Tunnelling and Underground Space Technology 30 (2012) 110123

Contents lists available at SciVerse ScienceDirect

Tunnelling and Underground Space Technology

Study of various models for estimation of penetration rate of hard rock TBMs
Ebrahim Farrokh a,, Jamal Rostami a, Chris Laughton b
a b

EME Department, The Pennsylvania State University, University Park, USA Laughton Associates, Texas, Austin, USA

a r t i c l e

i n f o

a b s t r a c t
Various approaches for predicting penetration rate of hard rock tunnel boring machines (TBMs) have been studied by researchers since the early stages of TBM application in the 1950s. These studies resulted in the development of several penetration prediction models. For evaluation and validation of a model, it is important to test its predictive capability on new projects. A model should include parameters for machine specications and ground conditions. The model validation process can reveal problems that an existing model may have in providing an accurate estimate for a given combination of specications and conditions. This paper offers a brief review and discusses the capabilities of some of the more commonly used TBM performance prediction models. To evaluate the accuracy of these models, the predicted rates are compared with recorded TBM penetration rates in a database of recently completed tunnels. Comparison between predicted and recorded rates indicates that most of the existing models tend to overestimate TBM performance. This comparison highlights the on-going difculties the industry continues to experience in estimating penetration rate. Even the use of normalized penetration rate indices has not been able to provide higher accuracy expected in related predictions. This paper discusses the development of new models to support an improved level of predictive accuracy in penetration rate estimating. These models are based on the analysis of a comprehensive database of more than 300 TBM projects records. Analyses of performance information within this database provided for the development of simpler models that are focused on quantifying the inuence of primary factors, such as tunnel diameter, UCS, RPM, and rock type. These new models are introduced to provide alternative ways of penetration prediction. These models are especially useful at the planning stage of a tunneling project where TBMs can be used. These models also serve to provide secondary checks for other more in-depth analyses of TBM performance for an initial assessment of required boring time (inverse of penetration rate), and an estimate of utilization rate in an activity-based TBM model. 2012 Elsevier Ltd. All rights reserved.

Article history: Received 1 December 2010 Received in revised form 28 October 2011 Accepted 7 February 2012 Available online 21 March 2012 Keywords: Penetration rate Penetration rate per revolution Specic penetration Field penetration rate Specic excavation rate

1. Introduction TBM performance has a dominant impact on tunnel completion time and cost. A key component in the successful planning of TBM tunneling is the accurate prediction of TBM performance parameters, notably the penetration rate (PR, the rate of TBM penetration during boring times) and the advance rate (AR, the rate of TBM progress during a work time period). Over the past few decades, several studies have been carried out to develop more accurate and comprehensive TBM performance prediction models. Early applications of TBMs were mainly undertaken in relatively massive rocks. In such rock masses research modelers focused primarily on evaluating the inuence of intact rock properties on PR for a given set of TBM parameters. As the use of TBMs and related manufacturing technology has evolved, the range of application of the
Corresponding author. Tel.: +1 814 863 7606; fax: +1 814 865 3248.
E-mail address: [email protected] (E. Farrokh). 0886-7798/$ - see front matter 2012 Elsevier Ltd. All rights reserved. doi:10.1016/j.tust.2012.02.012

TBM has expanded. TBMs are now frequently used in a wider range of rock mass conditions. Since joints and discontinuities within a rock mass may impact TBM performance, a need for an improved penetration rate predictive model for TBMs operating in fractured rock units became evident. Many of the earlier models could not address the impact of discontinuities on TBM PR. Consequently, attempts were made to either modify existing models or develop new models that included rock mass parameters. These modied and new models can be categorized into four classes as shown in Table 1. In developing a practical model, researchers focused on including the rock mass parameters that were known to have the strongest inuence on TBM performance. One of the main challenges in developing predictive methods for TBM performance is accounting for the interaction between TBM and rock mass. To better model the complexity of this interaction, some researchers developed new tests and indices that were specically devised for TBM tunneling prognosis. Special testing to derive parameters for boreability, drillability, and

E. Farrokh et al. / Tunnelling and Underground Space Technology 30 (2012) 110123 Table 1 Empirical or eld-based TBM performance prediction models and their advantages and disadvantages. Example Simple models Graham (1976) Typical advantages  Easy to apply Typical disadvantages

111

Multiple parameters models Probabilistic models Computer-aided models

CSM (Rostami, 1993, 1997), NTNU (Bruland, 1998), QTBM (Barton, 1999) Laughton (1998) Neural network models (e.g. Alvarez et al., 2000, Alvarez, 2000)

 Accounting for both rock mass and TBM parameters  Relying on good database  Accounting for randomness and approximation  Relying on good database

 Might underestimate due to lack of joint parameters  Limited range of application  Several parameters  Complex relationships  Using uncommon tests  Lack of detailed information from a like-case tunnel  Complex underlying structure  Over tting  Usually not available in public domain

indentation were all developed to provide for an improved prediction of cutter penetration under a given set of TBMrock interaction conditions, such as the one described for NTNU modeling by Bruland (1998). Other researchers have gone one step further and have attempted to recreate the process of rock fragmentation in a laboratory setting through the use of disc cutters. The use of full scale linear cutting tests, such as those described by Rostami and Ozdemir (1993), Rostami (1997, 2008), Sato et al. (1991), Sanio (1985), and Ozdemir et al. (1978) provide researchers with an enhanced ability to match eld parameters to deliver more accurate TBM performance predictions. Only a few laboratories around the world are equipped to perform such tests. Where such testing is not possible, TBM performance predictions may need to be based on an adjustment of performance data taken from sites where a rock with similar strength properties was bored. These adjustments may introduce signicant errors into the estimating process. The amount and likelihood of errors being introduced depend on accuracy of the underlying model assumptions, and the quality and quantity of TBM related and ground conditions data. Simple models are easy to use since they include only a few basic parameters (e.g. rock compressive and tensile strength), but they can only offer a limited range of application; many of the parameters that inuence TBM performance in more variable ground conditions, such as rock mass properties (e.g. RQD and rock type), are unaccounted for in the modeling process. Probabilistic models offer a more complex methodology for estimating performance. These models should only be used when it can be demonstrated that the detailed information (e.g. probability distribution functions for various parameters) of a similar tunnel is available to support the prediction of TBM performance on a new project. These models use performance data collected from similar case histories. If there are signicant differences in ground conditions or technology choices between the new drive and case histories within the database, substantial errors are likely in using the model. Another potential problem, which is also common for computer-aided models, is that in practice, these models are rarely used for TBM performance prediction purposes, even though they offer several advantages over the other methods (e.g. having a higher correlation coefcient and taking complex formula structures wherever needed). Multiple parameters models use more project specic data (compared to simple models), and they are easier to apply (compared to probabilistic and computer-aided models). Therefore, these models are among the most-favored models used in TBM performance prediction. This paper compares the results of common TBM performance prediction models through the evaluation of their predictive abilities. For this purpose, a testing database of 17 recent tunnel projects was used to represent the new projects. This database includes information on key performance parameters for various geological zones encountered along the tunnel. The paper also discusses the development of a new model that

