Mesoscopic Parameter Calibration of Granite Particles Based on Neural Network and Discrete Element Triaxial Numerical Simulation Test

At present, discrete element numerical analysis has been widely used in geotechnical engineering. In order to obtain ideal numerical analysis results, this paper takes granite particles as an example. By simulating conventional triaxial compression experiment in PFC3D, the rolling resistance linear contact model is selected. The corresponding relationship between mesoscopic parameters (normal stiffness, stiffness ratio, friction coefficient, rolling friction coefficient) and macroscopic parameters (initial elastic modulus, peak intensity, Poisson's ratio, maximum compression and dilatancy angle) is explored. The calibration of mesoscopic parameters is realized based on BP neural network inversion. The accuracy of inversion results is verified by comparison test and numerical simulation. It provides some reference value for discrete element numerical analysis of granite particle parameter selection.