Slope Reliability Analysis Based on Deep Learning of Digital Images of Conditional Random Fields Using CNN

Considering the site investigation data and the spatial variability of soil strength, a deep learning model of conditional random field (CRF) characteristics is proposed for reliability analysis of slope stability. The random fields of soil slope are discretized by Karhunen-Loeve Expansion (KLE) method and constrained by the Cone Penetration Tests (CPTs) using the kriging interpolation method. The discretization results are converted into digital images. Then, a Convolutional Neural Network (CNN) surrogate model is established to approach the implicit relationship between the images and responses of the performance function. Based on the surrogate model, the reliability index of the slope is calculated. Finally, the effectiveness of proposed method is demonstrated by a single layer saturated clay slope under undrained condition. The result shows that the proposed method can reduce the spatial uncertainty, and significantly reduce the computational cost of slope reliability analysis.

Keywords: Spatial variability, Slope reliability analysis, Convolutional Neural Networks, Digital image, Surrogate model, Conditional random field.