5) ns General equation of vehicle’s rectilinear motion where: v – vehicle linear velocity; i – total ratio of on a flat route has a form of [1, 2]: c driving system; ns – engine rotational speed. Fn Ft  Fp  Fb (2.1) STEP III where: Fn – driving force on vehicle wheels; Ft – Attempt of vehicle free rundown. A quantity rolling resistances force; Fp – air resistances force; being measured is a delay af and ac of vehicle Fb – inertia resistances force; motion. The gist of measurement involves After its transformations, expression (2.1) has acceleration of the vehicle up to a possibly a form of: maximum speed, and then setting a gear-shifting 2 lever into neutral and recording a delay Fn fmg  JCx Av Gma (2.2) a f f (t) where: f – rolling resistances coefficient; m – until the vehicle has stopped. Motion resistances are vehicle weight; g – gravitational acceleration; U – determined at velocities below 10 km when air density defined by a formula: h H T influence of air resistances is ommittably small. Air U U T 0 (2.3) resistances are determined by delay a recorded at 0 H T c 0 T velocities of 75–20 km/h. where: U – air density in normal conditions = 1.189 3 STEP IV kg/m ; HT – air pressure at the moment of Determination of motion resistances coefficient measurements taking; H0 – reference pressure of from the following formula: 100 kPa; TT – air temperature at the moment of a f § I k · measurements taking;T0 – reference temperature of f ¨1 ¦ ¸ ¨ 2 ¸ (2.6) 293 K; Cx – air resistances coefficient; A – vehicle g © mrd ¹ face surface; a – vehicle momentary acceleration; where: af – vehicle motion delay; g – gravitational v – vehicle linear velocity; į – vibrating masses acceleration; Ik – driving wheel moment of inertia; coefficient defined by a formula: m – vehicle weight. I  I i 2K  6I It is worth stressing that the motion resistances G 1  s T c m k (2.4) 2 coefficient includes rolling resistances and frictions rd m in: wheel bearings, meshing of gear wheels and where: Is - moment of inertia of engine vibrating shafts joints. masses; IT – coefficient to undefined states of STEP V internal combustion engine; ic – total ratio of driving system; Șm– driving system efficiency; Ik – Determination of vehicle air resistances driving wheels moment of inertia; rd – real, coefficient from the following formula:

momentary dynamic radius of wheels. § m ·§ ac I k · C ¨ fg a ¦ ¸ 2.3. Gist of New Method x ¨ 2 ¸  c  2 (2.7) ¨ UAv ¸¨ mr ¸ The method being proposed bases on © ¹© d ¹ determination of engine torque on the basis of where: A – vehicle face surface; ac – vehicle motion acceleration recording while vehicle is in motion in delay; ȡ – air density. road conditions. A simultaneous measurement of A measurable quantity is vehicle motion delay ac fuel consumption allows for determining hourly fuel being recorded at velocities of 70 ÷ 20 km/h, in consumption in engine’s rotational speed function. It a rundown attempt being realized in STEP III. DIAGNOSTYKA’ 3(47)/2008 127 NIZIēSKI, KUPICZ, Method For Diagnosing Internal Combustion Engines With Automatic Ignition…

STEP VI 3. EXPERIMENTAL RESEARCH Measurement of vehicle acceleration and fuel 3.1. Research goal consumption is started upon a minimum velocity has The preliminary research goal was [5] to: become stable on third gear, and then through x define properties of signals being examined; a rapid maximum stepping on acceleration pedal the x determine an influence of factors being tested on vehicle gathers speed until a maximum velocity has values of physical quantities being measured; been achieved. Vehicle dislocation, rotational speed x determine defined research terms and conditions, of engine crank shaft and fuel consumption are required by developed method to control a state measured. and location of 4CTi90–1 BE6 engine damages. The goal of primary research has been specified A measured quantity is vehicle dislocation s p . as follows: The gist behind measuring that quantity involves x to select the best diagnostic parameters due to a rapid acceleration of the vehicle until a maximum states differentiation criterion; velocity on a give gear has been achieved and x determine boundary values of diagnostic recording of dislocation s f (t) . The vehicle p parameters for selected states of 4CTi90–1 BE6 d 2 s engine; motion acceleration is obtained as a p . p dt 2 x determine states–diagnostic parameters STEP VII couplings, that is, to obtain a diagnostic model of Having the following data: dynamic radius of 4CTi90–1 BE6 engine, type - informative model Is. circle r , motion resistances coefficient f, air d The purpose of laboratory research is to collect resistances coefficient and motion Cx data essential for comparing their values with values acceleration a p , it is possible to determine torque of results obtained in the primary experimental from the following dependence: research, and on their basis verify the developed method for diagnosing 4CTi90–1 BE6 engine. § mgfr  UC Av2r Gmr a · M ¨ d x d d p ¸ (2.8) 3.2. Research Program s ¨ i ¸ © cKm ¹ Preliminary diagnostic testing program for the engine has been presented on fig. 1. where: į – vibrating masses coefficient; ap – vehicle Primary diagnostic testing program for 4CTi90– motion acceleration; Ge – momentary hourly fuel consumption measured in acceleration process. 1 BE6 engine has been presented on fig. 2. 2.5. Summary Laboratory testing program for 4CTi90–1 BE6 Analysis of the vehicle acceleration diagnostic engine has been presented on fig. 3. model authorizes to formulate the following 3.3. Test stand and measuring apparatus conclusions: A test stand for diagnostic testing of 4CTi90–1 1. the basis for the new method of research and BE6 engine has been HONKER make vehicle. assessment of a state of internal combustion Individual signals have been recorded using the engine with automatic ignition is equation of following apparatus: vehicle rectilinear motion on a flat route (2.1), x optoelectronic sensor of linear velocity of DLS-1 and dependence (2.8). vehicle made by Corrsys-Datron company: 2. a set of parameters of a state of internal x sensor of rotational velocity of FS2–60 engine combustion engine with automatic ignition made by Keyence company; includes: x 116H type of fuel flow-meter made by Pierburg 3. engine torque Ms ; company, 4. hourly fuel consumption Ge ; x fuel temperature sensor; 5. diagnostic parameters of vehicle state and x data acquisition and processing station ȝEEP–10 internal combustion engine can also be as made by Corrsys-Datron company follows: 4. PRELIMINARY TESTS RESULTS 6. dynamic radius of circle - rd ; ANALYSIS 7. motion resistances coefficient f ; 4.1. Influence of whether conditions on values 8. delay af ; of physical quantities being measured 9. delay ac ; x Analysis of performed tests results has indicated 10. acceleration ap . essential differences in values of physical 11. it should be stressed that for the above- quantities being measured in diverse weather mentioned sets of diagnostic parameters to be conditions. In order to reduce uncertainty of symptoms of a state of internal combustion diagnosis made, an analysis has been conducted engine with automatic ignition, their usefulness of the weather conditions impact and correction should be proved in a sense of meeting the equations have been presented allowing for hereto criteria: sensitivity, uniqueness, stability referring obtained results to standard ambient and informativeness, using experimental conditions i.e. temp. 20º C, pressure 100 kPa. research. 128 DIAGNOSTYKA’ 3(47)/2008 NIZIēSKI, KUPICZ, Method For Diagnosing Internal Combustion Engines With Automatic Ignition…

Badania wstĊpne as average value and a standard deviation. This

D Badanie wáasnoĞci sygnaáów quantity characterizes a dispersion of results, that is, 1 i 2 a statistical component of uncertainty of the E Temperatura otoczenia ,..., 1 1 measurement result (A type). 2 j 2 j 1 2 j F CiĞn. atm ,..., ,..., ,..., Obtained testing results are convergent with data

1 2 k included in technical literature. Growing ambient G WilgotnoĞü ,..., ,..., temperature effects in declining air density, and thus 1 2 l H Pow. czoáowa ,..., ,..., its smaller amount is sucked into the engine

1 2 m I Opory toczenia cylinder, which worsens a level of fulfillment and ,..., ,...,

1 2 n a progress of the combustion process, and therefore J Kierunek jazdy ,..., ,..., reduces effective pressure and engine torque. 1 2 o K Dáug. odcinka pomiarowego ,..., ,..., The hereto considered process has been

1 2 p L Operator approximated with a straight line. Straight line ,..., ,..., inclination coefficient has been used to determine 1 2 r M Egzemplarz pojazdu ,..., ,..., a correction equation, leading the obtained

1 2 s N Stan techniczny characteristics of the internal combustion engine ,..., ,...,

1 2 t torque to normal temperature of 20º C. The O Mierzone parametry ,..., ,..., dependence takes the following form: 1 2 u P Liczba pomiarów ,..., ,..., M M  (0,4060*(T T)) sT0 s 0 1 2 w (4.1) ,..., where: M – torque corrected to temperature of 20º sT0 Zbiór wartoĞci parametrów diagnostycznych C; M – torque measured in ambient Fig. 1.Chart of preliminary diagnostic testing s program for 4CTi90–1 BE6 engine temperature T; T0 – reference temperature – 20ºC;

Badania zasadnicze Ambient temperature influences a temperature of fuel being delivered to engine injection pump. Stan obiektu Dz Injection apparatus had no temperature controlling Egzemplarz pojazdu E z device in the vehicle being tested. Therefore, 1 2 j differences in velocity characteristics of hourly fuel F z Uszkodzenia ,..., 1 2 5 consumption, being measured while testing in 1 2 k G z Mierzone wielkoĞci ,..., ,..., ,..., fizyczne various ambient temperatures, have been expected. 1 1 2 l 2 k Parametry Results of the performed analysis are presented on H z ,..., ,..., ,..., diagnostyczne 1 fig. 5. 1 2 m 2 n ,..., ,..., A correction equation to standard ambient temperature has been determined in analogical way as that for the torque. Zbiór wartoĞci parametrów Zbiór wartoĞci parametrów diagnostycznych diagnostycznych zdatnego silnika niezdatnych silników G G  (0,02142*(T T )) eT0 e 0 Fig. 2. Chart of primary diagnostic testing program (4.2) for 4CTi90–1 BE6 engine where: G – fuel consumption corrected do eT0 temperature of 20º C; – fuel consumption Badania Ge Laboratoryjne

DL Miejsce badaĔ measured in ambient temperature T; T0 – Hamownia silnikowa Hamownia podwoziowa reference temperature – 20º C; E Stan techniczny L Air pressure FL Mierzone parametry Corrected maximum torque values have been Ms Ge Ms Ge Ms Ge Ms Ge analyzed again due to ambient pressure value. GL Liczba pomiarów Comparative results have been presented on fig. 6. A correction equation of air pressure influence on k 1 2 , ...., 1 2 , ...., k 1 2 , ...., k 1 2 , ...., k torque characteristics has been determined in analogical way as that for the temperature: Zbiór wartoĞci parametrów diagnostycznych M M  (2,0042 *( p  p)) s k sT0 0 Fig. 3. Chart of diagnostic laboratory testing program (4.3) where: – torque corrected to normal conditions for 4CTi90–1 BE6 engine M s k Ambient temperature (temperature 20º C, air pressure 100 kPa); Tests performed at identical air pressure have – torque corrected to temperature 20º C; M s been separated out of a set of all measurement T0 results. Maximum values of internal combustion p0 – reference pressure (normal) reaching engine torque have been assumed for analysis. 100 kPa; p – air pressure while making Results have been presented on fig. 4. The measurements. maximum torque value has been marked on the chart DIAGNOSTYKA’ 3(47)/2008 129 NIZIēSKI, KUPICZ, Method For Diagnosing Internal Combustion Engines With Automatic Ignition…

Air Humidity significant influence on torque value. Statistical Relative air humidity depends on pressure and (A type) uncertainty of results of individual temperature. In order to become independent from measurements (performed in stable conditions) is those two quantities, for the purposes of this method, higher than differences effecting from the impact of a relative humidity has been measured and then proper humidity. converted to absolute humidity (so-called proper Correction equations of the impact of weather humidity) expressed as content of water in one conditions can also be determined on the basis of kilogram of dry air. Influence of absolute humidity analysis of multiple regressions. It allows for on torque results is presented on fig. 7. obtaining a single equation that contains influence of As indicated on the above figure, absolute all the factors being tested. humidity hereto applied measurement method has no

185 184 Moment obrotowy silnika-wpáyw temperatury 183 182

181

180

179

178 y = -0,4060x + 183,5327 177 176

175

Ms [Nm] 174

173

172

171

170

169 168

167

166

165 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

Temp [OC] Fig. 4. Influence of ambient temperature on maximum torque values of internal combustion engine 4CTi90–1 BE6

Godzinowe zuĪycie paliwa-wpáyw temperatury 16

15,5

15

14,5

14

13,5 y = -0,0214180461x + 14,7649078819

13 Ge [l/h] 12,5

12

11,5

11

10,5

10 -6 -4 -2 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 [obr/min] Fig. 5. Influence of ambient temperature on hourly fuel consumption of internal combustion engine 4CTi90–1 BE6 182 Moment obrotowy-wpáyw ciĞnienia otoczenia 181

180

179

178 177

176 175

174

173 Ms [Nm] 172

171 y = 2,0042x - 25,064 170

169

168

167

166 165 99,2 99,3 99,4 99,5 99,6 99,7 99,8 99,9 100 100,1 100,2 100,3 100,4 100,5 100,6 100,7 100,8 100,9 101 101,1 101,2 CiĞnienie atmosferyczne [kPa] Fig. 6. Influence of air pressure on maximum values of 4CTi90–1 BE6 internal combustion engine torque 130 DIAGNOSTYKA’ 3(47)/2008 NIZIēSKI, KUPICZ, Method For Diagnosing Internal Combustion Engines With Automatic Ignition…

182 Wpáyw wigotnoĞci bezwzglĊdnej na wartoĞü momentu obrotowego 181

180

179

178

177

176

175

174

173 Ms [Nm]

172

171

170 y = -0,0157893836x + 175,4819075662 169

168

167

166

165 3456789101112131415 [g/kg] Fig. 7. Impact of air absolute humidity on obtained values of maximum torque of internal combustion engine 4CTi90–1 BE6

Summary dew-point temperature (characterizing air On the basis of performed tests and analyses of humidity); DP0 – dew-point/reference weather conditions influence on values of physical temperature: 10º C, quantities being measured, it should be stated that: x for fuel consumption:

1. ambient temperature has essential impact on Gek Ge  (0,0394 (T0  T ))  0,8545 values of parameters characterizing a process from (4.5) * ( p0  p)  0,3013 (DP0  DP) a torque of internal combustion engine 4CTi90–1 BE6; where: Gek – fuel consumption corrected to normal 2. ambient pressure has a very essential influence on conditions (temperature 20º C, air pressure 100 values of parameters related to a torque of internal kPa, reference humidity defined as dew-point 10º combustion engine 4CTi90–1 BE6; C);Ge – fuel consumption measured in ambient 3. proper air humidity has a minor impact on temperature T, pressure p and humidity DP; parameters values determined for internal 4.2. Influence of motion conditions on values combustion engine 4CTi90–1 BE6 torque of physical quantities being measured characteristics; Rolling resistances 4. influence of weather conditions on parameters Rolling resistances of the vehicle being tested values related to fuel consumption by the engine is have been changed by reducing air pressure in immaterial; vehicle tires. Tests have been conducted in two 5. in order to obtain reliable testing results, torque variants: at pressure reaching 0.25 MPa and 0.42 and fuel consumption values being obtained MPa. They are extreme pressure values allowed for should be corrected to normal conditions; Goodyear tires G90 7.50 R16C that the vehicle 6. the most convenient method of correction to being tested has been equipped in. normal conditions is a use of multiple regression Trust ranges have been calculated for parameters following this dependence: equations recognizing the influence of all factors 2 2 being tested; s yn s yn yn  t yn yn  t (4.6) 7. percentage of explained volatility for weather D n D n conditions impact on torque value is very high and where: tD – variable t–Student for n-1 degrees of exceeds 90%, and for fuel consumption 76.43%; freedom, fulfilling the relation P(– tD t < tD)=1–D; finally, correction equations of weather conditions y – parameter average value; 2 – parameter influence have the following form: n syn x for the torque: variance defined by dependence: N M M (0,5066 (T T )) 2,1704* 2 2 1 sk s  0   s s (y y ) y yn ¦ n  n (4.7) (4.4) n 1 n 1 ( p0  p)  0,1435 (DP0  DP) States differentiation test has been applied for the where: – torque corrected to normal conditions M s k purposes of this study, comparing average values of (temperature 20º C, air pressure 100 kPa, parameters and trust ranges for a state of usability reference humidity defined as dew-point 10º C); and non-usability. Damage identifiability has been Ms – torque measure in ambient temperature T, defined via trivalent assessment of a parameter. The pressure p and humidity DP; p0 – reference „0” value has been assumed for no identifiability. If pressure (normal) reaching 100 kPa; p – air a value of a tested parameter keeps declining pressure while making measurements; T0 – because of damage, a checking result is „-1”, and if reference temperature: 20º C; T – ambient it is rising „1”. Procedure of proceeding is presented temperature while making measurements; DP – on the model: DIAGNOSTYKA’ 3(47)/2008 131 NIZIēSKI, KUPICZ, Method For Diagnosing Internal Combustion Engines With Automatic Ignition…

1. definition whether a change in the parameter is local road section Halinow – Cisie with asphalt identifiable upon considering the trust range: surface and on military airfield Sochaczew with 2 2 concrete surface. Honker vehicle with EURO-3 s y0 s yn y0  yn ! t 0  t n engine version has been the testing object in both D n D n 0 n (4.8) cases. Measurements have been taken with an interval of two days with slightly different weather where: y – parameter average value corresponding 0 conditions, therefore results have been corrected to

to a state of usability; yn – parameter average normal conditions, following equations of (4.4) and value corresponding to a state of n-th non- (4.5). s 2 The performed tests results indicate that there are usability; t y0 – half of trust range of no significant differences in the proposed values of D 0 n 0 diagnostic parameters. parameter corresponding to a state of usability; Vehicle face surface s 2 The analysis of vehicle face surface impact on t yn – half of trust range of parameter Dn n measurements results has been conducted on the n basis of measurements made on Honker and corresponding to a state of n-th non-usability. Scorpion vehicles. The face surface in the first tested 2. if checking result is a number different from „0” object has reached 3.55 m2, while in the second case definition whether a value of the tested 3.82 m2. Characteristics have been compared and parameter has declined or maybe grown as corrected to normal conditions. a result of damage. Values of parameters related to engine torque Analysis of the presented results indicates that tire slightly differ. Upon considering trust ranges, there pressure, and effectively vehicle rolling resistances, are no differences between average values. This have no influence on testing results. It allows to allows saying that the assumed methodology formulate a conclusion that the assumed including in its calculations real resistances in methodology, recognizing real vehicle rolling vehicle motion is a correct one. The states resistances (determined in a free rundown attempt), differentiation test detects differences in parameters make the measurement result independent from related to fuel consumption: consumption at motion resistances impact. rotational speed of maximum torque Ge , and Driving direction M maximum consumption Gemax. The reason behind Impact of driving direction on results of this situation is probably a production dispersion of diagnostic parameters being obtained has been tested engines. The two vehicles originated from different by conducting a series of measurements in both production batches: Honker from year 2002, while driving directions. While conducting tests, wind the vehicle marked with a cryptonym of Scorpion velocity has exceeded 2 m/s lengthwise of the from year 2003. measured section. Analysis has bee carried out 4.3. Conclusions separately for measurements performed for driving Based on performed preliminary rests and against the wind and with the wind. analyses of results obtained, it should be said that: The torque process analysis indicates that 1. results repeatability tests has indicated that the parameters values slightly differ. Trust levels remain proposed parameters are characterized with at the same unchanged level. Noticeable is a shift of repeatability for the engine usability state; a maximum torque into directions of lower rotational 2. point and range estimations of research results speed, and a faster accumulation of a torque in range indicate that average values of the proposed 1800 - 2500 rotations per minute. This results from parameters are characteristic for the usability phenomena occurring during cooperation of turbo- state and individual damages, which proves their compressor with the engine. Higher motion usefulness for building a diagnostic model of resistance (adverse wind) cause higher load on the internal combustion engine; engine, higher combustion pressure, and thus 3. weather conditions have been found to have combustion fumes that generate higher supercharge essential impact on engine torque values, but pressure at low rotational velocities. Therefore, wile a small influence on fuel consumption average driving against the wind torque is higher than that at value; low rotational velocities; however, upon considering 4. based on many variances uniqueness test, it has trust ranges the differences are insubstantial. been found that there are no basis for rejecting The states differentiation test does not show any a hypothesis about equality of variances in tested significant differences in values of parameters, parameters for usability state, but the torque determined on the basis of measurements made with variance value differs for individual damages; the wind and against the wind. This proves 5. on the basis of correlation analysis, a dependence a correctness of the assumed testing methodology. has been found of average torque on weather Surface type of measured section conditions; Research tests on influence of a surface type have been conducted on two measurement sections: 132 DIAGNOSTYKA’ 3(47)/2008 NIZIēSKI, KUPICZ, Method For Diagnosing Internal Combustion Engines With Automatic Ignition…

6. an analysis has been made of weather conditions measurements have been taken during each of the impact and regression analysis and equations sessions. Characteristics of usable engine have been have been generated correcting the testing results presented on fig. 8. ÷ fig. 9. The charts also present to normal ambient conditions; trust ranges for proposed diagnostic parameters and 7. results of motion conditions influence allow for characteristics curves equations. Trust range of formulating the following conclusions: a parameter has been determined from dependence x impact of rolling resistances on values of (4.6). physical quantities being measured is immaterial; Table 1 presents values of proposed diagnostic x driving direction has no influence on values of parameters of usable 4CTi90–1 BE6 engine, in physical quantities being measured; EURO2 version, corrected to normal ambient x type of surface has no influence on values of conditions, and their trust ranges. physical quantities being measured; 5.2. Influence of damages on values of physical x vehicle face surface has no influence on values quantities being measured of physical quantities being measured; Weak links of the engine being tested, which 8. diagnostic parameters for usable internal have been determined on the basis of vehicles combustion engine have been proposed and maintenance monitoring at military bases, have been determined; assumed as a criterion for selection of damages to be 9. states differentiation test (4.4) has been proposed diagnosed. Table 2 presents weak links of internal that contains trust ranges for states of usability combustion engine 4CTi90–1 BE6 and non-usability. This has resulted in a trivalent Testing results of individual damages impact on assessment of internal combustion engine states. values of proposed diagnostic parameters, corrected to normal conditions, have been presented in 5. PRIMARY RESEARCH RESULTS a comparison with usable engine parameters. Trust ANALYSIS ranges of proposed diagnostic parameters have been 5.1. Characteristics of usable engine calculated for individual damages and the states Characteristics of usable 4CTi90–1 BE6 engine differentiation test has been performed based on have been determined based on 8 measuring sessions dependence (4.4). conducted in various weather conditions. Some 30 Results have been presented on fig. 10 ÷ fig. 11.

