Prediction of tunnel convergence using Artificial Neural Networks

https://doi.org/10.1016/j.tust.2011.11.002Get rights and content

Abstract

This research intends to develop a method based on Artificial Neural Network (ANN) for prediction of convergence in tunnels. In this respect, data sets of the convergence monitored in different section of a tunnel and geomechanical and geological parameters obtained through site investigations and laboratory tests are introduced to an ANN model. This data is used to estimate the unknown non-linear relationship between the rock parameters and convergence. Tunnel convergence model is developed, calibrated and tested using the above data from the perspective of Ghomroud water conveyance tunnel in Iran. The dominating rock masses in this case are metamorphic and sedimentary and are considered to be of weak to fair quality. In this tunnel there were some problems due to the convergence and instability of the tunnel. The tunnel boring machine has had several stoppages including a few major delays related to being trapped in squeezing ground and also delays due to face collapses. In order to predict the tunnel convergence a Multi-Layer Perceptron (MLP) analysis is used. A four-layer feed-forward back propagation neural network with topology 9-35-28-1 was found to be optimum. Simultaneously, the methods Radial Basis Function (RBF) analysis as another approach of ANN and Multi-Variable Regression (MVR) as a linear regression using statistical approach are used to analyze the problem and the results are compared. As a result, the MLP proposed model predicted values closer to the measured ones with an acceptable range of correlation. After the calibration and assessment of the ANN model, a parametric study is also carried out to estimate the intensity of the impact of the geological and rock mechanics parameters on tunnel convergence. It is observed that C, Φ, E and UCS parameters are the most effective factors and σt is the least effective one. Concluding remark is the proposed model appears to be a suitable tool for the prediction of convergence in the unexcavated zones of the tunnel as well as in new tunnels to be excavated in the similar geological environment. The results show that an appropriately trained neural network can reliably predict the convergence in tunnels.

Highlights

► Ghomroud tunnel located about 400 km South-West Tehran is the target of our study. ► In this tunnel TBM stoppages due to trapping in bad conditions have occurred. ► Convergence data was processed by intelligence based and statistical approaches. ► Methods include: Two ANN based, MLP and RBF and a statistics based, MVR methods. ► Convergence prediction, using geotechnical and geological data is the outcome.

Introduction

Excavation of tunnels for variety of applications is dominating worldwide. Convergence of the tunnels is one of the issues that affects the performance of the tunnel both during the excavation and afterwards. High rate of convergence can result in variety of problems. One of them, encountered during mechanized excavations, could be the trapping and jamming of the boring machine due to the convergence, which results in reduction of safety and the advance rate and consequently increase of the operation cost.

Creating underground spaces, produces stresses tending to close the cavity particularly at the vicinity of the excavation front (Kontogianni et al., 2008). In addition, the stress status of rock mass around a tunnel is strongly affected by the excavation-support process of the tunnel construction (Pan and Dong, 1991). Consequently, tunnel excavation produces not only stress changes to the ground, but also strain to the support lining, leading to the extension of the convergence after support installation. Kontogianni et al. (2006) reported that convergence recorded after the excavation, is assigned to firstly, strain resulting from the progressive tunnel front advance (face advance effect) and secondly, the time-dependent behavior of the material (creep effect).

Tunnel convergence monitoring, including deformation measurements and convergence measurements will provide data that can be used to determine whether the tunnel is in a stable or unstable structural situation. In this respect, instrumentation and monitoring the behavior of the excavation during and after construction is the best way to have a realistic judgment of the stability of the structure (Moosavi and Khazaei, 2003). The prediction of the magnitude and development of the expected convergence have a significant impact on a number of issues during tunnel excavation. While excavation and support methods are influenced by the magnitude of convergence, the over-excavation has to be adjusted to the expected convergence also to guarantee the clearance after stabilization (Grossauer et al., 2005).

Fig. 1 illustrates four typical curves of tunnel convergence. As shown in this figure, plot 1 represents a stable situation. Plot 2 depicts a situation where the tunnel after incurring a period of convergence, goes toward stability. Convergence rate in plot 3 is constant over time. Finally, plot 4 demonstrates the increase of convergence rate over time.

