Advanced Model-based Diagnosis of Internal Combustion Engines - A short survey of research results -
Prof. Dr.-Ing. R. Isermann
Technische Universität Darmstadt Institut für Automatisierungstechnik und Mechatronik FG Regelungstechnik und Prozessautomatisierung www.rtm.tu-darmstadt.de/rtp.html
R.I. 2018 Advanced Model based Diagnosis of Internal Combustion Engines 1. Introduction 2. Fault detection and diagnosis methods
3. Diagnosis of diesel engines 31 Intake system 3.2 Common rail injection system and combustion 3.3 Turbocharger
4. Diagnosis of gasoline engines 4.1 Intake system 4.2 Fuel system (high pressure system) 4.3 Combustion
5. Conclusions
R.I. 2018 1. Introduction
• Increasing complexity of combustion engines: • More components • More sensors and actuators • More control functions • More sensitive combustion processes → Improved monitoring and fault diagnosis capability required
R.I. 2018 1. Introduction
• Increasing complexity of combustion engines: • More components • More sensors and actuators • More control functions • More sensitive combustion processes → Improved monitoring and fault diagnosis capability required • On-board diagnosis functions (OBD, since 1988) • Driven by legislation • limited to emission related (large) deviations from normal behaviour • Only signal changes and plausibility tests • Off-board diagnosis systems in service-stations: • Workshop testers , connected to the OBD-plug • allow more OEM-designed diagnosis functions (not all possible faults) • Improvement of diagnosis functions should enable: • Detection and localisation of defective components • Detection of faults: small, incipient, abrupt, intermittant • Appropriate distribution of on-board and off-board diagnosis functions • Predictive maintenance • Appropriate reconfiguration to emergency operation
R.I. 2018
1. Introduction
• Increasing complexity of combustion engines: • More components • More sensors and actuators • More control functions • More sensitive combustion processes → Improved monitoring and fault diagnosis capability required • On-board diagnosis functions (OBD) • Driven by legislation • limited to emission related (large) deviations from normal behaviour • Only signal changes and plausibility tests • Off-board diagnosis systems in service-stations: • Workshop testers connected to OBD plug • allow more OEM designed diagnosis functions (do not diagnose all possible faults) • Improvement of diagnosis functions should enable: • Detection and localisation of defective components • Detection of faults: small, incipient, abrupt, intermittant • Appropriate distribution of on-board and off-board diagnosis functions • Appropriate reconfiguration to emergency operation • Future aspects, challenges: • Maintenance on demand • Remote (tele)-diagnosis during driving • Reconfiguration to fault tolerant functions
R.I. 2018
Advanced Model based Diagnosis of Internal Combustion Engines 1. Introduction 2. Fault detection and diagnosis methods
3. Diagnosis of diesel engines 31 Intake system 3.2 Common rail injection system and combustion 3.3 Turbocharger
4. Diagnosis of gasoline engines 4.1 Intake system 4.2 Fuel system (high pressure system) 4.3 Combustion
5. Conclusions
R.I. 2018 2. Fault detection methods
Fault Detection Methods
Conventional Model-Based Methods Methods
Limit /Trend. Plausibility Analysis of Analysis of Checking Checks Signal Models Process Models
OBD (if emission relevant)
Statistical Time Monitoring Correlation Spektral Wavelet Parameter State Parity Evaluation Analysis Analysis Analysis Estimation Estimation Equations
R.I. 2018 Model-based fault detection and diagnosis
I) signal model-based fault detection II) process model-based fault detection
→ Use of single sensor signals only → Use of precise relations between two or more sensor signals
R.I. 2018 Model based fault detection for combustion engines Faults
