Application of rock mass classification systems for performance estimation of rock TBMs using regression tree and artificial intelligence algorithms
Introduction
Use of Tunnel boring machine (TBM) has become the preferred tunnelling method in rock for various sizes, ground conditions, and application. This is due to higher speed, lower labour intensity, safer work environment, and lower cost that TBM tunnelling offers. Estimating the performance of TBM in various grounds is a vital part of tunnel design and selection of the most appropriate machine type and specifications. During the past two decades, numerous TBM performance prediction models have been proposed. This include theoretical models based on the process of rock fragmentation in a laboratory setting and the use of full scale linear cutting tests, and the empirical models based on field observation of machine performance. CSM model (Rostami, 1997, Rostami, 2013) is an example of the first group whereas Norwegian and Field Penetration Index (FPI) models are the most frequently used examples of the empirical models. The theoretical models do not completely represent the real rock mass conditions, in particular joints, as the TBM disc cutters would encounter in the field. Thus, Yagiz, 2002, Ramezanzadeh, 2005 have recommended modifications to the original CSM model to integrate rock mass properties as input parameters into the model to account for joints and discontinuities, but with limited success.
The early empirical models used a single intact rock parameter, mainly uniaxial compressive (UCS) or tensile strength (BTS-Brazilian Tensile Strength), to define the ground and many of the parameters that influence TBM performance in more variable ground conditions, such as rock mass properties, including RQD (Rock Quality Designation), joint spacing, etc. were excluded. Hence, latest empirical models considered several rock and rock mass parameters along with machine design and operational parameters in the models to estimate machine performance since machine specifications, i.e. operational thrust and torque have major influence on TBM performance. The Norwegian (NTNU) model uses specialized rock drillability/boreability indices, such as Drilling Rate Index (DRI) and rock mass properties including joint spacing and orientation to estimate TBM performance (Bruland, 1998, Blindheim, 1979, Dahl et al., 2012). The indices are not commonly available in many projects unless specifically ordered for borebility assessments.
Field Penetration Index (FPI) has been introduced by Nelson et al. (1983) and has been subsequently used as a means for predicting the performance of TBMs. For instance, Hassanpour et al. (2011), has used FPI estimated as a function of RQD and UCS to develop new equations and charts for TBM performance prediction. Also, a new model has been presented for prediction of TBM performance in blocky rock conditions (FPIblocky) based on UCS and volumetric joint account (JV) (Delisio and Zhao, 2014). Recently, Salimi et al., 2016c, Salimi et al., 2017 have proposed new empirical equations for prediction of TBM performance from FPI, which is calculated from UCS and joint spacing Js or alternatively UCS, RQD, and Jc.
Other empirical models have also been introduced to correlate TBM performance to the common rock mass classification systems, including Rock Mass Rating (RMR), Rock Mass Quality Index (Q) and Geological Strength Index (GSI) due to simplicity and worldwide acceptance/availability of the classification systems in general engineering practices, such as underground mining and construction (Hassanpour et al., 2009, Hassanpour et al., 2010, Khademi Hamidi et al., 2010). The performance of the models based on RMR, Q, and GSI has had limited success in accurately predicting TBM performance.
Apart from empirical and theoretical models, the use of artificial intelligence (AI) techniques has received widespread attention in TBM performance prediction. Several techniques, such as artificial neural network (ANN), fuzzy logic, adaptive neuro-fuzzy inference system (ANFIS), particle swarm optimization (PSO), and support vector machine (SVM) have been used for this purpose and for estimating penetration rate (PR) and advance rate (AR) (Mahdevari et al., 2014, Yagiz et al., 2009, Benardos and Kaliampakos, 2004, Alvarez Grima et al., 2000). Flexible nature of the AI techniques makes them powerful tools in approximating and solving engineering problems, more specifically when the problem is highly complex and nonlinear. The results of most of these studies have been ‘‘black box’’ programs that show high correlation between their predicted rates and actual machine performance but cannot be used in estimating machine performance in other projects.