can be used for the estimation of penetration rate. The new model is generated on the basis of the analysis of data from more than 300 TBM projects records (named as general database) from around the world. In other words, the TBM performance records in the general database are used to generate the new formulae and the testing database projects are treated as new projects where an estimate is needed (i.e. the results of predictions are compared with the actual eld performance to evaluate the validity and accuracy of the predictions and to objectively assess the predictive capabilities of a proposed set of new formulas). 2. Description of the TBM eld performance databases Two separate databases were compiled from the review of various technical sources. The rst database (general database) was assembled with the objective of developing a new performance model. The second database was developed to support model validation work. 2.1. General database The database on TBM eld performance contains different levels of information which denes the tunnel, rock mass conditions, and TBM performance parameters over the full length of a tunnel drive, within discrete geological zones, or short tunnel reaches. The general database contains data on more than 260 tunnel projects and includes over 300 data sets. This database is the continuation of an existing database developed at the University of Texas at Austin (UTA) (Nelson et al., 1994). Additional projects were added to the original sets from other available sources found in the literature. This database contains diameter from 1.63 to 11.74 m. TBM projects compiled in the database were completed between 1966 and 2004. An effort was also made to complete missing data elds within the database by checking many sources and published literature. In comparison to the original UTA databases, the updated database contains more detailed information and several new, more recently completed tunnel projects. This new database includes bored tunnel records with a total length of over 1500 km. Table 2 lists main parameters included in this database. The original database of Nelson et al. (1994), included data on 640 TBM projects. Data from the UTA database was compiled from diverse sources, including literature survey, manufacture records, and detail project records. Parameters for the database were recorded either as directly reported in documents or as estimated based on references (Laughton, 1998). The original database contained four levels of information. The rst three levels contain progressively more detailed information for a tunnel project over shorter spatial increments. Each zone is categorized based on a general geological structure and similar rock material characteristics. This increased level of detail continues down to a mining cell, which

112

E. Farrokh et al. / Tunnelling and Underground Space Technology 30 (2012) 110123

Table 2 Summary of the parameters included in the developed database. Layout parameters Project Name Location Application type Tunnel diameter Zone length Slope Construction access Tunneling hole through year Rock mass parameters Rock type Geological variability Quartz content UCS RQD/CFF Water condition Extreme mining areas Equipment parameters TBM type General spec. of TBM General spec. of BU TBM condition Type of mucking system Thrust Power Torque RPM Cutter size Cutter diameter Performance parameters Project time Support Type Penetration Rate Advance Rate Utilization factor Cutter wear rate Activity time distribution

is dened as a 10 m length within the tunnel. The fourth database level provides information required to model the TBM mechanical availability and the performance of key mining cycle activities; penetration, mucking and support installation. A schematic representation of the rst three database levels is shown in Table 3. Part of the general database used in current study is a subset of UTA database where more complete entries were available. Each data level can be linked to the other levels by means of a Drive Reference Number (DRN) that is unique for each tunneling project. For this study, only level two of the database was used. The data in this level contains average parameters for the zones of the tunnel or the different geological units along the tunnel. It contains 209 records and includes data on tunnel geometry, TBM characteristics, and geological parameters reported as numerical data, ranks, and categories. This database account for a total length of over 800 km bored length.

This database was one of the rst databases compiled for the purpose of predicting TBM advance rate. The database parameters dened rock type and related rock mass parameters referenced by reference to the recommendations of the International Society of Rock Mechanics (ISRM), including the Basic Geotechnical Description (BGD), and common rock mass classication systems (Laughton, 1998). It should be noted that in categorizing the rock types in this database, only the dominant rock type in each zone was noted. 2.2. Testing database To verify the predictive capability of existing or new models a testing database including 17 hard rock TBM projects was developed. These projects provided detailed information for TBM performance in each geological zone. TBM diameter for these projects ranged from 2.6 m to 11.8 m. As would be expected, the format

Table 3 Database levels and schematics of distance increments in each level (modied from Nelson et al., 1999). Database level Level 1 Geology description for the whole tunnel Distance increments

The whole tunnel drive

Tunnel Drive
Level 2 Geological variation characterized by zone parameters Tunnel

zone

Zone 1
Level 3 Geological variation characterized by zone and unit cell parameters Unit cell within a tunnel

Zone 2

Zone 3

Zone 4

Zone 5

zone

Unite Cell within Tunnel Zone

E. Farrokh et al. / Tunnelling and Underground Space Technology 30 (2012) 110123

113

and the extent of reported TBM operational parameters and geological data varied signicantly from project to project. Table 4 shows the summary information reported for these projects, notably including description of the TBM type, diameter and site geology.

3. Objective or target parameter A general review of previous TBM penetration rate models shows that, depending on the analytical approach adopted, different target penetration parameter were used to support the modeling of TBM penetration rate. Typically, the models predict performance parameters such as PR and PRev (penetration per revolution of the cutterhead). Other indexed parameters used in some of the prediction models include Specic Penetration (SP), and Field Penetration Index (FPI). These parameters are dened in Table 5. The purpose of using SP (Alber, 2000) is to combine the thrust and cutter head rate of rotation with penetration rate so that the penetration rate can be normalized against variations in rock mass strengths (As shown by Alber, 2000). Alber (2000) noted a general correlation between rock mass strength and SP without including any strength factor for the correlation. FPI was also introduced to be a normalized measure of penetration rate, specied in terms of applied thrust on cutters for a specic PRev in different geological conditions (Hamilton and Dollinger, 1979; Nelson et al., 1983; Klein et al., 1995). Nelson et al. (1983) offered a relationship between FPI and total hardness (Ht). Klein et al. (1995) presented correlations between FPI and intact rock and rock mass parameters with the primary goal of presenting different classes for different ground conditions in four tunnels. More recently, Hassanpour et al. (2009a,b) and Khademi et al. (2010) used FPI as an objective variable in a multiple regression setting to offer a new way of estimating the rate of penetration. In fact, SP is the inverse of FPI. The advantage of using these combined or normalized parameters is that they can be applied on machines of various sizes. These
Table 4 General information of the tunnel projects in testing database. No. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 Total Project name Ghomroud (Iran) Karaj (Iran) Zagros (Iran) Golab (Iran) Maen (Italy) Pieve (Italy) Varzo (Italy) Queens (USA) Milyang (S. Korea) Manapouri (Newzealand) S. Manhattan (USA) KCRC D 320-First Tube (Hong Kong) KCRC D 320-Second Tube (Hong Kong) Frasnadello-Pilot (Italy) Antea-Pilot (Italy) Frasnadello-Main (Italy) Antea-Main (Italy) Tunnel length (km)a 22.8 15.7 5.3 0.7 1.7 6.3 6.2 0.2 0.6 1.7 5.6 1.3 1.4 1.6 0.7 1.6 0.7 73.6 TBM type DS DS DS DS Open DS DS Open Open Open Open Mixed Shield Mixed Shield Open Open SS SS

