A novel ML-DEM algorithm for predicting particle motion in rotary drums

Abstract

The discrete element method (DEM) is a widely used approach for studying the behavior of particles in industrial equipment, including rotary drums. Although DEM is highly accurate and efficient, it suffers from the computational cost in simulations. The primary objective of this research is to reduce the computational costs of DEM by introducing a novel machine learning (ML) approach based on a deep neural network for predicting particle behavior in rotary drums. The proposed approach utilizes a continuous convolution operator in a neural network. To evaluate its effectiveness, the results of the proposed ML-DEM approach were compared quantitatively and qualitatively with the experimental data and the conventional DEM results. It was shown that in addition to its high accuracy, the proposed approach reduces the computational costs by approximately 35 % and 65 % compared to the conventional DEM simulations on GPU and CPU (with 8 processors), respectively. Furthermore, to ensure the comprehensive and independent validation of the proposed algorithm, the study investigated the effects of various parameters such as drum rotational speed and fill ratio on lateral entropy-based mixing, circulation time, and velocity profile in the active layer. The results were then compared with those obtained using the conventional DEM and found to be in good agreement. This new algorithm can serve as a starting point for reducing computational costs in simulating particle motion in granular systems.