185 183 Moment obrotowy silnika z przedziaáem ufnoĞci 181 179 177 175 173 171 169 167 165 163 161 159 157 155 153 Ms [Nm] Ms 151 149 y = 6E-18x6 - 1E-13x5 + 1E-09x4 - 4E-06x3 + 0,0086x2 - 9,846x + 4607,5 147 145 143 141 139 137 135 133 131 129 127 125 1700 1800 1900 2000 2100 2200 2300 2400 2500 2600 2700 2800 2900 3000 3100 3200 3300 3400 3500 3600 3700 3800 3900 [obr/min] Fig. 8. Torque characteristics of usable 4CTi90–1 BE6 internal combustion engine in EURO2 version 23 Ge zdatnego silnika po korekcji z 22 y = -1E-19x6 + 7E-16x ufno5 + 3E-12xĞci 4 - 4E-08x3 + 0,0001x2 - 0,2218x + 126,55 21

20

19

18

17

16

15

14 l/h 13

12

11

10

9

8

7

6

5 1700 1800 1900 2000 2100 2200 2300 2400 2500 2600 2700 2800 2900 3000 3100 3200 3300 3400 3500 3600 3700 3800 3900 obr/min Fig. 9. Fuel consumption characteristics of usable 4CTi90–1 BE6 internal combustion engine in EURO2 version DIAGNOSTYKA’ 3(47)/2008 133 NIZIēSKI, KUPICZ, Method For Diagnosing Internal Combustion Engines With Automatic Ignition…

Table 1. Proposed diagnostic parameters of usable Table 2. Weak links of 4CTi90–1 BE6 engine 4CTi90–1 BE6 internal combustion engine Non- Marki Value of Half of No Element- Damage probability usability ng diagnostic trust range state parameter 0 1 injection pump 0.1 w1 Average. torque [Nm] M 160.64 ± 2.28 0 s 2 turbo-compressor 0.1 w2

Max torque [Nm] Ms 175.11 ± 1.02 0 3 air filter 0.4 w3

Torque at min. speed 0 Ms0 137.31 ± 5.07 4 injectors 0.3 w [Nm] 4 Torque at max speed 5 cylinder damage 0.05 w 0 M 140.07 ± 1.84 5 [Nm] si 6 timing gear system 0.05 w 0 Rotational speed of 6 max. torque nM 2600 ± 21 [rotations/min] Based on primary experimental research that has M s0 C 0.7842 ± 0.0415 been conducted in traction conditions, it should be M M1 max said that: M si C 0.7999 ± 0.0112 1. states differentiation test with trivalent M M2 s max assessment of state has been prepared; Average fuel Ge 16.14 ± 0.74 2. a set of diagnostic parameters has been consumption [l/h] proposed and their values have been determined Consumption at for usable 4CTi90–1 BE6 engine; rotational speed of GeM 15.48 ± 0.35 max. torque [l/h] 3. values of parameters have been determined Consumption at min. corresponding to six selected damages; G 8.89 ± 0.51 rotational speed [l/h] emin 4. states differentiation test has been made for Consumption at max. introduced individual damages and a damage Gemax 21.49 ± 0.48 speed [l/h] identifiability has been determined; GeM CGe1 0.5744 ± 0.0343 Gemin

Gemax CGe2 0.7204 ± 0.0306 GeM

201 196 191 186 181 Ms pompa 176 Ms turbo 171 Ms filtr 166 Ms cylinder 161 Ms zawory 156 Ms zdatny 151 Ms wtryskiwacze 146 141 Ms [Nm] 136 131 126 121 116 111 106 101 96 91 86 1700 1800 1900 2000 2100 2200 2300 2400 2500 2600 2700 2800 2900 3000 3100 3200 3300 3400 3500 3600 3700 3800 3900 obr/min Fig. 10. A comparison of torque characteristics for usable engine and for individual damages

24 23 22 21 20 19 18 17 16 15 14

Ge [l/h] Ge pompa 13 Ge turbo Ge filtr 12 Ge cylinder 11 Ge zawory 10 Ge zdatny Ge wtryskiwacze 9 8 7 6 5 1700 1800 1900 2000 2100 2200 2300 2400 2500 2600 2700 2800 2900 3000 3100 3200 3300 3400 3500 3600 3700 3800 3900 obr/min Fig. 11. A comparison of torque characteristics for usable engine and for individual damages 134 DIAGNOSTYKA’ 3(47)/2008 NIZIēSKI, KUPICZ, Method For Diagnosing Internal Combustion Engines With Automatic Ignition…

Table 3 presents results of the states characterized with the highest precision – accuracy differentiation test for individual damages. An of diagnosis being made. Its disadvantage is a high analysis of Table 3 indicates that: cost and labor consumption when taking 1. results of the states differentiation test for measurements. It requires fuel consumption individual damages vary for specific damages, measurement, which is associated with a purchase of which proves that parameters have been expensive apparatus and time-consuming assembly correctly matched; inside the vehicle. 2. the most information providing parameters are: Those diagnostic parameters in the model are as follows: M s , Ms Ms0, for which a change in parameter 1. average torque value is differentiable in all tested damage M s cases; 2. maximum torque value Msmax at rotational speed 3. the least information providing parameters are: nM; CGe2 – no differentiability for four damages, CM1 3. torque value Ms0 at minimum rotational speed (see item 6.1) – no differentiability for four n0=nmin; damages, 4. torque value Msi at maximum rotational speed 4. the easiest identifiable are: injection pump ni=nmax ; damage – a change in all parameters besides 5. placement of maximum torque on rotational CM2 and CGe2, and injectors damage – a change speed axis - rotational speed of maximum in all parameters besides nM and CGe2 , however, torque nM; the value of all parameters keeps growing for 6. a ratio of torque at minimum rotational speed to this type of damage; maximum torque: 5. the most difficult identifiable is a damage M C s0 involving a leaky combustion chamber. In this M1 (6.1) M max case, only three parameters change: , M M s s 7. a ratio of torque at maximum rotational speed to and Ms0 (see item 6.1); maximum torque: 6. Table 3 is information model (diagnostic M C si matrix) enabling a control of a state and location M 2 (6.2) M s max of selected damages of 4CTi90–1 BE6 engine. 8. average fuel consumption Ge ; 9. fuel consumption G at rotational speed of Table 3. Information model (diagnostic matrix) of eM maximum torque n ; internal combustion engine 4CTi90-1 BE6 M 10. consumption at minimum rotational speed Gemin; Marking Pump Tur Air Inject Cylinder Leaky Usable corre bo Filter ors combusti 11. consumption at maximum speed Gemax ; ctor on 12. ratio of fuel consumption at maximum torque to chamber a minimum consumption:

M s 0 -1 -1 -1 1 -1 -1 GeM CGe1 (6.3) Ms 0 -1 -1 -1 1 -1 -1 Gemin

Ms0 0 -1 -1 -1 1 -1 -1 13. ratio of maximum fuel consumption to consumption at maximum torque: Msi 0 -1 -1 -1 1 -1 0 Gemax nM 0 -1 1 0 0 -1 0 CGe2 (6.4) GeM C M1 0 1 0 0 1 0 0 The research results obtained are compared to CM2 0 0 1 -1 1 0 0 values of quantities of usable-model engine and then Ge 0 -1 0 0 1 -1 0 a parameter value is being determined:

GeM 0 -1 1 0 1 -1 0 x „0” – no changes (the parameter value

Gemin 0 -1 -1 -1 1 -1 0 corresponds to the state of usefulness); x „1” – the parameter value has been suspended Gemax 0 -1 0 -1 1 -1 0 compared to the value corresponding to the state C Ge1 0 1 -1 0 1 0 0 of usefulness; C Ge2 0 0 1 1 0 0 0 x „-1” – the parameter value has declined compared to the state of usefulness. 6. INTERNAL COMBUSTION ENGINE If all parameters have „zero” value, then object is 4CTI90–1 BE6 DIAGNOSTIC METHOD usable. If „1” or „-1” has been recorded in the results sheet, then a set of results of parameters checks is 6.1. Diagnostic Model compared to combinations of results characteristic Based on experimental research analysis results, for individual damages. A non-usability state is three variants have been developed of diagnostic determined on that basis and damage is located. model of 4CTi90–1 BE6 engine. This paper will Then location of the damage is verified through present the most extended variant. The variant is a comparison of absolute values of obtained DIAGNOSTYKA’ 3(47)/2008 135 NIZIēSKI, KUPICZ, Method For Diagnosing Internal Combustion Engines With Automatic Ignition… parameters with values corresponding to individual 7,35 Nm) and (nM =2640± 67 rotations/min) and damages. Accuracy (probability) of diagnosis is (CM1=0,7731± 0,0681) and (CM2=0,5584± 0,0363) calculated on the basis of a number of parameters and ( Ge =14,80± 0,68 l/h) and (GeM =14,98± 0,20 that reflect changes with their character („0”, „1” or l/h) and (Gemin = 7,88± ± 0,07 l/h) and (Gemax = „-1”) and absolute value of parameters describing 18,08± 0,45 l/h) and (CGe1= 0,5263± 0,0154) and 0 a given damage. (CGe2=0,8281± 0,0363) then the state is w3 ; State control and 4CTi90–1 BE6 engine damages location algorithm, developed on the basis of model If ( M = 1) and (Ms = 1) and (Ms0= 1) and (Msi = 1) I, has the following form: s State control: and (nM = 0) and (CM1= 1) and (CM2= 1) and (Ge =1) If ( = 0) and (M = 0) and (M = 0) and (M = 0) and (GeM = 1) and (Gemin = 1) and (Gemax = 1) and M s s s0 si 0 (CGe1= 1) and (CGe2= 0) then the state is w4 ; and (nM = 0) and (CM1= 0) and (CM2= 0) and (Ge =0) and (GeM = 0) and (Gemin = 0) and (Gemax = 0) and Supplementary test aiming at higher certainty of 1 (CGe1= 0) and (CGe2= 0) then the state is w ; diagnosis Location of damage: If ( M =187,76± 1,51 Nm) and (Ms =195,14± 1,53 If ( = –1) and (M = –1) and (M = –1) and (M = s M s s s0 si Nm) and (Ms0= 184,33± 1,59 Nm) and (Msi=169,85± –1) and (n = –1) and (C = 1) and (C = 0) and M M1 M2 1,98 Nm) and (nM =2620± 39 rotations/min) and (Ge = -1) and (GeM = –1) and (Gemin = –1) and (Gemax (CM1=0,9446± 0,0054) and (CM2=0,8704± 0,0205) = –1) and (CGe1= 1) and (CGe2= 0) then the state is and ( Ge =18,89± 0,70 l/h) and (GeM =17,83± 0,08 0 w1 ; l/h) and (Gemin =13,07± 0,09 l/h) and (Gemax = 23,49± Supplementary test aiming at higher certainty of 0,08 l/h) and (CGe1=0,7328± 0,0121) and 0 diagnosis (CGe2=0,7591± 0,0123) then the state is w4 ; If ( =126,59± 1,48 Nm) and (M =134,44± 1,59 M s s If ( =–1) and (M = –1) and (M = –1) and (M = Nm) and (Ms0= 127,85± 2,03 Nm) and (Msi=109,56± M s s s0 si

3,32 Nm) and (nM =2440± 109 rotations/min) and –1) and (nM = –1) and (CM1= 0) and (CM2= 0) and (C =0,9510± 0,0159) and (C =0,8150± 0,0211) M1 M2 (Ge =–1) and (GeM = –1) and (Gemin = –1) and (Gemax and ( Ge =12,39± 0,54 l/h) and (GeM =10,94± 0,57 = –1) and (CGe1= 0) and (CGe2= 0) then the state is 0 l/h) and (Gemin=7,28±0,50l/h) and (Gemax = 16,70± w5 ; 0,09 l/h) and (CGe1= 0,6653± 0,0440) and Supplementary test aiming at higher certainty of 0 (CGe2=0,6552± 0,0504) then the state is w1 ; diagnosis If ( =104,54± 1,32 Nm) and (M =111,07± 2,77 M s s If ( M = –1) and (Ms = –1) and (Ms0= –1) and (Msi = s Nm) and (Ms0= 89,27± 2,33 Nm) and (Msi=98.02± –1) and (nM = 1) and (CM1= 0) and (CM2= 1) and 2,39 Nm) and (nM =2460± 115 rotations/min) and (Ge =0) and (GeM = 1) and (Gemin = –1) and (Gemax = (CM1=0,8037± 0,0248) and (CM2=0,8825± 0,0494) 0) and (CGe1= –1) and (CGe2= 1) then the state is and ( Ge =11,41± 0,58 l/h) and (GeM =10,92± 0,08 0 w2 ; l/h) and (Gemin =6,60± 0,12 l/h) and (Gemax = 15,29± Supplementary test aiming at higher certainty of 0,16 l/h) and (CGe1=0,6048± 0,0340) and 0 diagnosis (CGe2=0,7141± 0,0497) then the state is w5 ; If ( =133,54± 1,81 Nm) and (M =144,08± 2,33 M s s If ( =–1) and (Ms = –1) and (Ms0= –1) and (Msi = Nm) and (Ms0= 113,04± 1,88 Nm) and (Msi=125,33± M s 1,80 Nm) and (nM =3100± 83 rotations/min) and 0) and (nM = 0) and (CM1= 0) and (CM2= 0) and (Ge = (CM1=0,7846± 0,0115) and (CM2=0,8699± 0,0130) 0) and (GeM = 0) and (Gemin = 0) and (Gemax = 0) and 0 and ( Ge =15,11± 0,83 l/h) and (GeM =17,77± 0,24 (CGe1= 0) and (CGe2= 0) then the state is w6 ; l/h) and (Gemin = 7,95± 0,19 l/h) and (Gemax = 20,94± Supplementary test aiming at higher certainty of ± 0,16 l/h) and (CGe1= 0,4473± 0,0186) and diagnosis 0 (CGe2=0,8485± 0,0328) then the state is w2 ; If ( =153,50± 2,21 Nm) and (M =167,19± 1,24 M s s If ( M =–1) and (Ms = –1) and (Ms0= –1) and (Msi = s Nm) and (Ms0= 127,22± 1,49 Nm) and (Msi=136,42± –1) and (nM = 0) and (CM1= 0) and (CM2= –1) and 2,93 Nm) and (nM =2620± 32 rotations/min) and (Ge =0) and (GeM = 0) and (Gemin = –1) and (Gemax = (CM1=0,7609± 0,0086) and (CM2=0,8159± 0,0165) –1) and (CGe1= 0) and (CGe2= 1) then the state is and ( Ge =15,99± 0,79 l/h) and (GeM =15,34± 0,12 0 w3 ; l/h) and (Gemin =8,36± 0,12 l/h) and (Gemax = 21,78± 0,11 l/h) and (CGe1=0,5451± 0,0090) and 0 Supplementary test aiming at higher certainty of (CGe2=0,7040± 0,0135) then the state is w6 ; diagnosis If ( =132,78± 4,02 Nm) and (M =157,10± 4,47 Probability of a accurate diagnosis is calculated M s s on the basis of the following expression: Nm) and (Ms0= 121,46± 2,07 Nm) and (Msi=87,72± 136 DIAGNOSTYKA’ 3(47)/2008 NIZIēSKI, KUPICZ, Method For Diagnosing Internal Combustion Engines With Automatic Ignition…

n1 method for diagnosing internal combustion pd *100% t pdgr (6.5) n engine in traction conditions; 9. based on performed tests and analyses of their where: n1 – a number of diagnostic parameters whose value is convergent with a value set for a results, a three-variant 4CTi90–1 BE6 engine given damage; n – total number of diagnostic diagnostic model has been developed; 10. diagnostic models have become the basis for parameters; p – boundary probability of dgr development of algorithms to make a diagnosis – diagnosis accuracy . pdgr 0,85 assessment of a state and location of 4CTi90–1 6.2. Method Verification BE6 engine damages; The method for diagnosing internal combustion 11. research conducting conditions have been engines with automatic ignition, based on torque presented that guarantee maintenance of assumed measurement in traction conditions, has been probability to make accurate diagnosis; verified using: 12. the method has been verified on the basis of: x tests of smokiness of usable engine fumes and at a comparison of results to combustion fumes selected damages; smokiness research results, and through tests of a vehicle with introduced damages in traction x traction tests of a vehicle with usable engine and with selected damages. conditions. On the basis of verification tests it Accuracy of a diagnosis made in torque has been found that the diagnosis accuracy measurement method in traction conditions is an probability obtained in the course of verification assessment criterion. Verification tests results allow is not lower than 85%. for making a conclusion that diagnosis accuracy probability is above 85%. REFERENCES

7. SUMMARY AND FINAL CONCLUSIONS 1. ArczyĔski S.: Mechanika ruchu samochodu. WNT, Warszawa 1994. Summing up this paper on „Method for 2. DĊbicki M.: Teoria samochodu, WNT, Warszawa diagnosing internal combustion engines with 1969. automatic ignition based on torque measurement 3. Hebda M., NiziĔski S., Pelc H.: Podstawy in traction conditions,” the following should be diagnostyki pojazdów mechanicznych. WKà, stated: Warszawa 1984. 1. analysis and assessment of existing diagnostic 4. NiziĔski S. and others: Diagnostyka samochodów methods have been conducted and on the basis of osobowych i ciĊĪarowych. Dom Wydawniczy which it has been found that there is a need to Bellona, Warszawa 1999. develop a new diagnostic method allowing for 5. NiziĔski S., Michalski R.: Diagnostyka obiektów making a diagnosis of internal combustion technicznych. Wydawnictwo Instytutu engine in real road conditions and that may Technologii Eksploatacji. Warszawa-Sulejówek- become applicable to both board and external Olsztyn-Radom 2002. diagnostic systems; 6. ĩóátowki B.: Podstawy diagnostyki maszyn. 2. theoretical assumptions have been presented of Wydawnictwa Uczelniane Akademii Techniczno- a new diagnostic method the gist of which is Rolniczej w Bydgoszczy, Bydgoszcz 1996. vehicle acceleration process in traction conditions. A set of state parameters includes: Prof. dr hab. inĪ. Stanisáaw engine torque Ms and hourly fuel consumption NIZIēSKI jest pracownikiem Ge; naukowym Katedry Budowy 3. algorithm for determining diagnostic parameters Eksploatacji Pojazdów has been developed; i Maszyn UWM w Olsztynie 4. algorithm for research method structure and oraz Wojskowego Instytutu assessment of states of internal combustion Techniki Pancernej engines with automatic ignition has been i Samochodowej w Sulejówku. developed: Jest wieloletnim czáonkiem 5. preliminary experimental research have been Sekcji Podstaw Eksploatacji KBM PAN oraz conducted under which the following have been Polskiego Naukowo-Technicznego Towarzystwa realized: Diagnostyki Technicznej. Jego zainteresowania 6. primary experimental research in traction naukowe obejmują zagadnienia dotyczące logistyki conditions have been conducted; w systemach dziaáania, eksploatacji i diagnostyki 7. primary experimental research in conditions of obiektów technicznych, szczególnie pojazdów engine and chassis test benches have been mechanicznych. Jest autorem i wspóáautorem wielu conducted: prac naukowych i dydaktycznych. 8. a total uncertainty of torque measurement and fuel consumption have been estimated under the DIAGNOSTYKA’ 3(47)/2008 137 BATKO, Weight And Significance Of Object Dynamic Models In Constructing Monitoring Systems

WEIGHT AND SIGNIFICANCE OF OBJECT DYNAMIC MODELS IN CONSTRUCTING MONITORING SYSTEMS

Wojciech BATKO Department of Mechanics and Vibroacoustics, AGH University of Science and Technology, Al. Mickiewicza 30, 30-059 Kraków, [email protected]

Summary The author’s approach to tasks of constructing monitoring systems of the state changes of machines, devices and structures is presented in the paper. The weight and significance of the identification process of the monitored object dynamic model – as a factor which in an universal and coherent way preserves the realisation process of several tasks assigned to this system operations – is emphasised. Purposefulness of applying methods of the technical stability testing for estimation of dynamic behaviours of the monitored object, was indicated. Its usefulness for the selection of the proper monitoring symptoms, for determining their criterion values and for the estimation of their exceeding the permissible values - are shown in the paper.

Keywords: monitoring of machine technical state, the algorithms of monitoring procedures, modelling of dynamic state, stability, signal processing and analysis.

ROLA I ZNACZENIE MODELI DYNAMICZNYCH OBIEKTU W BUDOWIE SYSTEMÓW MONITORUJĄCYCH

Streszczenie W artykule zaproponowano autorskie podejĞcie do zadaĔ konstruowania systemów monitorujących zmiany stanów maszyn, urządzeĔ, konstrukcji. Zwrócono uwagĊ na rolĊ i znaczenie procesu identyfikacji modelu dynamicznego monitorowanego obiektu, jako czynnika, który w sposób uniwersalny i spójny zabezpiecza proces wykonawczy szeregu zadaĔ przypisanych funkcjonowaniu tych systemów. W szczególnoĞci, wskazano na celowoĞü zaangaĪowanie metod badania statecznoĞci technicznej do oceny zachowaĔ dynamicznych kontrolowanego obiektu. Pokazano jej uĪytecznoĞü dla procesu realizacji wyboru wáaĞciwych w procesie monitorowania symptomów, okreĞlenia dla nich wartoĞci kryterialnych, jak i tworzenia procedur ocen wyjĞcia ich poza dopuszczalne granice.

