Three-Dimensional Discrete Element Modeling for the Angle of Repose of Granular Materials: Artificial Intelligence and Machine Learning Approach

Abstract

This research studies the calibration of contact parameters for Johnson-Kendall-Roberts (JKR) model using machine learning (ML) algorithms. Multiple linear regression (MLR), support vector regression (SVR), decision trees (DT), and extreme gradient boost (XGBoost) were used. The angle of repose (AoR) of granular piles was measured, and a DEM model was built to simulate the experiment. After calibration, the model was used to generate a database that was used to train the ML algorithms. All algorithms exhibited high coefficients of determination (R²) and low errors. Additionally, the study discussed the effect of the different features on the accuracy of the models and presented a feature importance analysis for the different ML algorithms. Finally, a simplified method was suggested to calibrate the contact parameters using the XGboost method. The method was able to estimate the contact parameters that resulted in accurately determining the AoR of a selected sandy soil.