Robust Bayesian parameter calibration for DEM simulations of shear cell test

Shear test devices are commonly employed to characterise the mechanical behaviour of (cohesive) particulate materials. In simulations, Discrete Element Method (DEM) models require correct material parameter values for accurate predictions in a broad variety of process applications. In the present work, two open-source codes implementing robust Bayesian calibration (GrainLearning) and DEM (MFIX-DEM) are effectively integrated into a framework to investigate the capability of DEM to represent the key features of experimental shear cell test data on a model granular material. Material properties, geometric characteristics and operating procedures were carefully matched by using a periodic box subjected to stress- and displacement-controlled walls. Properties such as the static friction coefficient, rolling friction coefficient, and cohesive energy in DEM are varied in a statistically consistent way to capture the experimental data of the yield loci collected on spherical glass beads. With the calibration process for standard DEM, it was found that, under the conditions investigated, no combination of reasonable parameter values yielded sufficient shear resistance. So, in analogy with approaches documented in the literature, spherical particles with partially and fully locked rotation have been considered, with only the latter case being successful. The sequential Bayesian filtering approach allowed appropriate DEM parameters to be found along with their uncertainty quantification. Using these values, various other experimental features were captured with good agreement, including the evolution of the shear stress during loading and the dilation behaviour. Finally, DEM results were processed to examine the formation of the shear displacement vertical profile from the elastic to the yielding stage.