The objective of this study is to evaluate the feasibility of using common rock mass classification systems to estimate TBM Field Penetration Index (FPI) as an indicator of machine performance. This could lead to development of new models to predict the performance of TBM in various ground conditions. To meet this objective, multivariable regression analysis and artificial intelligence algorithms which are known as “white box” solutions, including Classification & Regression Tree (CART) and Genetic Programming (GP) due to offering pertinent graphs (diagrams) and mathematical equations, have been used. As such, compiled field data from seven tunnelling projects, containing Zagros water conveyance tunnel, Lot 2 in Iran (Hassanpour et al., 2009); Ghomrood water conveyance tunnel, Lots 3 & 4 in Iran (S.C.E. Company., 2004, Hassanpour et al., 2011); Karaj water conveyance tunnel, Lot 1 in Iran (Hassanpour et al., 2010); Golab conveyance water tunnel in Iran (Fatemi et al., 2016, Imensazan consulting engineers (ICE), 2009); Maroshi-Ruparel water supply tunnel, Mumbai India (Jain et al., 2014, Jain, 2014); Manapouri second tailrace tunnel, New Zealand (URS Company, 2003) and Lötschberg Base Tunnel in Switzerland (Delisio and Zhao, 2014, Delisio, 2014) were considered and used to establish a new model for TBM performance prediction.
Section snippets
Description of the projects & data collection
In order to develop a more accurate TBM performance prediction model that can be applied in different geological conditions and evaluate the common rock mass classification systems, including RMR, Q, and GSI, data from various projects with different rock mass conditions have been obtained from pertinent research groups and compiled in a database. The database on TBM field performance contains different levels of information which defines the tunnel, rock mass conditions, and TBM performance
TBM performance & rock mass classification systems
Over the years, many rock mass classification systems have been presented in mining and civil engineering. As noted by Bieniawski (1989), a rock mass classification scheme is designated to classify the rock masses, provide a basis for estimating deformation and strength properties, supply quantitative data for support estimation, and present a platform for quantitative communication between exploration, design, and construction groups. Also, these models are commonly employed in many empirical
TBM performance evaluation using (RMR) classification system
Various TBM performance indices have been introduced and used by many researchers to assess the boreability of a rock mass. This includes penetration rate (PR), net penetration rate or penetration per one cutterhead revolution (PRev), FPI, Specific Penetration (SP, inverse of FPI) which are the composition of penetration per revolution (that accounts for TBM size and RPM) and cutterload (which represents machine thrust). The purpose of using SP (Alber, 2000) or FPI (Nelson et al., 1983) is to
Developing new empirical equations
Empirical equations have great importance during the early stages of rock excavation and design works since they are more practical compared to extensive theoretical analyses. In this study, regression analysis was used to examine correlations between TBM performance parameters and geomechanical parameters in the database. Both types of multi-variable regression analysis (linear and non-linear form) were used to develop empirical equations.
Artificial intelligence methods & TBM performance
Due to variation in geotechnical conditions and machine operational parameters in tunnelling, artificial intelligence (AI) based models have been successfully employed by some researchers to develop TBM performance prediction models. One of the early works in this field was by Alvarez Grima et al. (2000) where neuro-fuzzy method for TBM performance prediction was introduced. Benardos and Kaliampakos (2004) proposed an ANN model by using data of 1077 m of Athens Metro tunnel in Greece. Yagiz et
Comparison of developed model
The performance of the proposed models were evaluated according to statistical criteria, such as correlation coefficient (R2), the root mean square error (RMSE), mean absolute deviation (MAD), mean absolute percentage error (MAPE), mean square error (MSE) and variance account for (VAF). Description of these statistical parameters can be found in text books and statistical software tutorials. The results of applying these models are summarized in Table 8. The results show that CART offers the
Sensitivity analysis of developed models
Sensitivity analysis examines the variability of the outputs of a model by changing the input variables within a prescribed range. This process allows for better understanding of the relationship between the input parameters and the output results and facilitates recognition of possible errors caused by models when the input parameters vary beyond an intended range. In addition, this analysis accommodates validating the results and identification of critical or uninfluential parameters in the
Conclusion
While various empirical models for TBM performance prediction have been introduced, many lack the correct context of input geotechnical parameters to account for the possible range of variation of such parameters. Rock mass classifications have been used for TBM performance prediction, often with low accuracy since they were originally developed for rock load and ground support design and not TBM excavation. Among the most commonly used classification systems, RMR offer better results with
Acknowledgement
First author extends sincere thanks to Prof. T.N. Singh and Drs. Jafar Hassanpour, Prasnna Jain, Amir Asad Fatemi, Andrea Delisio, Masoud Zare Naghadehi for sharing their data for developing the TBM database used in this study.
