DEM data-driven modeling of repose angle of granular materials

Repose angle is an important property of granular materials and is usually simulated using Discrete Element Method (DEM). However, DEM simulation is computationally intensive and is thus unsuitable for online applications where parameters are frequently changed. To solve this problem, we propose a DEM data-driven modeling method for fast prediction of repose angle. Firstly, variables affecting the repose angle are analyzed; by Latin hypercube sampling of parameter spaces, 100 sets of DEM simulations are performed to generate data of repose angle. Based on these data, a support vector machine (SVM) model is then established and trained for fast prediction of repose angle under various conditions. Tests and comparison show that the reposed angle predicted by the SVM model is close to the DEM simulation result while the required computing time is greatly decreased (from 43.8 hours to 0.17 seconds), and it outperforms BP neural network and Kriging interpolation method in terms of prediction accuracy. The SVM model for repose angle is also verified by physical experiments, with prediction error less than ± 1 °. The established model can replace DEM, and is suitable for applications where fast prediction of repose angle is required.