parameters also account for the rate of cutterhead rotation (RPM), which is typically directly related to cutterhead diameter. Stevenson (1999) introduced Specic Excavation Rate (SER) as the excavated volume per revolution divided by thrust per cutter to combine SP and the tunnel cross-sectional area. One benet of using a normalized penetration rate is to allow the inclusion of a TBM-rock mass interaction factor. Laughton (1998) maintained that SP is not an appropriate parameter for TBM performance prediction purposes as it does not reect the true non-linear nature of the PRev:Fn (penetration-cutter load) trend. Fig. 1 shows a typical nonlinear relationship between PRev and Fn (cutter load) and a typical assumption for SP (straight line). As can be seen, the slopes of these two lines might be different drastically. As such, the accuracy of TBM penetration rate estimates based on SP or FPI may contain a signicant error if an SP from one tunnel drive is used directly to predict TBM performance on a new tunnel drive where a different cutter load is to be applied. This refers to the fact that for the estimation of the actual PR from these indices, the model require an assumed level of the cutter load. However, the applicability of a certain cutter load on a given TBM requires a close examination of TBM power and torque capacity. In other words, the cutter load rating of a given cutter size cannot be reliably used for PR estimation since in reality the TBM may not have sufcient power to turn the head at the required thrust level. The difference between the results are especially high if the applied cutter load is near the range of the threshold thrust where there is a change of slope in the PRev-Fn curve. Table 6 shows some examples of the signicant differences that can exist between nominal cutterhead thrust and RPM capacities and operating values in various geological settings. Even in relatively hard rock (Diorite, Meta-Volcanic) common operating thrust and RPM values can be far away from the design values. Part of the reason for such large discrepancies is related to the fact that there are interactions between the strength of the rock, the thrust level, and TBM cutterhead design parameters. This issue is addressed in the Colorado School of Mines (CSM) model (Rostami and Ozdemir, 1993; Rostami, 1997). This model was based on the results of

TBM diameter (m) 4.5 4.65 6.73 4.5 4.2 4.05 4.05 7.06 2.6 10.05 3.84 8.75 8.75 3.9 3.9 11.8 11.8

Geology Metamorphic rocks Pyroclastic rocks Limestone, Marl, Shale Diorite, Schist Meta gabbro, Meta basite, Schist Schist, Granite, Diorite Gneiss Granite Granite, Andesite Gneiss Gneiss, Schist Granite Granite Dolomite, Limestone and Argillite Dolomite Dolomite, Limestone and Argillite Dolomite

Reference
b

Hassanpour, 2009 Hassanpour, 2009

Sapigni et al., 2002 Sapigni et al., 2002 Sapigni et al., 2002 Ramezanzadeh, Kim, 2010 Kim, 2010
b

Ramezanzadeh, 2005 Ramezanzadeh, 2005 Barla and Pelizza, 2000 Barla and Pelizza, 2000 Barla and Pelizza, 2000 Barla and Pelizza, 2000

DS: double shield, SS: single shield. a With available data. b Data from authors (Farrokh et al., 2006, 2011; Farrokh and Rostami, 2007, 2008, 2009).

114 Table 5 TBM penetration rates. Description PR PRev SP FPI SER Penetration rate Penetration rate per revolution Specic penetration Field penetration index Specic excavation rate

E. Farrokh et al. / Tunnelling and Underground Space Technology 30 (2012) 110123

Typical unit m/h mm/rev (mm/rev)/(kN/ cutter) (kN/cutter)/(mm/ rev) (m3/rev)/(kN/ cutter)

Formula (1000 PR)/(60 RPM) PRev/Fn Fn/PRev A SP

A: Tunnel cross section area, Fn: Normal force per cutter.

PRev, mm Operating Point Line Slope= SP, mm/kN PRev Predictor

net Fn delivered to the cutterhead may be signicantly less than the nominal cutter load calculated from applied machine thrust. As Laughton (1998) notes, SP and FPI are preferred for the evaluation of TBM performance in the eld, where TBM is operated over the critical thrust (e.g. massive rocks with higher UCS value). In softer, more jointed rock masses, using these thrust-normalized indices are not necessarily the critical performance factors and there are several other parameters that can come into play, which overshadow the effects of the TBM operational thrust or cutterload. Therefore, normalized factors, FPI and SP, should be used with caution before providing a penetration rate estimate for a new project. Meanwhile for the development of alternative models, PR can be used in statistical studies to allow for an analysis of the real relationship between the various parameters, including TBM diameter. This is the approach that is used in the following sections of this paper.

4. Evaluation of existing penetration rate models The purpose of the study described in this section is to test the capability of some of the more commonly used or recently developed TBM performance prediction models as applied over a range of tunnel diameters and rock mass conditions. Table 7 summarizes more commonly used TBM performance prediction models. These models have been developed since the early 1970s. Among the models in Table 7, 12 models (with given formulae) were selected for evaluation. These models were selected based on the availability of the required information logged in the testing database of TBM eld performance. The graphs in Fig. 3 compare the actual and predicted PRs for the selected models. A 45 dashed line (1:1line) represents the line where predicted and actual rates are the same. Points plotted above the dashed line indicate an over-estimate of PR by predicting models. It should be noted that the upper cluster of the CSM model plot is attributed to pump limitations (e.g. max. propel rate of the hydraulic thrust cylinders during the stroke extension). This means that with the proper information on the applied cutter load, the estimated PR can be adjusted. The three models of Cassinelli et al. (1982), Innaurato et al. (1991), and Yagiz (2002) tend to underestimate TBM performance. One potential problem of these particular models may be related to the absence of any parameter that would account for a variation of tunnel diameter. The absence of a diameter parameter may be a result of the limited number of case histories that were referenced in the development of these models. The penetration rates of larger diameter TBMs are generally lower than those of smaller TBMs. Yagiz (2002) modied the CSM model using eld data collected on the performance of a relatively largediameter TBM tunnel (Queens Tunnel with the diameter of 7.06 m) and does not include a parametric adjustment for TBM diameter. As such, the corresponding graph in Fig. 3 shows the variation of this model for the Queens TBM or its equivalent in other ground conditions. The results of the remaining comparisons shown in Fig. 3, indicate that these models also generally tend to overestimate the PR. The percent differences between predicted and observed values can be more than 100%. Some likely causes of the tendency of models to overestimate PR are listed as follows:  Limited database use in the development of the models.  Exclusion of inuential parameters, such as tunnel diameter, from the model due to the limited range of TBM diameters in the original databases used for development of the models or the omission of key physical relationships between TBM parameters (i.e. RPM and hence PR are strongly related to TBM diameter).