Faults Faults Diesel Engine N Fault Detection U Actu- Sen- Y ators sors Fourier transformation, statistical features (standard deviation) Parameter estimation Parity equations State variable estimation Signal or Process Models Neural networks Features Fault Detection
Nominal Residual Fault Diagnosis Fault Diagnosis Behavior Generation - Fault Loc alization Threshold Supervision - Fault Type Geometric/statistical methods Detec tion Symptoms - Fault Size Fault-symptom-trees of Changes - etc. Fuzzy-rules Neuro-/Neuro-Fuzzy-approaches
R.I. 2018 Advanced Model based Diagnosis of Internal Combustion Engines 1. Introduction 2. Fault detection and diagnosis methods
3. Diagnosis of diesel engines 31 Intake system 3.2 Common rail injection system and combustion 3.3 Turbocharger
4. Diagnosis of gasoline engines 4.1 Intake system 4.2 Fuel system (high pressure system) 4.3 Combustion
5. Conclusions
R.I. 2018 Engine testbench
. Asynchronous machine 160kW . Automation with dSpace systems . >15 pressure and temperature sensors . Fuel flow measurement . Exhaust gas measurement AVL: . FTIR . MicroSoot . Opazimeter . SmokeMeter . NOx-sensor
R.I. 2018 Investigated diesel engines: 1) 2 l, 4 cyl, 74 kW, 230 Nm 2)1.9 l, 4 cyl, 110 kW, 315 Nm Measurements at test bench:
R.I. 2018 pat Intake S Tat Detection vol. efficiency pipes m methods S air amplitude() p2,I Intake system S Turbo com- phase() p2,E uSFA pressor Samplitude() m p L 2,I S phase() mL Intercooler T2,I Sboosttemp Engine uSFA SFA Swirl SSFA actuator control actuator flaps system mB Injection Com-
EB system bustion
ne M i Engine Toil
mechanics Tw EGR uEGR valve pT, Exhaust EGR EGR gas uETC ETC recirculat. n actuator O2 Exhaust turbine
Exhaust pipes
R.I. 2018 Air path with possible, implemented faults
Turbine power EGR- Inlet manifold Valve- Fault Swirl port Filling port Restriction Intercooler
Compressor Manifold T2,I Leakage temperature sensor
Crank Open case vent joining Manifold p pressure 2,I sensor Air mass sensor Swirl flaps mair,sensor actuator
Air filter Built-in faults Fresh air
R.I. 2018 Modeling of nonlinear intake behavior Physical Modeling: ideal displacement pump Identification with net model (LOLIMOT)
p2,I Continuous Function: v fn eng, 2, I T 2,I p3
mair,eng
]
- Volumetric efficiency[ Volumetric neng
Modeling • Dynamic flow effects includes: • Charge heating • exhaust gas effects Volumetric mm&& air,, eng air eng Distribution of measuring points across efficiency: v p the identified model m&air, th 1 2,I nVeng D 2 RT 2,I
R.I. 2018 FourierModeling analysis of crank angleModeling of crank of angle Crank Angle synchronous Synchronous Boost boost Pressure pressure Oscillation oscillation 2 Boost pressure 1.9 [bar] 1.8 Amplitude: n = 2000min-1 eng 1.7 0 180 360 540 720 Phase: Crank angle [°CA] p2,I p 2, I fA n eng, 2, I cos 2 f n eng , 2, I p 180KW p LOLIMOT-
Modeling:0.15 40
0.1 20
[bar]
KW]
° [
Phase 0
Amplitude 0.05
-20 0 3 2.5 4000 4000 3000 2 3000 2 1.5 2000 2000 1 1000 1 1000
R.I. 2018 Calculation of fault residuals with parity equations Input 1) Residual Vol. Efficiency variables
rvv f neng, 2, I v Reference Real model process 2) Residual amplitude boost pressure oscillation Reference meassured r A f n , model signal Difference process signal AAppp2,I eng2, I or feature or feature
3) Residual phase Residual boost pressure oscillation Jump-cut Ausblendung function, if r f n , falls Voraus- ppp2,I eng2, I preconditions aren’tsetzungen fullfilled 4) Residual amplitude air mass flow oscillation Low pass (optional) r A f n , AAmm&&m&air, sensor eng2, I Dead-zone 5) Residual phase air mass flow oscillation r f n , Threshold value mm&&m&air, sensor eng2, I Symptom
R.I. 2018 Results f Online Fault Detection at Steady Operating Point Residual deflections of the intake system for different faults
Operating point: 2000min-1, 130Nm, boost pressure: 1.5bar, air mass flow 165kg/h
EGR valve open 0.4 Removed tube of the 0.3 Leakages after intercooler crank case vent 5mm 4mm 7mm r [-] 0.2 v 0.1 0 -0.1
