POLITECNICO DI TORINO Repository ISTITUZIONALE

View metadata, citation and similar papers at core.ac.uk brought to you by CORE

provided by PORTO@iris (Publications Open Repository TOrino - Politecnico di Torino)

Prediction of performance and cutter wear in rock TBM: Application to Koralm tunnel project

Original Prediction of performance and cutter wear in rock TBM: Application to Koralm tunnel project / Brino, Gabriele; Peila, Daniele; Steidl, Arnold; Fasching, Florian. - In: GEAM. GEOINGEGNERIA AMBIENTALE E MINERARIA. - ISSN 1121- 9041. - 145:2(2015), pp. 37-58.

Availability: This version is available at: 11583/2642312 since: 2016-09-29T11:28:53Z

Publisher: Patron Editore S.r.l.

Published DOI:

Terms of use: openAccess This article is made available under terms and conditions as specified in the corresponding bibliographic description in the repository

Publisher copyright

(Article begins on next page)

04 August 2020 GEOINGEGNERIA E ATTIVITÀ ESTRATTIVA

Gabriele Brino* Prediction of performance Daniele Peila* Arnold Steidl** and cutter wear in rock TBM: Florian Fasching** application to Koralm tunnel * DIATI, Politecnico di Torino **3G Gruppe Geotechnik Graz project 1. Introduction Excavation by Tunnel Boring Machines is the tunnelling method most frequently used nowadays in long infrastructural projects, in a wide range of geological conditions. In the last 40 years, many predic- Tunnel Boring Machines are tion models were developed to estimate TBM performance and cutter wear, using as input geological parameters. The research gives an overview of the existing penetration models for hard rock TBMs, nowadays a common and fruitful identifies the most frequently used input parameters and summarizes the characteristics of the da- option a tunnel designer can adopt, tasets on which the models are based on. Theoretical background is tested through the example of and technological advancements Koralm tunnel project, a 32.9-km-long base tunnel in Austria, in a 1000-m-long portion of the South allowed using them in a wide range tube in the construction lot KAT 2. The outcomes shows that the estimation of the penetration is of geological conditions, from hard reasonably accurate when applying models that are based on a database consistent with the project rock to loose soil. data, especially in terms of geology and typology of machine used in the excavation. The article pro- From the adoption of disc cutter poses a design method for a system of TBM data analysis and prediction at the construction stage, based on a back-analysis process about machine data in different geological conditions. The method- tools by James Robbins in 1956, TBM ology can be applied in any other project and the system is particularly useful in long tunnels, in which productivity improved year by year. a continuous improvement of the ability of prediction can have an effective impact on time and costs. Today peaks of advance rate per day Keywords: TBM, penetration, wear, utilization, prediction can go over 100 m/day, but, in adverse geologic conditions, a production of Previsione delle prestazioni e dell’usura degli utensili di una TBM da roccia: applica- less than 50 m/month may be experi- zione al progetto del Koralm tunnel. Lo scavo con Tunnel Boring Machine (TBM) è il metodo di enced (Barla and Pelizza, 2000). scavo più utilizzato per grandi progetti infrastrutturali al giorno d’oggi, in un’ampia varietà di condizioni geologiche. Negli ultimi 40 anni molti modelli predittivi sono stati sviluppati per stimare le presta- In the last 40 years, several au- zioni della TBM e l’usura degli utensili, usando come input i parametri geologici. Questa ricerca forni- thors (Cassinelli et al., 1982; In- sce una panoramica dei modelli esistenti per la stima della penetrazione degli utensili per una TBM in naurato et al., 1990; Rostami and roccia, identifica i parametri di ingresso più frequentemente utilizzati e sintetizza le caratteristiche dei Ozdemir, 1993; Gehring, 1995; Al- database da cui sono stati ricavati i modelli. I modelli teorici sono stati testati attraverso il caso studio ber, 1996; Bruland, 1998; Alvarez del progetto del Koralm Tunnel, una galleria di base lunga 32.9 km in costruzione in Austria, in una Grima et al., 2000; Barton, 2000; porzione della canna Sud di lunghezza 1000 m, facente parte del lotto KAT 2. I risultati mostrano che Ribacchi and Lembo Fazio, 2005; la stima è ragionevolmente accurata se si applicano modelli che su fondano su una base di dati con Yagiz, 2008; Gong and Zhao, 2009; caratteristiche simili a quelle del progetto, specialmente in termini di geologia e tipologia di macchina usato per lo scavo. L’articolo propone un metodo di progetto per la creazione di un sistema di analisi Hamidi et al., 2010; Hassanpour et dei dati della TBM e di previsione in fase costruttiva, basata su un processo di back-analysis di dati al., 2011; Farrokh et al., 2012) pro- macchina raccolti in differenti condizioni geologiche. La metodologia può essere applicata in qualsiasi posed models that allow to predict altro progetto, e il sistema è particolarmente utile in tunnel di grande lunghezza, nei quali una miglio- TBM performance and cutter wear. ria continua dell’abilità di previsione può avere un impatto rilevante in termini di tempo e costi. The aim of the work is to under- Parole chiave: TBM, penetrazione, usura, utilizzazione, previsione stand if the quality of the prediction, obtained through the adoption of Prévision de la performance et de la consommation des molettes pour un tunnelier à roche dure : application au projet du tunnel du Koralm. Le creusement avec un tunnelier these models, can be influenced by à roche dure est désormais la méthode de construction plus utilisée dans les projets d'infrastructures the characteristics of the databases longues, sur une fourchette très large de conditions géologiques. Dans les derniers 40 ans, beaucoup models in literature are based on (Bot- de modèles prévisionnels pour l'estimation de la performance et de la consommation des molettes tero and Peila, 2005). Data analysed pour un tunnelier à roche dure ont été développés, utilisant les paramètres géologiques comme don- in the study has been recorded during nées d'entrée. La recherche donne une vision globale des modèles existants en matière de pénétra- the excavation of the Koralm tunnel, tion pour les tunneliers à roche dure, elle identifie les paramètres d'entrée plus utilisés et elle synthé- a railway tunnel excavated in hard tise les caractéristiques des séries de données à la base du modèle. Les hypothèses théoriques sont testées sur l'exemple du projet du tunnel du Koralm, un tunnel de base autrichien de 32,9 km de rock in the southern Austria. longueur, et notamment sur une section de 1000 m de longueur du tube sud du lot constructif KAT 2. In mechanized tunnelling by Les résultats montrent que l'estimation de la pénétration est raisonnablement précise quand le TBM, the process of rock fragmen- modèle appliqué est basé sur une base de données cohérente avec les paramètres du projet, spé- tation is guaranteed by cutters, steel cialement en terme de géologie et de type de tunnelier utilisé pour le creusement. discs that are capable of dealing with

Geoingegneria Ambientale e Mineraria, Anno LII, n. 2, agosto 2015, 37-50 37 GEORESOURCES AND MINING