Thrust Value

Intercept

Fn, kN Critical Thrust

Fig. 1. Relationship between penetration per revolution (PRev) and cutter normal force (Fn) (modied from Laughton, 1998).

Table 6 TBM operational parameters in different settings. Tunnel name Rock type RPM Design max Ghomroud III&IV Ghomroud III&IV Ghomroud III&IV Ghomroud III&IV Karaj Zagros Golab Maena Milyang Queens
a

Fn (kN) (Gross) Applied 11 6 10.3 8 7 6 8.6 10 8.3 Design max $220 $220 $220 $220 $220 $220 $220 $220 $195 $320 Applied 170 90 130 100 150 100 125 155 160 250

Limestone Slate MetaVolcanic Graphite Schist Tuff Limestone Diorite Serpentinite Granite Granitic Gneiss

12 12 12 12 11 11 12 11 13 8.3

Obtained from Sapigni et al. database (2002).

full-scale cutting tests, where the relationship between PRev and Fn follows a curve that is closer to actual cutting behavior of a disc cutter. The different relationships between the rock strength, cutter thrust, and TBM parameters were very well dened by Frenzel et al. (2008) and are shown in Fig. 2. A key reason for observing lower thrusts in the eld might be related to the TBM encountering weaker rock mass conditions. In this regard, a big problem for shielded TBMs is that the Fn value used in the predictive calculation is a gross value and does not include any friction losses. The

E. Farrokh et al. / Tunnelling and Underground Space Technology 30 (2012) 110123

115

 Use of inappropriate, inferred, estimated, or inaccurate parameters.  Lack of adequate TBM operating information, especially in weaker rock masses.  Use of too many parameters. In applying the model interactions between the many parameters may result in unrealistic result. The need for multiple parameters may also oblige the estimator to guess at multiple missing input values.  The mere fact that the predictions are based on the machines installed capacities (cutter load, power, etc.) whereas in reality machines are operated at lower thrust levels to cope with other eld parameters. To provide an example of the latter problem, consider the model introduced by Khademi et al. (2010). The bivariate analysis shows that UCS by itself accounted for 70% of the variation of the FPI. Adding three more parameters led to only a marginal increase in R2 from 0.7 to 0.77. This means that the effects of the additional parameters were largely overshadowed by UCS. It is worth noting

Fig. 2. Operating limits of a TBM with 1700 disc cutters at different rock strengths after Frenzel et al. (2008).

Table 7 TBM performance models. Author/model Tarkoy Roxborough and Phillips Graham Ozdemir et al. Farmer and Glossop Cassinelli et al. Snowdon et al. Lislerud et al. Nelson et al. Bamford Sanio Hughes Boyd Sato et al. Innaurato et al. Rostami and Ozdemir Sundin and Wanstedt Howarth Rostami Bruland Barton Cheema Alvarez Yagiz Ribacchi and Lembo Fazio Ramezanzadeh et al. Gong Hassanpour et al. Year 1973 1975 1976 1978 1980 1982 1982 1983 1983 1984 1985 1986 1986 1991 1991 1993 1994 1986 1997 1998 1999 1999 2000 2002 2005 Comment For limestone, shale, sandstone, quartzite, orthoquartzite, schist, dolomite with total hardness of 2242 and penetration rate of 0.0763.716 m/h For UCS of 70205 MPa, Tensile strength of 5.513.8 MPa, Cutter tip width of 11.419 mm, Cutter diameter range 382432 mm PRev = 3940 Fn/UCS Based on The Robbins Company data in granite, quartzite, schist, and shale Based on six tunnel projects data. PRev = 624 Fn/TS Using RSR. PR = -0.0059 RSR + 1.59 A Formula to demonstrate relationship among normal force, rolling force and penetration per revolution Based on excavation records in Norway in shale, limestone, gneiss, basalt Based on information of four tunnels in sedimentary rocks Based on data of tunneling in claystone on the Thompson project in Australia for bedding spacing range 0.30.5 m Effect of foliation on penetration rate For sandstone and penetration of up to 10 mm/rev On the basis of cutterhead power, specic energy, and tunnel cross section area Followed Sanios work and used the same approach, but on a rotary cutting machine Updated version of the method presented by Cassinelli, see above. Based on 112 homogeneous sections. No information is provided on the number of bored tunnels. PR = UCS0.437 0.047 RSR + 3.15 CSM model. On the basis of LCM tests For granite, micaschist, gneiss with UCS range 65200 MPa, point load range 19 MPa, CAI = 1.95.9, toughness of 2.23.3 Based on information of excavation in sandstone and marble with Fn = 3.16 kN and RPM = 14 Updated CSM model. On the basis of LCM tests. PRev = f (Fn, Fr) NTNU model. PRev = [Mekv/M1]b
0:2 QTBM model. PR = 5 Q TBM Based on information of one project to modify CSM model Neuro-Fuzzy modeling Based on information of one project to modify CSM model. PR = 0.859 + RFI + BI + 0.0969 PRevCSM; RFI = 1.44 Log(a) 0.0187 JS; BI = 0.0157 Ps 0:66 SP = 250 UCS cm

UCScm = UCS exp((RMR 100)/18) 2005 2005 2009a and 2009b 2010
0:37 Based on information of 11 projects to modify CSM model. PRev = PRevCSM exp(1.80.0031 JS-0.0065a)

Based on information of one project Based on information of two projects. FPI = 0.425 RMCI + 11.28 RMCI = 0.01 UCS RQD2/3 Based on information of one project. FPI = 4.161 + 0.091 UCS + 0.077 RQD + 0.117 Jc + 1.077 Log(a)

Khademi et al.

