Machine learning approaches for stability prediction of rectangular tunnels in natural clays based on MLP and RBF neural networks

Abstract

In underground space technology, the issue of tunnel stability is a fundamental concern that significantly causes catastrophe. Owing to sedimentation and deposition processes, the strengths of clays are anisotropic, where the magnitudes of undrained shear strengths in the vertical and horizontal directions are different. The anisotropic undrained shear (AUS) model is effective at considering the anisotropy of clayey soils when analyzing geotechnical stability issues. This study aims to assess the stability of rectangular tunnels by adjusting the dimensionless overburden factor, cover-depth ratio, and width-depth ratio in clay with various anisotropic strength ratios. The stability analysis of these tunnels involves employing finite element limit analysis and the AUS model to identify the planes of soil collapse in response to the aforementioned variations. In addition, this study presents the development of soft-computing models utilizing artificial neural networks (ANNs) to forecast the stability of rectangular tunnels across various combinations of input parameters. The findings of this study are presented in the form of design charts, tables, and soft-computing models to facilitate practical applications.

Keywords

• Anisotropic clay
• Rectangular tunnel
• Finite element limit analysis
• Limit analysis
• Artificial neural networks