Artificial neural networks-based surrogate model for predicting the slope stability and large deformation failure consequences

Predicting slope failure consequences typically requires large-deformation numerical simulations, which are computationally intensive and impractical for routine risk assessments. In this study, we develop a surrogate modeling framework based on two coupled Artificial Neural Networks (ANNs) to estimate slope stability and rapidly predict the consequences of large-deformation slope failure. The second ANN, which is used to evaluate large-deformation failure consequences, is activated only when the slope is classified as unstable by the first ANN stability model. The ANN model is trained on datasets constructed from slope stability/instability charts, composed of groups of dimensionless parameters obtained from slope analyses conducted using the Material Point Method (MPM). For given soil strength properties and slope geometry, the trained ANN can quickly predict the factor of safety ( F s ) and key slope failure indicators, including run-out distance, sliding depth, influence distance, and sliding volume. The developed ANN model is first validated against a two-dimensional homogeneous slope case, followed by additional numerical cases to assess its robustness and engineering applicability. The model’s performance is quantified using the coefficient of determination ( R 2 ) and root mean square error ( RMSE ). A very good predictive accuracy, with R 2 values exceeding 0.99 and RMSE values below 0.15, is achieved on the testing dataset. The ANN model reliably reproduces both slope stability and large-deformation failure characteristics through the case studies. These findings highlight the ANN model as a rapid and reliable tool for quantitative slope risk assessment. Furthermore, an intuitive graphical user interface is developed to facilitate the practical deployment of the trained model, ensuring its accessibility and usability in real-world engineering applications.