Machine learning-based identification of segment joint failure in underground tunnels.

Abstract

Shield tunnels that reside deep within soft soil are subject to longitudinal differential settlement and structural deformation during long-term operation. Longitudinal deformation can be classified into two modes: bending and dislocation deformation. The failure of bolts and engineering treatment techniques differ between these two modes. Therefore, it is imperative to accurately identify the tunnel's longitudinal deformation mode to determine the validity of the segment joint and implement appropriate engineering treatment. Traditional methods for detecting dislocation or opening suffer from high labour costs. To address this issue, this study presents an innovative identification method using a back-propagation neural network (BPNN) to detect segment joint failure in underground tunnels. First, this study collects the tunnel settlement curves of various subways located in the East China soft soil area, and it calculates tunnel settlement-dislocation and settlement-opening datasets using the equivalent axial stiffness model. A corresponding BPNN regression model is subsequently established, and the new settlement curve is the input to this regression model to predict the dislocation and opening, thereby determining the validity of the segment joint. The efficiency of this method is demonstrated through its successful application to the Hangzhou Metro Tunnel. This article is part of the theme issue 'Artificial intelligence in failure analysis of transportation infrastructure and materials'.