Prediction of tunneling-induced ground movement with the multi-layer perceptron

https://doi.org/10.1016/j.tust.2005.07.001Get rights and content

Abstract

This paper presents a method to predict ground movement around tunnels with artificial neural networks. Surface settlement above a tunnel and horizontal ground movement due to a tunnel construction are predicted with the help of input variables that have direct physical significance. A MATLAB based multi-layer backpropagation neural network model is developed, trained and tested with parameters obtained from the detailed investigation of different tunnel projects published in literature. The settlement is taken as a function of tunnel diameter, depth to the tunnel axis, normalized volume loss, soil strength, groundwater characteristics and construction methods. The output variables are settlement and trough width. Parameters for the prediction of horizontal ground movement include diameter to depth ratio (D/Z), unit weight of soil and cohesion. The neural network demonstrated a promising result and predicted the desired goal fairly successfully.

Introduction

In urban areas, it is essential to protect existing adjacent structures and underground facilities from damage due to tunneling. In order to minimize the risk, a tunnel engineer needs to be able to make reliable prediction of ground deformations induced by tunneling. Numerous investigations have been conducted in recent years to predict the settlement associated with tunneling; the selection of appropriate method depends on the complexity of the problems. Many environmental factors associated with tunneling have recently led to a considerable research effort being devoted to the study of settlements caused by tunneling through soft ground.

Progress has been made in recent years in the ability to predict ground movement, but the state-of-the-art is deficient in many ways. On the basis of detailed investigation, a viable approach for the prediction of the ground movement is necessary, and an artificial neural network (ANN) comes in handy to fulfill this approach.

In this paper, measurements of settlement and ground movements recorded in different tunnel projects have been reviewed and analyzed. The data from these case studies were used to train and test the developed neural network model to enable prediction of the magnitude of settlements and ground movements with the help of input variables that have direct physical significance.

Section snippets

Artificial neural network

Artificial neural networks (ANN) are networks of highly interconnected neural computing elements that have the ability to respond to input stimuli and to learn to adapt to the environment. ANN include two working phases, the phase of learning and that of recall. During the learning phase, known data sets are commonly used as a training signal in input and output layers. The recall phase is performed by one pass using the weight obtained in the learning phase. ANN is now a well-established tool

Theoretical background

Construction of a tunnel in soft ground brings about a change in the state of stress and corresponding strains and displacement around the tunnel opening. Therefore, some degree of soil deformation may always be induced. To make reliable forecast of the inevitable ground movements and settlements associated with every design or construction procedures, proper consideration as to type of soil, groundwater conditions geometry and depth of tunnel, etc. is necessary (Thongyot, 1995).

Parameters for surface settlement

Prediction of ground settlement above tunnels has long been a subject of research. A number of comprehensive works have been done based on empirical relationships; some of these relationships are presented in Table 1. Based on these existing models, a number of input parameters are identified namely D, Z, Vs, soil type (soil strength). Effect of groundwater level and construction method on soil behavior has also been incorporated. Although the hydraulic conductivity of clay is small, the

Prediction of maximum settlement

In order to define the relative strength of parameters used in the prediction of maximum tunnel settlement, a comprehensive parametric studies were conducted. At first three parameters namely tunnel depth (Z), diameter (D) and cohesion (Cu) were considered. The four layered neural network with three input parameters, five neurons in hidden layer 1, nine neurons in hidden layer 2 and single output neuron was trained and tested. The network trained with a learning rate of 0.5 conversed in 20,000

Conclusion

To minimize overall project costs and the risk of damage or accident as a result of ground movements due to tunneling, the engineer who designs the project needs to be able to make reliable predictions of the extent and amount of settlement that are likely to arise in various conditions. To fulfill the strategy, a BPNN was developed and used to predict the ground behavior around soft ground tunnels. The neural network demonstrated a promising result and predicted the settlement, width of

References (29)

  • J. Ghaboussi et al.

    New nested adaptive neural networks (nann) for constitutive modeling

    Computers and Geotechnics

    (1998)
  • I.M. Lee et al.

    Prediction of pile bearing capacity using artificial neural networks

    Computers and Geotechnics

    (1996)
  • W. Yoshikoshi et al.