L'article propose une méthode de conception pour un système d'analyse des données du tunnelier 1.1.1. Analytical models et de prévision en phase de construction, basé sur un procès d'analyse à rebours des données machine en conditions géologiques différentes. La méthode peut être appliquée sur d'autres projets et le système est particulièrement utile pour Analytical performance predic- des longs tunnels, où l'amélioration continue de la capacité prévisionnelle peut avoir un retour effec- tion models are based on the study of tif sur les délais et sur les coûts. the rock fragmentation process with Mots-clés: TBM, pénétration, consommation, utilisation, prévision. mechanical tools. From the results of full-size laboratory tests, as linear the high cutter loads required for discs in the same position, due to cutting and punch penetration, the hard rock and keeping a constant cutterhead design. This solution is authors individuated forces acting production and high abrasion resist- used when the load applied to the on the cutter, and thrust, torque and ance. cutter and/or the time of applica- power requirement of the TBM are The thrust acting on a single tool tion of the force is not sufficient for obtained from a balance of forces. consists of two forces: a component the detachment of a complete chip Analytical models usually com- of the force normal to the face, FN, (Rostami and Ozdemir, 1993; Bru- bine intact rock characteristics and a force tangential to the face, land, 1998). (mainly uniaxial compressive and the rolling force FR. tensile strength, ıc and ıt) with in- Two different processes can cause formation about the cutter (e.g. cut- rock breakage, when the TBM is 1.1. Prediction models ter diameter Ɏdisc, spacing scutters, tip pressed against the face, respective- width wtip, and thrust per cutter). ly single pass and multiple cutting Prediction models produce fo- They do not take in deep account process. Single pass cutting process recast in terms of penetration rate rock mass characteristics that recent consists in the detachment of rock (PR), that is the penetration of the studies (Bruland 1998; Ramezanza- chips, due to the action of two disc cutterhead per revolution, utiliza- deh, 2006; Gong and Zhao, 2007) cutters and is the process that more tion factor (UF), i.e. percentage of proved to be very important to un- frequently can occur. time in which the machine is actual- derstand TBM performance, as joint The effect of cutters on the rock ly working, and tools wear (Peila and frequency and joint orientation. The has been studied through full-size Pelizza, 2009, Peila, 2009). most frequently used analytical mo- tests, as the linear cutting test equip- These predictive methods can be dels had been modified recently to ment (Rostami and Ozdemir, 1993). classified in two types, considering overcome this limitation and they When a single disc is pressed against the hypotheses and the source of can be considered semi-empirical. the face, it creates a pulverised bulb data each model is based on: analyti- The most used ones are: of rock immediately below the cut- cal and empirical models. Analytical Colorado School of Mines (CSM) ter, the so-called crushed zone, and models are studies that start from model was initially developed by radial tension cracks propagation theoretical assumptions, e.g. theori- Rostami and Ozdemir (1993), starts (fig. 1). es about force balance and stress di- starting from the results of linear Another possibility for rock stribution, while empirical models are cutting tests and mathematical as- fragmentation, especially in case based on the back-analysis of data sumption on rock fragmentation of hard rock, is the multiple cutting from a database collected in tunnel process. It considers the formation process. The chip is generated by projects, and equations proposed de- of a pressure bulb of crushed rock the multiple passage of different rive from their statistical treatment. immediately under the tip, assumed to be circular; the pressure decrease on the size, while maximum stress concentration is present immediately under the cutter. Parameters considered in the model are ıc, ıt, Ɏdisc, wtip, scutter and rotational speed vrot. Ramezanzadeh (2006) integrated these studies taking into account joint spacing and the an- gle between the plane of weakness and the tunnel axis Į. The study developed by Frenzel (2010) exten- ded CSM model to include cutter Fig. 1. Chip formation and detachment under the effect of a cutter disc (Bruland, 1998). wear prediction, by correlating cut- Formazione e distacco del chip di roccia per effetto dell’azione del disco (Bruland, 1998). ter life, Cerchar Abrasivity Index

38 Agosto 2015 GEOINGEGNERIA E ATTIVITÀ ESTRATTIVA

(CAI), and disk diameter with tools correlations with CLI, quartz con- els predict penetration rate, four of consumption. tent (q), and lithology. them give a solution for cutter wear Gehring model (1985) was deve- Non-linear multiple regression prediction, and others four allow loped in collaboration with Voest analysis is often exploited in em- to foresee the utilization factor. As Alpine, a machines manufacturer. pirical models to find correlations shown in tab. 1 and tab. 2, geolog- Base equation proposed by the au- between parameters (Alber, 1996; ical parameters most considered in thor correlated the penetration rate Yagiz, 2008; Gong and Zhao, 2009; prediction models are ıc and Js, with to ıc and FN. Last version of the mo- Hassanpour et al., 2011; Farrokh et more than 70% of occurrences, and del included five correction factors al., 2012). Other authors (Alvarez Į (more than 60%). Also lithology, that considered respectively fracture Grima et al., 2000) tried to apply ıt, and RQD are considered in more energy, joint spacing, state of stress, other strategies, as the neuro-fuzzy than 35% of the cases. cutter diameter, and cutter spacing: model, based on data clustering and CAI and quartz content are con- these factors derived from empirical a back-propagation algorithm. sidered in those models that study correlations. The model included an Another group of models corre- also the cutter wear, whereas joint empirical equation that allows cutter late machine performance and cut- condition, GSI, porosity, stress, wa- wear prediction as a function of CAI. ter wear with rock mass classification ter content and rock density register systems), primarily because they are a low frequency. accepted and known worldwide in More than 40% of the models 1.1.2. Empirical models rock mechanics. studied are influenced by the thrust Correlations with Rock Mass In- per cutter FN. Nowadays, new empirical models dex, RMR, were proposed by Cass- In any case, low frequency of ma- are continuously developed, because inelli et al. (1982), Innaurato et al. chine parameters is registered, also of the great number of parameters (1990), Ribacchi and Lembo Fazio because all models referring to rock influencing TBM performance, and (2005), Hamidi et al. (2010), and mass classifications are simple and the high variability related to spe- Hassanpour et al. (2011); Rock Mass they do not take into account the cific on-field conditions. Data col- Quality Index for TBM, QTBM, was interaction between the machine lection programs could involve both introduced by Barton (2000); a cor- and the face: CSM (1993), NTNU rock face mapping and rock coring relation with Geological Strength (1998) and Gehring (1995) models program on-field, with significant Index (GSI) was proposed by Has- provide a most complete study of all variations among models analysed, sanpour et al. (2011). parameters involved in the process while TBM performance and ma- Globally all the 15 discussed mod- (tab. 3). chine data datasets in hard rock tunnelling are almost standardised. Tab. 1. Prediction models: intact rock parameters. One of the most important em- Modelli predittivi: parametri della roccia intatta. pirical model for performance, uti- Model Intact rock parameters lization factor and cutter wear pre- Rock ıc ıt CAI q ȡ diction is Norwegian University of CSM (1993, 2006, 2010) x xxxx Sciences and Technology (NTNU) model. Last version of NTNU model Gehring (1995) x x was proposed by Bruland (1998) in NTNU (1998) x x x x its doctoral thesis. Alber (1996) x x x The methodology proposed needs Yagiz (2008) x x x a series of laboratory tests that allows Gong, Zhao (2009) x x to define three NTNU/SINTEF in- Hassanpour et al. (2011) x x dices, namely Drilling Rate Index Farrokh et al. (2012) x x (DRI), Bit Wear Index (BWI), and Cutter Life Index (CLI). Alvarez G. et al. (2000) x Penetration rate prediction is Cassinelli et al. (1982) x based on intact rock and rock mass Innaurato et al. (1990) x characteristics (DRI, Į, joint fre- Barton (2000) x x quency Jf, and porosity n) as well as Ribacchi, Lembo Fazio (2005) x x machine parameters, e.g. FN, Ɏdisc, Hamidi et al. (2010) v , s , and number of cutters rot cutters Hassanpour et al. (2011) x ncutters). The model allows cutter wear and UF prediction, thanks to Frequency 5 13 5431

