DEM parameter calibration based on multi-objective Bayesian optimization and prior physical information

The discrete element method (DEM) is proving to be a reliable tool for studying the behavior of granular materials and has been increasingly used in recent years. The accuracy of a DEM model depends heavily on the accuracy of the particle property parameters chosen which is of vital importance for studying the mechanical properties of granular materials. However, the existing DEM parameter calibration methods are limited in terms of applicability, and the trial-and-error method remains the most common way for DEM parameter calibration. This paper presents a novel calibration method for DEM parameters using the multi-objective tree-structured parzen estimator algorithm based on prior physical information (MOTPE-PPI). The MOTPE-PPI does not rely on the training datasets and may optimize with every single test, significantly reducing the computational efforts for DEM simulation. Moreover, MOTPE-PPI is suitable for a variety of contact models and damping parameters in DEM simulation, showing robust applicability and practical feasibility. Taking an example, the DEM parameters of sandy gravel material collected from Dashixia rockfill dam in China are calibrated using MOTPE-PPI in the paper. The prior physical information is obtained through a series of triaxial loading–unloading tests, single-particle crushing tests, and literature research. Seven parameters in the rolling resistance linear contact model and breakage model are considered, and the optimization process takes only 25 iterations. Through quantitative comparison with existing parameter calibration methods, the high efficiency and wide applicability of the DEM parameter calibration method proposed in this study. The calibrated DEM parameters are used to investigate the hysteretic behavior and deformation characteristics of the granular material, revealing that the accumulation of plastic strain and resilient modulus is related to confining pressure, stress level, and the number of cycles.