Sáowa kluczowe: monitoring stanu technicznego, algorytmy procesu monitorującego, modelowanie stanu dynamicznego, statecznoĞü, analiza i przetwarzanie sygnaáów

1. INTRODUCTION a certain automation of the diagnostic changes occurring in the monitored machine - are more and Monitoring of state changes of machines and more often implemented in solutions. Their devices – corresponding to the recognition of early realisations are based on various model formalisms stages of defects and identification of disturbances determining their functioning as well as on the in their operations – is a complex research task. In knowledge collected during their operations. The practice, it is realised the most often via measuring proper selection of algorithms in the construction and tracing dynamic changes of processes occurring process is decisive for the universality and in the object under inspection. This is done by the functionality of the monitoring system. properly selected sensors in the monitoring system. Currently, in the collection of construction rules The construction process is related to: of the monitoring systems available in the market, x Selection of measuring rules and recording of there is none coherent theory enabling the proper the determined process variables, connection of conditions of loosing the ability to x Selection of the conversion methods for safe operations with the rules of selecting recognition of causes of the observed states, diagnostic symptoms. Constructional and x Assuming relevant analytical algorithms of exploitational features of the inspected object their trends, corresponding to the requirements of undisturbed x Establishing criterion references enabling the estimation of the monitored processes and proper classification of the recognised states. prediction processes with a determined time Procedures of filtration of measurement advance - are not sufficiently taken into account in disturbances, algorithms of change predictions of the construction of monitoring systems. Trials of controlled signals or expert systems allowing for looking for solutions, in which the essence of operating of the monitoring system constitute 138 DIAGNOSTYKA’ 3(47)/2008 BATKO, Weight And Significance Of Object Dynamic Models In Constructing Monitoring Systems relationships between the state of the inspected where En denotes the linearly normalised system and the form of the monitored signal [3, 4, n-dimensional space. 7] conjugated with the criteria selection for the When using the identified model of the alarm signals [5, 6] or for switching off - are not monitored object it is possible to recognise numerous. theoretically its behaviour under the influence of Thus, it seems that an important realisation disturbances. The deviations of the inspected approach to the construction of monitoring systems processes versus the programmed ones and caused is its coupling with the identification of the either by parameter changes or by variability of dynamic model of the inspected object and with the inputs acting on the object - are classified. analysis of its technical stability [11]. Such Taking into account their model connections approach ensures the possibility of undertaking and with the measuring observations the Kalman [15] solving several tasks in the building process of the model formalism can be assigned to them, ensuring monitoring system. The frame of constructional the realisation process of the optimal filtration and operations related to the building of the monitoring prediction tasks of the monitored diagnostic signals. system, it means related to the selection of proper The identified model of the inspected object diagnostic symptoms, of the levels of their given by Equation (1), under conditions of acting quantification for the purpose of diagnostics random disturbances, allows - in addition - the decisions or the requirements of the filtration of estimation of the time when the analysed operation measuring disturbances – can be based on them. enters into zones assigned to alarm states [19, 20]. The aim of this paper is to indicate certain The assigned for them analyses can be related to the possibilities within this scope. It does not pretend to existing analytical methods of the dynamic systems show all results of the author researching these stochastic trajectories, including solutions based on problems, but aims to indicate the existence of the the method of „functionals supporting the universal platform giving the possibility of obtainable zone„. The analysis of the monitoring undertaking and solving several tasks in process of the state changes of the hydrodynamic constructing monitoring systems – referring to the bearing [1, 2] being the part of the dynamic system: common modelling formalisation. ‘Rotor – Bearings – Supports – Foundation’ can constitute a good example. 2. OBJECT MODELLING – THE BASE The model describing dynamic behaviours of FOR CONSTRUCTING MONITORING the object, when its steady state is disturbed, can SYSTEM become the basic tool at constructing the monitoring system. It generates indications Several tasks determining functioning of concerning the symptoms selection for the monitoring systems, can be reduced to the problem observation of the object state changes as well as of analysis the possible trajectories of the inspected the selection criterion values protecting the dynamic system under the influence of disturbances classification of undesirable state changes. in the exploitation process, caused by undesirable Realisation of this operations results from analysis exploitation inputs (loads and external of dynamic behaviours of the object done when disturbances), or by changes of structural testing its stability. Various understanding of the parameters (it means: mass, elastic and damping stability notion and methods of its testing properly parameters) due to failures. related to the required functional properties of the They determine the possible solutions of the set monitored object – can be applied [11, 13, 14]. of differential equations modelling state changes in The criterion of the technical stability [11] can the monitored object and related to them the be a good testing criterion for realisation of several estimation of changes of the monitored dynamic tasks, which appear in processes of constructing processes. monitoring systems. It widens the notation of Their forms result from the selected model stability in the Lagrange’s and Lapunow’s meaning formalisation of the inspected object, it means, into conditions, which can appear in the from the description of the set of elements being in investigated technical reality. They seem to be a mutual cooperation with the surroundings, which especially essential for selecting the method of the is characterised by a set of measurable features monitored state changes realisation in the inspected (containing the total information on the state of the object. It takes into account the quantification of inspected object), which are changing with time. motion disturbances characteristic for the tasks Generally the model describing the cooperation assigned by the monitoring system. They are – from can be represented by a pair (X , f), in which X the point of view of the correctly functioning denotes the vector of state space, while f is a vector inspected object - conditioned by the assumed of this space representation: permissible perturbations of input forces, or object x characteristics as well as by the permissible = f (t, x , x ,... .,x ) , i=1,2,... .,n (1) xi i 1 2 n changes of initial conditions – related to the in which functions f i are determined in a space: acceptable changes of the structure parameters of t t 0, (x1, ... .,xn)  G  En the monitored object. DIAGNOSTYKA’ 3(47)/2008 139 BATKO, Weight And Significance Of Object Dynamic Models In Constructing Monitoring Systems

In the case when the solutions of the technical exits zone Z has to remain in zone : for stability theory are applied for the realisation the ttT00dd , where Tt 0 is a time of motion. At monitoring of the object state it is necessary to such condition we are dealing with the technical assume: stability in a finite time. x Permissible deviations of the motion trajectory Realisation of the monitoring process of from the steady state (related to the safe changes of the inspected object state can be also exploitation of the analysed object); looked for on the grounds of the stochastic x Permissible range of the initial conditions technical stability definition introduced by changes; W. Bogusz [11]. x Levels of expected external and internal It is formulated as follows, in reference to the disturbances permanently acting on the object described by the model: inspected object, which dynamic behaviours are x described by the equation: x f ( x, t, [ (t,D ) ) (6) x x f ( x, t )  R (x, t ) (2) This model takes into account random in which x, f, U are vectors determined in space disturbances D, acting on it, in a form of a certain n stochastic process [ (,t D ) of the following ƒ : features: x ªf1 º ªR º ª 1 º 1 [(t,D ) = » R (t ,x,D )» (7) « » « » « » sup x f R x G x = « 2 » f = « 2 » R= « 2 » (3)  « ... » «...» « ... » In reference to such conditions of functioning of « » « » « » the inspected object, two zones : and Z contained ¬xn ¼ ¬fn ¼ ¬R n ¼ in En are defined, such that zone Z is limited and and their functions f(t,x) R (t,x) are determined in open, : is limited, closed and contains origin of the zone contained in the (1)n  dimensional coordinates and also takes into account the space: condition Z :. Let us assume that the positive number H satisfying inequality 0 H 1 exists. txGE!0, (4) n Initial conditions for tt are x()tx , while It is assumed that input functions of 0 00 permanently acting disturbances R (t,x) in the zone the solution assigned to them: x(,tt00 , x ). If each

(4) is limited: solution x(,tt00 , x ) of initial conditions belonging R (t ,x) G (5) to zone Z belongs also to zone : with the probability higher than 1 H , the analysed object is where G is a positive number, while x denotes technically, stochastically stable it means: the Euklides norm of vector R(t,x). stochastically stabilised versus zones Z, : At this type of model dynamic behaviour the and process [ ()t with the probability 1 H : notion of technical stability for the system modelled P { x( t, t 0 , x 0 ) :} > 1 - H for x0 Z (8) is understood as follows: The presented above mathematical Let there be two zones : and Z - contained in formalisation of the problem of monitoring state G. Zone : is closed, bounded and contains the changes of the object described by model (2), origin of the system, while Z is open and contained supplemented by relevant statements facilitating within :. technical stability testing, can be used as the Let us assume, that the solution of the analysed approach methodology to constructing monitoring system (2) is x(),t of an initial condition: system. Thus, it forms the logical frames for x()tx . solving problems occurring at searching for the 00 proper constructions of monitoring systems. If for every x0 belonging to Z, x(t) remains in 3. METHODS OF TESTING THE zone : for ttt 0 , at the disturbance function satisfying inequality (5), the system (2) is TECHNICAL STABILITY – TOOLS IN CONSTRUCTING MONITORING technically stable due to zones Z, and the limited, SYSTEMS constantly acting disturbances (5). According to this definition of the technical Realisation of the idea of utilising solutions of stability, each motion trajectory, which exits zone the technical stability theory in the process of Z, has to remain in zone : for ttt 0. constructing monitoring systems [5, 6] means For the monitored objects, for which introducing – into their solutions – algorithms of instantaneous exits of the monitored signals outside testing conditions of loosing the technical stability levels considered permissible - are safe, the notion by the inspected object. This requires the of the technical stability can be weakened into the identification of the dynamic model of the condition, in which each trajectory of motion which monitored object, being the grounds for 140 DIAGNOSTYKA’ 3(47)/2008 BATKO, Weight And Significance Of Object Dynamic Models In Constructing Monitoring Systems comparative references of the observed states. law given in paper [15], since it provides an Algorithms of testing the technical stability enable estimation of the velocity with which the dynamics their classification. The main determinants of the of the monitored system proceeds to the zero monitoring process are identifications of the equilibrium point. dynamic model of the inspected object. The The law states that, in the case when the identification process should be conjugated with the Lapunow’s function: selected methods and algorithms of testing the Vx()! 0,xƒn , x z 0, is such that Vx() 0 technical stability, which can become the useful n tool for the recognition of alarm states in the for xxƒ,0 z at which the following relation inspected object. occurs: Solutions included in the so-called ‘topologic’ J max(Vx ( ) / Vx ( )), x ƒzn , x 0 methods of solutions of differential equations can x be applied in the constructing of monitoring systems. They require examination of trajectories, it the inequality: bt / means curves [ x(),txty () ] on the phase plane Vxt(,)d Vx (,0) e (11) x, x , related to the dynamic model of the which allows to estimate the velocity, with which the analysed dynamic system proceeds to the zero monitored object. Their analysis can constitute the equilibrium point – is the true one. basis for determination of the permissible dynamic The results of this law enable to select properly behaviour of the monitored object, at the described the delay time of the monitoring system for reacting level of permanently acting disturbances and for an instantaneous disturbance. allowable perturbations of the steady state. Those Another way of investigating properties of the perturbations are related to changes of the monitored trajectories – estimating the dynamic parameters value, which might be significant for system stability – is using two functions [12]: appearing of defects, including their early stages. x x x x The investigating procedures are based on ) (x, y ) = x y + x y ; < (x, y ) = x y - x y certain topologic facts, related to certain invariants ) (12) of homomorphic rearrangements, formulated in the The first function is a derivative of the square form of statements. They enable the qualitative vector of the distance of the point on the trajectory estimation of dynamic behaviours of the analysed from the origin of coordinates, while the second is object and conditions for the technical stability loss the value of the phase velocity moment versus the - corresponding to them. origin of coordinates, which positive or negative The Lapunow’s method [11, 16, 17] is the most determinant allows to estimate the character of the often applied procedure. The properties of the monitored motion. Their values, positive or scalar function Vxt(,) - properly selected for the negative, enable assigning to the trajectory points dynamic description of the inspected object - are the directions characteristic for input, output or slip used. Investigating its derivative along solutions points versus the analysed curve, which allows to (behaviours) of the system (2) determines its determine G and : zones within the monitored stability. phase space. This law states: if there is a scalar function Assumptions for building quantifiers of the Vxt(,) of class C1 determined for each x and monitored trajectories properties (from the point of view of their stability) can be also looked for on the t t 0 in the vicinity of the equilibrium point, basis of the topologic retracting method, WaĪewski meeting the conditions: [18]. In this method, developed by W. Bogusz [9], x V (x, t) > 0 , for x z 0 the zones bounded by curves of input and output x points of the system (2) from areas considered to be along solutions (4) , for x V (x, t) d 0 permissible - are constructed. x  G - Z (9) As it results from the synthetic presentation of x V (x1 , t1 ) < V (x2 , t 2 ) for x1  Z and x2 the examination methods of the technical stability  G - : ; t 1 < t 2 [9], their application in constructing the monitoring then the object described by (2) and meeting system is related to three tasks [5]: conditions (3-5) is technically stabilised. 1. Identification of the dynamic model of the Referring the results of this law to the problem monitored construction node of the object under of realisation assumptions for the monitoring inspection. systems of the machine state, the Lapunow’s 2. Creation of the measuring instrumentation function Vxt(,) should be formulated and by protecting observations of trajectory changes of means of the experimentally determined trajectories the dynamic behaviour of the monitored object x, y of the object under testing, the conditions given node, determined by the measurement: by Equation (9) verified. x(),txty () Essential, helpful element in the process of 3. Constructing quantifiers for the monitored building a state classifier can be the results of the courses by the implementation of algorithm of DIAGNOSTYKA’ 3(47)/2008 141 BATKO, Weight And Significance Of Object Dynamic Models In Constructing Monitoring Systems

examining the technical stability, based either values as well as formation of estimation on the method of the Lapunow’s function, or procedures of their exceeding the permissible two functions )(x, y) and \(x, y), or the values. retracting method. As it seen from the short review given above, The solution of the first and second task does several problems related to constructing the not generate significant difficulties in the realisation assigned algorithms remains still unknown and of the monitoring system. More difficult is constitutes an interesting research field. The author assigning of positively or negatively determined hopes, that the indicated research idea will be Lapunow’s function to the dynamics describing the developing and the results will produce more monitored object, given by the identified set of effective rules of monitoring the state changes in differential equations. machines, structures and devices. An essential element of estimation the usefulness of phase trajectories in the defect 5. REFERENCES recognition process is the requirement of tracing changes in the trajectory shape of the inspected 1. Banek T., Batko W.: Time Estimation of processes. Its realisation, in the case of using the Disturbance Occurrences in Monitoring currently available – in the market – monitoring Systems. Zeitschriftt f. angew. Math. u..Mech. systems for inspecting vibrations of machines, (ZAMM), 77, s.1-8, 1977. devices or constructions, would require the 2. Banek T., Batko W.: Method of Supporting reorganisation of their functioning. There would be Functional in the Estimation of Alarm a need of securing the possibility of measuring Conditions in Systems of Vibration Monitoring. phase trajectories, which means measuring Mechanika Teoretyczna i Stosowana No 4, Vol. displacements and vibration velocities on the 32, 1994. pp .931-944. selected constructional nodes of machines. Analysis 3. Banek T., Batko W.: Estimation of Disturbances of their changes is the basic information carrier, in Monitoring Systems, Publishing House AGO, which enables tracing the technical stability loss of Kraków, 1997 (in Polish). the monitored object, caused by inadmissible 4. Batko W.: Development of monitoring solutions disturbances in its proper functioning. weakly-sensitive for disturbances. Report of the Thus, an essential task in the estimation of KBN Project No PB 582/3/91. 1991-1993 usefulness of the technical stability theory in the (Warsaw University of Technology) (in Polish). development of diagnostic systems is the 5. Batko W.: Stability in Technical Diagnostics, verification of behaviours of phase trajectories of Diagnostyka No 4 (44), 2007, p. 69-72. the monitored objects under conditions of defect 6. Batko W.: Technical stability – a new modeling occurrences [9, 10]. Computer simulation perspective for building solutions of monitoring experiments provide certain directions and systems for machine state. Zagadnienia information [8] on changes of phase trajectories Eksploatacji Maszyn, z. 3 (151), Vol. 42, s.147- caused by defects. 157, 2007. 7. Batko W., Banek T.: Noise filtering in 4. CONCLUSIONS monitoring systems. Journal of Applied Mathematics and Computer Science 1993,vol.3, The author’s approach to tasks of designing and No.3, 509-517. constructing of the systems monitoring changes of 8. Batko W., Dąbrowski Z., KiciĔski J.: Nonlinear machines, devices and structure states – is phenomenal in Vibroacoustics Diagnostic, presented in the paper. The attention was directed Publishing House of the Institute of exploitation towards the importance of the dynamic model Technology (ITE) Radom 2008. (in Polish). identification process of the monitored object. Its 9. Batko W., Majkut L.: Damage Identification in presence in the constructing procedure of the Prestressed Structures Using Phase monitoring system preserves - in an universal and Trajectories, Diagnostyka, No 4(44), 2007, p. coherent way - the realisation process of diagnostic 63-68. tasks. It enables the possibility of the simulation 10. Batko W., Majkut L.: The Phase Trajectories as base formation for the recognition of inefficiency of the New Diagnostic Discriminates of Foundry the monitored objects, protects the optimal filtration Machines and Devices Usability, Archives of and prediction of changes in the processes under Metallurgy and Materials, Vol. 52, 2007. inspection and formulates methodological 11. Bogusz W.: Technical Stability. PWN, guidelines for the realisation of the monitoring Warszawa, 1972 (in Polish). process. 12. Bogusz W.: A Two Tensor Method for Application of methods of technical stability Investigation Nonlinear Systems, Proceedings of investigations for assessments of dynamic Vibration Problems, Warszawa, 2, No 3, 1961. behaviours of the tested object generates selection 13. Busáowicz M.: Resistant Stability of Linear indications of symptoms necessary in the Dynamic Stationary Systems with Delays, monitoring system, determination their criteria Publishing House of the Biaáystok Technical 142 DIAGNOSTYKA’ 3(47)/2008 BATKO, Weight And Significance Of Object Dynamic Models In Constructing Monitoring Systems

University, Warszawa-Biaáystok, 2002 (in 20. Zabczyk J.: Exit Problem and Control Theory. Polish). Systems and Control Letters, 6, 1985, pp.163- 14. Byrski W.: Observations and Control in 172. Dynamic Systems. Scientific and Didacting Publishing House, AGH, Kraków, 2007 (in Polish). Prof. dr hab. inĪ. Wojciech 15. Kalman R. E.: A new approach tolinear filtering BATKO – Head of and prediction problems. Jour. of Basic Eng. Department of Mechanics Trans. Of ASME, vo. 82D, 1960. and Vibroacoustics of the 16. Mitkowski W.: Stabilisation of Dynamic University of Science and Systems. Publishing House of AGH, Kraków, Technology (AGH) in 1991 (in Polish). Krakow. He is researching 17. Stability Problems in Discrete Systems, Team the problems of dynamics of work, Publications of IPPT PAN, No 49, 1970 mechanical objects and (in Polish). related to them analyses and 18. WaĪewski T.: Sur un principe topologique de modeling of vibroacoustic processes for the needs l’examen de l’allure asymptotique de integrals of constructing diagnostic systems. He is an author de equations differentiates ordinaries, Ann. of more than 200 papers, including authoring or co- Soc. Polon. Math., No. 20, 1947. authoring of 12 books. He is a member of the Machine Building Committee and the Committee of 19. Zabczyk J.: Stable Dynamical Systems under Acoustics of the Polish Academy of Sciences. Small Perturbations. Preprint 353, December 1985, Institute of Mathematics PAN. DIAGNOSTYKA’ 3(47)/2008 143 UHL, Mechatronics In Diagnostics

MECHATRONICS IN DIAGNOSTICS

Tadeusz UHL

AGH University of Science and Technology, al. Mickiewicza 30, 30-059 Kraków, email: [email protected]

Sumamry This paper deals with current problems of diagnosing mechanical structures. The structural diagnostics have been presented as an interdisciplinary area integrating the fields of mechanics, electronics, computer science and materials engineering. The analogy between mechatronics and structural diagnostics will be discussed. The diagnostic methods contain three basic elements: measurements, including sensors designing and signal processing, procedures and diagnostic algorithms as well as structure properties prediction. The main issues within monitoring and structural diagnostics will be presented in the paper.

Keywords: mechatronics, diagnostics, structural health monitoring.

MECHATRONIKA W DIAGNOSTYCE

Streszczenie W pracy przedstawiono wspóáczesna problematykĊ diagnozowania konstrukcji mechanicznych. Przedstawiono diagnostykĊ konstrukcji jako dziedzinĊ interdyscyplinarną, stanowiącą integracjĊ mechaniki, elektroniki, informatyki i inĪynierii materiaáowej. Pokazana zostanie analogia pomiĊdzy mechatroniką, a diagnostyką konstrukcji. Metody diagnostyki zawierają trzy podstawowe elementy; pomiary, w tym konstrukcja czujników, przetwarzanie sygnaáów; metody i algorytmy oraz przewidywanie wáasnoĞci konstrukcji. W ramach referatu przedstawione zostaną podstawowe problemy z dziedziny monitorowania i diagnostyki konstrukcji.

Sáowa kluczowe: mechatronika, diagnostyka, monitorowanie stanu konstrukcji.