10 Thresholds 0 of fault r [mbar] -10 detection Ap Swirl flaps actuator Restriction -20 filling ports closed between -30 intercooler and engine
20
10 r [kg/h] 0 Am& -10
-20 0 50 100 150 200 Time [sec]
R.I. 2018 Fault-Symptom-Table for Fault Diagnosis Fault-symptom table for the intake system
Faults Symptoms S S S S S Am& Ap m& p Removed tube of the crank case vent + o o o o Leakage between - o o o o intercooler and engine Restriction between o - - + + intercooler and engine Swirl flaps actuator, filling port is closed o - - o o
EGR-valve: stuck at open ++ - - o o
Leaky EGR-valve + o o o o
Legend: Symptom volumetric efficiency S ++ Symptom responds intense positive S Symptom amplitude air mass flow oscillation Am& + Symptom responds positive Symptom amplitude boost pressure oscillation
- Symptom responds negative S Symptom phase air mass flow oscillation m& o Symptom does not respond S Symptom phase boost pressure oscillation p → Different fault types can be isolated
R.I. 2018 Advanced Model based Diagnosis of Internal Combustion Engines 1. Introduction 2. Fault detection and diagnosis methods
3. Diagnosis of diesel engines 31 Intake system 3.2 Common rail injection system and combustion 3.3 Turbocharger
4. Diagnosis of gasoline engines 4.1 Intake system 4.2 Fuel system (high pressure system) 4.3 Combustion
5. Conclusions
R.I. 2018 Investigation of a Common Rail Injection System (3 plungers, Bosch CP1H)
high pressure pump manipulation current of
and metering valve pressure control valve Ipcv • manipulation current
of metering valve Imv common rail pressure pcr • engine speed neng • crank angle φ pressure common rail control valve pressure sensor
electric fuel pump
Opel 1.9 l DTH Z19 injectors components 5 measured variables
R.I. 2018 Characteristics of the Common Rail Pressure Sensor Signal
Mass balance of the common rail:
(1)
R.I. 2018 Mean Value Model of the Common Rail Pressure The pressure gradient in the common rail follows the equation
Ecr: bulk modulus of the fuel in the common rail, Vcr: volume of the common rail
Mean value over the period 2160°C (third cycle) averages the oscillations:
α: angle for which the computation is performed
In the fault free case this mean value is a function of • the orifice cross sections of metering valve and pressure control valve • the engine speed • the injector volume flow residual 1 “rail pressure mean value”:
R.I. 2018 Characteristic of the Common Rail Pressure Sensor Signal Common rail pressure in a constant operation point is mainly influenced by • the crank angle dependent injections • crank angle synchronous fuel delivery of the high pressure pump periodical oscillations of the rail pressure Idealized illustration of the oscillations for a constant operation point identicalvolume flows volumeto the flows toinjectorsunbalanced the τinj τbank τsup injectorsunbalancedinjections and fuel identical fuel unbalanceddelivery delivery of the fuel delivery three pistons τpist τhpp τsup Periods of rail pressure oscillations:
main injection cycle: τinj = 180°CA second injection cycle (unbalanced injections): τbank = 720°CA main pump cycle: τpist = 180°CA second pump cycle (unbalanced fuel delivery): τhpp = 540°CA (3 pistons) third injection and pump cycle through superposition: τsup = 2160°CA = 3x4x180°CA period τ = 1/ Ω (angle frequency)
R.I. 2018 Common Rail Pressure Signal Analysis – Fourier Amplitudes, normal behavior
complex Fourier coefficients: ↓ 540
(2)
amplitudes: (3)
R.I. 2018 Common Rail Pressure Signal Analysis – Fourier Amplitudes (with unbalanced injections)
↓ 720
pressure wave resulting from injection
R.I. 2018
Rail pressure uniformity analysis
Periodic signal:
Injektoren τ τ τ
inj bank sup
-
Uniformity analysis: durch Hoch
τpist τhpp τsup
druckpumpe Schwingungsanregung Uniformity analysis of rail pressure signal:
Task: injector or pump fault?