Agosto 2015 39 GEORESOURCES AND MINING

Tab. 2. Prediction models: rock mass parameters. genfurt. The high-speed railway is Modelli predittivi: parametri dell’ammasso roccioso. one of the projects involved in the Model Rock mass parameters EU program Trans-European Tran- Js RQD Jw Jc Į ı,w n GSI sport Network (TEN-T); in parti- CSM (1993, 2006, 2010) X x cular, Koralm railway is part of the Gehring (1995) X x Baltic-Adriatic Corridor VI Helsin- ki/Gdànsk – Padova/Bologna. NTNU (1998) X x x Koralm tunnel crosses the Alber (1996) x x mountains that divide Styria and Yagiz (2008) X x Carinthia, from Frauental an der Gong, Zhao (2009) X x Lassnitz to St. Andrä. Hassanpour et al. (2011) x The portion of alignment investigated is part of KAT 2 construc- Farrokh et al. (2012) X tion lot, in particular the area of the Alvarez G. et al. (2000) X Middle Austro-Alpine crystalline Cassinelli et al. (1982) X x x x x complex (ﬁg. 2). The crystalline Innaurato et al. (1990) X x x x x complex, with an overburden up to Barton (2000) X x xxxx approximatively 1,200 m, consists Ribacchi, Lembo Fazio (2005) x of a thick poly-metamorphic se- quence. Main deposit is para-gneiss, Hamidi et al. (2010) x x x with inclusions of marbles, amphi- Hassanpour et al. (2011) x x x x x x bolites, eclogites, quartzites, ortho- Frequency 11 6 4 2 10 5 1 3 gneisses, and pegmatites. Slightly fractured hard rock sections with Tab. 3. Prediction models: disc and cutterhead parameters. highly to extremely abrasive rocks Modelli predittivi: parametri dell’utensile a disco e della testa di scavo. dominate. Ground water inﬂows Model Disc Cutterhead with occasional high water pressu-

FN ĭdisc wtip vlim ĭTBM ncutters scutters vrot res are usually related to areas with CSM (1993, 2006, 2010) xxxxxx x intense fracturing, damage zones or et al. Gehring (1995) x x faults (Moritz , 2011; Wagner et al., 2009). NTNU (1998) x x x x x x Koralm tunnel is excavated by Alber (1996) two double shielded TBMs Aker- Yagiz (2008) Wirth TB 993 E/TS, one per tube. Gong, Zhao (2009) The cutter head is driven by electric Hassanpour et al. (2011) motors with an overall power rating Farrokh et al. (2012) x x of 4,800 kW, maximum cutting torque is 30,000 kNm. Two cylin- Alvarez G. et al. (2000) x x x ders provide gripping thrust, with Cassinelli et al. (1982) a maximum total clumping force of Innaurato et al. (1990) 115,000 kN, corresponding to a ma- Barton (2000) x x x ximum contact pressure of 4.7 MPa. Ribacchi, Lembo Fazio (2005) The cutterhead has a maximum Hamidi et al. (2010) boring diameter of 9.93 m (being 20 cm the maximum over-proﬁle Hassanpour et al. (2011) allowed), and total shield length is Frequency 641142 3 3 13,860 mm, with a ratio Ɏcutterhead/ Lshield = 1.396. The cutterhead is 2. Materials and methods 2.1. Koralm tunnel project composed of 80 17” discs (77 at the face and 3 external gauge cutters), The prediction models listed in The Koralm tunnel project is a and average spacing between two previous chapter were tested throu- 32.9-km-long railway double-tube cutters is 65 mm. Maximum rota- gh the example of Koralm tunnel tunnel, part of the Koralm railway tional speed is 6 rpm. project, in particular a 1000-m-long connecting the two main cities in All parameters are summarised in section of the South tube. the Southern Austria, Graz and Kla- tab. 4.

40 Agosto 2015 GEOINGEGNERIA E ATTIVITÀ ESTRATTIVA

Fig. 2. Geological section and construction lots at Koralm tunnel: gneiss- and mica-schist (Gn-Mi), marble (Ma), ﬁne-grain gneiss (Fi Gn), coarse-grain gneiss (G Gn), siltstone (S), sandstone (SS) and paragneiss (P Gn) (Harer and Koinig, 2010). Sezioni geologiche e lotti costruttivi nel Koralm tunnel: gneiss- e mica-scisti (Gn-Mi), marmi (Ma), gneiss a grana ﬁne (Fi Gn), gneiss a grana grossa (G Gn), limi (S), arenarie (SS) e paragneiss (P Gn) (Harer e Koinig, 2010).

Tab. 4. AkerWirth TB 993 E/TS character- tigated in the study is the area of prediction models. istics. the south tube in KAT2 lot between Machine producers provide a Riepilogo caratteristiche AkerWirth TB 993 chainage 6,900 and 7,900, calculat- maximum value of thrust a cutter E/TS. ed from Leibenfeld construction site can sustain during its life. Parameter Value (from chainage 51,153 to 52,153 of FN,max is often used as input pa- the whole line, fig. 2). The portion rameter, but not necessarily TBM Cutterhead diameter 9.93 m of alignment was excavated from is driven at its full capacity, for dif- Shield length 13.86 m January to March 2014. ferent causes, e.g., geological con- Total number of cutters 80 The segment was selected on the ditions, operational limits, machine base of the variability of rock mass limitations (Rostami, 2008; Barton, Number of active cutters 77 characteristics along the chainage 2000). Av. cutter spacing 65 mm and the high percentage of homo- A different approach is applied Discs diameter 17” (432 mm) geneous sections. Knowing that pre- in the analysis performed on Ko- Max. thrust per cutter 250-267 kN diction models consider the face as ralm tunnel database at construction homogeneous, only 37 over a total stage, based on back-analysis of F Tip width 25 mm N amount of 46 sections were consid- per cutter (Fcutter) distribution as a Max. rotational speed 6 rpm ered, on the 1000-m-long portion of function of one geological parame- Max. total thrust 115,000 kN alignment studied. ter. Input parameters come from ge- Correlations between FN and rock Installed cutterhead power 4800 kW ological-geomechanical surveys mass classifications (GSI, RMR) Max. cutting torque 30000 kNm performed daily at the face, in par- highlight the variability of the thrust ticular a weighted average of pa- per cutter along the alignment, as rameters, based on the percentage well as FN distribution as a function 2.2. Data analysis and of the deposit. ıt and CAI values of the uniaxial rock mass strength. methodology come from laboratory tests report- In fig. 3 a flow chart is proposed in ed in the geological, hydro-geo- order to obtain a continuously up- All computations and plots have logical and geotechnical report dated TBM performance prediction been processed with MathWorks (BCG consult et al., 2009), while and analysis: MATLAB® release 2008a. A series NTNU drillability indices have – definition of a database composed of scripts have been created ad-hoc been derived from literature stud- of geological survey parameters to process and filter data; the system ies (Bruland, 1998), in absence of and TBM machine data; of scripts has been designed in or- laboratory test data. An additional – back-analysis regarding variabi- der to be easily applied to any other hypothesis is an isotropic state of lity of FN with rock mass classi- project. stress. Mean values of each class fication (e.g. GSI) or rock com- The section of the tunnel inves- were used as input parameters for pressive strength, and exploring

