A calibration framework of DEM variables using genetic algorithms

Huy Q. Do, Alejandro M. Aragón, and Dingena L. Schott,
Faculty of Mechanical, Maritime and Materials Engineering, Delft University of Technology

The discrete element method (DEM) is a widely accepted method for modeling the behavior of granular materials. However, a major barrier to the effective use of DEM is calibration. It is time consuming and a challenge to select appropriate microscopic variables and their values so that simulations can accurately reproduce the behavior of real systems. In this research, a procedure for fast and effective automated calibration using genetic algorithms is proposed. This approach is successfully demonstrated for single- and multi-objective optimization calibration problems. Five unknown variables, i.e., rolling and sliding friction coefficients, size, density, and Young's modulus of particles are determined based on the known bulk properties of density, AoR, and discharging time. The Pareto-optimal front visualizes the best compromise between the two conflicting objectives: the model error and the simulation time. For large scale applications, this is important to obtain efficient DEM variables with minimum model errors subject to specific simulation time constraints.