Nomenclature: PR: penetration rate, PRev: penetration per revolution, SP: specic penetration rate, FPI: eld penetration index, Fn: cutter normal force, Fr: cutter rolling force, UCS: uniaxial compressive strength of intact rock, TS: tensile strength, UCScm: rock mass uniaxial compressive strength, RSR: rock structure rating, RMR: rock mass rating, RQD: rock quality designation, Mekv: equivalent cutter thrust (kN/cutter), M1: critical cutter thrust (kN/cutter), that is necessary thrust to achieve 1 mm/rev, b = penetration coefcient, a: the angle between the tunnel axis and the planes of weakness, Ps = peak slope index (obtained from punch penetration test), Fs/Js = fracture/joint spacing, RFI: rock fracture index, BI: brittleness index, Jc: RMR joint condition partial rating, QTBM: Barton rock mass quality rating for TBM driven tunnels, RMCI: rock mass cuttability index, LCM: linear cutting machine.

116

E. Farrokh et al. / Tunnelling and Underground Space Technology 30 (2012) 110123

Fig. 3. Results of comparison for the selected models.

that even with very detailed information on selected sections of a tunnel drive; some models still fall short of yielding satisfactory results. From this perspective, although all these models were calibrated to eld performance data at one level or another, their predictive power is limited. These models can only be expected to

provide realistic predictive results when the new set of conditions is very similar to the conditions on which the models were based. The Neuro-Fuzzy model developed by Alvarez et al. (2000) is another analytical approach for PR prediction. This model used four major rules as described by Alvarez et al. (2000). Each rule

E. Farrokh et al. / Tunnelling and Underground Space Technology 30 (2012) 110123

117

is a linear combination of ve parameters. These rules were obtained based on the TakagiSugeno fuzzy method (Takagi and Sugeno, 1985) in combination with the least square method. Fig. 4 depicts the recreated Neuro-Fuzzy model rules (r1r4) using Excel spreadsheet formulae expressions:  r1: If CFF is low and UCS is medium and RPM is medium and Thr/c is medium and Dsize is large, then PR = 0.7288 CFF 0.01444 UCS + 0.1076 RPM + 0.001287 Thr/ c + 0.006937 Dsize + 0.937.  r2: If CFF is medium to high and UCS is medium and RPM is high and Thr/c is medium and Dsize is medium, then PR = 1.95 CFF + 0.05495 UCS + 0.13 RPM + 0.03825 Thr/ c + 0.04546 Dsize 24.65.  r3: If CFF is very high and UCS is very low and RPM is low and Thr/c is low and Dsize is small, then PR = 9.639 CFF + 0.1399 UCS + 3.332 RPM + 0.0511 Thr/ c 0.009726 Dsize + 1.319.  r4: If CFF is medium to high and UCS is high and RPM is high and Thr/c is low and Dsize is large, then PR = 1.459 CFF + 0.06171 UCS + 1.943 RPM + 0.3512 Thr/c 0.2676 Dsize 8.085. In these formulas, CFF is the core fracture frequency, UCS is the uniaxial compressive strength in MPa, RPM is the cutterhead revolutions per minute, Thr/c is the thrust per cutter in kN, and Dsize is the cutter diameter size in mm. The calculation methodology is explained in detail in Alvarez and Babuska (1999). When applying these rules within the testing database, the rules applied in isolation do not generate reasonable PR values. The application of rules number 24 are particularly suspect as they yield results that are mostly negative. Furthermore, for certain set of parameters this model yields negative value (see Fig. 5). The occurrence of negative results may indicate that although the fuzzy

logic models may yield better overall performance accuracy, the interpretation and use of such models might be more problematic. 5. Proposed new model Two different objective parameters were used to develop a new model for the estimation of TBM penetration rate under a given set of ground conditions. The rst analysis involved the analysis of penetration per revolution (PRev). Obviously, the penetration rate (PR) can be obtained from multiplication of PRev and cutterhead revolution per minute (RPM). The second analysis entailed the direct analysis of penetration rate (PR). As noted at the beginning of the paper, the different tunnel records within the database were heterogeneous. Only a limited number of records could be used in performing the above analyses, thus reducing the population size that was used in the statistical analyses. As for statistical process used in this study, the rst step involved bivariate analyses between penetration rate and other parameters to identify inuential parameters. Second step included multiple regression analyses to develop a best t combination of key parameters that demonstrate the strongest correlation to PR. Finally, the prediction capability of the new model was veried using the testing database information. 5.1. Modeling approach The compiled database includes numerical, ranked, and categorical data. Dealing with categorical data in the statistical analysis can be problematic. In one form, it is possible to assign different numbers to different categories to convert the qualitative or descriptive ranking information into quantitative values (ranking) as used by Alvarez et al. (2000). Fig. 6 shows a histogram of TBM diameters in the database. As can be seen, the 36 m diameter

Fig. 4. Recreated example of the Neuro-Fuzzy model recreated in an Excel sheet from Alvarez et al. (2000).

118

E. Farrokh et al. / Tunnelling and Underground Space Technology 30 (2012) 110123

Fig. 5. An example of the recreated Neuro-Fuzzy model (Alvarez et al., 2000) yielding negative value.

range is common for a large number of projects, and it is the most popular range in the database. The graphs in Figs. 79, show that tunnel diameter is related to TBM characteristics and performance. PR and PRev were chosen as the objective parameters. These parameters are commonly reported more than the other performance measures such as FPI, SP, and SER. Conversion of the PRev values to FPI or SP based on machines installed power and cutter load could be erroneous due to operation of the machines below their nominal capacities and lack of measured parameters for calculation of these indices. Fig. 10 depicts the correlation between some of the most important dependent variables and the penetration rate (PR) or penetration per revolution (PRev). Coefcients of determination (R2) of the bivariate analyses (Fig. 10) show that PR-UCS has the strongest correlation. It should be noted that R2(adj) is the adjusted R2 for the number of parameters in the model, and the number of data points used for regression analysis.

As can be seen in Fig. 10, PR and PRev are at maximum at lower UCS levels. As UCS increases, PR and PRev decrease. This is a logical trend and is in agreement with several research studies such as those reported by Laughton (1998), Robbins (1992), Hassanpour et al. (2009a,b), and Khademi et al. (2010). However, this is not the whole story since with lower UCS values, the operators tend to reduce the thrust thus the correlation in the graphs is not as strong as it should truly be. Since different rock textures (cementation and grain size and shape) affect the penetration rate, such properties should also be taken into account in PR studies. In these analyses, seven rock type categories, as proposed by Hoek and Brown (1980) and Stevenson (1999), were adopted. These rock types are listed in Table 8. The rst four classes are for Sedimentary Rocks. The fth, sixth, and seventh classes are for Metamorphic Rocks, Granitic Rocks, and Volcanic Rocks respectively. It should be mentioned, Gneiss

Fig. 6. Histogram of tunnel diameter in the database.