Agosto 2015 41 GEORESOURCES AND MINING

3. Results 3.1. Performance

PR, ROP and SP prediction was investigated by applying prediction models to geological parameters in homogeneous sections. PR prediction performed on 37 sections are summarized in fig. 5 and tab. 6. Each line represents the prediction obtened by adopting a specific model, estimated by using FN values obtained from ıc class statistics (see tab. 5). Several models provide an accu- Fig. 3. Current method at design stage, and proposed new methodology for conti- rate prediction of PR on the whole nuously updated TBM performance and cutter wear prediction at construction stage. database; in particular NTNU mod- Metodo attuale a livello di progetto e nuova metodologia proposta per una previsione con- el (1998), CSM modified model tinuamente aggiornata delle prestazioni e dell’usura durante la fase di costruzione. (1993), Q-TBM model (2000), and the equations proposed by Farrokh the influence of lithology on the cation does not provide a heteroge- (2012), and Hassanpour et al. (2011) thrust per cutter; neous distribution in terms of sec- provide a little underestimation; a – using as geological inputs the data tions per class. little overestimation can be found collected during a survey in a by adopting Gehring model (1995), short-term view, or predicted ge- and Hamidi’s equation (2010). From ological parameters for next por- the chart, one can notice that the tions of chainage; – obtaining 25-th, 50-th and 75- th percentile of FN as a function of rock mass classification in the portion of alignment considered; – computing all prediction models through a software, or a spre- adsheet, designed for the task; – comparing TBM performance with prediction, and evaluation by experienced engineers and ge- ologists, taking into accounts also outcomes from exploration methods; – analysing the distribution of the errors, and deciding best models for given geological conditions; – updated estimation of costs and Fig. 4. Boxplot, Fcutter as a function of ıc. time. Boxplot, Fcutter in funzione di ıc. In Koralm tunnel case, the distribution exploited in the method is Tab. 5. Fcutter as a function of ıc, table of statistics. Fcutter as a function of ıc, reported in Fcutter in funzione di ıc, tabella riassuntiva. fig. 4, because it includes an adequate Par. Group Nr. Sec. F [kN/cutter] number of sections in each category cutter (respectively 19 sections in the class μ ı 25th Med. 75th

50-100 MPa, 13 in the class 100-150 ıc 50 – 100 MPa 19 177.52 44.00 154.98 187.51 213.07 MPa, and 5 sections in the class >150 100 – 150 MPa 15 212.93 34.31 209.45 220.76 227.41 MPa, as reported in tab. 5). On the other hand, GSI classiﬁ- > 150 MPa 5 243.58 16.00 228.74 249.36 256.58

42 Agosto 2015 GEOINGEGNERIA E ATTIVITÀ ESTRATTIVA

the premises to perform a short-term analysis. In order to understand the prediction accuracy, three flags variables (”Accepted”, ”Underestimation”, and ”Overestimation”) define the behaviour of the model referred to a given section. If the range of performance predicted falls within the range of data recorded, model can be considered satisfactory, otherwise it is recognized as an underestimation or an overestimation. Model error MEik, i.e. the error of k-th model at i-th section, is calculated as the normalized difference be- Fig. 5. PR prediction outcomes on the whole dataset (37 sections); diamonds cor- tween mean FN predicted and mean th th th F recorded, multiplied for the ratio respond to PR obtained by 25 , 50 and 75 percentile of FN, star symbol to mean N value. Mean value from TBM data in the section is reported with a solid line, median between 75th and 25th percentile: with a dash-dot line, and 25th/75th percentile with two dashed lines. F + FF− Risultati della previsione di PR sull’intero campione (37 sezioni); i quadri corrispondono ai ME =⋅ik, pred ik,, pred ik rec ik F − F valori di PR corrispondenti al 25° percentile, la mediana e il 75° percentile di FN, la stella al ik, pred ik, rec valore ottenuto con la media. Il valor medio dei dati TBM nella sezione di studio è riportato con una linea continua, la mediana con una linea tratto punto e i percentili 25° e 75° con where: + th due linee tratteggiate. – Fik,pred is the 75 percentile of the predicted thrust per cutter Tab. 6. PR prediction, summary table. at section i-th referred to model Previsione di PR, tabella riassuntiva. k-th; – th Model Acc. Under Over TME – Fik,pred is the 25 percentile of the CSM mod., 1993, 2006 13 23 1 0.52 predicted thrust per cutter at sec- Gehring, 1995 19 7 11 0.29 tion i-th referred to model k-th; – F is the mean thrust per cut- NTNU, 1998 6 22 9 0.57 ik, pred ter predicted at section i-th refer- Alber, 1996 13 2 22 0.98 red to model k-th; F Yagiz, 2008 4 23 10 0.38 – ik, rec is the mean thrust per cutter Gong, Zhao, 2007 3 34 0 1.83 recorded at section i-th. Hassanpour1, 2011 13 18 6 0.36 Tab. 6 shows the outcomes of the computation obtained for the 37 Farrokh, 2012 8 16 13 0.32 sections analysed. Alvarez Grima, 2000 10 22 5 >>1 Mean model error referred to Cassinelli, 1982 2 31 4 0.45 model k-th TMEk is the average val- Innaurato, 1990 6 22 9 0.28 ue of |ME| in the N sections considered (in this case N=37), calcu- Barton, 2000 12 23 2 0.84 lated as follows: Ribacchi, 2005 2 17 18 0.88 ∑N ME Hamidi, 2010 18 1 18 0.62 i=1 ik TMEk = Hassanpour2, 2011 6 27 4 0.85 N Hassanpour3, 2011 9 6 22 0.84 Most accurate models in terms of Hassanpour4, 2011 4 27 6 0.78 accepted cases and TME are CSM modified model, proposed by Ros- equations proposed by Innaurato et However, a deeper analysis of the tami and Ozdemir (1993), and cor- al. (1990), Cassinelli et al. (1982), prediction for each section is need- rected by adding the influence of and Yagiz (2008) are independent of ed, in order to understand which fractures (Ramezanzadeh, 2006), thrust per cutter. sections are critical, and if there are Gehring model (1995), Q-TBM

