Bayesian Calibration of GPU–based DEM meso-mechanics Part II: Calibration of the granular meso-structure

The spatial configuration is a significant contributing factor to dry granular materials' shear strength and dilative properties. This paper studies the effects of the particle size distribution (PSD) and the initial void ratio of a meso-structure through the drained triaxial compression using DEM. A novel multi-threaded two-level optimization is applied to the Force-Biased Generation (FGB) algorithm to enable the PSD and initial void ratio to be varied independently and studied systematically. The GPU-based Periodic Boundary Conditions (PBC) algorithm is developed in the CoSim-DEM framework and used to generate numerous statistical samples of triaxial responses in parallel. The optimized-FBG algorithm is used to study the effects of the initial void ratio and PSD of granular meso-structures for the DEM triaxial simulations. An increase in size polydispersity leads to an increase in shear strength at a constant initial void ratio. Sensitivity analysis of the parameters under DEM triaxial compression is performed. It is shown that dense samples are more sensitive to material parameters than loose samples. The interparticle friction angle monotonically increases in proportion to the volumetric strain for the dense sample, but interestingly, monotonically decreases for the loose sample. Next, Bayesian calibration is performed on synthesized DEM triaxial compressions. The macro responses are inverted, and the initial void ratio converged to the true value, other parameters are however difficult to recover due to the strong linear correlation between the material parameters and the mesostructure configuration. Finally, the micro-parameters are inferred for the drained triaxial condition of fordry graded sand for various confining pressures. The inverted macro responses are more accurate when including the PSD in the inversion process but may augment the values of the other parameters.