1. INTRODUCTION and operational solutions typical of this new field of structure testing. SHM constitutes a combination of New terms have recently appeared in the such fields as sensorics, teleinformatics, electronics technical literature: SHM (Structural Health (microprocessor engineering in particular), Monitoring) and structural damage detection [1, 3, mechanics and materials engineering [1, 7]. It has 48]. Structural damage consists in such changes been schematically presented in fig. 1. introduced into structure properties that have led or Systems implementing SHM processes should be will lead in the future to the loss of operating embedded in the structure and act independently. properties. This term includes continuous Installation and use of such systems bring two kinds monitoring of structure material condition (in real of benefits: time) for structure elements as well as for the entire x for design engineers: it enables introduction of structure conducted during its operation. The structural changes so that the breakdown structure material condition should meet the probability is minimal, requirements specified in the design process. The x for users: guidelines for such management of requirements concerning material condition must structure use to minimize breakdown risk, take into account changes caused by normal wear treating the structure as a part of a bigger system. and tear during operation, changes caused by the As it is shown in the presented diagram of SHM influence of environment in which the structure is process, its structure is very similar to mechatronics used and accidental events affecting material and it is completed with issues related to materials condition. Due to the fact that the monitoring is engineering. carried out in a continuous way during operation, The main functions of SHM system are as full operation history is available and these data can follows: be used to forecast condition, breakdowns and time 1. Monitoring of structural coherence (e.g. material of structure safe use. In several titles, SHM is condition); defined as a new method of Non-Destructive Testing x A function connected with the type of physical [3, 15, 35]. The novelty of this approach consists in phenomenon that constitutes a quantity continuous monitoring of material condition while measured by sensors and a failure mode; operating a given device. It requires some structural 144 DIAGNOSTYKA’ 3(47)/2008 UHL, Mechatronics In Diagnostics

x A function connected with the type of physical Two fundamental approaches may be currently phenomenon used by built-in sensors for signal listed in this area: statistical and based on image generation; recognition techniques. They support the decision- 2. Monitoring of structure wear and tear – load making process on the current condition of the cycles, influence of environment; monitored structure. Due to the fact that 3. Prognosis of residual lifetime of the structure contemporary SHM systems are embedded in the (function1+function2+damage mechanisms); structure, the selection of inference algorithms and 4. Managing the structure condition (maintenance analysis of available measurement information must activities, repair works, etc.). be specialized and possible to implement on processors with limited capacity. A review of these procedures can be found in the next chapter of this Material Mechanics paper. The highest level of SHM systems contains Science procedures of forecasting condition of a monitored object. It is, undoubtedly, the most difficult and SHM error-prone process implemented within SHM systems. The suggested diagnostic procedure has been presented in figure 2. Informatics Electronics As the above diagram shows, the crucial role in the diagnostic process is played by two models: a global model describing structure behavior and a local model within simulation of material behavior, its degradation or other phenomena Fig. 1. Diagram of fields’ integration in SHM connected with loss of properties. systems Application of such systems is encouraged by: x possibility of avoiding catastrophic breakdowns, As a result of this multifunctionality, SHM x possibility of optimizing the operational process systems have a hierarchical structure [9, 19, 28]. The (minimization of emergency shutdowns), first layer of the SHM system is a monitoring layer x obtaining information for design engineers, determined by the type of physical phenomenon which is necessary for structure modification, monitored by sensors and connected with structure x possibility of minimizing maintenance costs and damage and by the type of physical phenomenon increase in the device availability thanks to used by sensors in order to generate a signal, most applying condition-based repair methodology frequently an electric one, containing information and avoiding disassembly and replacement of about damage and subject to recording and undamaged or unworn elements, processing. A few or a few dozen sensors usually x possibility of avoiding operator’s mistakes in work in such a system. They are connected and structure condition assessment. create a network of sensors measuring also environmental factors that affect the structure STRUCTURE condition and operation. Most frequently considered Actuator Signal processing / system structural sensors currently include: light pipe /Sensor identification / model updating technology-based sensors [2, 49], the so-called

Bragg gratings used in aviation and construction Model of the Damage detection Loads industry, wireless sensors and their networks [4, 11, structure (FE) and localization identification 13, 18, 21, 23, 47, 51, 59], sensors in the form of piezoelectric elements embedded in the structure Model of materials SIMULATION (multiscale) [11, 18, 23, 47] and vision systems based on classic cameras [21, 59]. Damage prognosis/ Data from all the sensors and historical data rest life prognosis along with data from other, similar structures enable Fig. 2. Diagram of SHM process a fusion of diagnostic information on the structure condition (signal fusion). By linking this information Such systems are used in air force and civil with data from the base on structure damage and aviation, military equipment, building infrastructure wear phenomena, a forecast of its condition and data as well as critical machines in the industry (e.g. on the range of necessary repair can be obtained. power engineering, chemistry, etc.) [1, 37, 48]. More and more frequently, simulation systems are Economic reasons constitute a very significant applied for these purposes. The systems enable factor influencing the universality of SHM systems’ |a very fast generation of results similar to those use. Such justification can be found in several papers obtained from the network of sensors on the basis of [1]; it consists in comparing maintenance costs and known damage models (structure virtual operation). structure availability. In the case of a structure with Methodology of data collection and selection of no SHM systems installed, the costs increase with information on an object condition constitutes operation time but the structure availability a crucial element of SHM procedures [15, 19, 44]. DIAGNOSTYKA’ 3(47)/2008 145 UHL, Mechatronics In Diagnostics decreases. Installation of such systems enables to keep constant maintenance costs accompanied by Sensors/ actuators constant availability of the structure. There is, however, one precondition for applying SHM systems which limits universality of their Microprocessor / embedded implementation. The cost of the system itself must be lower than the positive economic result connected Algorithms with its use. Software The necessity to reduce the costs of SHM systems is currently related to the application of SHM Design intelligent materials and structures [3]. They enable NDT to integrate the structure and built-in sensors into one system. To make such activities effective, they need to be undertaken at the stage of structure designing. Tendencies to apply intelligent structures have been observed since early 80s, particularly in aviation and construction industry. Their Fig. 3. Diagram of structural health monitoring characteristic feature is adaptation of these (SHM) process – basic elements of the process structures’ properties to operational conditions. This adaptation takes place independently in intelligent What is typical of SHM systems, it is their structures. Intelligent structures include: structures integration with a structure. Very often, sensoric sensitive to operational conditions, structures elements constitute a part of the structure performing controllable within their properties and auto- also other functions; it is usually made by using adaptive structures adjusting their properties to the intelligent materials. operational needs. In practice, homogeneous Summing up the above review of the issues materials frequently applied in structures are connected with SHM, one may say that it is a new replaced by composite materials or other interdisciplinary field concentrating such detailed multimaterials (materials composed of layers with areas of science as mechatronics, materials different physical properties). Intelligent materials engineering, electronics, computer science, physics, and structures include: structures with controlled optics and many other, applied in operation of geometry (shape), structures with controlled structures in aviation, construction industry, vibrations and structures with controlled condition. automotive industry and power engineering, The last structure type, in particular, is closely including nuclear power engineering. The range of connected with SHM techniques [1, 4, 27, 40, 56]. It application is wider and wider. is most often expressed in the form of embedded sensors made of intelligent materials or embedded 2. SENSOR TECHNOLOGY FOR actuators aiming to soften the effects of STRUCTURE DIAGNOSTICS a breakdown. The purpose of such actuators is to generate the structure deformations in order to Development of SHM methods and applications reduce stresses in the areas of their concentration results from a rapid development of measurement [34, 55]. Nowadays, scientists are searching for techniques enabling collection, archivisation and phenomena and methods of their measurement that processing of measured signals. The crucial role in would enable to conduct continuous structural health these systems is played by sets of sensors that enable monitoring through monitoring of the structure to directly measure the quantities connected with material condition. material and structure degradation phenomena. Research within SHM systems’ development is Sensors of this type include: optical sensors, very often inspired by discoveries in biology and piezoelectric sensors and contactless sensors used biomimetics [8]. Very similar research is conducted more and more frequently to test and analyze in medicine and SHM methods’ development [16]. structural health. Another method of sensors’ SHM systems can be applied not only during classification is the manner of transferring structure operation but also during its manufacture, measurement results to the system of data collection transport and assembly. They enable proper and analysis. This division includes wireless sensors management of structure lifetime through a suitable and sensors connected to the system by means of selection of mission that can be performed and cables. Due to the size of monitored structures and maintenance activities it needs to safely meet all the nature of currently built monitoring systems in the requirements imposed. More and more frequently, form of dispersed layouts, application of wireless specialists search for methods that could be used in sensor networks is more and more frequent. each stage of a product lifetime, as most Optical sensors [2] used in SHM are based on economically effective and easiest to implement. light pipe techniques whose development is connected with rapid growth in application of light pipes in telecommunications. The advantages of light pipes used for measurements in SHM systems 146 DIAGNOSTYKA’ 3(47)/2008 UHL, Mechatronics In Diagnostics include: resistance to electromagnetic interference, where: [ is photoelasticity constant (1.13 pm/PH for very small weight, high definition, wide transfer the light 1550 nm), H is average strain on the grating band, large information capacity and relatively low length. cost. Application of light pipes to implement sensors The measurement of strains consists in tracking can be divided into three categories: the displacement of wave length for the given 1. Sensors based on light intensity measurement grating. 2. Sensors based on phase modulation The initial modulation of a light pipe is given in (interferometers) the production process. For one light pipe, one can 3. Sensors based on light wave length modulation. place as many as 3000 gratings, obtaining the The first group of sensors includes proximitors to possibility of distinguishing from which grating the measure displacements of whirling shafts [37]. given wave length comes. In practice, the used strain These sensors enable to measure displacements with sensors have a definition (theoretically) of approx. a definition better than 3 Pm. Another example of a few nH, however due to the influence of a light intensity measurement-based sensor is temperature, which should be compensated, their a pressure sensor, in which the light pipe is located definition amounts to 1 PH. Particular solutions between two rough surfaces subject to loading differ in the method of measurement of wave length resulting from differential pressure between them. displacement. The most frequently used ones As a result of surface irregularity, the light pipe is include: WDM interferometer, Fabry – Perota filter, bent and one can observe bigger optical losses than optoacoustic tuned filter, CCD linear sensor, hybrid in case of unbendable light pipe. Thus, it is possible tuned filter. The choice of measurement method to measure pressure. Various light sources as well as depends on the accuracy required and frequency of various light detectors are applied in these sensors. changes of the measured quantity. In the majority of Another group of sensors enable currently the Bragg grating cases, the measuring band is limited to most detailed laboratory distance measurements. 100 Hz. This frequency is completely sufficient to The principle consists in measuring the power of monitor structure condition. Light pipe sensors with optical signal of a monochromatic light wave Bragg gratings are used as elements embedded in reaching the detector inside the light pipe. The composite structures, in reinforced concrete power of the signal of two waves overlapping each structures to monitor strains in reinforcement. The other and running inside the light pipe (a wave advantages of sensors constructed in this way generated by the source and a reflected wave) include a possibility of application in explosive depends on the phase displacement between the environments, long-lasting stability of work, small waves and for a displacement by 180o it may equal dimensions, small weight, possibility to multiplex zero. The change in one wave path against the other sensors on one light pipe (usually, 10 Bragg gratings results in their mutual phase displacement and it on one light pipe), relatively low cost of the sensor indirectly changes the power of the signal reaching and resistance to strong electromagnetic fields. Light detector. Light pipe, interferometric sensors enable pipe sensors are used to measure distribution of distance measurement with a definition of up to structure strains in such objects as bridges, aircrafts 10 nm. and ships. Possibility of tracking local changes in Another area of activity of light pipe-based strains’ distribution constitutes an advantage as sensors is application of Bragg gratings [3]. Bragg regards SHM systems. Several structure damages in grating placed inside the light pipe, most frequently the initial nucleation stage do not modify global on the length of 10 mm, constitutes a periodic strains in a measurable way but only cause some modulation of refraction coefficient. It leads to light slight local changes. wave diffraction and consequently, to the change in Research on development of sensors using light the optical wave length according to the diffraction pipes with Bragg gratings on them is conducted very (Bragg’s) law for a wave running in the light pipe: intensively all over the world. These sensors have Ob 2neT a wide range of described commercial applications in SHM systems. where: O is the length of a light wave after going b Other sensors largely used in SHM systems are, through Bragg grating, ne is the averaging refraction due to the possibility of easy integration with coefficient, T is the period of light pipe properties’ monitored structure, piezoelectric sensors [1]. modulation. The techniques using piezoelectric sensors may In the case of a change in properties’ modulation be classified into three groups: (Bragg grating period), the length of a light wave is 1. Passive techniques, also changed. It means that Bragg grating acts as an 2. Active techniques, optical filter whose parameters depend on elongation 3. Mixed techniques. or shortening of the light pipe measuring length. The These techniques have been schematically dependence between relative strains and change in presented in fig. 4. wave length may be expressed in the following form:

'Ob Ob[H DIAGNOSTYKA’ 3(47)/2008 147 UHL, Mechatronics In Diagnostics

piezoelectric constant, Ho is permittivity of free Mechanical systems Active system space. External The efficiency of piezoelectric materials can be measurement Actuators measured by means of electromechanical coupling systems coefficient K2, determined from the following Czujnik dependence: e2 Mechanical systems K 2 Passive system EH m External where: e is piezoelectric constant, E is Young’s measurement Sensors systems module, e is piezoelectric material permittivity. The advantage of piezoelectric sensors is that Embedded they generate loads with no power supply, and the system disadvantage is that they work only with dynamic Embedded loads and below Curie temperature. In practice, measurements piezoelectric materials are not ideal dielectrics and systems this is the reason for difficulties in measuring very Power supp low frequencies (leakage conductance of loads). Piezoelectric sensors play a significant part in Fig. 4. Schematic presentation of typical structural health monitoring. They are applied to SHM systems measure structure vibrations within low frequencies (0- 20 kHz), within ultrasound frequencies (above The SHM systems presented in fig. 4 use various 20 kHz), as well as those connected with acoustic physical phenomena taking place in structures. The emission (above 100 kHz). In continuous monitoring phenomena most frequently applied in SHM systems systems, which are typical of SHM, piezoelectric include: acoustic emission [1], propagation of sensors are embedded in the structure. Such ultrasounds (including Lamb waves) in structure solutions are used in both methods based on [17] and a change in electromechanical impedance vibration measurement and active methods of as a result of damage [34]. condition monitoring based on generation and One of the advantages of applying piezoelectric measurement of waves in structures. elements is the possibility to use them as sensoric In case of many monitored structures, the cost of elements measuring structure response and monitoring system is high due to the necessity of simultaneously, as elements exciting vibrations installing cables that supply sensors and enable within high frequencies. Excitation conducted by signal transmission from the sensor to the data piezoelectric systems is characterized by relatively collection station. In order to reduce these costs, considerable force but small displacement. several companies and research laboratories work on Piezoelectric sensors are made of materials wireless transmission of data [28, 51]. Recently, showing piezoelectricity effect connected with there has been a significant increase of interest in electric charges generation as a result of stresses wireless sensors. Moreover, we can observe taking place in the material. Such properties are irregular development of such sensors’ production typical of quartz crystals and different kinds of technology. ceramic materials, the most popular ones include Modern wireless sensors constitute an lead zirconate titanate (PZT), barium titanate, independent system of data collection with a built-in polyvinylidene fluoride (PVDF). microprocessor dedicated to measuring data For a typical piezoelectric element with the processing and to managing wireless communication shape of a disc of d thickness, W voltage generated as [51]. The sensors are very often equipped with a result of stress is expressed by the following systems to recover energy from vibrations by means formula: of piezoelectric effect [42, 43]. Such a system is V gdW integrated with a sensoric element which enables to where: g is the voltage coefficient defined as a ratio minimize the amount of information sent. An outline of voltage to stress for a unitary thickness of a disc. of a wireless sensor has been presented in fig. 5. The dependence between dipole moment and Power mechanical deformation of a piezoelectric element Supply may be expressed by means of the following formulas: Sensors elements + Micro- Actuators elements A/D Converter computer + D/A Converter V EH  eEo

D H o Eo  eH Communication module- receiver / where: V is mechanical stress, H is relative strain, Eo transmitter is electric field intensity, D is density of electric field flux, E is material Young’s module, e is Fig. 5. Schematic structure of a wireless sensor 148 DIAGNOSTYKA’ 3(47)/2008 UHL, Mechatronics In Diagnostics

Technological development of wireless sensors applied to structural health monitoring goes towards Problem formulation implementation of monitoring algorithms directly in the sensors and of the so-called local SHM Frequency range setting algorithms. Operational modal analysis 3. RESEARCH METHODS IN STRUCTURE DIAGNOSTICS Influence of structural Influence of excitation properties V>>0 forces s=0 There are three groups of SHM methods: vibration measurement based methods, methods Damage localization based on testing wave phenomena and methods using other phenomena accompanying structure operation, e.g. generation of a changeable Experimental modal analysis, Inverse identification problem- temperature field as a result of structure damage scaled modal parameters load identification [46]. The use of active methods is more and more estimation, modification simulation frequent. These methods consist in generating phenomena whose course is influenced by potential damage in the tested structure. The basic methods Service/ maintenance within structure diagnostics include: • Methods based on vibrations measurement (low- Source of excitation frequency, structure modal parameters) [6, 8, 12, Structural modification modification, energy transfer path modification 40, 36] • Methods based on measurement of elastic waves Fig. 6. Diagram of structure monitoring and propagation (the most frequently used ones diagnostics method based on a modal model. include Lamb waves) [22, 1, 48, 17, 38, 39, 61]; • Methods based on strain measurement (more and One of the main methods in SHM practice is more often, with the use of light pipe technique – a method based on tracking changes in structure Bragg grating) [2]; modal parameters, as it is shown in fig. 6. However, • Methods based on measurement of other in many practical cases, such environmental phenomena (more and more frequent parameters as a change in ambient temperature, application of active thermovision) [46]; humidity, etc., have a greater influence on structure • Methods based on impedance measurement parameters than the arising damage. Therefore, this (measurement of input function and response in influence should be filtered out. A modal filter one point) [34]. constitutes one of the methods that enable to filter The listed methods are not the only ones used in environmental parameters’ influence on structure practice. They are, however, most frequently applied modal parameters. The basis of modal filter theory methods with existing and available monitoring has been presented for the first time in [31, 57, 60]. systems. Its application to diagnose structure condition has A model based method schematically presented been presented in [8, 33]. The idea of modal filter in fig. 6 is one of the methods more and more often method consists in transforming physical applied in structure monitoring [32]. coordinates (generalized) into modal coordinates This method is more and more widely used due connected with particular forms of structure to development of operating modal analysis vibrations. For this purpose, new modal vectors have techniques and loads identification methods basing been defined – the so-called reciprocal modal on the solution to the identification inverse problem vectors. According to the definition, reciprocal [15, 52, 54, 50]. modal vectors are orthogonal towards all modal vectors except for one on which the filter is tuned. 4. DAMAGE DETECTION ALGORITHM Thanks to that, modal filter may be used to disintegrate the system response to particular modal Several various algorithms used to detect damage in structures have been described in literature. As coordinates Kr. T regards their use in SHM systems, these should be Kr Z

Rrpp j ˜1 (2) DIAGNOSTYKA’ 3(47)/2008 149 UHL, Mechatronics In Diagnostics

In the next step, it is assumed that the transition can be decentralized which means that it can be used spectral function Hpp(Z) for the system with one locally for particular sensors independently. degree of freedom is expressed in the following Secondly, it does not require great design power. form: This algorithm is dedicated to be used in embedded systems. The algorithm first step is estimation of R R * parameters of AR (ARX) models, which can be rpp rpp (3) H pp Z  conducted using the least squares method, known jZ  O j * r Z  Or from the linear systems theory: where: O  r pole of the system. T 1 T U 4 ) ) ) Y (7) For the given frequency range, the above >@ transition function is determined by the vector of where: 4is parameters’ vector, F is regression k value: matrix containing results of input function T measurements, Y is output vector. H Z H Z H Z  H Z (4) pp >@pp 1 pp 2 pp k Thanks to the knowledge of regressive model Assuming that there is one input function in the parameters for an undamaged structure and during system and that the response is measured in N points operation, their difference may be determined. on the structure, the matrix of transition function has Various kinds of metrics are applied here. One of the the value k x N; possibilities is application of difference standard

ªH1 Z1 H 2 Z1 H N Z1 º deviation. If it exceeds a certain, specified level, «H Z H Z H Z » (5) damage arises in the structure. « 1 2 2 2 N 2 » H kN Z In another approach [6, 24, 45], basing on the «  » « » knowledge of AR (ARX) model parameters, discrete ¬H1 Zk H 2 Zk H N Zk ¼ system poles are determined solving a characteristic On the basis of the transition function matrix equation. Basing on the knowledge of these poles, formulated in this way, the matrix of reciprocal structure modal parameters are determined in the modal vectors