Agosto 2015 43 GEORESOURCES AND MINING model by Barton (2000), the equa- Its dataset consists on rock samples The first equation by Hassanpour tion proposed by Farrokh et al. with high ıc (100-300 MPa) from et al. (2011, fig. 6e) and the equation (2012), first equation proposed by South-African and South-Korean proposed by Farrokh et al. (2012, fig. Hassanpour et al. (2011), and the projects in different rocks (granite, 6f) have good results in all compres- equation proposed by Hamidi et al. sandstone, gneiss, amphibolite), so sive strength classes. Many empirical (2010). the variety of lithology considered models based on rock mass classi- The number of section whose in the database allows better results. fication show a better response at performance is over- and under-esti- Error plot shows good results at low high values of ıc, as Q-TBM model mated is balanced in Gehring mod- compressive strength, while at high (2000), and the equations by Innau- el (1995), as in equations proposed ıc there is an underestimation, prob- rato et al. (1990), Cassinelli et al. by Hassanpour et al. (2011) and by ably because the improvement of (1982), and Ribacchi and Lembo Farrokh et al. (2012). Models pro- discs performance at high compres- Fazio (2005). posed by CSM (1993) and by Barton sive strength. Best models in terms of error and (2000) are subjected to errors of un- The importance of lithology accepted cases were chosen for each derestimation in many cases, while crossed in comparison with model range of compressive strength and equation by Hamidi et al. (2010) database is clear in case of the equa- reported in tabs. 7, 8, and 9: usually overestimates the perfor- tion proposed by Hamidi et al. (2010, – Gehring model (1995), Al- mance. fig. 6d): the dataset is composed of ber equation, Hassanpour et al. NTNU model, 1998 version, pro- records from a tunnel excavated by (2011) equation, and Hamidi et posed by Bruland, and the equation a double shielded TBM in metamor- al. (2010) equation, if 50 < ıc < proposed by Innaurato et al. (1990) phic rocks, exactly as in Koralm tun- 100 MPa; present low error, but they have a nel KAT 2. The same observations – CSM model modified (1993), low number of accepted cases, and can be done for the equations pro- Gehring model (1995), and equa- they usually underestimates Ak- posed by Alber (1996), and by Has- tions proposed by Hassanpour et er-Wirth machine performance. sanpour et al. (2011) in their studies, al. (2011) and Innaurato et al. Other equations have a higher in great part performed in metamor- (1990) if 100 < ıc < 150 MPa; threshold of error and are often un- phic and sedimentary rocks. – NTNU model (1998), and equa- balanced. The class-by-class analysis (tab. 7, Tab. 7. PR prediction by the filter ”ıc-based”, 50-100 MPa group. 8, 9, and 10) reveals that some mod- Previsione di PR con il filtro ”ıc-based”, gruppo 50-100 MPa. els are more appropriated to certain Model Acc. Under Over TME geological conditions, and they do not produce good results if extrapo- CSM mod., 1993, 2006 8 11 0 0.63 lated. Gehring, 1995 14 0 5 0.22 As shown in fig. 6a and in fig. 6b, NTNU, 1998 3 16 0 0.66 NTNU (1998) and CSM (1993) Alber, 1996 13 2 4 0.31 models have a low degree of accu- Yagiz, 2008 0 19 0 0.38 racy for low ıc, while they improve their accuracy for ıc > 100 MPa with Gong, Zhao, 2007 0 19 0 2.96 absolute values of TME always sensi- Hassanpour1, 2011 8 11 0 0.34 bly lower than 0.5. This effect is due to datasets prediction models com- Farrokh, 2012 3 13 3 0.23 putation is based on. NTNU model Alvarez Grima, 2000 6 13 0 0.37 (1998) comes from the analysis of Cassinelli, 1982 0 19 0 0.57 tunnels excavated in Scandinavian igneous rocks, with high uniaxial Innaurato, 1990 2 14 3 0.21 compressive strength, and this effect Barton, 2000 9 10 0 0.85 can be seen in the categories ”100- Ribacchi, 2005 0 6 13 1.33 150 MPa” and ”> 150 MPa”, where Hamidi, 2010 16 0 3 0.20 TME goes closer to zero. CSM model (1993) prediction is less influenced Hassanpour2, 2011 6 11 2 0.68 by ıc, if compared to NTNU (1998). Hassanpour3, 2011 7 5 7 0.54 The model proposed by Gehring Hassanpour4, 2011 4 11 4 0.64 (1995, fig. 6c) is the one that provide predictions closer to data recorded. Av. 50 – 100 MPa 15 0 4 0.22

44 Agosto 2015 GEOINGEGNERIA E ATTIVITÀ ESTRATTIVA

Fig. 6. Error boxplots for PR prediction on the whole dataset (37 sections). Boxplots riguardo gli errori nella previsione di PR sull’intero campione (37 sezioni).

Tab. 8. PR prediction by the ﬁlter ”ıc-based”, 100-150 MPa group. Previsione di PR con il ﬁltro ”ıc-based”, gruppo 100-150 MPa. Model Acc. Under Over TME Model Acc. Under Over TME CSM mod., 1993, 2006 4 8 1 0.41 Cassinelli, 1982 1 9 3 0.38 Gehring, 1995 4 4 5 0.42 Innaurato, 1990 3 5 5 0.43 NTNU, 1998 1 5 7 0.45 Barton, 2000 1 10 2 1.03 Alber, 1996 0 0 13 1.63 Ribacchi, 2005 1 7 5 0.45 Yagiz, 2008 2 4 7 0.42 Hamidi, 2010 2 1 10 1.14 Gong, Zhao, 2007 1 12 0 0.73 Hassanpour2, 2011 0 11 2 1.17 Hassanpour1, 2011 4 4 5 0.43 Hassanpour3, 2011 2 1 10 1.29 Farrokh, 2012 2 3 8 0.51 Hassanpour4, 2011 0 11 2 1.11 Alvarez Grima, 2000 4 4 5 >>1 Av. 100 – 150 MPa 4 5 4 0.39

