Differentiable graph neural network simulator for forward and inverse modeling of multi-layered slope system with multiple material properties

Graph neural network simulators (GNS) have emerged as a computationally efficient tool for simulating granular flows. Previous efforts have been limited to simplified homogeneous geometries characterized only by the friction angle, which does not reflect the complexity of realistic slopes encountered in engineering practice. This study introduces a differentiable GNS framework designed for multi-layered slope systems comprising both forward and inverse modeling components. The forward component relies on a fine-tuned GNS that incorporates both friction angle and cohesion. Its performance is demonstrated through column collapse and multi-layered slope runout simulations, where the GNS replicates multi-material flow dynamics while achieving significant computational speedup over the Material Point Method (MPM). The inverse modeling component leverages the trained GNS, reverse-mode automatic differentiation, and L-BFGS-B optimization to infer material properties from a target runout geometry. Its performance is demonstrated by back-calculating the material strengths that led to failure-induced runout in a dam system composed of multiple materials. Results are obtained within minutes and show good agreement with the target strength values. The framework introduced in this study provides an efficient approach for forward runout assessments and inverse strength back-calculation in realistic slope systems.