    Prediction of ground settlements asscociated with shield tunneling

    Soils and Foundations

    (1978)
  • Agrawal, G., Frost, J.D., Chameau, C.L.A., 1994. Data analysis and modeling using artificial neural network. In:...
  • J.H. Atkinson et al.

    Subsidence above shallow tunnels in soft ground

    ASCE Journal of Geotechnical and Geoenvironmental Engineering

    (1979)
  • P.B. Attewell et al.

    Ground deformations resulting from shield tunneling in London Clay

    Canadian Geotechnical Journal

    (1974)
  • E.B. Baum et al.

    What size net gives valid generalization?

    (1994)
  • W. Clough et al.

    Design and performance of excavations and tunnels in softclay

    Soft Clay Engineering

    (1981)
  • B.M. Das

    Principles of Foundation Engineering

    (1999)
  • F. Fausett

    Fundamentals of Neural Networks: Architectures, Algorithms and Applications

    (1994)
  • A.C. Goh et al.

    Estimation of lateral wall movements in braced excavations using neural networks

    Canadian Geotechnical Journal

    (1995)
  • S. Hayakin

    Neural Networks: A Comprehensive Foundation

    (1998)
  • C.H. Juang et al.

    Cpt-based liquefaction evaluation using artificial neural networks

    Journal of Computer-aided Civil and Infrastructure Engineering

    (1999)
  • C.H. Juang et al.

    Cpt-based liquefaction analysis. Part i: Determination of limit state function

    Geotechnique

    (2000)
  • Cited by (137)

    • Application of machine learning, deep learning and optimization algorithms in geoengineering and geoscience: Comprehensive review and future challenge

      2022, Gondwana Research
      Citation Excerpt :

      To prevent or mitigate the disasters and ensure the geotechnical structures with sufficient safety (Ou et al., 2022), accurate predictions for occurrence of geohazards and reliability analysis for geotechnical structures are urgently needed. In recent years, with rapid development of AI, ML, DL, and OA techniques, these methods are gradually widely adopted to conduct the assessment for the occurrence probability of geohazards and the destroy or deformation of geotechnical structures with sufficient accuracy and efficiency (Darabi et al., 2012; Li et al., 2014; Liang et al., 2012; Luo et al., 2019; Mohammadi et al., 2015; Neaupane and Adhikari, 2006; Shi et al., 2019; Solomatine and Xue, 2004; Zhan et al., 2021). Although there have been some high-level overviews of ML for the prediction of some specific areas in geotechnical engineering such as slopes, tunnels, deep braced excavations (Sheil et al., 2020; Shreyas and Dey, 2019; Zhang et al., 2020a), yet a more comprehensive review of recent research progress and applications of ML, DL and OA in geotechnical engineering and geoscience is quite limited.

    • Auto machine learning-based modelling and prediction of excavation-induced tunnel displacement

      2022, Journal of Rock Mechanics and Geotechnical Engineering
      Citation Excerpt :

      As an emerging method, machine learning (ML) has been widely concerned in geotechnical engineering, due to the characteristic of high efficiency, first-class generalisation performance, and high-dimensional problem solving ability. Various kinds of ML algorithms have been utilised in specific fields, such as slope and landslides (Liu et al., 2014, 2020; Xu and Niu, 2018), pile settlement (Azizkandi et al., 2014; Armaghani et al., 2018, 2020), characterisation of soil properties (Kurnaz and Kaya, 2018; Ching and Phoon, 2019; Cheng et al., 2020b; Zhang et al., 2020b, 2021a), retaining wall deflection brought about by deep braced excavation (Kung et al., 2007; Ji et al., 2014; Goudjil and Arabet, 2021; Zhang et al., 2021b), ground surface settlement induced by tunnelling (Neaupane and Adhikari, 2006; Pourtaghi and Lotfollahi-Yaghin, 2012; Kohestani et al., 2017; Moeinossadat et al., 2018; Zhang et al., 2019a), and subsurface stratification from limited boreholes (Shi and Wang, 2021a, b). To be specific, Zhang et al. (2020b) developed a non-parametric ensemble artificial intelligence approach to calculate the compression modulus Es of soft clay based on a gradient-boosted regression tree algorithm.

    View all citing articles on Scopus
    1

    Former Graduate Student, SIIT, Thammasat University.

    View full text