• Injectors: signal is periodic with 720°CA = 4·180°CA (second injection cycle)
•
• High pressure pump signal is periodic with 540°CA = 3·180°CA (3 pistons), (second pump cycle)
• Pressure measurements: filtered with a moving average filter over several periods (1440, 2160 oCA)
R.I. 2018
Rail pressure residuals
Periodic function:
Injektoren τ τ τ
inj bank sup
-
Uniformity analysis: difference of phase durch
shifted signals Hoch
τpist τhpp τsup
druckpumpe Schwingungsanregung Uniformity analysis of rail pressure signal::
• residual 2 “unequal pump delivery”
• residual 3 “unequal injections”:
R.I. 2018 High Pressure Pump Volume Flow Analysis (no injection, overrun state)
Characterization of the mean pump flow oscillation over 3x180°CA = 540°CA by the pressure signal’s variance:
The variance is intended to detect faults in the low pressure part (influences all three pistons equally). Depends on the • rail pressure • engine speed • position of the metering valve
residual 4 “high pressure pump volume flow”: (13) R.I. 2018 Residuals for different faults (sampling with 1 CA degree)
Rail pressure mean value Raildruckmittelwert
Unequal fuel delivery Gleiche Pumpmengen
Unequal injections Gleiche Einspritzmengen
High pressure pump volume flow Hochdruck-Pumpen- Volumenstrom
R.I. 2018 Fault Symptom Table for the Common Rail System Faults Symptoms S =f (r ) Diagnosability * limit *
Sinj,1 Sinj,2 Sinj,3 Sinj,4 general special
F1: low delivery quantity of one pump piston 0 + 0 +
F2: reduced injection quantity of one injector 0 0 + 0
F3: pressure loss in front of high pressure pump - 0 0 +/- – (e.g. a plugged fuel filter) F4: pressure in front of high pressure pump too + 0 0 +/- – high (e.g. a faulty metering valve) F5: opening of the pressure control valve is too - 0 0 0 large F6: opening of the pressure control valve is too + 0 + 0 small F7: pressure sensor signal is too high + 0 0 - –
F8: pressure sensor signal is too low - 0 0 + – Diagnosability: different patterns show: general : fault is generally diagnosable (isolable) special : fault is diagnosable, if several operation points are taken into account
R.I. 2018 Advanced Model based Diagnosis of Internal Combustion Engines 1. Introduction 2. Fault detection and diagnosis methods
3. Diagnosis of diesel engines 31 Intake system 3.2 Common rail injection system and combustion 3.3 Turbocharger
4. Diagnosis of gasoline engines 4.1 Intake system 4.2 Fuel system (high pressure system) 4.3 Combustion
5. Conclusions
R.I. 2018 Faults: Investigated Faults 1. Blow-by tube removed 15: Compr. blades damaged 2. Leak exh. 9mm 7,8: sVGT blocked 3. Leak. Int. 5mm 4. Leak. Int. 5mm 1: blowby tube 5. Leak. Int. max 6. Restriction int. 7. sVGT bl. open
12: air filter clogged 8. sVGT bl. Mid 9. No fan airflow intercooler 10,11: HPEGR Pos. 10. hpegr bl. mid 9: no fan airflow 3,4,5: leakage 11. hpegr bl. closed 12. Air filter clogged 13. Swirl fl. pos. 0 2: leakage 14. Swirl fl. pos 75 6: restriction 15. Compr. blades 13,14: swirl flap pos.