Various methods for prediction in engineering applications have been suggested by others. One of the methods is the application of artificial intelligence based method (Masulli and Studer, 1999). In this research, an Artificial Neural Network (ANN) model is planned to be used to identify dependencies between the convergence and the geotechnical and geological conditions encountered.

ANN has emerged as a powerful tool for analyzing mining and geotechnical problems (Kapageridis, 2002, Bizjak and Petkovsek, 2004, Khandelwal et al., 2004). They are capable of learning non-linear functional mapping and are suitable for adapting to complex functions (Chan et al., 2006). Several researchers have used ANN in a number of tunnelling (Santos and Celestino, 2008) and geotechincs related problems (Kapageridis, 2002) effectively and successfully.

In this prediction task, the basic problem to be solved is the approximation of an unknown function from the observation of a number of known input–output data sets to predict convergence during and after tunnel excavation. In order to predict the tunnel convergence a Multi-Layer Perceptron (MLP) analysis is used. Simultaneously, the methods Radial Basis Function (RBF) and Multi-Variable Regression (MVR) are used to analyze the problem and the results are compared. The model considers a wide range of data including geotechnical parameters, geological information and monitoring data at different sections of the tunnel. The proposed model appears to be a suitable tool for the prediction of convergence in the unexcavated zones of the tunnel as well as in new tunnels to be excavated in the similar geological environment. In practice, prior to the excavation, as much related data as practicable should be collected to have a realistic view of the zone to be excavated. This data can be used to predict the convergence as well using the suggested model.

Section snippets

Problem description and the history

The most important problem associated with using a shielded Tunnel Boring Machine (TBM) in weak rocks is the probable ground squeezing and as a result, TBM jamming potential (Farrokh et al., 2006). In this situation, it is expected that the amount of imposed friction force due to ground pressures on the shields will be higher than the TBM thrust force. Hence, the ground pressure evaluation along the tunnel path is recognized to be necessary. Tunnel convergence magnitude, on the other hand,

Case description

Ghomroud tunnel as part of a water conveyance system is mainly located in the Lorestan province, Iran, between the cities Aligoudarz and Golpayegan as shown in Fig. 2. The system is expected to convey some 120 × 106 m3 water per year (23 m3/s) from Dez river branches to the Golpayegan dam (Mahab Ghodss Consulting Engineers, 2003). The system includes five deviation dams, two long tunnels about 36 and 9 km long and a water conveyance channel. The excavation of the tunnel under question (36 km long)

Artificial Neural Networks

An ANN model is a mathematical or computational model that is inspired by the structure and/or functional aspects of biological neural networks and is in fact an emulation of biological neural system.

Neural network analysis can be used to handle non-linear problems that are not well suited to be touched by classical analysis methods. An ANN is a computing system consisting of a highly interconnected set of simple information processing elements called units or nodes or neurons. The arrangement

Tunnel convergence monitoring system

Convergence is controlled by the time-dependent rock mass behavior and the excavation and support history. It plays an important role in the modern tunnelling methods (Sulem et al., 1987). In these tunnels, usually a few hours after a cyclic excavation, convergence control points are established and measurements are taken, usually with total station instruments and convergence meters. Such measurements should be repeated frequently, usually based on a daily program until the points are fully

Regression analysis

In a regression analysis the goal is to build a model that relates the independent variables to dependent variables with minimum possible error (Liitiainen et al., 2009). In statistics, Multi-Variable Regression (MVR) is an approach to model the relationship between a dependent variable and a set of independent variables. In MVR, data are modeled using linear functions and unknown model parameters are estimated from the data. In our research, a relationship between the dependent variable of

Artificial Neural Network analysis

Since the linear regression using statistical approach, as mentioned above, yields a weak correlation coefficient, a second approach using ANN technique is considered to estimate the potential non-linear relationship between the parameters. A non-linear regression, in principle, is to search for the optimized value of a non-linear object function.

Results and discussion

After building several MLP models based on trial and error, the best result of each model, listed in Table 3, are compared and the one with maximum of R2 and minimum of MSEREG is chosen. The best R2 and MSEREG are obtained 0.936 and 0.084 respectively, which represents a reasonable accuracy.