6. SUMMARY 11. Fritzen C. P., Mengelkamp G., Guemes A.: A CFRP Plate with Piezo-electric Actuators and Basing on the presented review of structural Sensors as Self Diagnosing Intelligent Structure, monitoring and diagnostics issues, this research area in Proceedings of the International Conference can be treated as an interdisciplinary field in which on Noise and Vibration Engineering, Leuven, correct integration of a monitoring system with Belgium, 2002, pp. 185 191. structure plays a more and more significant role. 12. Fugate M. L., Sohn H., Farrar C. R.: Vibration- More and more frequently, implementation of this Based Damage Detection Using Statistical process involves application of built-in systems that Process Control, Mechanical Systems and Signal work independently and are fully embedded in the Processing, 2001, 15 (4) pp.707–721. monitored object. Therefore, structural monitoring, 13. Giurgiutiu V., Buli X., Cuc A.: Dual Use of diagnostics and mechatronics have several common Traveling and Standing Lamb Waves for features whereas methods and tools used in Structural Health Monitoring, in Proceedings of mechatronics are largely applied in monitoring and the 23rd International Modal Analysis diagnostics. Conference (IMAC XXIII), Orlando, Florida, USA, 2005, pp. 361 – 368 BIBLIOGRAPHY 14. Hermans L., Van der Auweraer H., Coppens P., Mathieu L.: Modal Parameter Extraction from 1. Adams D. E.: Health Monitoring of Structural In-Operation Data, XV IMAC, Orlando, pp. Materials and Components, Wiley, Chichester, 531–539, 1996. 2007. 15. Inman D. J., Farrar C. R., Steffan V. Lopes (ed.): 2. Ansari F.: Fibre Optic Sensors for Construction Damage Prognosis for Aerospace, Civil and Material and Bridges, Lancaster, PA, Mechanical Systems, Wiley, New York, 2005. Technomic, 1998. 16. Karbhari V. M., Guan H., Sikorsky C.: Web- 3. Balageas D., C. P. Fritzen, A. Gumes (Ed.), based Structural Health Monitoring of a FRP Structural Health Monitoring, ISTE, London, Composite Bridge, in Proceedings of the 2006. International Conference on Structural Health 4. Basheer M. R., Rao V., Derriso M.: Self- Monitoring and Intelligent Infrastructure, Tokyo, organizing Wireless Sensor Networks for Japan, November 13–15, Vol. 1, 2003, pp.217– Structural Health Monitoring, in Proceedings of 225. the 4th International Workshop on Structural 17. Kessler S. S., Spearing S. M., Soutis C.: Damage Health Monitoring, Stanford, CA, September 15– Detection in Composite Materials using Lamb 17, 2003, 1193–1206. Wave Methods, Smart Materials and Structures, 5. Beynon J. H., Das S., Howard I. C., Shterenlikht 2002, 11(2) pp. 269–278. A.: Extending the Local Approach to Fracture; 18. Kim J. S., Vinoy K. J., Varadan V. K.: Wireless Methods for Direct Incorporation of Health Monitoring of Cracks in Structures with Microstructural Effects into Finite Element MEMS–IDT Sensors, in Smart Structures and Model of Fracture, Proc. Conf. ASME2002, ed., Materials, San Diego, CA, March 18, Brust F. W., Vancouver, 2002, 229-237. Proceedings of the SPIE, Vol. 4700, 2002, pp. 6. Bogacz M., Uhl T.: Real-time Modal Model 342–353. Identification and Its Application for Damage 19. Klein L. A.: Sensor and Data Fusion: Concepts Detection, Proceedings of ISMA 2004 and Application, SPIE Press, 1999. Conference, Leuven, (2004), pp. 1065- 1075. 20. Klepka A., Uhl T.: Application of Wavelet 7. Chang P. C., Flatau, A., Liu S. C.: Review Transform to Identification of Modal Parameters Paper: Health Monitoring of Civil of Nonstationary Systems, J. of Theoretical and Infrastructure, Structural Health Monitoring, Applied Mech., Vol. 43, (2005), pp. 277-296. Vol. 2 (3), pp. 257-267, 2003 21. Kohut P, Holak K, Uhl T.: Structural Health 8. Deraemaeker A., Preumont A.: Vibration-based Monitoring Using a Vision System, Materials of Damage Detection Using Large Array Sensors XIVth School of Modal Analysis, Cracow, 2008. and Spatial Filters, Mechanical Systems and 22. Leonard K. R., Malyarenko M. K.: Hinders, Signal Processing, Volume 20, 2006. Issue 7, Ultrasonic Lamb Wave Tomography, Inverse 1615-1630. Problems, vol. 18, 2002, pp. 1795–1808. 9. Doebling S. W., Farrar C. R., Prime M. B.: A 23. Lin M., Chang F.-K.: The Manufacture of Summary Review of Vibration-based Damage Composite Structures with a Built-in Network of Identification Methods, Shock and Vibration Piezoceramics, Composites Science and Digest, Vol. 30, No. 2, 1998, pp. 91–105. Technology 62 (2002), pp. 919–939. 10. Farrar C. R., Sohn H., Robertson A. N.: 24. Ljung L., Söderström T.: Theory and Practice of Application of Nonlinear System Identification to Recursive Identification, MIT Press, Cambridge, Structural Health Monitoring, in Proceedings of 1985. the 2nd European Workshop on Structural Health 25. Lu Y., Gao F.: A Novel Time-Domain Auto- Monitoring, Munich, Germany, 2004, pp. 59– Regressive Model for Structural Damage 67. DIAGNOSTYKA’ 3(47)/2008 151 UHL, Mechatronics In Diagnostics

Diagnosis, Journal of Sound and Vibration, Vol. Detection in Metallic Structures, Proceedings of 283, 2005, pp. 1031–1049. the Third European Workshop on Structural 26. Lucero J., Taha M. M. R.: A Wavelet-Aided Health Monitoring, Granada, 2006, pp.275–282. Fuzzy Damage Detection Algorithm for 40. Silva J. M. M., Gomes A. J. M. A.: Crack Structural Health Monitoring, in Proceedings of Identification of Simple Structural Elements the 23rd International Modal Analysis through the Use of Natural Frequency Conference (IMAC XXIII), Orlando, Florida, Variations: The Inverse Problem, in Proceedings USA, 2005, pp.. 78 – 84. of the 12th International Modal Analysis 27. Luo H., Hanagud S.: Dynamic Learning Rate Conference (IMAC XII), Honolulu, Hawaii, Neural Network Training and Composite USA, 1994, pp. 1728–1735. Structural Damage Detection, AIAA Journal, 41. Simonian S. S.: Inverse Problems in Structural Vol. 35, 1997, pp. 1522–1527. Dynamics, Int. Journal on Numerical Methods in 28. Lynch J. P.: An Overview of Wireless Structural Engineering, vol. 17, pp. 357-365, 1981. Health Monitoring for Civil Structures, Phil. 42. Sodano H. A., Inman D. J., Park G.: A Review of Transactions R. Society A, vol. 365, 2007, pp. Power Harvesting from Vibration Using 345-372. Piezoelectric Materials, Shock and Vibration 29. Lynch J. P., Sundararajan A., Law K. H., Digest, Vol. 36, No. 3, 2004, pp.197–205. Kiremidjian A. S., Carryer E.: Embedding 43. Sodano H. A., Park G., Leo D. J., Inman D. J.: Damage Detection Algorithms in a Wireless Use of Piezoelectric Energy Harvesting Devices Sensing Unit for Operational Power Efficiency, for Charging Batteries, in Smart Structures and Smart Materials and Structures, Vol. 13, No. 4, Materials: Smart Sensor Technology and 2004, pp. 800–810. Measurement Systems, San Diego, CA, March 30. Friswell M. I., Penny J. E. T.: Crack Modeling 3–6, Proceedings of the SPIE, Vol. 5050, 2003, for Structural Health Monitoring, International pp.101–108. Journal of Structural Health Monitoring 2002, 1, 44. Sohn H., Farrar C.: Damage Diagnosis Using pp. 139–148. Time Series Analysis of Vibrating Signals, Smart 31. Meirovitch L., Baruh H.: Control of Self-adjoint Materials and Structures, vol.10, no.3, 2001, pp. Distributed Parameter System, Journal of 446-451. Guidance Control and Dynamics, 8 (6), 60-66. 45. Sohn H., Worden K., Farrar C. R.: Statistical 1982. Damage Classification under Changing 32. Mendrok K., Uhl T.: Overview of Modal Model Environmental and Operational Conditions, Based Damage Detection Methods, Proceedings Journal of Intelligent Material Systems and of 2004 ISMA, Leuven, Belgium. Structures, 2003, 13(9) pp. 561–574. 33. Mendrok K., Uhl T.: Application of Modal Filter 46. Szwedo M., Uhl T., Bednarz J.: Application of for Damage Detection. Proceedings of III Vibrothermography for Damage Detection in ECCOMAS Thematic Conference on Smart Mechanical Structure, Proc. of IV EW SHM, Structures and Materials, Gdansk 2007., Poland Cracow, 2008. 34. Park G., Sohn H., Farrar C., Inman D. J.: 47. Spencer B. F. Jr., Ruiz-Sandoval M. E., Kurata Overview of Piezoelectric Impedance Based N.: Smart Sensing Technology: Opportunities Health Monitoring and Path Forward, Shock and Challenges, Journal of Structural Control and Vibration Digest, vol. 35, 2003, pp. 451-463. and Health Monitoring, Vol. 11, No. 4, 2004, pp. 35. Park J. M., Kong J. W., Kim D. S., Yoon D. J.: 349–368. Nondestructive Damage Detection and 48. Staszewski W., Boller C., Tomlinson G. (Ed.): Interfacial Evaluation of Single-Fibers/ Epoxy Health Monitoring of Aerospace Structures, Composites using PZT, PVDF and P(VDF-TrFE) Wiley, Chichester, 2004. Copolymer Sensors, Composites Science and 49. Todd M. D., Nighols J. M.., Trickey S. T., Technology, Vol. 65, 2005, pp. 241–256. Seaver M., Nichols C. J., Virgin L. N.: Bragg 36. Parloo E., Guillaume P., Van Overmeire M.: Grating Based Fiber Optic Sensors in Structural Damage Assessment Using Mode Shape Health Monitoring, Phil. Trans. R. Soc. A, vol. Sensitivities, Mechanical Systems and Signal 365, 2007, pp. 317-345. Processing, Vol. 17, 2003, pp. 499–518. 50. Trujillo D. M.: Application of Dynamic 37. Pi Y. L., Mickleborough N. C.: Modal Programming to the General Inverse Problem, Identification of Vibrating Structures Using International Journal on Numerical Methods in ARMA Model, Journal of Engineering Mechanics Engineering, vol.23, pp.613-624, 1987. 115, 1989. 51. Uhl T., Hanc A., Tworkowski K., SĊkiewicz à.: 38. Ryles M., McDonald I., Ngau F. H., Staszewski Wireless Sensor Network Based Bridge W. J.: Ultrasonic Wave Modulations for Damage Monitoring System, Key Engineering Materials, Detection in Metallic Structures, Proceedings of 2007 vol. 347 pp. 499–504. the Third European Workshop on Structural 52. Uhl T., Lisowski W., Kurowski P.: In-operation Health Monitoring, Granada, 2006, pp.275–282. Modal Analysis and Its Application, AGH, 39. Ryles M., I. McDonald, F. H. Ngau, Staszewski Cracow, 2001. W. J.: Ultrasonic Wave Modulations for Damage 152 DIAGNOSTYKA’ 3(47)/2008 UHL, Mechatronics In Diagnostics

53. Uhl T., Petko M., Peeters B.: Embedded Systems Prof. Tadeusz UHL, Eng. for Real Time Flight Flutter Testing, in was born in Nowy Sącz in Structural Health Monitoring 2007, ed. Fu-Kuo 1956. He graduated from Chang, DSTech, Lancaster,2007, pp. 272-280. AGH University of Science 54. Uhl T., Mendrok K.: Identification Inverse and Technology in Cracow Problem and Its Application, ITE, Radom, 2005. and obtained the following 55. Wang K., Inman D. J., Farrar C. R, Modeling academic titles: PhD in 1983, and Analysis of a Cracked Composite Cantilever associate professor in 1990, Beam Vibrating in Coupled Bending and and professor in 1997. Torsion, Journal of Sound and Vibration, Vol. Currently, Prof. Uhl holds the 284, 2005, 23–49. Chair of Robotics and Mechatronics at AGH 56. Wu F., Chang F-K.: A Built-in Active Sensing Department of Mechanical Engineering and Diagnostic System for Civil Infrastructure Robotics. He is the author of over 400 publications, Systems, in Smart Structures and Materials, San including 16 textbooks and monographs. He deals Diego, CA, March 5–7, Proceedings of the SPIE, with the issues of structural dynamics, machine Vol.4330, 2001, pp.27–35. diagnostics, models identification, mechatronics and 57. Yan A. M., Kerschen G., De Boe P., Golinval J. construction of monitoring systems. C.: Structural Damage Diagnosis under Varying Environmental Conditions, Part I: A Linear Analysis, Mechanical Systems and Signal Processing, Vol. 19, 2005, pp. 847–864. 58. Zang C., Friswell M. I., Imregun M.: Structural Health Monitoring and Damage Assessment Using Measured FRFs from Multiple Sensors, Part II: Decision Making with RBF Networks, in Proceedings of the 5th International Conference on Damage Assessment of Structures (DAMAS 2003), Southampton, England, UK, 2003, pp.141–148. 59. Zaurin R., Catbas F. N.: Computer Vision Oriented Framework for Structural Health Monitoring of Bridges, in IMAC XXV. Society for Experimental Mechanics, 2007, Orlando, FL, pp. 213-218.] 60. Zhang Q., Allemang R. J., Brown D. L.: Modal Filter: Concept and Applications, Proceedings of International Modal Analysis Conference, Kissimmee, 1990, pp. 487-496. 61. Zhongqing S, Lin Y, Ye L.: Guided Lamb Waves for Identification of Damage Incomposite Structures: A review, Journal of Sound and Vibration vol. 295 (2006) pp.753–780 DIAGNOSTYKA’ 3(47)/2008 153 DĄBROWSKI, About Need For Using Nonlinear Models In Vibroacoustic Diagnostics

ABOUT NEED FOR USING NONLINEAR MODELS IN VIBROACOUSTIC DIAGNOSTICS

Zbigniew DĄBROWSKI

Institute of Machine Design Fundamentals, Warsaw University of Technology Narbutta 84, 02-524 Warsaw, fax: +48 22 234 8622, [email protected]

Summary The paper discussed the thesis that in vibroacoustic diagnostics using the nonlinear mathematical model is necessary. Natural specification of machine degradation is the frequency response function change and “additional” inputs related to defects remain usually as self-exited vibrations, which require the application of nonlinear description. Very often a new machine can be described with an adequate accuracy by a linear model. During exploitation certain nonlinear disturbances related to wear and tear - occur. Thus, an observation of nonlinear effects allows solving a diagnostic task.

Keywords: nonlinear effects, vibroacoustic diagnostic.

O KONIECZNOĝCI STOSOWANIA W DIAGNOSTYCE WIBROAKUSTYCZNEJ MODELI NIELINIOWYCH

Streszczenie W artykule przedyskutowano potrzebĊ stosowania w diagnostyce wibroakustycznej modeli nieliniowych. Naturalnym opisem degradacji maszyny jest zmiana funkcji odpowiedzi czĊstotliwoĞciowej, a „dodatkowe” wymuszenia związane z uszkodzeniami pozostają z reguáy jako drgania samowzbudne, co z zaáoĪenia wymaga zastosowania opisu nieliniowego. Bardzo czĊsto nową maszynĊ moĪemy opisaü z dobrą dokáadnoĞcią modelem liniowym. W trakcie eksploatacji pojawiają siĊ i wzrastają nieliniowe zaburzenia związane ze zuĪyciem. Tym samym obserwacja efektów nieliniowych pozwala na rozwiązanie zadania diagnostycznego.

Sáowa kluczowe: efekty nieliniowe, diagnostyka wibroakustyczna.

The problem of assessment of the machine state process occurs, are considered in the paper. by means of vibrations and noise analysis is based It should be assumed that those phenomena – from the theoretical side – on the statement that, are accompanied by increases of dissipation energy the vibroacoustic energy dissipation increases during in other processes, mainly thermal ones, but also the machine exploitation. Therefore a certain electric and hydraulic (e.g. an oil leakage from vibration or noise measure should exist, which in the the damaged bearing can cause damping moment - when further exploitation is dangerous of vibrations). In the complicated structure – exceeds the permissible value. Such reasoning of the mechanical system the effect of an apparent results from the adaptation of a model assuming “self repairing” can occur and it may periodically an increase of dissipated energy during wear and lower the amount of dissipated energy and change tear as well as on the assumption that energy proportions in between dissipation forms. of parasitic vibroacoustic processes is proportional It is presented pictorially in Figure 1. to the total dissipation of energy. Now-a-days the If, according to this short reasoning, we assume assumption of a general increase of energy that there are cases when an increased level consumed is not doubtful. We may assume that this of vibrations and noises (in the whole observable is the proved law of nature. However, there is still range or in the selected bands) is not proportional a problem of developing the easiest mathematical to a wear, it will still not indicate that vibroacoustic notation and looking for the “optimal” model. processes are insensitive to it. The assumption It is well known, that the statement arises, that the change of proportion in between of proportionality of vibroacoustic energy individual forms of the dissipated energy can to the total parasitic energy is a simplification from have its representation in the observed form which might be – and actually are – exemptions. of vibrations (even when the level remains Examples, where a periodical lowering constant or lowered). This phenomenon of the vibration level indicates a dangerous defect was investigated at analysing defects of rolling and where the “waving” of a trend of changing bearings, where – at the constant general level values being the measure of the vibroacoustic – the proportions between the dominating 154 DIAGNOSTYKA’ 3(47)/2008 DĄBROWSKI, About Need For Using Nonlinear Models In Vibroacoustic Diagnostics amplitudes in the spectrum were changing. Similar spectra, “a” and “b”, concern “normal” working results were obtained at checking hydraulic elements conditions. Thus, neither a loss of stability nor a loss when investigating a multi-symptom index of continuity of solution as a function of parameter (vibrations + heat) and at diagnostics of a tool wear changes occurs. Such situation would correspond in the machining process. The mentioned to the essential defect of a device. During the time phenomena have been observed during passive elapsing between both observations the system as well as active diagnostic experiments. Having an evolution could have occurred and probably have accurate recording of several symptoms – during done so. This evolution manifested itself by small the whole life-time – for the statistically changes of coefficients; it means elements of matrix representative sample of specimens one can of inertia, damping and rigidity and by establish significant dependencies. However, the an occurrence of new – generally weak – excitations difficult problem of assuming the proper model still and also by eventual decreasing of previous forces. remains to be solved. This last phenomenon can be omitted as being

Different Forms of of a low probability. Let us assume further, that Diagnostic Symptoms Energy Dissipation Energy Dissipation there are m poly-harmonic inputs (where m < n). E S E 1 1 The amplitude spectrum for a linear system

E2 S2 is formed as the result of the following transformation: E S 3 3 m

Yf() Pfiijjj ()˜ H (,,,) fmkc j 1y n (1) Vibroacoustic VA ¦ i 1 Energy Symptom where: Pi – Fourier transform of excitation moments, ? Hi – Transmittance.

Machine Condition Total Energy Symptom Each transmittance, and more accurate its module – it means the coefficient of amplification, has as many extremes as degrees Fig. 1. Utilising the diagnostic symptom of freedom of the system (natural frequencies). in a machine diagnostics. The set of transmittances is explicitly determined Let us discuss the example presented in Figure 2 by 3·m parameters. Equation (1) must be satisfied showing the spectrum of vibration accelerations for each frequency of the output spectrum. of the rolling bearing. An average value and average At assuming a spectrum discretisation into k square value are the same as measurement accuracy. elements we have k+2 equations. An addition of number 2 results from the postulate of the average a) b)

0.08 0.08 and average square value conservation. 0.07 0.07 At the assumption that the system inputs were not 0.06 0.06 (a) (b) changed and that all coefficients (parameters) 0.05 Y 0.05 Y ) ) f f ( ( 2 0.04 2 0.04 < < of mass, rigidity and dissipation could have changed, 0.03 0.03 the minimum number of degrees of freedom 0.02 0.02 0.01 < 2 0.01 of the linear system enabling transformation 0 0 0 100 200 300 400 500 0 100 200 300 400 500 ()ab () f [Hz] f [Hz] of the response Yf()o Yf ()equals: Fig. 2. Example of „iso-energy” evolution k  2 n (2) of a spectral power density of the vibration 3 acceleration during the machine wear (rolling bearing). Equation (2) regardless of the fact, that it requires generating a huge number of equations, A frequency distribution is – of course has only a formal meaning. It determines precisely – completely different. Taking into account only the minimum number of degrees of freedom the module, or more precisely the spectral power enabling – at the preserved model structure density and disregarding phase shifts, let us discuss – the possibility of the given change which transformation should be applied to the of the frequency response structure without dynamic system in order to transform spectrum “a” changing inputs and at assuming the application into spectrum “b”. Let us assume, at first, that of outside forces in each degree of freedom the system is linear and discrete. Both spectra were and the possibility of a free selection obtained from the observation of the machine steady of all coefficients. Physical realisability state, it means under conditions when inputs of such system is practically impossible. are stable processes of the dominating participation In the actual diagnostic tasks, investigating of determined and periodical components. Let us an evolution of a device, the change of a dynamic neglect – for the time being – random disturbances. response structure depends on changes of the Let the system has n degrees of freedom. Both insignificant part of parameters (the ones responsible DIAGNOSTYKA’ 3(47)/2008 155 DĄBROWSKI, About Need For Using Nonlinear Models In Vibroacoustic Diagnostics for the defect). Thus, the number of the necessary really means: for the cost of a tremendously equations should be multiplied by coefficient O , increased number of motion equations) could find defined by the ratio of invariable elements (degrees the solution of the problem by assuming of freedom), to the ones in which certain number simultaneous decrease of primary inputs caused of parameters are changing (usually not all) by other means of energy loss due to wear, but and by coefficient O , which denotes the ratio the model obtained in such manner would 1 not be identifiable. of all degrees of freedom to the ones receiving Natural specification of machine degradation energy from outside. When the number of spectral is the frequency response function (transmittance) lines, taken into account, is limited to 100 change and “additional” inputs related to defects the number of degrees of freedom of the system remain usually as self-exited vibrations, which should amount to several thousands. This require the application of nonlinear description, etc. is the condition for the identificability of the system; The presented above considerations lead it means that changes of actual parameters (in our to the conclusion that a classic diagnostic task very example: state variables) should be transformable often requires application of a nonlinear description. by the determined and mutually explicit procedure The following postulate can be formulated into the model parameters changes. However, on the basis of numerous papers: A new (“after regardless of the mentioned difficulties, there an initial usage”) machine can be described with is a possibility of obtaining such solution. Therefore an adequate accuracy by a linear model. During the application of the linear software FEM exploitation certain nonlinear disturbances related has allowed to solve several problems described to wear and tear – occur. Thus, an observation analytically as nonlinear. Simplifying a little of nonlinear effects allows solving a diagnostic the consideration, we can state that the presented task. operations are based on reducing the globally This postulate is also true for technical devices, nonlinear influences to locally linear ones which operations require a nonlinear description and the increase of the number of degrees from the ‘very beginning’ (e.g. piston-and-crank of freedom of the system results from decreasing mechanism). In such situation we will observe the zone considered to be the “local” area – up an increase of nonlinear disturbances. Besides, to the determined limit of error. This linearisation “a diagnostic” model does not need to be a fully corresponds de facto to the approximation of a curve “dynamic” model. by a certain number of segments. Let us solve now a simple example. An ordinary Let us return now to our assumptions. We have differential equation of the 2nd order in a form assumed at the beginning, that the response of a simple harmonic oscillator – is given: Yf a was the result of linear transformations x Z2 xPt () of inputs, which means that the sum of inputs 0 and outputs from the system should fill the same with an input: frequency bandwidth. Thus, each harmonic component of the input corresponds to one and only Pt() P cos2: t. one harmonic component of the response, which An evident singular solution is a well-known usually has different amplitude, slightly changed “school type” dependency: frequency (because of damping) and different phase shift (what was not taken into account in our P x()tt 22 cos2: (3) reasoning) but surely will not decompose into Z0 : a sequence of components. In the discussed example a Let us check whether finding the singular no spectrum of the system response Yf nor solution of a frequency being equal e.g. to the half Yf b can be obtained at inputs of a smaller of the input frequency : in the form: x Atcos: number of components. In diagnostic tests – during is possible. By substitution we obtain the following machine exploitation - new harmonic components equation: appear very often and the value of their amplitude 22 Acos::tPt (Z0 ) cos2:, (4) is considered the symptom. Such transformation in a linear system requires applying a new which satisfaction for each t requires zeroing excitation, which – in the diagnostic test – would of the input amplitude at the arbitrary amplitude need a postulate that each defect (wear) constitutes of response. Thus, it leads to an obvious triviality. such force. This is an evident contradiction. Let us assume the possibility of modification We assumed at the beginning, that the of the basing equation by introducing an arbitrary transformation of spectral concentration nonlinear function of variable x(t) and let us check Yf()ab()o Yf () () is isoenergetic (< 2 =const), whether now obtaining the response of the frequency equal half of the input frequency is possible. whereas each defect would have been described The task is as follows: by the energy “inflow” from outside. However, supporters of a linear description for any cost (which 156 DIAGNOSTYKA’ 3(47)/2008 DĄBROWSKI, About Need For Using Nonlinear Models In Vibroacoustic Diagnostics