R.I. 2018 Turbocharger power model
ntc
p1 p2c Compressor Compressor T1 power torque PC T3 MC
Friction 2π/ 1/s ntc torque Jtc MF
p3 ntc p4 M Turbine Turbine T T 3 torque power PT T1
svgt ntc ntc
R.I. 2018 Modeling of turbocharger power and speed
• Compressor power: a 1 1 p a P m& c T * 2c 1 c c p,1 a c p1 Additional Models • Turbine power: e 1 . p e P m c T * 1 4 t t p, e t , aero 3 p3 • Friction power: 22 Pf(2 ) K f n tc
• Turbocharger speed: 1 PPP n& ()2 t c f tc 2 nJ tc tc R.I. 2018
Simulation of turbine, compressor and friction power
16000 Pt Pc 14000 Pf
12000
10000
8000 power in W 6000
4000
2000
0 500 1000 1500 2000 2500 time in s
R.I. 2018 generation of turbocharger residuals Calculated outputs of power models Measured model inputs Thermodynamic turbocharger model
Calculated model outputs
measured outputs
Stationary Reference Power Models
R.I. 2018 Fault-symptom table for the turbocharger, with air and exhaust path for a medium operation region
Faults Symptoms
Spt Spc Spf Sn Sp2 Sp3 Sp4 A F1 Restriction air 0 0 0 0 ─ 0 ─ filter Air F2 Blow-by tube 0 0 ─ ─ ─ 0 ─ Path removed F3 Leakage 0 0 + + ─ ─ 0 intake 5 mm F4 Leakage + + + + ─ ─ 0 intake 7 mm F5 No cooling 0 0 0 0 ─ 0 + airflow of the intercooler F6 Restriction 0 0 ─ ─ ─ 0 ─ behind intercooler F7 Leakage 0 0 0 0 ─ 0 ─ B exhaust Exhaust F8 HP-EGR valve ─ ─ ─ ─ ─ + 0 blocked path closed C F9 Compressor 0 0 0 0 ─ 0 0 blades Turbo- damaged charger F10 VGT blocked + + + + 0 0 ─ middle position Symptoms are changes of residuals which exceed thresholds →The faults show different symptoms and can therefore be diagnosed, Sidorow et al. (2011)
R.I. 2018 Modular structure for a model based overall fault diagnosis of diesel engines
Auxiliary Detection Diagnosis ECU Actuators Components Sensors Symptoms units Modules Modules pa ,Ta ,T1 S U int,i flp intake flaps intake system mair ,s flp ,sth
Detection Uth throttle air compressor „intake system“ p ,T U fp 2E 2E fuel pump air cooler Diagnosis S „engine“ injection system prail , Iinj inj,i U inj Detection „injection“ combustion p cyl Scom,i valve train tank ventilation actuator engine U mechanics Detection vtr „combustion“
U egr segr , pegr ,Tegr Engine Control EGR valve(s) EGR path(s) Unit
Sexh,i U t turbine exhaust gas st ,T3 actuator(s) turbine Detection „exhaust system“ Diagnosis p ,v „exhaust“ 4 O2 exhaust system
Detection Seat,i U c ,T ,T cv exhaust gas NOx bDPF aDPF „exhaust aftertreatment gas aftertreatment“ Scol,i scool ,Tcool Detection coolant pump coolant valve cooling system „cooling“ U oilp Diagnosis „cooling Slub,i & oil pump lubrication Toil , poil Detection lubrication“ oil pump actuator system „lubrication“
R.I. 2018 Advanced Model based Diagnosis of Internal Combustion Engines 1. Introduction 2. Fault detection and diagnosis methods
3. Diagnosis of diesel engines 31 Intake system 3.2 Common rail injection system and combustion 3.3 Turbocharger
4. Diagnosis of gasoline engines 4.1 Intake system 4.2 Fuel system (high pressure system) 4.3 Combustion
5. Conclusions
R.I. 2018 Used engine sensors (series production) and investigated faults (combined) 2 NO-storagecat Oxycat
(continous) 1 Goal: Diagnosis of faults,like: 1. Leakage intake system, 4 before throttle (after HFM)
2. Leakage intake system, after throttle
3
dynamometer
3. Less fuel injection in one cylinder
VW FSI 1,6 l, 81 kW, 155 Nm 4. Increased EGR FVV-project, M. Leykauf 5 5. Reduced ignition energy
R.I. 2018 Advanced Model based Diagnosis of Internal Combustion Engines 1. Introduction 2. Fault detection and diagnosis methods
