Numerical and artificial neural network analyses of ground surface settlement of tunnel in saturated soil

Abstract

The main purpose of this research is to predict the ground surface settlement in tunneling of a single circular tunnel with simultaneous changes in the mechanical properties of soil and geometrical properties of the tunnel section. In this research, numerical and parametric analysis of circular tunneling in frictional–cohesive saturated soil has been investigated using 2D finite element method by ABAQUS. In other words, the behavior of ground surface, considering to change the different values of depth-to-diameter ratio (H/D), soil cohesion, internal friction angle, permeability coefficient, and the influence of these variables on settlement of surface in each model, has been separately evaluated. Then, a multilayer perceptron (MLP) artificial neural network is designed to predict the ground surface settlement. MLP is a type of feedforward artificial neural network utilizing backpropagation technique for training phase, and the Levenberg–Marquardt method is used to reduce the errors and the distance between the network outputs and finite element method results. There are some independent variables in the input layer and a dependent variable in the output layer. The middle layer consists of seven neurons. Finally, the high potential of the artificial neural network with a correlation coefficient of 0.98 is shown in the prediction of ground surface settlement.