Prediction of Granular Velocity in Three-dimensional Avalanche Systems Based on Artificial Neural Networks

Obtaining particle velocities in three-dimensional granular systems is usually based on numerical simulations. However, when simulating large-scale complex granular systems, this approach faces challenges that include time-consuming processes and high computational costs. To address these limitations, this study integrates numerical simulations with artificial neural networks (ANN). We first simulate the granular avalanche process in a three-dimensional horizontal rotating drum using numerical simulations. Subsequently, by entering the static initial configuration of the granular system solely into a physics-informed ANN model with physical constraints, we successfully predicted the velocity of all particles in the system at the next moment, achieving a Pearson correlation coefficient of 0.837. Furthermore, the proposed model demonstrates robust performance when applied to systems with varying contact potentials, particle size dispersity levels, and different static initial configurations, thus validating the enhanced robustness and generalization capability of the optimized ANN framework.