Agosto 2015 45 GEORESOURCES AND MINING

Tab. 9. PR prediction by the filter ”ıc-based”, > 150 MPa group. tion (2010). In fact, these models Previsione di PR con il filtro ”ıc-based”, gruppo > 150 MPa. take into account excavation in sed- Model Acc. Under Over TME imentary rocks with high drillability, CSM mod., 1993, 2006 1 4 0 0.40 as shale and limestone. Gehring, 1995 1 3 1 0.24 NTNU, 1998 2 1 2 0.57 3.2. Utilization factor Alber, 1996 0 0 5 1.78 Utilization factor (UF) can be Yagiz, 2008 2 0 3 0.28 predicted by using four different ap- Gong, Zhao, 2007 2 3 0 0.43 proaches: NTNU model by Bruland Hassanpour1, 2011 1 3 1 0.22 (1998), the correlation proposed Farrokh, 2012 3 0 2 0.20 by Alber (1996), the equation proposed by Innaurato et al. (1990), and Alvarez Grima, 2000 0 5 0 >> 1 Q-TBM model by Barton (2000), Cassinelli, 1982 1 3 1 0.22 reported in tabs. 11, 12, 13, and 14. Innaurato, 1990 1 3 1 0.18 The analysis is similar to the one performed for PR, even if, in Barton, 2000 2 3 0 0.31 this case, UF can be calculated just Ribacchi, 2005 1 4 0 0.28 on a portion of alignment and not Hamidi, 2010 0 0 5 0.88 continuously. In this case, the test Hassanpour2, 2011 0 5 0 0.65 is considered passed if data stay in a range of 25% from the mean value. Hassanpour3, 2011 0 0 5 0.83 What emerges from the analysis Hassanpour4, 2011 0 5 0 0.46 on the whole database is that NTNU Av. > 150 MPa 2 1 2 0.21 (1998) and Q-TBM (2000) models allow a successful prediction of the utilization factor of the machine in tions proposed by Gong and Zhao reliability, 2/5 sections correctly a long-term perspective. The equa- (2007), Farrokh et al. (2012) and predicted, but they still have a low tion found by Innaurato et al. (1990) Q-TBM model (2000), if ıc > 150 error. Worst results derive from sec- overestimates the utilization factor, MPa. tions with ıc ranging from 100 to but it must be said it is a simple mod- Looking at the whole database, 150 MPa, just 4/13 sections correctly el, based on a polynomial correlation the prediction is acceptable in 21/37 assessed and a higher error. with RMR classification. sections, corresponding to 56.8% of Marble is the main deposit in two Relationship proposed by Alber the sections analysed (tab. 10). sections (309 and 310), and in this (1996) induces an underestimation Results for low level of compres- case the underestimation can be due in UF prediction, even if deeper sive strength are satisfactory (15/19 to the fact the great majority of this analysis should be performed; in sections correctly predicted), thanks models does not take into account fact, it is based on the in-situ state of to good outcomes coming from the rock drillability, very high in mar- stress, here assumed isotropic. best models in terms of accepted ble (Bruland, 1998). A better as- Following the same methodology values and TME, as Gehring model sessment is provided by models that proposed for machine performance (1995) and the equation proposed take into account DRI, i.e. NTNU analysis, NTNU model (1998) lead by Hamidi et al. (2010). The anal- (1998) and Barton (2000) models, to an absolute value of TME lower ysis for high ıc level presents lower as well as using Hamidi et al.’s equa- than 0.5 in all ıc ranges. A constant overestimation can Tab. 10. Summary, PR prediction by the filter ”ıc-based. be seen in UF values calculated with Riassunto, previsione di PR con il filtro ”ıc-based”. the equation proposed by Innaurato Model Acc. Under Over TME et al. (1990), especially when considering high values of ıc (tab. 13). Av 50 – 100 MPa 15 0 4 0.22 This trend is justified from the Av. 100 – 150 MPa 4 5 4 0.39 characteristics of the dataset, i.e. Av. > 150 MPa 2 1 2 0.21 tunnels excavated by open shield TBMs. The choice of an open shield Filter “ıc-based” 21 6 10 0.28 TBM is linked to a high rock sta-

46 Agosto 2015 GEOINGEGNERIA E ATTIVITÀ ESTRATTIVA

Tab. 11. UF prediction by the filter ”ıc-based”, 50-100 MPa group. – Alber (1996) equation and Q- Previsione di UF con il filtro ”ı -based”, gruppo 50-100 MPa. c TBM (2000) model, if 100 < ıc < Model Acc. Under Over TME 150 MPa; NTNU, 1998 13 3 3 0.71 – NTNU (1998) and Q-TBM (2000) models, and equation pro- Alber, 1996 7 11 1 0.75 posed by Alber, if ıc > 150 MPa. Innaurato, 1990 9 1 9 0.84 Prediction is within the range of Barton, 2000 9 7 3 0.94 UF in 29 cases, over a total number Av. 50 – 100 MPa 14 2 3 0.76 of 37 sections, corresponding to the 78.4 % of sections analysed. Results are balanced for each class of ı , Tab. 12. UF prediction by the filter ”ıc-based”, 100-150 MPa group. c Previsione di UF con il filtro ”ıc-based”, gruppo 100-150 MPa. even if prediction for low level of compressive strength present a quite Model Acc. Under Over TME high value of TME, if compared with NTNU, 1998 9 0 4 0.25 other groups. It has to be noticed UF Alber, 1996 12 0 1 0.16 is often overestimated in the portion Innaurato, 1990 8 0 5 0.34 of alignment under analysis. Barton, 2000 9 1 3 0.20 Av. 100 – 150 MPa 12 1 0 0.13 3.3. Cutter wear

Tab. 13. UF prediction by the filter ”ıc-based”, > 150 MPa group. The average tunnel length cove- Previsione di UF con il filtro ”ıc-based”, gruppo > 150 MPa. red by the TBM till the consumption Model Acc. Under Over TME of a cutter occurs (Ltot) is the para- NTNU, 1998 2 1 2 0.33 meter predicted by 4 different models: CSM model (1993), in parti- Alber, 1996 2 1 2 0.22 cular the work proposed by Frenzel Innaurato, 1990 1 0 4 0.53 (2010, 2011, et al., 2008); Gehring Barton, 2000 4 0 1 0.29 model (1995); NTNU model by Av. > 150 MPa 3 0 2 0.28 Bruland (1998), the correlation proposed by Alber (1996). As said before, lithology abrasive- Tab. 14. Summary, UF prediction by the filter ”ıc-based. Riassunto, previsione di UF con il filtro ”ıc-based”. ness and ıc are the parameters that have a higher influence on cutter Model Acc. Under Over TME consumption, as well as CAI. CSM mod.(Frenzel), 1993, 2006, 2010 5 14 0 0.69 The results of laboratory tests on Gehring, 1995 0 19 0 0.79 rock samples (tab. 15) highlight how NTNU, 1998 7 12 0 0.60 marble has a low abrasiveness in comparison with schist and gneiss, Alber, 1996 0 19 0 0.96 Wear analysis is highly focused on Av. 50 – 100 MPa 7 12 0 0.65 mica-schist and gneiss-schist lithology, representing the most common bility, so a higher UF is a natural (2000) models, and Innaurato et geology formation in Koralm Tun- consequence in case of a portion al. (1990) equation, if 50 < ıc < nel. of alignment excavated in double 100 MPa; As discussed in Section 4.2, Ltot shield mode. Equation proposed by Innaurato et al. (1990) and Q-TBM Tab. 15. CAI and quartz content from laboratory tests on Koralm Tunnel rock samples. (2000) model present a higher var- CAI e contenuto in quarzo nei test di laboratorio sui campioni di roccia del Koralm tunnel. iability, if compared with all other models. Rock CAI q Best models in terms of error and μ ı min max adequate predictions, grouped by Mica- and gneiss-schist 3.51 0.83 3 45 their range of compressive strength, Fine grain gneiss 4.03 0.71 3 90 are: – NTNU (1998) and Q-TBM Marble 2.19 0.92 0 21

