An efficient reliability assessment of coupled FDM-DEM slope stability using physics-informed feature-based surrogate machine learning techniques
Year: 2026
Type: article
Abstract: Reliable slope stability evaluations require variability for the soil parameters involved in hydromechanical processes that lead to failure. The traditional methods, which are based on either continuum or discrete-element methods, often fail to capture complete hydromechanical interactions. With this motive, a coupled FLAC3D-PFC3D framework that combines continuum-scale stress changes with discrete micromechanical behaviors was developed in this work. However, the high computational costs of these coupled models for many random sample simulations limit their use in large-scale reliability studies. To solve this problem, a surrogate-assisted reliability framework was developed. First, 500 coupled simulations were performed in FLAC3D with varying shear strength parameters to establish a baseline dataset for the factor of safety (FoS). These results were subsequently used to train surrogate models, i.e., Gaussian process regression (GPR), support vector regression (SVR), a broad learning system (BLS), and a stacking ensemble. At this stage, physics-informed features derived from Mohr‒Coulomb shear strength theory were incorporated to enhance the generalization and reliability of the model. Once trained, the surrogates were applied to 9500 midpoint Latin hypercube (LHS) samples, yielding 10,000 FoS evaluations for uncertainty and reliability analysis purposes. The surrogates reproduced the reference FoS distributions accurately and at low computational cost, while providing consistent reliability estimates, with a probability of failure P f = 4.6–5.1% and a reliability index β = 1.64–1.69. A binary stability classification (FoS ≤ 1.0) further confirmed the predictive robustness of the model (ROC-AUC > 0.96). These results demonstrate that integrating coupled hydromechanical modeling with physics-informed surrogate learning enables efficient and reliable probabilistic slope stability analyses, supporting risk-informed geotechnical design work.
Source: Computers and Geotechnics
Authors: Hilal Ahmad, Zhang Yinghua, Mehtab Alam, Majid Khan, Naqvi Syed Junaid Haider
Language: English
Subfield: Safety, Risk, Reliability and Quality
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $3,640
OpenAlex ID: W7155806719