The RBF model was used by some researchers in many geotechnical fields, for example advance rates in tunnelling construction was predicted using RBF by Lau et al. (2010). In our research, for comparison of

Parametric study

After the calibration and assessment of the ANN model, a parametric study is conducted to estimate the impact of the geological and rock mechanical parameters on tunnel convergence goal. In this procedure the relative influence of each parameter on the tunnel convergence is identified. Parametric study is performed varying the desired parameter and maintaining fixed values for the other parameters. The parameter to be analyzed varied between its minimum and maximum values of the data set. The

Application of the proposed model

Convergence monitoring plays an important role in the observational or “design-as-you-go” approach for underground spaces. In particular, it can be considered as a primary field measurement, because it is not only a readily recordable indicator of the overall ground response, but its magnitude constitutes a very useful parameter for the evaluation of tunnel stability (Indraratna and Kaiser, 1990, Sulem et al., 1987).

In order to evaluate the performance of the proposed model, a section of tunnel

Conclusions

According to the results obtained from this research, it can be said that ANN is a useful means to predict a tunnel convergence. For this study, a MLP neural network with nine inputs, was developed and used for prediction of the tunnel convergence. The optimum ANN architecture has been found to have nine neurons in the input layer, two hidden layers with 35 and 28 neurons, respectively, and one neuron in the output layer.

In this research three methods namely, MVR, RBF and MLP are used and

Acknowledgements

The authors are greatly thankful to Dr. M. Saniee for consultation, to Mrs. N. Ramezankhani for editing and reviewing the manuscript and to Mr. M. Mahdevar and Mr. H. Khosravi for cooperation in site investigations. Also Sahel and Mahab Ghodss Consulting Engineers are highly appreciated for supplying data and pertaining information.

References (48)

  • C.Y. Kim et al.

    Neural network based prediction of ground surface settlements due to tunnelling

    Comput. Geotech.

    (2001)
  • V. Kontogianni et al.

    Predictions and observations of convergence in shallow tunnels: case histories in Greece

    Eng. Geol.

    (2002)
  • V. Kontogianni et al.

    What is the contribution of time-dependent deformation in tunnel convergence?

    Eng. Geol.

    (2006)
  • V. Kontogianni et al.

    Delayed failure at the Messochora tunnel, Greece

    Tunnel. Underg. Space Technol.

    (2008)
  • S.C. Lau et al.

    Applying radial basis function neural networks to estimate next-cycle production rates in tunnelling construction

    Tunnel. Underg. Space Technol.

    (2010)
  • S.S. Leu et al.

    Data mining for tunnel support stability: neural network approach

    Automat. Constr.

    (2001)
  • E. Liitiainen et al.

    Residual variance estimation in machine learning

    Neurocomputing

    (2009)
  • Y.W. Pan et al.

    Time-dependent Tunnel Convergence II. Advance Rate and Tunnel-Support Interaction

    Int. J. Rock Mech. Min. Sci. Geomech. Abstr.

    (1991)
  • O.J. Santos et al.

    Artificial neural networks analysis of Sao Paulo subway tunnel settlement data

    Tunnel. Underg. Space Technol.

    (2008)
  • J. Sulem et al.

    Closure analysis in deep tunnels

    Int. J. Rock Mech. Min. Sci. Geomech. Abstr.

    (1987)
  • J. Sun et al.

    Rock mechanics and rock engineering in China: developments and current state-of-the-art

    Int. J. Rock Mech. Min. Sci.

    (2000)
  • T. Unlu et al.

    Effect of Poisson’s ratio on the normalized radial displacements occurring around the face of a circular tunnel

    Tunnel. Underg. Space Technol.

    (2003)
  • T.T. Wang et al.

    An experience of tunnelling in mudstone area in southwestern Taiwan

    Tunnel. Underg. Space Technol.

    (2002)
  • G.V. Cybenko

    Approximation by superpositions of a sigmoidal function

    Math. Control Signal System (MCSS)

    (1989)
  • Cited by (128)

    View all citing articles on Scopus
    View full text