A calibration method for ore bonded particle model based on deep learning neural network

Abstract

In the simulation of ore particle crushing, the most important step is to calibrate the ore parameters, which directly affects the accuracy of subsequent simulations. In order to use the discrete element method (DEM) to simulate the crushing process of ore in the crusher, an accurate simulation based on the bonded particle model of large ore particles is required. The application of EDEM is very mature in discrete element method, and it is well established in various contact theories. In this paper, through the uniaxial compression test, the EDEM is used to simulate the normal stiffness per unit area, shear stiffness per unit area, critical normal stress, critical shear stress, particle density, shear modulus, and the bonded disk radius, in order to explore how these parameters influence the compressive strength and average gradient of stress-strain of the particle model. On this basis, a calibration method for the parameters of bonded particle model based on the deep learning neural network structure is proposed. The results show that the compressive strength and the average stress-strain of gradient of the bonded particle model are determined by the combined effect of the normal stiffness per unit area, shear stiffness per unit area, critical normal stress and the critical shear stress. While the above four parameters remain unchanged, the bonded disk radius of the filling particles plays a dominant role. In addition, this paper verifies the reliability and effectiveness of calibration methods by setting up 8 sets of experiments outside the test set and comparing them with existing calibration methods. The advantage of this method is that reasonable calibration parameters can be given quickly and accurately for the bonded particle model with different compressive strength and shear modulus, and the error of this method is smaller than the existing calibration methods.