Fig. 7. Relationship between installed or nominal torque and tunnel diameter.

E. Farrokh et al. / Tunnelling and Underground Space Technology 30 (2012) 110123

119

identied and reported. A transformation of the objective parameter was made wherever it was necessary to correct for normality in the regression model (i.e. Ln(PRev) for Eq. (1), and Ln(PR) for Eq. (3)). The results of these analyses are shown in Fig. 11 and Eq. (1).

PRev Exp0:41 0:404 D 0:027 D2 0:0691 RTc 0:00431 UCS 0:0902 RQDc 0:000893 Fn R2 63% 1

PR

PRev RPM 60 1000

Fig. 8. Relationship between nominal RPM and tunnel diameter.

where D is tunnel diameter in m, RTc is rock type numerical code (1 for G and GN, 2 for MV, 3 for SLK, 5 for C), UCS is uniaxial compressive strength in MPa, RQDc is RQD numerical code (Table 9), and Fn is disc cutter normal force in kN. It should be noted that at larger diameters, the TBM face area increases signicantly allowing for a more efcient cutting process, higher penetration per revolution can be achieved. Since the case histories typically do not report the operating thrust; we assume that the TBMs operated close to the design thrust values a realistic assumption in harder rocks. The installed thrust capacity was used in the calculation of thrust per cutter (Fn). Eq. (2) can be used to convert PRev to PR. Fig. 12 depicts comparative results of Eqs. (1) and (2) results with the actual values within the test database. As can be seen in Fig. 12, the modeled PR values show good correlation with actual values. Furthermore, the new model yields good improvement in predictive capacity over the other models evaluated. Fig. 13 shows the regression analysis performed on the general database for PR evaluation using the logarithm of the PR (Ln(PR)). The resulting PR can be expressed by a power function as in Eq. (3). The regression equation is:

PR

:133 :183 Fn0:186 RQD0 RT 0 RPM0:363 D5:47 exp0:046 D2 c c

5:64 UCS0:248 exp1:58 D

R2

58%
Fig. 9. Relationship between installed or nominal thrust and tunnel diameter.

3
where D is tunnel diameter in m, RTc is rock type code (1 for G and GN, 2 for MV, 3 for SLK, 5 for C), UCS is uniaxial compressive strength in MPa, RQDc is RQD code (Table 9), Fn is disc cutter normal force in kN, and RPM is revolution per minute. Eq.(3) was obtained by performing a log transformation of most of the model parameters. Fig. 14 shows comparative results of Eq. (3) with actual values from the testing database. The comparison between the results in Figs. 12 and 14 indicates that, overall, PR formula yields better results and using normalized penetration rates for the purpose of TBM performance prediction may yield less reliable results, even when PRev is estimated and adjusted with the use of real RPM. It should be noted that Eqs. (1) and (3) may produce higher errors in estimating PR values in highly jointed rock masses due to lack of exact RQD value or lack of other rock mass parameters (e.g. joint orientation). In general, the proposed new models offer better results than those of the previously discussed models and indicate a potential for better PR prediction. Further study of this approach will require the development of new databases that report additional TBM and ground condition factors. This effort will require worldwide coordination and close collaboration between project partners, including contractors, engineers, and owners. Data to be reported should include: Recent tunnel information for various TBM types. Detailed rock mass classication information such as RMR or RQD, Q, and GSI for a range of tunnel sizes and rock types.

(GN) is inherently metamorphic, but it is typically closer to granite in terms of its behavior, especially where foliation is less pronounced. For this reason, it was categorized as GN in this analysis. As can be seen in Fig. 10, when the rock type is taken into consideration in the analysis, a good relationship can be established between rock strength and PR/PRev. The graphs show that in general a higher PR/PRev is achieved in sedimentary rocks, and a lower PR/PRev is achieved in igneous rocks. These results are in agreement with the results proposed by Laughton (1998) and Robbins (1992) for different rock types. Core fracture frequency (CFF) data was the only rock mass parameter that was available for all the records from Nelson et al. database (1994) and some of the records of the general database. Basically, this factor is in close relationship with RQD and refers to the frequency of rock mass fractures. Table 9 shows the approximate relationship between CFF and RQD and the numerical codes used for subsequent analyses.

5.2. Multivariate regression analysis Multivariate regression analyses with PRev as the objective parameter were performed using Minitab 16 (Minitab Inc., 2010). These analyses allowed PR results to be projected over different machine sizes, with PRev being multiplied by RPM to determine PR. Based on these analyses, best-t linear regressions were

120

E. Farrokh et al. / Tunnelling and Underground Space Technology 30 (2012) 110123

Fig. 10. Correlation between PR/PRev and other parameters (rock type abbreviations from Table 8, for rock type refers to records with missing rock type, for PRev refers to outlier).

Table 8 Rock type categorization in database (modied from Hoek and Brown, 1980). Rock type Claystone, mudstone, marl, slate, phyllite, argillite Sandstone, siltstone, conglomerate, quartzite Limestone, chalk, dolomite, marble Karstic Limestone Metamorphic rocks such as gneiss and schist Coarse igneous such as granite and diorite Fine volcanic such as basalt, tuff, and andesite Code C S L K M G V

Recording and use of actual operational parameters for a range of tunnel sizes and rock types. 6. Discussion and conclusions Comparisons between predicted and actual TBM performance indicated that most of the existing predictive models, especially the simple ones, cannot offer accurate estimates of TBM performance for new projects. A study of these models indicates that using more parameters in a model will not always guarantee improved results due to lack of inherent limits of the initial models. In some cases, existing models do not include important parameters; as such they cannot adequately distinguish between the

Detailed specication for various TBM types, for a range of tunnel sizes and rock types.

E. Farrokh et al. / Tunnelling and Underground Space Technology 30 (2012) 110123 Table 9 CFF categorization. CFF Less than 8 fractures/m 812 fractures/m 1216 fractures/m Code S or 1 M or 2 H or 3 Description Low frequency Medium frequency High frequency Corresponding RQD range 90100 6090 <60

121

Fig. 13. Results of linear regression analysis for PR (response is Ln(PR)).

Fig. 11. Results of linear regression analysis for PRev (response is Ln(PRev)).

ground conditions and job constraints that control TBM performance. To achieve more accurate estimates, a database of TBM eld performance was compiled and subjected to a statistical analysis in order to derive new equations and models for improving performance prediction for hard rock TBMs. The proposed models offer better accuracy than existing models. Part of the reason for the model improvement may be related to the ability to utilize a larger database of TBM eld performance records that includes a wider range of tunnel diameters and ground conditions. Limited numbers of rock mass parameters in the database and the lack of accurate records of actual TBM operational parameters are among the reasons for the scatter in the results of the new model. Expanding the existing databases with more parameters and adding more details on the different geological zones should allow for more improvement in the results and a higher degree of accuracy and reliability in model prediction.

