Jump to Main Content
Calibration of DEM models for irregular particles based on experimental design method and bulk experiments
- Zhou, Haili, Hu, Zhanqi, Chen, Jigang, Lv, Xuan, Xie, Nan
- Powder technology 2018 v.332 pp. 210-223
- Poisson distribution, algorithms, angle of repose, bulk density, calibration, computer software, experimental design, graphene, models, powders
- Discrete element method (DEM) is widely used in particle modeling and related applications. DEM modeling requires a large number of micro particle parameters, which are difficult to measure for irregular particles, thus the parameter calibration method is introduced. This paper presents a new DEM parameter calibration method: taking the angle of repose and bulk density of expanded graphite (EG) particles as macro responses, the particle density, sliding frictions, coefficients of restitution and Poisson's ratio as micro variables, the micro particle parameters are calibrated with adaptive simulated annealing (ASA) optimization algorithm by establishing radial basis function neural network (RBFNN) approximate models between the micro variables and the macro responses based on optimal Latin hypercube sampling and virtual simulation experiments implemented in the commercial software EDEM. A baffle lift method is used to measure the angle of repose, and the fuzzy membership function is used to fit the measurements of the responses considering the dispersion of the experimental values of irregular particles. In addition, based on the membership degree, experimental truth values are defined, optimization conditions are determined and calibration results are evaluated. Finally, the dynamic angle of repose of the EG particles is obtained by a rotary cylinder experiment to further verify the calibrated parameters. The results show that: a) incorporating the measurement errors in the calibration process can help to improve the calibration efficiency without compromising the calibration accuracy; b) with high precision, the approximate model established by RBFNN can accurately predict the macro response values as an effective alternative to DEM model; c) the calibration results are a set of feasible solutions with each calibrated parameter distributed over a small range; d) the membership degree of the validated DEM results to the experimental truth values is high and the simulation result of the dynamic angle of repose is in good agreement with the experimental data, indicating that the calibrated micro DEM parameters can truly reproduce the particle stacking behavior, which proves the effectiveness and reliability of the proposed method.