Agosto 2015 47 GEORESOURCES AND MINING variability is taken into account by wear. NTNU prediction entails an for low level of compressive strength assuming a range of acceptability underestimation at lower ıc and an present frequently underestimation, equal to ± 25% of the mean value. overestimation at medium-high ıc. and a quite high value of TME, if What emerges from the analysis Furthermore, the wear prediction compared with other groups; bet- on the whole database (tab. 14, 17, distribution as a function of uniaxial ter results can be seen in rock with and 18) is that CSM (1993) and compressive strength for this model medium-high/very high values of ıc NTNU (1998) models allow a bet- shows a higher variability than other (100-150 MPa). ter prediction of disc wear in a long- models. Sections that present marble li- term perspective, while the appro- Best models in terms of error and thology are characterized by pre- aches proposed by Gehring (1995) correct evaluation, grouped by their dicted values constantly minor in and by Alber (1996) are not able to range of compressive strength, are: consumption. As widely explained give a reliable answer to cutter con- – CSM (1993) and NTNU (1998) in Section 2.2.2, NTNU (1998) sumption prediction; these models models, if 50 < ıc < 100 MPa; prediction of cutter wear is the most underestimate Ltot recorded along – NTNU model (1998), if 100 < ıc complete, because it is sensible to the alignment. < 150 MPa; rock abrasiveness, drillability, quartz Detailed prediction is capital in – CSM (1993) and NTNU (1998) content, in addition to machine and cutter wear analysis: the number of models, if ıc > 150 MPa. performance parameter (Bruland, accepted cases is remarkable just in Prediction is acceptable in 18 1998). In these cases, best assess- CSM (1993) and NTNU (1998) cases, over a total number of 37 sec- ment is provided by NTNU model models. tions, corresponding to 48.7 % of the (1998), while in other cases, where Prediction from all models, sections analysed, a result not com- abrasiveness is higher, also other except from NTNU model (1998), pletely satisfactory if compared with models provide good outcomes. produce an underestimation of disc PR and UF estimation. Prediction Total mean error of ”ıc-based” ﬁl- ter is 0.50, and error threshold can

Tab. 16. Ltot prediction by the ﬁlter ”ıc-based”, 100-150 MPa group. be even reduced by applying NTNU Previsione di Ltot con il ﬁltro ”ıc-based”, gruppo 100-150 MPa. model (1998) to rock sections com- Model Acc. Under Over TME posed of lithology that presents high CLI value (corresponding to low CSM mod.(Frenzel), 1993, 2006, 2010 2 11 0 0.52 abrasiveness), as marble or lime- Gehring, 1995 0 13 0 0.76 stone. NTNU, 1998 7 3 3 0.39 Alber, 1996 0 13 0 0.79 Av. 100 – 150 MPa 7 3 3 0.39 4. Conclusions

Tab. 17. Ltot prediction by the filter ”ıc-based”, > 150 MPa group. The analysis about application of Previsione di Ltot con il filtro ”ıc-based”, gruppo > 150 MPa. prediction models in the 1000-m- Model Acc. Under Over TME long portion of alignment of the Koralm tunnel project provides CSM mod.(Frenzel), 1993, 2006, 2010 0 5 0 0.52 satisfactory results, 21/37 sections Gehring, 1995 0 5 0 0.79 (56.8 %) were correctly predicted NTNU, 1998 3 0 2 0.34 with the methodology proposed, i.e. by grouping sections starting from a Alber, 1996 0 5 0 0.74 geological parameter. Results can be Av. > 150 MPa 4 1 0 0.24 summarized as follows: – at medium-low compressive Tab. 18. Summary, Ltot prediction by the filter ”ıc-based. strength (50 < ıc < 100 MPa), Riassunto, previsione di Ltot con il filtro ”ıc-based”. best models are the ones whe- Model Acc. Under Over TME re the database includes similar rocks, with outstanding results in Av 50 – 100 MPa 7 12 0 0.65 terms of accepted values (79 % of Av. 100 – 150 MPa 7 3 3 0.39 sections correctly predicted); Av. > 150 MPa 4 1 0 0.24 – medium-high compressive strength (100 < < 150 MPa) Filter “ı -based” 18 16 3 0.50 ıc c is the range with worst outcomes,