The analyses of the available data indicate that certain set of available parameters including, tunnel diameter, UCS, RPM, and rock type, account for approximately half of the PR and PRev variation. These analyses further indicate that UCS is the single most important rock parameter controlling PRev. Obviously, frequency and condition of jointing can have a dominant impact on TBM performance especially in harder rocks and further study of this phenomenon, notably including a quantitative representation of joint spacing and condition may be needed to improve model accuracy in harder rock units. There are other areas of concern related to the use of TBM design parameters in performance prediction. The TBMs in operation may only use a fraction of the torque and thrust capacity that is installed in the factory. In the development of new models or the modication of existing models, one should be able to evaluate performance over a range of physical scales, perhaps zone-by-zone analysis, which based on previous experiences can yield the best results. Additionally, one should always be mindful of the tradeoff between model complexity, accessibility of the relevant data and rock property tests for various models, and the desired accuracy and reliability of the model. Given all these issues, the proposed models should be used with caution on any new project since they may lack TBM and ground-related parameters that are specic to a project. The proposed new formulas can serve the purpose of offering preliminary TBM performance estimates. Used in conjunction with anticipated utilization rates, the models can be used to

Fig. 12. The comparative results of Eqs. (1) and (2) using testing database.

122

E. Farrokh et al. / Tunnelling and Underground Space Technology 30 (2012) 110123 Farrokh, E., Rostami, J., 2008. Correlation of tunnel convergence with TBM operational parameters and chip size in the Ghomroud tunnel, Iran. Tunnell. Undergr. Space Technol. 23 (6), 700710. Farrokh, E., Rostami, J., 2009. Effect of adverse geological condition on TBM operation in Ghomroud tunnel conveyance project. Tunnell. Undergr. Space Technol. 24 (4), 436446. Farrokh, E., Mortazavi, A., Shamsi, Gh., 2006. Evaluation of ground convergence and squeezing potential in the TBM driven Ghomroud tunnel project. Tunnell. Undergr. Space Technol. 21 (5), 504510. Farrokh, E., Rostami, J., Laughton, C., 2011. Analysis of unit supporting time and support installation time for Open TBMs. Rock Mech. Rock Eng. 44 (4), 431445. Frenzel, Ch., Ksling, H., Thuro, K., 2008. Factors inuencing disc cutter wear. Geomech. Tunnel. 1, 5560. Gong, Q.M., 2005. Development of a Rock Mass Characteristics Model for TBM Penetration Rate Prediction, PhD Thesis, Nanyang Technology University. Graham, P.C., 1976. Rock exploration for machine manufacturers. In: Bieniawski, Z.T. (Ed.), Exploration for Rock Engineering. Balkema, Johannesburg, pp. 173 180. Hamilton, W.H., , Dollinger, G.L., 1979. Optimizing tunnel boring machine and cutter design for greater boreability. In: RETC Proceedings, Atlanta, vol. 1, pp. 280296. Hassanpour, J., 2009. Investigation of the Effect of Engineering Geological Parameters on TBM Performance and Modications to Existing Prediction Models. Ph.D. Thesis, Tarbiat Modares University, Tehran, Iran. Hassanpour, J., Rostami, J., Khamehchiyan, M., Bruland, A., Tavakoli, H.R., 2009a. TBM Performance Analysis in Pyroclastic Rocks: A Case History of Karaj Water Conveyance Tunnel. Rock Mech. Rock Eng. Hassanpour, J., Rostami, J., Khamehchiyan, M., Bruland, A., 2009b. Development new equations for performance prediction. Geo Mech. Geoeng.: Int. J. 4 (4), 287297. Hoek, E., Brown, E.T., 1980. Underground Excavations in Rock. Instn. Min. Metall., London, 527 p. Howarth, D.F., 1986. Review of Rock Drillability and Boreability Assessment Methods. Trans. IMM Sect. A Min. Ind. The Institution of Mining and Metallurgy, London, vol. 9, pp. A191A202. Hughes, H.M., 1986. The relative cuttability of coal-measures stone. Min. Sci. Technol. 3 (2), 95109. Innaurato, N., Mancini, R., Rondena, E., Zaninetti, A., 1991. Forecasting and effective TBM performances in a rapid excavation of a tunnel in Italy. In: Proceedings of the Seventh International Congress ISRM, Aachen, pp. 10091014. Khademi Hamidi, J., Shahriar, K., Rezai, B., Rostami, J., 2010. Performance prediction of hard rock TBM using rock mass rating (RMR) system. Tunnell. Undergr. Space Technol. 25 (4), 333345. Kim, T., 2010, Private communication and data exchange. Klein, S., Schmoll, M., Avery, T., 1995. TBM performance at four hard rock tunnels in California. In: Proceedings of the Rapid Excavation & Tunnelling Conference, pp. 6175 (Chapter 4). Laughton, C., 1998. Evaluation and Prediction of Tunnel Boring Machine Performance in Variable Rock Masses. Ph.D. Thesis, The University of Texas, Austin, USA. Lislerud et al., 1983. Hard Rock Tunnel Boring, Project Report 183. Norwegian Institute of Technology, Trondheim. Minitab Inc., 2010. Minitab 16 Software. Nelson, P., ORourke, T.D., Kulhawy, F.H., 1983. Factors affecting TBM penetration rates in sedimentary rocks. In: Proceedings, 24th US Symposium on Rock Mechanics, Texas A&M, College Station, TX, pp. 227237. Nelson, P.P., Abad-Aljali, Y., Laughton, C., 1994. Tunnel Boring Machine Project Databases and Construction Simulation. Geotechnical Engineering Report GR94-4, Geotechnical Engineering Center, Department Civil Engineering, The University of Texas at Austin. Nelson, P.P., Abad-Aljali, Y., Laughton, C., 1999. Improved Strategies for TBM Performance Prediction and Project Management. RETC, pp. 963979. Ozdemir, L., Miller, R., Wang, F.D., 1978. Mechanical Tunnel Boring Prediction and Machine Design. Final Project Report to NSF APR73-07776-A03. Colorado School of Mines, Golden, Co. Ramezanzadeh, A., 2005. Performance Analysis and Development of New Models for Performance Prediction of Hard Rock TBMs in Rock Mass, Ph.D. Thesis, INSA, Lyon, France, p. 333. Ramezanzadeh, A., Rostami, J., Kastner, R., 2005. Inuence of Rock Mass Properties on Performance of Hard Rock TBMs. RETC, June 2729, Seattle, Washington, USA. Ribacchi, R., Lembo-Fazio, A., 2005. Inuence of rock mass parameters on the performance of a TBM in a Gneissic formation (Varzo tunnel). Rock Mech. Rock Eng. 38 (2), 105127. Robbins, R.J., 1992. Large diameter hard rock boring machines: state-of-the-art and development in view of alpine base tunnels. Felsbau 10 (2), 5662. Rostami, J., Ozdemir, L., 1993. . A new model for performance prediction of hard rock TBM. In: Bowerman, L.D., et al. (Eds.), Proceedings, Rapid Excavation and Tunneling Conference, pp. 793809. Rostami, J., 1997. Development of a Force Estimation Model for Rock Fragmentation with Disc Cutters through Theoretical Modeling and Physical Measurement of Crushed Zone Pressure. Ph.D. Thesis, Colorado School of Mines, Golden, Colorado, USA, p. 249. Rostami, J., 2008. Hard rock TBM cutterhead modeling for design and performance prediction. Geomechanik und Tunnelbau, Ernst & Sohn, (Austrian Journal of Geotechnical Eng.), January 2008.