Pcos2:o t xA cos : tx : Z2 xfxP  ( ) cos2 : t ways”, the most often in a series and infinite form. 0 . (5) fx() ? The fact, that the first and the second approximation are usually confirmed by experiments indicates that Proceeding in an identical fashion as previously nonlinear models are worth to be applied now a days and transforming the input we will obtain: when calculation tools are highly developed. 1 Features of nonlinear models are as follows: :AtAtfAtPt22cos2 :Z cos2 : ( cos : ) (cos 2 : 1) . (6) 0 2 1. Principle of superposition is not binding; 2. The system response can have and generally has This equation is much better and at the proper the different frequency distribution that the input; selection of f(x) function its identity satisfaction 3. System transmittances depend on themselves and is possible. E.g. assuming: on inputs (they are not system invariables); fx kx 2 1 4. Local transient states can occur (example of „bended” amplitude-frequency characteristics we will obtain: is known from each text-book on the vibration theory); ()cos2(cos1)22AtkAt 2 š Z0 : :  :  5. Resonance responses can occur at frequencies t being an arbitrary linear combination of input 1 Pt(cos222:œ: 1) Z and natural frequencies, including an effect 2 0 of the so-called internal resonance; PkA 2 . (7) 6. Self-exited vibrations can occur. Pk 2 Diagnostics specialists know all mentioned A 1 above features from the observation of actual objects. The situation presented in Fig. 2 This is a special case, relatively difficult – is a typical example of a frequency conversion. however possible at the appropriately selected Resonant increases of the amplitude in bands, kinematics – for the technical realisation. in which the vibration level in a new machine was It corresponds to the situation when the properly low, local losses of motion stability, generation selected input force amplitude of the resonance of self-exited vibrations, as well as strong frequency instead of increasing the first harmonic dependence of parasitic vibration processes “releases” vibrations of a different frequency. This, on the load – are typical symptoms frequently in turn, corresponds to changing the phase trajectory utilised in vibroacoustic diagnostics. into a certain limiting cycle and constitutes a certain Thus, purposefulness of application nonlinear form of self-exited oscillations, in which descriptions seems to be doubtless similarly a restitution function described by an even function as looking for vibration measures (more (“full” parabola) becomes an “amplifier”. However, precisely: methods and techniques of signal the discussion of the obtained results is not analysis) sensitive to the nonlinear disturbances significant in this case. The example should evolution. be treated as a mathematical “plaything”, which indicates – in a very simple manner – the possibility of generating, by a nonlinear system, the response REFERENCES of the frequency different than the input frequency. Many examples of the applications of this thesis and theoretical discuss as well you can find in book: 1. Batko W., Dąbrowski Z., KiciĔski J.: Nonlinear P cos :tA cos :1t + B cos 2:2 t +... Effects in Vibroacoustic Diagnostic, Institute for Sustainable Technologies  National : = : + H 1 Research Institute, 2008.

Fig. 3. Nonlinear system as frequency transformer. Prof. Zbigniew DĄBROWSKI, Ph. D., D. Sc is full professor in Much more physically real would be an example Institute of Machine Design of a double frequency response – according Fundamentals, Warsaw to the schematic presentation from Figure 3, University of Technology and but finding function f(x) is not simple, similarly head of the Vibroacoustic as a function causing a sub-harmonic response, division. His main professional for various amplitudes and frequencies of inputs. interest are machine diagnostic A solution of such problem requires an application and noise and vibration of analytical approximate methods or simulation minimising. He has authored methods, exactly the same as applied at solving about 200 technical papers (17 books). He’s a head nonlinear differential equations. Mathematical laws of Vibroacoustic Section of Acoustic Committee of are not to be avoided. A solution of nonlinear Polish Academy of Science. problem is obtained in an approximate form “in both DIAGNOSTYKA’ 3(47)/2008 157 RADKOWSKI: Vibroacoustic Monitoring Of Mechanical Systems For Proactive Maintenance

VIBROACOUSTIC MONITORING OF MECHANICAL SYSTEMS FOR PROACTIVE MAINTENANCE

Stanisáaw RADKOWSKI

Institute of Machine Design Fundamentals, Warsaw University of Technology ul. Narbutta 84, 02-524 Warsaw, [email protected]

Summary A relevantly defined operational strategy has decisive influence from the point of view of the ability to maintain and improve reliability and safety as well as from the point of view of maintaining manufacturing quality. The paper presents the main tasks and the method of implementing pro-active operation. Particular attention is paid to the issues of selection and adaptation of the methods of diagnosing the low-energy phases of defect development as well as use of a posteriori diagnostic information. Attention is paid to the importance of technical risk analysis.

Keywords: proactive maintenance, vibroacoustic diagnostic, low-energy stage of failure, bayesian updating.

MONITORING WIBROAKUSTYCZNY SYSTEMÓW MECHANICZNYCH W PROAKTYWNEJ STRATEGII EKSPLOATACJI

Streszczenie Odpowiednio okreĞlona strategia eksploatacji ma decydujący wpáyw na utrzymanie i poprawĊ niezawodnoĞci, bezpieczeĔstwa oraz utrzymanie jakoĞci produkcji. W pracy przedstawiono gáówne zadania i sposób realizacji proaktywnej eksploatacji. Szczególną uwagĊ zwrócono na zagadnienia doboru i adaptacji metod diagnozowania niskoenergetycznych faz rozwoju uszkodzeĔ oraz wykorzystania aposteriorycznej informacji diagnostycznej. Zwrócono takĪe uwagĊ na znaczenie prowadzenia analizy ryzyka technicznego.

Sáowa kluczowe: proaktywna strategia eksploatacji, diagnostyka wibroakustyczna, niskoenergetyczna faza rozwoju uszkodzenia, bayesowskie uaktualnianie.

1. INTRODUCTION being distinguished by its predictability, holistic approach and openness. The fact that the need for implementing the The predictive nature of the system means its principles of harmonious development is treated as ability to forecast the technical condition and the a rule of the development of modern economy quality of realization of functional tasks by the creates a whole series of new challenges for more system and by its individual elements. and more fields of science and technology. From the The criteria of harmonious development and the point of view of machine construction and operation resulting necessity of holistic approach become the this means adoption of design and operation basis for planning the maintenance-and-repair principles which account for Life Cycle Engineering processes. The system’s open nature means both the while rejecting the current, reactive principles of possibility of using selected modules of the system design, operation, maintenance and management in specific applications as well as the possibility of which are focused on maximization of short-term using selected methods both within the system itself effect. Now the goal is to maximize the long-term as well as autonomously, for the tasks realized effects. This means adoption of an operational outside the system. This last feature is the outcome strategy whose integral elements include technical of the assumption that that system development and condition diagnosis as well as predictive models of adaptation are intended to fulfill the needs and meet functional tasks’ realization and principles of pro- the expectations of its users. This in particular active machine maintenance and operation. concerns accommodation of the subsystem Thus the defined strategy accounts for a whole responsible for data registration preprocessing, series of aspects of harmonious development, development and adaptation of technical condition starting from economic analysis of individual models, functionality assessment methods as well as lifecycles, ecological requirements, ergonomic the methods of optimization and analysis of requirements and cultural requirements, with the maintenance and repair processes. Due to the technical component of the management system predictive nature of the system and the inherent burden of uncertainty it is the analysis of probability 158 DIAGNOSTYKA’ 3(47)/2008 RADKOWSKI: Vibroacoustic Monitoring Of Mechanical Systems For Proactive Maintenance distribution, while analyzing the condition as well as in order to minimize and in particular to avoid the development and use of degradation process undesirable developments leading to serious models, that become particularly important. The consequences; development of a forecast of future scope of changes associated with the proposed events based on current observations and registered approach is illustrated by the definition of permanent changes of parameters which have been a technical object’s maintenance process which says detected by analyzing the results and the that it is a combination of all technical and measurements collected in the database. The last associated administrative actions taken during an item is particularly important when monitoring the object’s lifetime in order to maintain that object or condition of mechanical elements and units as well its element in a condition enabling performance of as the remaining components which are subject to the expected functional tasks. The predictive nature degradation and wear and tear for which the of the proposed strategy defines in each case the detection of early phases of defect development may scope and the nature of the activities. In practice help prevent the occurrence of the catastrophic phase various scopes of analysis are applied, depending on of defect development, including destruction of the the method of defining the control, accounting for or whole system. not accounting for the consequences of failures or Thus reduction of the uncertainty of reliability accidents [1]. estimations becomes a critical issue in the process of making the decisions which are intended to Sympton value of damage ensure technical safety of the system and minimize the costs. Proactive Reactive One of the essential methods of reducing the maintenance maintenance epistemological uncertainty is to develop models Failure and diagnose the degradation and wear and tear Repair processes, thus reducing the variance of evaluations of the residual period until the occurrence of Condition a catastrophic defect. Prenucleation monitoring and and early stage maintenance Realization of this goal calls for assessment of Development of failure failure detection mode structural reliability of the system while accounting for detection and analysis of degradation processes affecting all the components during the previous and Operating time current period of use. This requires development of Fig. 1. Comparison of technical diagnostics goals in a relevant database containing information on proactive maintenance versus reactive maintenance potential defects of the system’s components, knowledge gathered based on the experience From this point of view it is the implementation acquired by relevantly trained personnel as well as of proactive operational strategy that becomes procedures which account for the feedback and particularly important. As is presented by Fig. 1, the adaptation changes occurring in the system. essence of such an approach boils down to In the process of estimating the probability of of anticipation of preventive actions, in equal degree defect occurrence, the above enables us to account prior to defect emergence as well as during the for the influence of operational conditions on the period of development of low-energy phases of possibility of defect occurrence, the influence of defects. This calls for developing and adapting earlier defects, quality, scope and intervals between relevant methods of diagnosis which are supported inspections, the probability of defect occurrence in by relevant diagnostic models. specific time in the future. The consequences are estimated in a similar manner, especially the 2. CHARAKTERISTICS OF A PROACTIVE magnitude of loss and the probability of worst case OPERATIONAL STRATEGY scenario occuring. Thus developed risk matrix serves as the basis for defining risk category, priority Let us note that the essence of thus defined and scope of inspections, ways of changing the a strategy is the extensive use of monitoring, architecture of the monitoring system. diagnosis, forecasting and decision-making models Idea of proactive maintenence system algorithm, for creating the possibilities of taking maintenance- was presented in literature [1, 2, 3] (Fig. 2). and-repair actions while anticipating problems [2]. This denotes the need for developing and applying advanced monitoring, presentation of information on emergency states and values, selection of methods and means enabling monitoring and on-line inference in a way enabling early detection of growing disturbances and extracting from general signals the information on anomalies in operation which are characteristic of defects; controlling the defects and taking corrective actions by the operator DIAGNOSTYKA’ 3(47)/2008 159 RADKOWSKI: Vibroacoustic Monitoring Of Mechanical Systems For Proactive Maintenance

conditions on the possibility of defect occurrence, Monitoring and Functioning and the influence of earlier defects, the quality and the diagnosis monitoring data process scope of operational inspections, the probability of Current state of the process Object state evolution defect occurrence in specified time in the future. Current technical state Past parameter of process Current functioning condition Past functioning conditions The consequences are estimated in a similar way, Prognosis especially the gravity of loss and the probability of process Future state of the process the worst-case scenario occurring. Thus, the Operating knowledge Mean time before/between Expected functionig developed risk matrix serves as the basis for failure conditions Remaining life of object determining the risk category, the priority and scope

Direct/indirect Aided-decision of the inspection as well as the method and Level of maintenance making process technical risk cost directions of changes in the operation maintenance procedures. An exemplary division into categories is presented in [5] (Fig. 3). Selected maintenance activity Fig. 2. Architecture of the proactive maintenance CATEGORY 3 CATEGORY 4 system

Let us note that estimation and modelling of the degradation process is one of the most effective CATEGORY 3 methods of defect development anticipation and maintenance of system operation in terms of nominal parameters. In reality such an approach

denotes compilation of several conventional of Probability the top event methods of forecasting – probabilistic behavioral CATEGORY 1 CATEGORY 2 models and event models in particular. Probabilistic Seventh behavior and degradation models enable analysis of Fig. 3. Risk categories the type and extent of uncertainty which conditions forecasting reliability. Event models are a kind of The following scope of activities is assigned to a combination between the contemplated models and each category respectively: the actual system and they make up the basis for x category 1 – (acceptable), constructing and analyzing causal models which x category 2 – (acceptable, inspection required), enable assessment of degradation and determination x category 3 – (undesirable), of the optimum scenario of maintenance-and-repair x category 4 – (not acceptable). work. The influence of various degradation processes, Let us note that subsequent stages of the including wear, cracking and corrosion, is modeled forecastng procedure make references to various in the second phase. models and types of knowledge. The a priori In the to-date contemplated models the knowledge gathered from experiments serves as the probability of defect occurrence was being defined basis for developing stochastic models of on the assumption of invariability of examined degradation processes while technical diagnosis and distributions during operation of the object. In process monitoring are used for developing the reality, as a result of wear and tear processes and indicators of an object’s technical condition. associated changes of conditions of mating elements Interaction between the elements of the and kinematic pairs, we observed conditional monitored object is described and forecasted with probability distributions, however the relationship the use of cause-and-effect relations based on laws demonstrated itself both in quantitative terms of physics in the form of model-aided forecast. In (change of the parameters of probability density accordance with the definition [3], behavioral function) and in qualitative terms (change of the models contain both the description of functioning function describing the distribution). In addition the as well as the dynamics of a system. In the first case degradation processes accompanying the the system is analyzed as a set of many processes performance of functional tasks can cause similar analyzed from the point of view of flow of materials, variability of distributions of the probability which energy and information. The purpose of the system describes load capacity. In this case one can expect dynamics model is to offer description of conditions that the location of the separating line and the in which failures occur – in this case the modeling probability of defect occurrence will not only process points to two phases – in the first one the depend on the time of operation of an object but on cause-and-effect relations are defined between the the new dynamic feedbacks in the system, associated degradation process, the cause and the in particular with the development of non-linear consequences. This is often done with the use of the relations and non-stationary disturbance. EMECA method [4]. The estimation of probability of defect occurrence accounts for the influence of operating 160 DIAGNOSTYKA’ 3(47)/2008 RADKOWSKI: Vibroacoustic Monitoring Of Mechanical Systems For Proactive Maintenance

3. VIBROACOUSTIC SIGNAL AS THE quantification of energetic disturbances occurring in SOURCE OF DIAGNOSTIC INFORMATION the case of defect initiation.

Bispectrum (diag), Eps=0, Sigma=0.5, Delta=0, A1=1, A2=1, f1=100, f2=70 The central issue is how to extract the relevant 25 diagnostic information and use it in the forecasting process. Thus the research focuses in particular on 20 the methods of analyzing the relations between various frequency bands and their links to various 15 types of defects or phase of their development. The value of the information contained in the 10Amplitude bi-spectrum consists of, among others, the fact that it enables examination of statistical relations between 5 individual components of the spectrum as well as to 0 detect the components generated as a result of 0 50 100 150 200 250 300 350 400 occurrence of non-linear effects and the additional Bifrequency [Hz] feedback associated with the emerging defects. This Fig. 4. Illustration of diagonal bi-spectrum in the results from the fact that in contrast with the power case of the third order non-linearity spectrum, which is positive and real, the bi-spectrum Bispectrum (diag), Eps=0, Sigma=0, Delta=0.5, A1=1, A2=1, f1=100, f2=70 function is a complex value which retains the 1500 information on both the distribution of power among individual components of the spectrum as well as the changes of phase. Let us note that the bi- 1000 spectrum enables one to determine the relations between essential frequencies of the examined dynamic system. High value of bi-spectrum for Amplitude 500 defined pairs of frequency and combinations of their sums or differences will point to the existence of frequency coupling between them. This may mean 0 0 50 100 150 200 250 300 350 400 that the contemplated frequencies, being the Bifrequency [Hz] components of the sums, have a common generator, Fig. 5. Illustration of diagonal bi-spectrum in the which in the presence of non-linearity of higher case of the fourth order non-linearity order may lead to synthesizing the aforementioned new frequency components. Let us assume that the degree of damage D is the

Bispectrum (diag), Eps=0.7, Sigma=0, Delta=0, A1=1, A2=1, f1=100, f2=70 dissipated variable that covers the changes of the 5 structure’s condition due wear and tear: 4.5

4 dE (4, D ) 3.5 d 0

3 wE (4, D ) wE (4, D ) (1) d 0 dD  d 0 d4 2.5 wD w4

Amplitude 2 1.5 where: 1 df (D,4,J (4)) 0.5 dEd 0 d4 0 50 100 150 200 250 300 350 400 Bifrequency [Hz] J (4) - the parameter describing how big a part of Fig. 3. Illustration of diagonal bi-spectrum the dissipated energy dEd is responsible for in the case of square non-linearity structural changes, 4 - operating time. An example of using the diagonal bi-spectral Bearing in mind the possibility of diagnosis of measure for the qualitative and quantitative the origin and the development of low-energy phases determination of the effect of non-linearity evolution of defect formation, when the extent of the original is presented in Fig. 3÷5. defect can be different in each case, let us analyze The confirmation of the importance of the this issue more precisely. changes of phase coupling is offered by the analysis To examine this problem let us recall here the of the process of diagnostic information generation two-parameter isothermal energy dissipation model during the low-energy phases of development [6]. proposed by Najar [7] where: While attempting to develop a model oriented on such defects one should on the one hand consider the dE dE  dE Tds V dD (2) issue of examining the signal's parameters from the ds d dq 4 point of view of their sensitivity of to low-energy where: changes of the signal and, on the other, the issue of DIAGNOSTYKA’ 3(47)/2008 161 RADKOWSKI: Vibroacoustic Monitoring Of Mechanical Systems For Proactive Maintenance

dE - energy transformed into heat, k(D) a bD (7) dq  dE - energy responsible for internal ds For damage of small magnitude the linear structural changes, approximation seems to be sufficient and enables T - temperature, description of defects whose emergence is ds - growth of entropy. characterized by small growth of defect energy (see The expression (2) shows that the growth of the Figure 6). dissipated variable D is attributable to the dE part ds of energy, which is the dissipated part of dEd energy, that causes the growth of entropy ds. The role of the multiplier determining the relation between the increments of dissipated variable and the entropy is played by the dissipation stress V 4 . The assumption of T constans results in independence of dissipation-related loss dE dE , thus following integration the ds 4s expression (2) takes the following form:

E T s (3) D – normalized degree of failure ds ' Fig. 6. Change of energy of defect development The derivative of defect development energy for small D 1 2 related to D, when E f (D0 ) d EH , means the 2 Thus while defining the set of diagnostic boundary value of deformation energy and takes the parameters we should pay attention to the need for following form: selecting such a criterion so that it will be possible to identify defects whose emergence is characterized dE E (D )(1 D )(1 k)D k ds f 0 f (4) by small growth of defect-related energy. 1k 1k dD D f  D0 While contemplating this issue let us assume that vibroacoustic signal is real and meets the cause- and-effect requirement, which means that it can be For a defined initial defect of D and for 0 the base for creating an analytical signal. a defect leading to damage D f , relationship (5) will In accordance with the theory of analytical have the following form: functions, the real and the imaginary components are the functions of two variables and meet Cauchy- Riemann requirements. dEd (D) s (1 k)E (k)D k (5) While contemplating this issue let us assume that dD D0 , f vibroacoustic signal is real and meets the cause- and-effect requirement, which means that it can be Let us note that parameter E is an D0 , f the base for creating an analytical signal. exponential function of power k, similarly as the In accordance with the theory of analytical whole derivative. While referring to the second rule functions, the real and the imaginary components are of thermodynamics for irreversible processes we the functions of two variables and meet will assume the following in the contemplated Cauchy-Riemann requirements. model: Please be reminded that the analysis of the run of the analytical signal will be conducted while relying dE (D) on the observation of changes of the length of vector ds t 0 (6) A and phase angle : dD M z(x, y)  jv(x, y) A(cosM  j sinM) (8) Thus for the assumed model to be able to fulfill condition (6), the exponent must meet the Thus, requirement of k d 1 . In addition, while referring to z x(W ), y(W ) A(W ) cosM(W ) (9) the rule of minimization of dissipated energy, the Q (x(W ), y(W ) A(W ) sin M(W ) conditions of permissible wear process show that the change of exponent k is possible as the defect means that the signal measured is the orthogonal develops. projection of vector A on real axis. To examine this problem let us assume that the Ultimately, while exploiting the Cauchy-Riemann exponent shows a straight line dependence on the conditions for variables A(W ) and M(W ) we will extent of damage: obtain: 162 DIAGNOSTYKA’ 3(47)/2008 RADKOWSKI: Vibroacoustic Monitoring Of Mechanical Systems For Proactive Maintenance