3. Diagnosis of diesel engines 31 Intake system 3.2 Common rail injection system and combustion 3.3 Turbocharger
4. Diagnosis of gasoline engines 4.1 Intake system 4.2 Fuel system (high pressure system) 4.3 Combustion
5. Conclusions
R.I. 2018 Modular structure of model based diagnosis for gasoline engines
Input signals Sensors Symptoms
u Throttle Throttle Fault detection S Air mass flow . Intake system u m Tumble Tumble air Intake S system Exhaust system Intake pressure p 2
(u ) T time synchr. S CS Camshaft 2 Output Lambda evaluation Controler T0 = 10ms Diagnosis
Combustion Fault detection Fuel System system with operating SRail pressure Fault: ECU High pressure modes fuel and n (in overrun state) pump eng combustion S type Homogeneous system Rail pressure Fault- size (t i ) p rail Injectors (injection) Stratified crankangle synchr. Symptom- u m S CV Control valve B evaluation Speed signal 1°CA Table
(l ) bef.Cat t ignition Ignition system (l after Cat ) Exhaust system (NOx ) ( ): not used for diagnosis
u (T ) EGR EGR-valve 3
R.I. 2018 Fault diagnosis with fault-symptom table (FSI) Nonlinear process model based Signal model based
Symptoms S1 S2 S3 S4 S5 S6 Mode homogen. Rail Rail stratified Air Output Speed Manifold pressure pressure mass Lambda signal press. (overrun (injec- flow ampl. contr. Fault state) tion) Leak before throttle - o - Leak after throttle (2mm) - + - Leak after throttle (3mm) - ++ - Increased EGR-mass flow o + d (with restriction) Less injection mass in one d + - cylinder Less pump fuel supply - o o () () Reduced ignition energy d o -- +: positive symptom -: negative symptom o: no symptom change d: don‘t care : applicable , : non applicable All investigated 6 faults are isolable (injection and ignition faults can be separated) R.I. 2018
Conclusions
• Combustion engines: Strongly nonlinear reciprocating multi-input multi-output processes (MIMO) • Simplification for diagnosis by considering of engine modules with parallel multi-input single - output models (MISO) • Fault detection by symptom generation with parity equations for output and input variables: • Primary residual = measured output/input – reference model output/input AND/OR • Secondary residual = calculated feature – reference feature (amplitude, frequency,…) • Reference models for normal behavior: – Process models (Input/output-behavior) • Physical models • Semi-physical models (physical based structure and identification) • Identified input-output models (parameter estimation, local linear neural networks) • Models used for calculating physical based features (amplitudes ,torques ,masses,...) – Signal models (periodic): • Single frequencies: amplitudes, phases • Multi frequencies: Correlation, Fourier analysis (FFT), Wavelet analysis
• Generation of many symptoms allows a large fault detection coverage and fault diagnosis • Fault diagnosis: e.g. developed as fuzzy inference system • Most important: Physical/engineering understanding, modeling and experiments
R.I. 2018 Conclusions Future aspects, challenges:
Advanced fault detection and diagnosis is required for:
• Maintenance on demand
• Fault prediction
• Remote (tele)-diagnosis during driving
• Reconfiguration to fault tolerant functions (In the context of automatic driving the powertrain may become a safety relevant system)
R.I. 2018 R. Isermann Combustion Engine Diagnosis Springer-Verlag, Heidelberg, Berlin 2017 ATZ/MTZ Fachbuch
R.I. 2018 END
R.I. 2018