Uncertainty propagation in discrete element models using PDEM

The Discrete Element Method (DEM) for the analysis of cohesive granular materials, such as concrete, requires the use of contact interaction models with several parameters. Despite the existence of a number of approaches to estimate these parameters, uncertainty in their values is inevitable. This uncertainty propagates across the system, giving rise to uncertainty in the measured response. It is important to estimate this uncertainty and to check and control dispersion in the results in order to improve the predictive ability of the DEM model. This usually requires running a very large number of simulations to effectively sample the space of random param-eters and thereby obtain the probability density function (PDF) of the response. Such an approach is not feasible for DEM-based simulations on account of the inordinate computational cost. The present paper provides an alternate approach based on the Probability Density Evolution Method (PDEM). PDEM, which requires fewer sampling points, is employed for the fast and efficient generation of the PDF. The approach is demonstrated through a DEM model developed for the slump test, widely used to determine the workability of fresh concrete. The simulation results are analysed using PDEM, enabling the PDF of the measured response to be determined from the results of a limited number of trials. Moreover, the dispersion in the DEM results is found to be low, with a relatively small coefficient of variation, thereby attesting to the predictive ability of the DEM model.

Automated summary

This paper introduces an alternate approach based on the Probability Density Evolution Method (PDEM) for the fast and efficient generation of the Probability Density Function (PDF) of the Discrete Element Method (DEM) model. The approach is demonstrated through a DEM model developed for the slump test, and the results show that PDEM can determine the PDF of the measured response from a limited number of trials, with low dispersion in the DEM results and strong predictive ability.