Intelligent construction method for rock slope fracture network model based on discrete element and neural network

To address the engineering challenge of accurately characterizing complex internal fracture networks in high-steep rock slopes, this study proposes an intelligent Discrete Fracture Network (DFN) construction method integrating field adit data and intelligent algorithms. Taking the Dadu River Chengdu Bank rock slope as a case study, spatial distribution characteristics of adit fractures were obtained through engineering geological surveys. A 3D DFN generation model framework was established using Monte Carlo stochastic principles. Multiple numerical simulation experiments were designed using 3DEC software to generate a data sample library encompassing geometric parameters such as fracture orientation, trace length, and spatial position. A mapping relationship between adit fracture characteristic parameters and DFN model parameters was constructed using Back Propagation (BP) and Cascade Correlation (CC) neural networks. Through model structure optimization and hyperparameter training, optimal prediction models were obtained. Results indicate that the exploration adit fracture image format features a resolution of 600 × 9000 pixels, green color, and 70% transparency for fracture traces. Under a stratified modeling strategy, prediction errors for different lithological units were 18.4% for the four-layer BP network in diorite regions, 8.0% for the three-layer BP network in cataclastic rock regions, and 10.3% for the CC network in granite regions, validating the applicability of the intelligent algorithms. This method achieves high-precision inversion of fracture networks using limited adit data, providing a novel approach for modeling complex rock mass structures. The findings have been applied to the Dadu River Grand Bridge slope engineering project.