Performance of artificial neural network and convolutional neural network on slope failure prediction using data from the random finite element method

The random finite element method has been widely used to evaluate slope uncertainty and reliability. To determine the probability of failure, the safety factor sampling often involves the Monte Carlo method. This process is a time-consuming task that requires many computational resources. In this study, we propose a pretrained model to directly estimate the safety factor and the trace of the slope slip surface by using machine learning techniques. Each safety factor associated with its random field conditions can be quickly predicted to evaluate the probability of failure. Furthermore, the potential slip surfaces can be examined through the predicted strain contour. Herein, artificial neural networks and convolutional neural networks are both implemented in the factor of safety and slip surface predictions. The results show that the convolutional neural network model is more accurate than the artificial neural network model when the scenario is more complicated (i.e., considering the additional random field of soil properties). Additionally, the convolutional neural network model considers the spatial relationships of input data, which is an appropriate method to address random field problems. This method effectively shortens the time compared with the traditional random finite element method.

Automated summary

In this study, a pretrained model using machine learning techniques is proposed to directly estimate the safety factor and slip surface trace, providing a quick evaluation of the probability of failure. The convolutional neural network model outperforms the artificial neural network model, especially when considering additional random fields. This method effectively reduces the computation time compared to the traditional random finite element method.