Fig. 14. The comparative results of Eq. (3) using testing database.

estimate TBM advance rates for the early stage of a project planning process. The model can also be used to calculate the TBM boring time in the models under development for estimation of the utilization rate, based on activity time allocations. The authors strongly recommend the use combination of models to ensure a higher degree of condence in the development of nal estimates. Preference should be given to using models that best match key project criteria relative to the ground conditions and TBM type. In summary, the proposed PR predictive models provide alternative methodologies that can be used at the planning stage of a TBM drive when project-specic information is limited or as a double check on the results of more detailed predictive approaches based on the use of existing, more complex models.

References
Alber, M., 2000. Advance rates of hard rock TBMs and their effects on projects economics. Tunnell. Undergr. Space Technol. 15 (1), 5584. Alvarez Grima, M., Babuska, R., 1999. Fuzzy model for the prediction of unconned compressive strength of rock samples. Int. J. Rock Mech. Min. Sci 36 (3), 339 349. Alvarez Grima, M., 2000. Neuro-Fuzzy Modeling in Engineering Geology: Applications to Mechanical Rock Excavation, Rock Strength Estimation and Geological Mapping. Balkema, Rotterdam. Alvarez Grima, M.P.A., Bruines, P.A., Verhoef, P.N.W., 2000. Modeling tunnel boring machine performance by neuro-fuzzy methods. Tunnell. Undergr. Space Technol. 15 (3), 259269. Bamford, W.F., 1984. Rock test indices are being successfully correlated with tunnel boring machine performance. In: Proceedings of the 5th Australian Tunneling Conference, Melbourne, vol. 2, 1984, pp. 922. Barla, G., Pelizza, S., 2000. TBM tunneling in difcult ground conditions, Proceedings of GeoEng 2000. In: Proceedings of the International Conference on Geotechnical & Geological Engineering, Melbourne, November 1924, 2000. Technomic Publishing Company, Lancaster, pp. 329354. Barton, N., 1999. TBM performance in rock using QTBM. Tunnels Tunnell. Int. 31, 4148. Boyd, R.J., 1986. Hard rock continuous mining machine: mobile miner MM-120. In: Howarth, D.F., et al. (Ed.), Rock Excavation Engineering Seminar, Dept. Mining and Met. Eng, University of Queensland. Bruland, A., 1998. Hard Rock Tunnel Boring. Ph.D. Thesis, Norwegian University of Science and Technology, Trondheim. Cassinelli, F., Cina, S., Innaurato, N., Mancini, R., Sampaolo, A., 1982. Power Consumption and Metal Wear in Tunnel-boring Machines: Analysis of Tunnel-Boring Operation in Hard Rock. Tunneling 82, London, Inst. Min. Metall., pp. 7381. Cheema, S., 1999. Development of a Rock Mass Boreability Index for the Performance of Tunnel Boring Machines. Doctoral Dissertation, Department of Mining Engineering, Colorado School of Mines, Golden, CO, USA. Farmer, I.W., Glossop, N.H., 1980. Mechanics of disc cutter penetration. Tunnels Tunnell. Int. 12 (6), 2225. Farrokh, E., Rostami, J., 2007. The Relationship between Tunnel Convergence and TBM Operational Parameters and Chip Size for Double Shield TBMs, RETC, Canada, pp. 10941108 (Chapter 88).

E. Farrokh et al. / Tunnelling and Underground Space Technology 30 (2012) 110123 Roxborough, F.F., Phillips, H.R., 1975. Rock excavation by disc cutter. Int. J. Rock Mech. Min. Sci. Geomech. Abstr. 12, 361366. Sanio, H.P., 1985. Prediction of the performance of disc cutters in anisotropic rock. Int. J. Rock Mech. Min. Sci. Geomech. Abstr. 22, 153161. Sapigni, M., Berti, M., Behtaz, E., Busillo, A., Cardone, G., 2002. TBM performance estimation using rock mass classication. Int. J. Rock Mech. Min. Sci. 39, 771788. Sato, K., Gong, F., Itakura, K., 1991. Prediction of disc cutter performance using a circular rock cutting ring. In: Proceedings, The rst International Mine Mechanization and Automation Symposium, Colorado School of Mines, Golden, Colorado, USA. Snowdon, R.A., Ryley, M.D., Temporal, J., 1982. A study of disc cutting in selected British rocks. Int. J. Rock Mech. Min. Sci. 19, 107121. Stevenson Garry, W., 1999. Empirical estimates of TBM performance in hard rock. In: Proc. RETC, AIME, pp. 9931006.

123

Sundin, N.O., Wanstedt, S., 1994. A boreability model for TBMs. In: Nelson, P., Laubach, S.E. (Eds.), Rock Mechanics Models and Measurements Challenges from Industry. Proceedings of the 1st North American Rock Mechanics Symposium, The University of Texas at Austin, Balkema, Rotterdam, pp. 311318. Takagi, T., Sugeno, M., 1985. Fuzzy identication of systems and its applications to modeling and control. IEEE Trans. Syst. Man Cybern. 15 (1), 116132. Tarkoy, P.J., 1973. Predicting TBM penetration rates in selected rock types. In: Proceedings, 9th Canadian Rock Mechanics Symposium, Montreal. Yagiz, S., 2002. Development of Rock Fracture and Brittleness Indices to Quantifying the Effects of Rock Mass Features and Toughness in the CSM Model Basic Penetration for Hard Rock Tunneling Machines. PhD Thesis. T-5605, Colorado School of Mines, CO, USA, p. 289.