References (50)
Advance rates of hard rock TBMs and their effects on project economics
Tunnell. Undergr. Space Technol.
(2000)- et al.
Modelling tunnel boring machine performance by Neuro-Fuzzy method
Int. J. Tunnell. Undergr. Space Technol.
(2000) - et al.
Modelling TBM performance with artificial neural networks
Tunnell. Undergr. Space Technol.
(2004) - et al.
Classifications of properties influencing the drillability of rocks, based on the NTNU/SINTEF test method
Int. J. Tunnell. Under. Space Technol.
(2012) - et al.
A new model for TBM performance prediction in blocky rock conditions
Int. J. Tunnel Under. Space Technol.
(2014) - et al.
Study of various models for estimation of penetration rate of hard rock TBMs
Int. J. Tunnel Under. Space Technol.
(2012) - et al.
Numerical simulation of influence of joint spacing on rock fragmentation by TBM cutters
Int. J. Tunnell. Undergr. Space Technol.
(2006) - et al.
In situ TBM penetration tests and rock mass boreability analysis in hard rock tunnels
Tunnell under. Space Technol.
(2007) - et al.
A new hard rock TBM performance prediction model for project planning
Tunnel Under. Space Technol.
(2011) - et al.
Performance characteristics of tunnel boring ma-chine in basalt and pyroclastic rocks of Deccan traps—a case study
Int. J. Rock Mech. Geotech. Eng.
(2014)
Performance prediction of hard rock TBM using Rock Mass Rating (RMR) system
Tunnel. Undergr. Space Technol.
A support vector regression model for predicting tunnel boring machine penetration rates
Int. J. Rock Mech. Min. Sci.
Study of pressure distribution within the crushed zone in the contact area between rock and disc cutters
Int. J. Rock Mech. Mining Sci.
Application of non-linear regression analysis and artificial intelligence algorithms for performance prediction of hard rock TBMs
J. Tunnel. Under. Space Technol.
TBM performance estimation using rock mass classification
Int. J. Rock Mech. Min. Sci.
Utilizing rock mass properties for predicting TBM performance in hard rock condition
Tunnell. Undergr. Space Technol.
Application of two non-linear prediction tools to the estimation of tunnel boring machine performance
Eng. Appl. Artif. Intell.
TBM Tunneling in Jointed and Faulted Rock
Engineering Rock Mass Classifications: A Complete Manual for Engineers and Geologists in Mining, Civil and Petroleum Engineering
Classification and Regression Trees
A representation for the adaptive generation of simple sequential programs
Evaluation of TBM performance prediction models and sensitivity analysis of input parameters
Int. J. Bull. Eng. Geol. Environ.
Cited by (64)
Autonomous steering control for tunnel boring machines
2024, Automation in ConstructionAn empirical method for estimating TBM penetration rate using tunnelling specific energy
2024, Tunnelling and Underground Space TechnologyAnalysis and prediction of small-diameter TBM performance in hard rock conditions
2024, Tunnelling and Underground Space TechnologyReal-time perception of rock–machine interaction information in TBM tunnelling using muck image analysis
2023, Tunnelling and Underground Space TechnologyTBM penetration rate prediction ensemble model based on full-scale linear cutting test
2023, Tunnelling and Underground Space Technology