Mohr-Coulomb strength and FDEM parameter determination of weathered granite via optimized neural network and deep learning

Abstract

The traditional finite discrete element method (FDEM) is less applied in practical engineering-scale modeling due to its complex input parameter calibration process, poor prediction ability of calibration techniques, and insufficient description of material plastic damage. This study proposed a novel FDEM model enriched with the Mohr-Coulomb (MC) model and applied the neural network method to standardize the input parameters. The method enables rapid calibration of input parameters for various rock materials and accurately predicts their plastic development and failure modes, thereby enhancing the adaptability of FDEM in complex engineering scenarios. First, the Newton-Raphson-Based Optimizer (NRBO)-Back Propagation Neural Network (BPNN) method is employed to establish the correspondence between input parameters and Unconfined Compressive Strength (UCS) and Brazilian Tensile Strength (BTS) results. Subsequently, the Non-dominated Sorting Genetic Algorithm II (NSGA-II) is used to manage and assign numerical parameters to match the target rock strength. Notably, by introducing failure mode parameters, the method refines the FDEM's description of different failure modes in weathered granites. Finally, the consistency between experimental and numerical results demonstrates the effectiveness of the proposed approach. This work successfully addresses the rapid calibration of FDEM input parameters using machine learning and overcomes the limitations of traditional models in describing the plastic development of rock materials.