Z 5ha-env-shape-par 5 dz dA dM cosM  AsinM (10) dW dW dW 4.5

As we expected the obtained relationship presents 4 an equation that enables the analysis of the measured signal while observing A and M . At the same time it 3.5 should be noted that for low-energy processes, when 3 we can neglect the changes of the vector’s length and value of shape parameter assume A # const , the whole information on the 2.5 changes of the measured signal is contained in the phase angle, or more precisely in the run of 2 0 50 100 150 200 momentary angular velocity. number of measurement While accounting for the obtained results of the Fig. 8. Parametr of shape in function of mesurement analysis of the process of low-energy defect number emergence and detection of diagnostic information associated with the changes of momentary values of Z 5ha-env-scale-par 60 amplitude and angular velocity, let us analyze the conditions that must be fulfilled by a diagnostic 55 model intended to enable observation of the 50 influence of such disturbance on the form of the 45 system’s dynamic response. 40

Z 5ha-obw pdf 35 30

value of scale parameter 25 0.1 20 0.08 15 0.06 0 50 100 150 200 measurement number pdf 0.04 Fig. 9. Parametr of scale in function of mesurement 0.02 number 0 200 100 In order to achieve higher efficiency in 100 application of the results of vibroacoustic diagnosis, 50 we should take into account, in a much bigger 0 number of measurement 0 amplitude degree, the individual vibroacoustic characteristics Fig. 7. Function of density probability distribution which as defined during preliminary measurements oh the envelope of 5th harmonics meshing frequency and analysis. Such an approach fully meets the requirements This illustrates the problem of effective use of of Bayes methodology [9], including: results of diagnostic observations in the task of - the possibility of adopting randomness of the diagnosis of early stages of defect development. examined parameter in a probabilistic model; While referring to [8] we quoted the example of - the possibility of obtaining the a posteriori estimate evolution of probability density distribution function of the parameter based on the observation and for the value of vibroacoustic signal envelope, measurement of a vibroacoustic signal as well as precisely the fifth harmonic frequency of meshing as the a priori distribution in accordance with the calculated for the width of the frequency band requirements of Bayes’ theorem; corresponding to twice the frequency of the input - selection of the optimum estimator of a parameter shaft (Fig. 7). Fig. 8 and Fig. 9 accordingly present in the sense of Bayes decision-making theory. the values of shape and scale parameters Assuming that detection, identification and corresponding to these changes. Let us note that both location of changes of vibroacoustic properties of parameters depend on defect development while the monitored object is the outcome of vibroacoustic their values, the value of shape factor in particular, monitoring, the a priori distribution of probability of does not change monotonously. parameters can be determined on the basis of pre- defined vibroacoustic characteristics. A good illustration of the discussed method of using the Bayes formula for evaluation of changes of distribution parameters based on the diagnostic information is offered by Cruse’s paper [10] in which Bayes theorem is used for determining the DIAGNOSTYKA’ 3(47)/2008 163 RADKOWSKI: Vibroacoustic Monitoring Of Mechanical Systems For Proactive Maintenance value of parameters describing the growth of f (D / a) f (a) f (a / D) v f (D / a) f (a) (14) fatigue-related defect while accounting for the f (D) observation of crack development. The essence of this approach involves updating In a similar way the probability density is of estimated parameters of the probabilistic model so proportional to the square root of Fisher’s as to achieve bigger alignment between the results of information matrix factor [12]: modeling and observations. In accordance with the above presented 1 2 assumptions it is assumed that unknown or uncertain f (a) v (det ,(a)) (15) parameters of distribution are random variables. Uncertainty of estimation of results can be linked to where: variability of random variables by means of Bayes ªw 2 ln f D / a º theorem [11]. ,(a) E« 2 » – is calculated as the Then, while assuming that we will be estimating ¬ wa ¼ the parameters of a priori distribution of parameter matrix of average second derivatives from the a of the function of probability density f(a) and that credibility function logarithm based on the results of D is an observation set enabling reduction of a priori the experiment. uncertainty on the condition of the results of the Thus formula (11) is finally written in the observation being included, we should be able to following way: conduct the estimation of a posteriori distribution parameters by means of the following formula: f (a / D) v L(D / a)(det ,(a))1 2 (16)

f (D / a) f (a) f (a / D) (11) f(D) f (D) 0.18 0.16 aposteriori Where: 0.14 distribution f f (D) f (D / a) f (a)da 0.12 apriori ³ distribution f 0.1 In addition we can assume that the denominator, which is described by the integral of the a posteriori 0.08 likelihood function probability density function, is constant and that 0.06 f(D/a) is the probability of observation which can be expressed by the credibility function. In such a case 0.04 equation (11) can have the following form: 0.02

0 f (a / D) K L[D / a] f (a) (12) 0 10 20 30 40 50 60 B ˜ ˜ D Fig. 10. An example of using diagnostic information where: in bayesian updating K B – standardizing constant L[D / a] – credibility function Thus the presented method of using the risk To be able to determine the probability of analysis method, supported by vibroacoustic a defect in the analyzed timeframe, the information diagnosis, in the process of making operational contained in the observations should account for decisions refers on the one hand to the definition of both, occurrence of a defect and non-occurrence of risk and the associated estimation of probability of a defect. For the exponential form of the function occurrence of undesirable incident as well as to describing the distribution, the credibility function estimation of the extent and value of loss will be noted in the following form: accompanying such an incident. On the other hand it n m refers to the possibility of using the results of a diagnostic experiment in the task of reducing the L>@D / a ––p(T f / a)u >@1 P(T f / a) (13) i 11j uncertainty related to estimation of parameters of where: a posteriori distribution of intensity of defects. Use n – denotes the set of detected defects of Bayesian models enables direct application of the m – denotes the set of events defining non- results of diagnostic observation in Bayesian existence of a defect. estimation of a posteriori distribution as well as Let us note that Bayes formula can be simplified tackling the problem of selecting the a priori by accounting for proportionality of a posteriori and distribution. The results of such an analysis are a priori distributions only : presented in Fig. 10. As has been demonstrated in this example, such use of diagnostic information enables one to solve the problem of determining the conditional probability distribution for the 164 DIAGNOSTYKA’ 3(47)/2008 RADKOWSKI: Vibroacoustic Monitoring Of Mechanical Systems For Proactive Maintenance

parameters of defect intensity while relying on the techniques. Computers in Industry, Vol. 53, results of the diagnostic experiment. 2006, pp. 569÷580. [4]. Radkowski S.: Safety technique fundamentals. 4. CONCLUSIONS Oficyna Wydawnicza Politechniki Warszawskiej, 2003 (in polish). While summing up the methodology of a pro- [5]. Radkowski S.: Use of the diagnostic information active system of operations based on risk evaluation, in safety oriented maintenance. Diagnostyka, attention should be drawn to the necessity of Vol. 2(38), 2006, pp 85÷92 (in polish). tackling the following issues: [6]. Radkowski S.: Nonlinearity and intermodulation - Identification of elements of the system as well as phenomena tracing as a method for detecting factors and persons responsible for controlling the early stages of gear failures. Insight Non- system and its functioning; destructive Testing and Condition Monitoring, - Defining a wide area system for control and August 2008 (in press). analysis of the system’s functioning; [7]. Najar J.: Continuous damage of in: Continuum - Development of the behavioral model of the damage mechanics theory and applications. system accounting for the physical mechanisms of Editors: Krajcinovic D., Lemaitre J. degradation processes affecting the system’s International Centre for Mechanical Sciences elements and their mutual interactions, Courses and Lectures, 295, 1991, pp. 234÷293. - Modeling of events, incidents and defects in the [8]. GumiĔski R., Radkowski S.: Use of aposteriori system in a way enabling inclusion of the results in diagnostic information in technical risk the decision-making processes, analysis. Diagnostyka (in press). - Development of a forecasting model enabling [9]. Grabski E., JaĨwiĔski J.: Bayesian methods in integration of the models of events with the reliability and diagnostics. Wydawnictwa behavioral model of the system, Komunikacji i àącznoĞci, Warszawa, 2001 (in - Developing a man-machine communication polish). program focused on psycho-physical capabilities [10]. Cruse T. A.: Reliability Based Mechanical of the operator. Design, Marcel Dekker, Inc, New York, 1997. The methodology of designing and [11]. GumiĔski R., Radkowski S.: Diagnostics and implementation of a proactive operational strategy, uncertainly decreasing of technical risk developed according to such assumptions, relies on evaluation. Materiaáy Szkoáy NiezawodnoĞci, a combination of probabilistic models of Szczyrk, 2006, pp 133÷140 (in polish). development of degradation processes and dynamic [12]. Jeffreys H. J.: Theory of Probability, Oxford monitoring architecture accounting for changes in Clarendon Press, 1961. the models of events as supplemented by developed man-machine communication systems. Prof. Stanisáaw RADKOWSKI, a professor in Institute of Machine Design Fundamentals of BIBLIOGRAPHY Warsaw University of Technology, head of Integrated Laboratory of the Mechatronics System [1]. Muller A., Suhner M. C., Iung B.: Maintenance of Automotive and Construction Machinery, alternative integration to prognosis process manager of the Technical engineering. Journal of Quality in Maintenance Diagnosis and Risk Analysis Engineering (JQME), Vol. 2, 2007. scientific team, Chairman of [2]. Muller A., Suhner M. C., Iung B.: Formalisation Polish Society of Technical of a new prognosis model for supporting Diagnosis. In his scientific proactive maintenance implementation on work he deals with industrial system, Reliability Engineering vibroacoustic diagnosis and & System Safety, Vol. 93, 2008, pp 234÷253. technical risk analysis, analysis [3]. Tian Han, Bo-Suk Yank. L Development of an of maintenance strategy e-maintenance system integrating advanced DIAGNOSTYKA’ 3(47)/2008 165 NIKOLENKO, OLEJNICHENKO: Substantiation of structure and parameters of hydraulic stands with …

SUBSTANTIATION OF STRUCTURE AND PARAMETERS OF HYDRAULIC STANDS WITH RECUPERATION OF CAPACITIES FOR DIAGNOSTICS OF ADJUSTABLE HYDROMACHINES

I. NIKOLENKO, A. OLEJNICHENKO

*National Academy of Environmental Protection and Resort Development **Joint-Stock Company “Strojhydravlika” (Odessa, Ukraine) Adress: 181, Kiyevskaja st., Simferopol, 95493, AR Krimea, Ukraine; tel.: +38-(0652)-277-417, e-mail: [email protected]

Summary Set of parameters which objectively characterize a technical condition of the hydromachine can be received at their bench tests. Considering the set forth above features of hydromachines, for reception of the maximal volume of the information at the minimal expenses of time and energy at tests apply stands with recuperation of capacities. Essentially new designs of stands with recuperation of capacities for test of hydraulic machines and mechanical units are offered and patented. The basic structures of test beds with recuperation capacities and bases of a choice of their parameters which allow testing adjustable hydromachines in a wide range of their parameters are considered. The developed stands allow raising quality of diagnosing of adjustable hydromachines essentially.

Keywords: hydromachine, stand, recuperation, control, test.

1. INTRODUCTION 2. OBJECTS AND PROBLEMS

Greater part of modern mobile technique - build-road, The feature of hydraulic drive of mobile agricultural, communal et al produced with hydraulic technique at tests and rolling is that variety of its drives a to 90% engine power is utillized in which. The kinds, and also different and difficult external technical level of modern gidrodrive is determined the environments are required by applications of parameters of his basic power aggregates – pumps and different types of tester equipment. Creation and motors. In spite of ap-preciation of value of hydraulic introduction of universal tester equipment drives, application of the managed power aggregates allows to execute on him a test and rolling of instead of fixed volume allows to extend an adjusting aggregates of different type in the wide range of range, decrease the swept volume of pump and increase office hours. his frequency of rotation to frequency of rotation of Stands for the tests of aggregates de bene billow of drive engine without intermediate mechanical esse are divided by two classes: with the direct transmis-sions, that on the whole diminishes sizes and stream of power and with rekuperatsiey of mass of transmission and promotes its reliability. power. Most modern are stands with One of important technological operations at making rekuperatsiey of power with a hydraulic drive. and repair of knots, aggregates and machines there are Advantages of hydraulic stands with their tests and rolling. Tests provide the estimation of rekuperatsiey of power it is been [2]: conforming to of producible good the requirements of - smooth regulation of speed and parameters of designer and normative and technical documents, and loading; also exposure of weak points in a construction and - change of a direction loading in the process of technology with the purpose of development of measures tests and rolling; on their removal. Rolling allows in the process of - reliable protection of examinee aggregates making and repair substantially to promote reliability from an overload; indexes, that promotes quality of producible wares in - possibility of complete automation of tests and same queue. For organization of these technological rolling, and also remote control the operations at a production it is necessary operatively and parameters of stand; objectively to estimate the state of aggregate which is - small labour intensiveness of setting of tested. Diagnostics allows to define the actual consisting equipment and examinee aggregates. of technical object, and also character of change of him The row of constructions of hydraulic stands of time [1]. Knowledge of the real state of aggregate at is developed with rekuperatsiey powers [3, 4, 5] tests allows to optimize time of tests and rolling, and also which allow to execute tests and rolling, both to expose and remove the lacks of construction, and also hydraulic equipment and mechanical technologies of making of aggregates, knots and details. aggregates. Hydraulic stands can be executed on the opened chart, when the suction hydraulic 166 DIAGNOSTYKA’ 3(47)/2008 NIKOLENKO, OLEJNICHENKO: Substantiation of structure and parameters of hydraulic stands with … line of pump and weathering gidrodvigatelya is that serve of working liquid of pump 6 was connected with a tank, or on the closed chart, when higher consumable expense by a motor 7. foregoing hydraulic lines are united between itself. Surplus of working liquid given a pump 6 given On fig. 1 the hydraulic chart of universal tent-bed test on weathering in a tank through a loading valve is presented with rekuperatsiey of power [3]. 12 and flowmeter 15. Change of tuning of A stand is contained by an engine 2, which is a drive for loading valve 14 provides the change of the managed pump 1, pressure hydraulic line 5 last pressure in a pressure highway 8, and, reported through the managed throttle 4 with loading consequently moment on the billows of motor 3, and suction hydraulic line with gidromashin 6 and 7. a tank 13. Coupled hydromachines 6 and 7 connected by Change of serve of pump 2 or tuning of between itself a pressure hydraulic line 8, and their shafts throttle 4 changes frequency of rotation of are mechanically connected by a transmission 10. In the loading motor 3, and consequently, through presented chart hydromachine 6 is a pump, and a transmission 11, changes frequency of rotation hydromachine 7 is a motors. Weathering hydraulic line of billows of gydromachines 6 and 7. Setting of of motors 9 and suction hydraulic line pump 6 and motor differential loading valve 14 with a management 7 connect with a tank 13. In addition, shafts of from pressure in the pressure highway of hydromachines 6 and 7 through a transmission 8 gydromachines 6 and 7, provides the changes 10 mechanically connected with the shaft of loading of moment on the billows of examinee motors 3. aggregates at their permanent frequency of rotation or provides the change of frequency of rotation of billows of examinee aggregates at a permanent moment.

3. MAIN SECTION

Hydraulic, kinematics and power calculations of hydraulic stand with rekuperatsiey of power executed the known methods. The feature of calculation is a choice of gear-ratio of transmission and determina-tion of drive engine power. Flow of working liquid of hydromachine working in the mode of pump:

Qɧ nɧ ˜ qɧ ˜Kɧ0 , (1)

where nɧ - frequency of rotation shaft of pump, - displacement of pump, - volumetric qɧ K ɧ0 efficiency of pump. Fig. 1. The principle hydraulic scheme of stand. In a kind recuperation of power on a stand will present the serve of pump in a kind:

Pressure hydraulic line 8 reported through Qɧ t (Qɦ /K0ɦ )  Qɧɤ D ˜Qɦ (2) a loading valve 11 and flowmeter 12 with a tank 13, and where Q - rate of flow in motor, K - pressure hydraulic line 5 through a loading valve 14 ɦ 0ɦ volumetric efficiency of motor, - flow of reported with a tank 13. Loading valve 14 executed Qɧɤ a differential, here his managing cavity is united with working liquid through a loading valve, pressure hydraulic line 8. Hydromachine 6, loading D - coefficient flow of pump. motor 3 and flowmeter 12 supplied the sensors of At such denotation D 1 is a relative angular speed 15. Pressure hydraulic line 5 and amount of working liquid, which does not pass 8 supplied the sensors of pressure. through gidromotor. A hydraulic stand works as follows. At including of Frequency of rotation shaft of pump is engine 2 pump 1 gives through a throttle 4 on determined from dependence: a pressure hydraulic line 5 working liquid in n n ˜i , (3) a loading motor 3, which drives to the rotation ɧ ɦ ɬ a transmission 10. Rotation through a transmission 10 where nɦ - frequency of rotation shaft of passed on the shaft of pump 6, which forces motor, i - a transfer relation of transmission. a working liquid on pressure hydraulic line 8 in ɬ a motor 7, which passes a rotation on a transmission 10, Taking into account dependences (1) (2) will what and recuperation of power at stand is provided. get Transmission 10 must provide speed of rotation of iɬ t D ˜ (qɦ / qɧ ) ˜ (K0ɦ /K0ɧ ) , (4) billow of pump higher, than at gidromotora 7, in order DIAGNOSTYKA’ 3(47)/2008 167 NIKOLENKO, OLEJNICHENKO: Substantiation of structure and parameters of hydraulic stands with …

In this case the required power drive of stand is with the purpose of diminishing of losses of determined on dependence: power in a transmission. Forcing power of 'p (5) machines on results in the decline of their Nɩ ˜(Qɧ ˜(1Kɧ )  Qɦ ˜(1Kɦ )  Qɧ ˜(D 1)), Km efficiency, that requires development of where 'p - an overfall of pressure on a loading valve, additional complex of structural and ɤ technological measures for providing of his , , - general efficiency of pump, motor and Kɧ K ɦ Kɬ increase and stabilizing. Application of the transmissions accordingly. automated facilities of diagnostics at the tests of Taking into account the accepted denotations and new or modernized constructions allows foregoing dependences will transform a formula (5) substantially to shorten time of polishing of pre- p ª q º production models to the serial making. ' ɦ , (6) N ɩ t ˜ nɧ ˜ qɧ ˜ «D Kɧ  ˜ 1K ɦ » The methods of determination efficiency of K q m ¬ ɧ ¼ rotor ɪɧdromaɫhinɭɿ are known, which are In case if q ɦ qɧ power of drive of stand on based on the theory of similarity [7]. In dependence (6) appears in a kind obedience to this theory, general efficiency of 'p pump is determined on dependence: N t ˜ n ˜ q ˜ (D 1K K ) , (7) 1 k ɩ ɧ ɧ ɧ ɦ  ɭ1˜V 1 , (8) Km K ɧ 1 k  k V For example, at the use as power aggregates of drive f 1 P1 1 where specific power from the losses of of stand axial piston hydromachine (APH) hydrodrive k ɭ1V 1  type GST-90 with nominal power pump and motor of working liquid; k  specific power which is 46 kWt and displacement 90 sm3, f 1 N ɦ N ɧ qɧ q ɦ lost on a dry friction; k V  specific power of K K 0,92, accept D 1,15. Consequently, P1 1 ɧ ɦ losses on a viscid friction in a pump; i t 1,15, and drive power of stand of 14,3 kVt. ɬ a criterion of similarity of flow V 1 p P ˜Zɧ  According to dependence (7) N t 0,31˜ N . At the tests ɩ ɧ of viscid liquid in an equivalent crack for of every aggregate individually on the opened chart the a pump, which is named the Zommerfel's total expenses of power make N ɩ t 92 kWt, that function, p  pressure in a general pressure consumable power at tests goes down in 6 times. Taking line, P  dynamic viscidity of working liquid, to account that a hydraulic drive is provided in a chart Z  angular speed of shaft of pump. (Fig. 1), a that decline of power will be a few less than. ɧ For drafting of diagnostic model of drive of stand the For motorts general efficiency is method of formalization of functioning is accepted on determined on dependence: 1 k k the basis of generalized three key element [6]. The power  f 2  P 2 V 2 , (9) K ɦ aggregates of hydraulic drive can be presented charts 1 k ɭ2 ˜V 2 which have three entrance variables, functionally related where k  specific power which is lost on to other elements. As variables for knots accepted: knot f 2 a dry friction in a motor; k  specific power taking of power, knot entrance of working environment P 2 V 2 and knot taking of working environment. Thus every of losses on a viscid friction in a motor; k V  element is described the system from 3 equalizations. ɭ2 2 One of signs of change of the technical state of specific power from the losses of working liquid in a motor;  a criterion of aggregates of hydrodrive, both in the conditions of V 2 p Z ˜Z ɦ exploitation and at tests and rolling there is power similarity of flow of viscid liquid in an efficiency. For the aggregates of by volume hydrodrive equivalent crack for a motor; select three indexes, which represent power efficiency: Z ɦ Q angular speed of motor billow. complete, volume and mehanical efficiency. The analysis Taking to account for (8) and (9) of these indexes at tests allows simply to determine the Z i ˜Z , that V i ˜V . technical state of aggregate, and the dynamics of their ɧ ɬ ɦ 1 p 2 changes allows operatively to set duration of tests, and The values coefficients of specific powers also to set reasons of low quality of good. Therefore as losses of energy in power aggregates are a basic sign of the state at the tests of power hydraulic determined on the characteristic sizes of aggregates efficiency is accepted. hydromachines: Important direction of perfection power aggregates of - from the losses of working liquid 3 hydraulic drives is an increase and stabilizing of C i § G · (10) k ɭ ¨ i ¸ efficiency in all of range of operating parameters. The ɭi ei ¨ q ¸ increase of technical level hydraulic drives of mobile © i ¹ where: C Q coefficient of losses; 0 d e d1 - technique is provided the increase of nominal pressure to ɭi i

30..45 MPa, by the increase of nominal frequency of adjusting parameter; G i  equivalent gap; rotation of pumps to frequency of rotation drive engine, qi  characteristic size of hydromachine. and also by application of the managed power aggregates q 3 i 2 ˜S 168 DIAGNOSTYKA’ 3(47)/2008 NIKOLENKO, OLEJNICHENKO: Substantiation of structure and parameters of hydraulic stands with …

 from losses on a dry friction: C ˜ q k Pi i , (11) Pi ei ˜G i where: coefficient of losses on a viscid friction. CP  In the above-mentioned dependences q for concrete i type of hydromachine the characteristic volumes q of i size are permanent, all of other are determined the structural features of units and details of hydromachines, and also technology of their making, as a rule are variable at tests. Investigation change of the state aggregate which is tested is a change of signs the state. Thus basic sources of uninvariance there is an object of diagnosing. As a result of break-in process details of aggregates takes a place changes of internal factors, that is reflected at the level general efficiency of aggregates. Descriptions of drive of stand at tests have statistical changeability the analysis of which allows to estimate the basic sign of diagnosing. For the analysis of the state of aggregates Fig. 2. The stand with recuperation of powers which test apply two indexes: mean value and dispersion for tests APH 403.112 of general efficiency of aggregates for period of their test. The size of mean value allows to set conforming to of aggregate the normative requirements, and an analysis of dispersion is time of break-in process.

