The focus of the study is to examine the undrained behavior of twin circular tunnels in anisotropic and nonhomogeneous clays. To consider the effect of anisotropic soil, the popular anisotropic undrained shear (AUS) failure criteria are adopted in the study while the nonhomogeneous behavior is represented by linearly increasing strength with depth. Using Broms and Bennermarks' stability number, this study investigates the dependence of the undrained stability number N on four dimensionless input parameters, namely the isotropic ratio (re), the undrained shear strength gradient (rho D/suTC0), the cover depth ratio (C/D), and the spacing ratio (S/D). The effects of these four design parameters on the failure mechanism are also examined graphically. After being verified with previously published works, the comprehensive 1080 numerical results are then utilized as the dataset to create several machine learning models, including artificial neural network (ANN), support vector machine (SVM), and multivariate adaptive regression splines (MARS). The evaluating process by optimizing hyper-parameters reveals that the MARS model is a top competitor, providing considerable regression accuracy with a simple predictive function. The sensitivity analysis has also uncovered that both rho D/suTC0 and C/D have significant influences on the undrained stability number N, while comparing to re and S/D. The present study would provide many practical insights to the problem of twin circular tunnels in anisotropic and nonhomogeneous clays.

来源平台：ARABIAN JOURNAL FOR SCIENCE AND ENGINEERING