Proceedings of the
9th International Symposium for Geotechnical Safety and Risk (ISGSR)
25 – 28 August 2025, Oslo, Norway
Editors: Zhongqiang Liu, Jian Dai and Kate Robinson
3D-CNN-Based Surrogate Modeling and Data Augmentation for 3D Slope Reliability in Spatially Variable Soils
1School of Civil Engineering, Chongqing University, Chongqing 400045, P. R. China.
2Marie SkBodowska-Curie Fellow, Department of Engineering, University of Cambridge, Cambridge CB3 0FA, United Kingdom.
3Department of Civil & Environmental Engineering, National University of Singapore, Singapore.
ABSTRACT
While the significance of spatially variable soil properties in slope stability assessment has been well understood, the implementation of three-dimensional (3D) probabilistic slope reliability assessment is still bottlenecked by its excessive computational time. This paper presents a novel surrogate model based on Convolutional Neural Networks (3D CNN) to replace the computationally demanding 3D random field finite element method (RFEM) for Monte-Carlo simulations. To enhance model performance with limited training data, a data augmentation technique has been developed based on the shear strength reduction (SSR) method. The methodology's effectiveness is illustrated using a 3D slope case with undrained shear strength as the random variable.
Keywords: 3D slope reliability assessment, Spatial variability, 3D random finite element method, Surrogate model, 3D Convolutional neural network, Data augmentation.