48 Agosto 2015 GEOINGEGNERIA E ATTIVITÀ ESTRATTIVA

just 31 % of accepted results, and entire length of the tunnel, provi- BGG Consult ZT GmbH, ÖBB-Inf- best models are a mix of empirical ding most reliable numerical results rastrucktur AG, 3G Gruppe Geo- and analytical models; that can integrate the outcomes here technik Graz ZT GmbH. Ausschrei- – at high compressive strength (ıc presented, integrating the system of bungsprojekt Koralmtunnel Baulos > 150 MPa), empirical models are TBM data analysis and prediction KAT 2; Geologie, Hydrogeologie und the ones that provide best results, at the construction stage proposed Geotechnik. Unpublished, 2009. but just 40 % of performance are in this work. Furthermore, predic- Bottero M. e Peila D. (2005) The use correctly predicted; tion should be enhanced strongly by of the Analytic Hierarchy Process for – if DRI is high (marble sections), applying a distribution of thrust as a the comparison between microtun- predictions can be performed by function of rock mass classifications, nelling and trench excavation. Tun- averaging models that take into in particular GA and GSI. These nelling and Underground Space account drillability, as NTNU results will allow a more accurate Technology, 20 (6), pp. 501-513. (Bruland, 1998) and Q-TBM prediction of TBM performance, Brino, G., 2015. Prediction of perfor- (Barton, 2000) models, and even- utilization factor and cutter wear: mance and cutter wear in rock TBM: tually models whose dataset is outcomes can have a remarkable im- application to Koralm tunnel project. while composed of these rocks pact both on technical side and on MSc thesis, Politecnico di Torino. (Hamidi et al.’s equation, 2010). economic side. Bruland, A., 1998. Hard Rock Tunnel Same philosophy was applied in The results can be used, in this Boring. PhD thesis, NTNU Trond- order to predict the utilization fac- case, as a prediction method to eva- heim Norwegian University of Sci- tor. UF predicted is within the ran- luate performance and cutter con- ences and Technology, Vol. 1-10. ge of values recorded in 29 cases, sumption for the excavation of the Cassinelli, F., Cina, S., Innaurato, N., over a total number of 37 sections, residual part of Koralm tunnel, but Mancini, R. e Sampaolo, A., 1982. corresponding to 78.4 % of sections the methodology can be applied in Power consumption and metal wear analysed. NTNU model by Bruland any other project. The system is par- in tunnel-boring machines: analysis (1998), the correlation proposed by ticularly useful in long tunnels, in of tunnel-boring operation in hard Alber (1996), and Q-TBM model by which a continuous improvement of rock, in: Tunnelling ’82, London, pp. Barton (2000) have good results in the ability of prediction, and of the 73-81. UF estimation. goodness of correlations found too, Cheda, J., Schuerch, R., Perazzelli, P. Cutter wear prediction was inve- can have an effective impact on time e Mezger, F., 2013. Performance of stigated too, even if results are less and costs. penetration models for hard rock satisfactory. TBMs in the case of the Gotthard Models allow a correct prediction Base Tunnel, in: World Tunnel Con- in 49.7 % of the cases, with better re- gress 2013 Geneva. Underground sults in the range medium-high/very Bibliography – the way to the future. Geneva, high values of ıc. CSM (1993) mo- Switzerland. Anagnostou, G. e del, in particular Frenzel’s equation Alber, M., 1996. Prediction of pene- Ehrbar, H. (Eds), International (2010, 2011), and NTNU model tration and utilization for hard rock Tunnelling Association ITA-AITES, by Bruland (1998) guarantee better TBMs, in Proceedings ISRM Interna- pp. 1-8. results; in particular, the Norwegian tional Symposium Eurock ’96, Turin, Farrokh, E., Rostami, J. e Laughton, C., model is able to predict with success Italy, Balkema, pp. 721-725. 2012. Study of various models for also the wear in case of low abrasive- Alvarez Grima, M., Bruines, P.A., and estimation of penetration rate of ness (marble deposits). Verhoeh, P.N.W., 2000. Modeling hard rock TBMs. Tunnelling and Un- The study proves how the choice Tunnel Boring Machine Performance derground Space Technology, 30, of the correct prediction model can by Neuro-Fuzzy Methods. Tun- pp. 110-123. have a severe impact on TBM per- nelling and Underground Space Frenzel, C., Käsling, H., and Thuro, K., formance prediction, in particular it Technology, 15(3), pp. 259-269. 2008, Factors Influencing Disc Cut- highlights the influence of the geo- Barla, G. e Pelizza, S., 2000. TBM tun- ter Wear. Geomechanik und Tun- logy. nelling in difficult ground conditions, nelbau, 1(1), pp. 55-60. Since this analysis regards only in: GeoEngineering 2000, Interna- Frenzel, C., 2010. Verschleisskos- 1000 m of the tunnel, a wider data- tional Conference on Geotechnical tenprognose für Schneidrollen bei base is needed in order to compare and Geological Engineering. Mel- maschinellen Tunnelvortrieben in TBM performance and investiga- bourne, Australia. Festgesteinen. Technical report, tions in a suitable way. In the futu- Barton, N., 2000. TBM Tunnelling in TU Munich. Ingenieurgeologie re, correlations could be improved Jointed and Faulted Rock. Balkema, Hydrogeologie Geothermie. Dr. by additional analyses regarding the Brookfield, Rotterdam. Friedrich Pfeil (Münchner Geow-

Agosto 2015 49 GEORESOURCES AND MINING

issenschaftliche Abhandlungen, mance prediction model for project Boring Machines, in: 6th Iranian Tun- Reihe B). planning. Tunnelling and Under- neling Conference. Frenzel, C., 2011, Leistungs- und Ver- ground Space Technology, 26, pp. Ramezanzadeh, A., 2006. Performance schleissprognose in der Praxis am 595-603. Analysis and Development of New Beispiel des CSM Modells, in: Pen- Innaurato, N., Mancini, R., Rondena, Model for Performance Predic- etrations- und Verschleissprognose R. e Zaninelli, A., 1991. Forecasting tion of Hard Rock TBMs in Rock beim TBM-Vortrieb im Fels, ETH and effective TBM performances in Mass. PhD thesis, Institut national Kolloquium. a rapid excavation of a tunnel in Ita- des sciences appliqués de Lyon, Gehring, K., 1995. Leistungs- und Ver- ly, in: Proceedings of the Seventh In- France. schleissprognosen im maschinellen ternational Congress ISRM, Aachen, Ribacchi, R. e Lembo Fazio, A., 2005. Tunnelbau. Felsbau, 13, pp. 439- pp. 1009-1014. Inﬂuence of Rock Mass Parameters 448. Moritz, B., Wagner, H., Mussger, K., on the Performance of a TBM in a Gong, Q.M. e Zhao, J., 2007. Inﬂuence Handke, D., and Harer, G., 2011. Gneissic Formation (Varzo Tunnel). of rock brittleness on TBM penetra- Criteria for the selection of tunnel- Rock Mechanics and Rock Engi- tion rate in Singapore Granite. Tun- ling method through the example neering, 38(2), pp. 105-127. nelling and Underground Space of the Koralm Tunnel. Geomechan- Rostami, J. e Ozdemir, L., 1993. A new Technology, 22, pp. 17-24. ics and Tunnelling, 4(4), pp. 305- model for performance prediction of Gong, Q.M. e Zhao, J., 2009. Deve- 315. hard rock TBM, in: Rapid Excavation lopment of a rock mass characteri- Oggeri, C. e Oreste, P., 2012. The and Tunnelling Conference 1993. stics model for TBM penetration rate Wear of Tunnel Boring Machine Boston, U.S.A., pp. 793 – 809. prediction. International Journal of Excavation Tools in Rock. American Rostami, J., 2008. Hard rock TBM Rock Mechanics and Mining Scien- Journal of Applied Sciences, 9(10), cutterhead modelling for design ces, 46, pp. 8-18. pp. 1606-1617. and performance prediction, Geo- Hamidi, J.K., Shahriar, K., Rezai, B. e Peila D. e Pelizza S. (2009). Ground mechanik und Tunnelbau, 1(1), pp. Rostami, J., 2010. Performance pre- probing and treatments in rock TBM 18-28. diction of hard rock TBM using Rock tunnel to overcome limiting condi- Yagiz, S., 2008. Utilizing rock mass Mass Rating (RMR) system. Tun- tions. Journal of Mining Science, properties for predicting TBM per- nelling and Underground Space 45(6), pp. 602-619. formance in hard rock condition. Technology, 25, pp. 333-345. Peila D. (2009). Indagini Preliminari Tunnelling and Underground Harer, G. e Koinig, J., 2010. Current nella Costruzione di Gallerie: Anali- Space Technology, 23, pp. 326-339. state of design, investigation and si della Letteratura Tecnica. Geoin- Wagner, H., Handke, D., Matter, J., construction works at the Koralm gegneria Ambientale e Mineraria, Fabbri, D. e Keiper, K., 2009. Con- Tunnel. Geomechanics and Tunnel- 128(3), pp. 23-44. cepts to overcome squeezing geo- ling, 3(2), pp. 155-162. Ramezanzadeh, A., Rostami, J. e Kast- logical conditions at the Koralm tun- Hassanpour, J. , Rostami, J. e Zhao, J, ner, R., 2003. Performance Predic- nel. Geomechanics and Tunnelling, 2011. A new hard rock TBM perfor- tion Models for Hard Rock Tunnel 2(5), pp. 601-611.

Acknowledgements The authors wish to thank ÖBB-Infrastruktur for the permission to use the data from the Koralm tunnel construction, lot KAT 2, and 3G Gruppe Geotechnik Graz ZT GmbH the active collaboration in this research.

50 Agosto 2015