4. RESULTS OF RESEARCHES

On the presented stand in joint-stock company «Strojhydravlika» managed bent axis APH trimot type 403.112 is tested. His issue accustoms from 2008 for application in the hydraulic transmissions of build- travelling and agricultural technique. The distinctive feature of new series is possibility of providing of the discrete adjusting of the displacement. Thus two workers 3 of volume, maximal qmax =112 sm , and the minimum displacement is set depending on the technical terms of Fig. 3. Setting of two hydromachines customer. By undeniable advantage of hydraulic drives 403.112 at the stand. with discretely managed motors is providing of two workings speeds, and also possibility of putting in these Treatment of values of parameters of tests speeds without the stop of machine, that promotes the allowed continuously to determine the signs of productivity and management quality substantially. the state of hydromachin: mean value and A standard pumping unit and original adjuster is dispersion of efficiency. The analysis of these applied in the construction of APH type 403.112. With signs allowed to define accordance of the the purpose of verification of capacity of gidromashiny technical state of aggregates the requirements of its tests were conducted on a stand with rekuperatsiey of technical document for development of power, which is presented on fig. 2, and setting two machines, and also to specify time of their managed machine 403.112 on a stand shown, on fig. 3. break-in process. The additional task of tests was verification of longevity by work of regulator of number of cycles of switching, which would provide his tireless durability at the base number of cycles of loadings, what corresponded continuous work of hydromachine during the set interval of time with switching of regulators of every aggregate. The given stand also can be applied at tests and break-in process mechanical units. In this case units which are tested are installed instead of transmission 10 (fig. 1). DIAGNOSTYKA’ 3(47)/2008 169 NIKOLENKO, OLEJNICHENKO: Substantiation of structure and parameters of hydraulic stands with …

The loading cycles APH 403.112 at the stand are REFERENCES presented on diagrams which are shown on fig. 4. On diagrams are shown Q1  rate of flow the hydromachine, 1. Sjiricin T.: Reliability of gidro- and which works in a mode of the motor, and Q2 - rate of pnevmodrives. M: Machinostroenie, 1981 flow the hydromachine, which works in a mode of the (in Russian). pump. Switching of operating modes was provided with 2. Volkov D., Nikolaev S.: Reliability of build hydraulic allocators which conditionally are not shown machines and equipment. M: Vysh. school., on the basic scheme fig. 1. 1979 (in Russian). 3. Gorbatij V., Nikolenko I. and other:. The 1 stand for tests of hydromachine with recuperation of power. – A.s. USSR ʋ 0,5 1606750. Bul. ʋ 42., 1990 (in Russian). Qi\Qmax 0 4. Gorbatij V., Nikolenko I. and other: The stand for tests of hydromachine. A.s. SSSR 0 0,5 1 1,5 2 2,5 3 3,5 4 4,5 5 ʋ 1613683. - Bul ʋ 46., 1990 (in t,c Russian). 5. Nikolenko I., Komarov A.: The hydraulic - Q1 - Q2 stand for test of mechanical units. Patent of Uraine ʋ 64366/ Bul. ʋ 2, 2004 (in Fig. 4. The loading cycles of hydromachines Ukrainian). 6. Berengard Ju., Gajcgori M.: Algorithm of The results of the conducted tests were confirmed by forming of mathematical model of a capacity and longevity of the developed construction of hydrodrive of arbitrary structure. managed variable displacement hydromachines 403.112. Engineering science. ʋ 1, 58 – 65., 1977 with the discrete adjuster. (in Russian). 7. Kondakov L., Nikitin G., Prokofiev V. and 5. CONCLUSIONS others: Machine-building hydrodrive. – Ɇ.: Machinostroenie, 1978 (in Russian). For providing of diagnosing of the technical state of power aggregates of hydraulic drives at their tests and break-in process it is suggested to apply hydraulic stands with recuperation of power. The row of constructions of such stands is developed. The method of calculation parameters of hydraulic stand with recuperation of power is offered. At the tests of hydraulic aggregates of expense of power on their conducting go down in some times. Continuous analysis of signs of the state of aggregates which test apply two indexes: mean value and dispersion of general efficiency of aggregates for period of their test. 170 DIAGNOSTYKA ’3(47)/2008 Warto przeczytaü / Worth to read

Editors kind of fault, i.e. cracks in shafts were documented Wojciech MOCZULSKI as well. Krzysztof CIUPKE The part ends with a concept of adapting the system to changing operating conditions and varying overall technical state of the machine. The second Knowledge acquisition part prepared by the group from the Institute of for hybrid systems of Fundamental Technological Research, Polish risk assessment and Academy of Sciences, presents other approach to critical machinery a specific kind of faults of rotating machinery which diagnosis are cracks in shafts of machines. The method involves a stochastic attempt that takes advantage ITE Radom 2008 of Monte-Carlo simulation. In order to detect and locate individual cracks the knowledge base of multiple cases of cracks developing at different The work contains the results of research locations was used. Moreover, some issues activity of seven teams from six research centers. concerning durability of cracked machine shafts Parallel proceeding of these teams has been part of were discussed. the research carried out within the framework of The third part of the monograph was elaborated the research project "Integrated dynamic system for by the scientific group from the Chair of Building risk assessment, diagnostics and control of technical and Operation of Vehicles and Machinery, the Uni- objects and processes" (project acronym: DIADYN) versity of Warmia and Mazury in Olsztyn. This supported by the Polish Ministry of Science and work was targeted at a system for operational risk Higher Education (formerly Polish Committee for assessment. Implementation of the system concerned Scientific Research) from year 2005 until 2008. The several objects operated in the food industry. The steering group of the project decided to manage the system includes identification of risks to be met project in the form of a package of groups of tasks within operation of the selected machinery. Further (GTs). The presented book integrates the results of on, a survey of methods of risk analysis was carried two GTs and has been mainly concerned on out. Finally, the method for assessment of knowledge acquisition for hybrid systems capable of operational effectiveness in the risk aspect was dealing with risk assessment and diagnostics of presented. critical machinery. The fourth part is the result of work of the team The book will be interesting for industrial from the Department of Vibroacoustics and engineers, scientists and students as well who Biodynamics of Systems, PoznaĔ University of wish to pursue the risk assessment and advanced Technology, and concerns a diagnostic agent for technical diagnostics issues of safety-critical critical machinery. The main issue was based on machinery. a multi-agent approach. The part includes The last paragraph explains the main objective examples of applications of the elements of the of the work that we have edited. The goal of the system to the diagnostics of rotating machinery teamwork of the seven research groups has been and vehicles. to address several issues of knowledge acquisition The next part prepared by the scientific group of for hybrid systems of risk assessment and critical the Department of Vehicles Engineering and machinery diagnosis. The book has been composed Transportation, the University of Technology and of parts that correspond to the issues of research Life Sciences in Bydgoszcz, concerns issues of the groups, each of them being an original insight to diagnostics through identification. This software- the title topic of the monograph. oriented approach takes advantage of the theory of The first part concerns a system for acquiring technical diagnostics. A novel software tool diagnostic knowledge of selected faults of critical developed within the DIADYN project was rotating machinery, and contains results of described. The part ends with the description of an collaboration between the research groups from the on-board diagnostic system for critical machinery. Department of Fundamentals of Machinery Design, Finally, the last part of the book, authored Silesian University of Technology at Gliwice, and by the group of the Institute of Fundamental from the Institute of Fluid Flow Machinery, Polish Technological Research, Polish Academy of Academy of Sciences. After dealing with the issues Sciences, is devoted to numerical identification of of knowledge acquisition with special attention paid loads. Two approaches were concerned: model- to knowledge discovery in databases, selected faults less identification based on pattern recognition in rotating machinery were described together with methods, and methods taking advantage of employed methods of their modeling and numerical models, which allowed reconstruction of simulation. Then the results of simulation load, and continuous identification of loads. Another experiments were presented in the form of problem addressed thorough the work concerns prototype diagnostic relations. Further on, a system optimal location of sensors upon the mechanical for knowledge acquisition was presented and several system. results of system verification for one distinguished DIAGNOSTYKA ’3(47)/2008 171 Warto przeczytaü / Worth to read

starzeniem fizycznym. W związku z tym istota Wojciech BATKO zagadnienia tkwi w tym, Īe naleĪy opisaü maszyny Zbigniew diagnostycznymi modelami nieliniowymi, które DĄBROWSKI mogą rozwiązaü problem diagnozowania Jan KICIēSKI (monitorowania), prognozowania i genezowania stanu maszyn, za pomocą symptomów wibroakustycznych. Zjawiska nieliniowe Biorąc pod uwagĊ niektóre przedstawione w diagnostyce informacje dotyczące diagnostyki wibroakustycznej technicznej maszyn naleĪy stwierdziü, Īe opiniowana praca Panów Profesorów Jest bardzo aktualna i wypeánia lukĊ dotyczącą tej trudnej problematyki. Autorzy ITE Radom 2008 podstawili sobie bardzo ambitne, trudne do realizacji zadanie: rozpatrzenie i dokonanie analizy niektórych Efektywne funkcjonowanie dowolnych zjawisk nieliniowych wystĊpujących w diagnostyce systemów dziaáania na przykáad: przemysáowych, maszyn. Konsekwentnie starają siĊ wykazaü rolniczych, wojskowych i innych jest uĪytecznoĞü róĪnego rodzaju dynamicznych modeli zdeterminowane stanem maszyn i urządzeĔ strukturalnych do opisu zjawisk nieliniowych, technicznych. Stanem zdatnoĞci funkcjonalnej których ewolucja moĪe stanowiü wiarygodny i zadaniowej obiektów technicznych naleĪy symptom diagnostyczny. Taki sposób postĊpowania sterowaü. NarzĊdziem sterowania są metody nazywają diagnostyką wedáug modelu. diagnostyki technicznej. W drugim rozdziale pracy Autorzy w sposób Maszyny to systemy techniczne z przepáywem wyczerpujący i precyzyjny przedstawili nastĊpujące masy, energii i informacji. Są to systemy zagadnienia: transformujące energiĊ z nieodáączną ich - pojĊcie nieliniowoĞci; dyssypacją: wewnĊtrzną i zewnĊtrzną. NoĞnikiem - diagnostykĊ maszyn wedáug modelu (defekty, informacji model teoretyczny ukáadu, symptomy, relacje o stanie maszyn są procesy: diagnostyczne); - robocze, w których zachodzi przetwarzanie - filtracja i predykcja obserwowanego sygnaáu; jednego rodzaju energii w inny lub jej - relacja sygnaáļ model. przenoszenie; Trzeci rozdziaá pracy jest poĞwiĊcony istotnemu - towarzyszące (resztkowe) na przykáad: zagadnieniu, tzn. identyfikacji dynamicznego modeli nagrzewanie siĊ elementów, zanieczyszczenie nieliniowego, która trafnie zdefiniowano oleju, a w szczególnoĞci procesy nastĊpująco: „identyfikacją modelu matematycznego wibroakustyczne. nazywamy wszelkie dziaáania, w wyniku których Niestety, pomimo wielkich dokonaĔ wielu proponowany model matematyczny odpowiada badaczy, na dzieĔ dzisiejszy nie poznano dokáadnie rzeczywistoĞci (obserwacji) w sensie jakoĞciowym natury generacji procesów wibroakustycznych, i iloĞciowym zgodnie z przyjĊtymi kryteriami zatem nie rozwiązano problemu podstawowego, to i zachowujĊ tĊ odpowiednioĞü przy przewidywanym jest przyporządkowania wybranych skáadowych zakresie dopuszczalnych zmian, to znaczy pozwala sygnaáu, konkretnym parom kinematycznym na wnioskowanie o aktualnym obserwowanym i mechanizmom maszyn. Przykáadem moĪe byü brak fragmencie rzeczywistoĞci, z zadana dokáadnoĞcią”. efektywnych metod wibroakustycznych badaĔ o Inne niektóre rozpatrzone zagadnienia tego ceny stanu: silników spalinowych, przekáadni rozdziaáu to: hydromecha-nicznych, czy mostów napĊdowych - identyfikacja modelu liniowego; pojazdów mechanicznych. Taki stan rzeczy wynika - rozwiązanie ukáadu sáaboliniowych równaĔ z faktu, Īe nie dysponujemy odpowiednimi róĪniczkowych zwyczajnych; modelami diagnostycznymi maszyn. Z reguáy - badania symulacyjne odpowiedzi nieliniowych stosowane są modele liniowe, które odzwierciedlają w warunkach zakáóceĔ; stan obiektów rzeczywistych z pewnym - identyfikacja relacji sygnaáļ model za przybliĪeniem. pomocą filtracji Kalmana. Model, w którym wystĊpują zaleĪnoĞci miĊdzy: W rozdziale czwartym pracy rozpatrzono proste parametrami sygnaáów diagnostycznych, masami, modele nieliniowe, ukáadów wirujących, elementami sprĊĪystymi, táumiącymi oraz a w szczególnoĞci: wymuszeniami nazywa siĊ diagnostycznym - nieliniowy model ruchu wirnika na sztywnych modelem strukturalnym obiektu technicznego. podporach; Modele diagnostyczne nie wyznaczają stanu - model wirnika na podporach elastycznych; obiektu, lecz są podstawą budowy okreĞlonych - model dynamiczny ukáadu przenoszenia mocy. algorytmów diagnozowania, za pomocą których W rozdziale piątym opracowania w sposób ustala siĊ dopiero jego stan. oryginalny przedstawiono problem W czasie eksploatacji maszyn pojawiają siĊ wykorzystywania filtracji Kalmana i narastają zaburzenia nieliniowe związane z ich w diagnozowaniu ruchów czopaw panwi áoĪyska 172 DIAGNOSTYKA ’3(47)/2008 Warto przeczytaü / Worth to read hydrodynamicznego. Rozwiązano zadanie eliminacji Przedstawiono zostaáa wáasna oryginalna metoda zakáóceĔ w systemie monitorującym drgania czopa okreĞlania wspóáczynników sztywnoĞci i táumienia waáu w panwi áoĪyska hydrodynamicznego, áoĪysk Ğlizgowych, oparta na modelach nieliniowych bĊdącego czĊĞcią systemu dynamicznego: „wirnik – áoĪysk Ğlizgowych i zmodyfikowanej metodzie áoĪyska – podpory  fundament”. Zaprezentowano perturbacji. Razem stanowi to bardzo wyniki symulacji filtracji zakáóceĔ i ich odniesienie zaawansowany, kompleksowy i wzajemnie spójny do rzeczywistych wyników pomiarów. Uzyskane model dynamiczny záoĪonego ukáady typu: linia wyniki wskazują nowy obszar moĪliwych w tym wirników z imperfekcjami-konstrukcja podpierająca, zakresie rozwiązaĔ. do analizy zarówno w zakresie liniowym, a przede Rozdziaá szósty pracy omawia zagadnienia wszystkim w zakresie nieliniowym. Nie jest znany funkcji koherencji jako miary zaburzenia inny odpowiednik takiego modelu. nieliniowego, lokalizacji uszkodzeĔ struktury Opracowany model ukáadu: linia wirników- kompozytowej obiektu. Stwierdzono, Īe analiza áóĪyska Ğlizgowe-konstrukcja podpierająca stanowi koherencyjna pozwala na obserwacjĊ propagacji zestaw wielu równaĔ i zaleĪnoĞci o bardzo záoĪonej uszkodzenia struktury masztu kompozytowego strukturze. Otwiera on nowe moĪliwoĞci z dokáadnoĞcią, co najmniej porównywalną z metodą w badaniach tego typu maszyn. rentgenograficzną i moĪe byü zarówno podstawą PracĊ koĔczy rozdziaá dziesiąty poĞwiĊcony identyfikacji modelu matematycznego jak okreĞleniu przedziaáu adekwatnoĞci opracowanego i znalezienia miary symptomowej uszkodzenia. modelu turbozespoáu. Wobec niemoĪliwoĞci W rozdziale siódmym Autorzy pokazali w jaki uzyskania analitycznego opisu „przeksztaácenia sposób moĪna wykorzystaü rozwiązania teorii odwrotnego”, jedynym sposobem osiągniĊcia statecznoĞci technicznej w procesie konstruowania pozytywnych efektów w zadaniu identyfikacji systemów monitorujących zmiany stanów: maszyn, modelu jest okreĞlenie przedziaáów, dla których urządzeĔ, konstrukcji, bądĨ procesów. Jak wykazaáy wyznaczone charakterystyki dynamiczne są badania symulacyjne, warto uwzglĊdniü wáaĞciwe (tzw. przedziaáów adekwatnoĞci). w konstrukcji systemu monitorującego, warunek Przedstawiono wyniki badaĔ dwóch obiektów: potrzeby kontrolowania zmian portretów fazowych. wirnika „laboratoryjnego” i turbozespoáu duĪej Mają one duĪą wraĪliwoĞü na zmiany stanu mocy. W badaniach teoretycznych nadzorowanego obiektu. PodkreĞlono, Īe nie da siĊ i eksperymentalnych wykorzystano eksploatacyjną zbudowaü nowoczesnego systemu monitorującego analizĊ modalną. Zostaáy okreĞlone przedziaáy bez podejĞcia systemowego, tzn. uwzglĊdniającego: adekwatnoĞci jak i charakterystyki dynamiczne wiedzĊ diagnostyczną, metody analizy dynamiki badanych obiektów. Wyniki obliczeĔ przedstawione badanego obiektu, analizĊ jego zachowaĔ zostaáy w formie kart diagnostycznych, nieliniowych i związanych z nimi ocen ich wspomagających okreĞlenie stanu badanych statecznoĞci. elementów maszyn wirnikowych. WaĪnym elementem pracy jest rozdziaá ósmy, Reasumując opiniĊ o pracy naleĪy stwierdziü, co w którym Autorzy przedstawili opracowany nastĊpuje: nieliniowy model diagnostyczny turbozespoáu, 1) ksiąĪka jest pierwszą krajową publikacją obejmującego miĊdzy innymi nastĊpujące poĞwiĊconą zjawiskom nieliniowym zagadnienia: w diagnostyce wibroakustycznej ; - model struktury maszyny wirnikowej i algorytm 2) praca wnosi nowe wartoĞci poznawcze realizacji obliczeĔ dynamicznych; i utylitarne do nauki o badaniach i ocenie stanu - model konstrukcji podpierającej; maszyn, w szczególnoĞci diagnostyki - transformacjĊ zespolonych charakterystyk wibroakustycznej; podatnoĞciowych konstrukcji podpierającej do 3) publikacja wytycza nowy kierunek dziaáaĔ rzeczywistych charakterystyk masowo  w badaniach wáaĞciwoĞci i zastosowania táumiąco - sztywnoĞciowych; procesów wibroakustycznych w iagnozowaniu - algorytm obliczeĔ nieliniowych ukáadu: linia (monitorowaniu), prognozowaniu i enezowaniu wirników  konstrukcja podpierająca stanów maszyn; z wykorzystaniem funkcji wagowych. 4) sądzĊ, Īe monografia bĊdzie podstawą dalszych Rozdziaá dziewiąty pracy Autorzy poĞwiĊcili prac opracowania nieliniowych modeli realizacji cyfrowej nieliniowoĞci modelu diagnostycznych i ich peánego wykorzystania turbozespoáu maáej i duĪej maszyny wirnikowej. w procesach utrzymania maszyn. Linia wirników skupia na sobie oddziaáywania wszystkich podzespoáów maszyny wirnikowej, tzn. oddziaáywania áoĪysk Ğlizgowych i konstrukcji podpierającej, wyraĪone w formie wspóáczynników sztywnoĞci i táumienia filmu olejowego i macierzy wspóáczynników sztywnoĞci i táumienia mas pozostaáych elementów.

Recenzenci / Reviewers: prof. dr hab. inĪ. Wojciech BATKO dr hab. inĪ. Henryk MADEJ, prof. Pĝ prof. dr hab. inĪ. Zbigniew DĄBROWSKI prof. dr hab. inĪ. Ryszard MICHALSKI dr hab. inĪ. Janusz GARDULSKI, prof. Pĝ prof. dr hab. inĪ. Wojciech MOCZULSKI prof. dr hab. inĪ. Jerzy GIRTLER prof. dr hab. inĪ. Stanisáaw NIZIēSKI doc. dr hab. Henryk KAħMIERCZAK dr hab. inĪ. Andrzej PIĉTAK, prof. UWM prof. dr hab. inĪ. Yevhen KHARCHENKO prof. dr hab. inĪ. Stanisáaw RADKOWSKI prof. dr hab. inĪ. JóĪef KORBICZ dr hab. inĪ. Henryk TYLICKI, prof. UTP dr hab. inĪ. Waldemar KUROWSKI, prof. PW prof. dr hab. inĪ. Andrzej WILK dr hab. inĪ. Bogdan àAZARZ, prof. Pĝ prof. dr hab. inĪ. Bogdan ĩÓàTOWSKI

Druk: Centrum Graficzne „GRYF”, ul. PieniĊĪnego 13/2, 10-003 Olsztyn, tel. / fax: 0895272430 Oprawa: Zakáad Poligraficzny, UWM Olsztyn, ul. Heweliusza 3, 10-724 Olsztyn tel. 089-523-45-06, fax: 089-523-47-37