Development of an Integrated DEM–CatBoost Model for Efficient Prediction of Liquid Bridge Forces in Wet Food Powder Systems

Abstract

Developing an accurate and computationally efficient contact model for discrete element method (DEM) simulation of wet food powders remains a significant challenge due to the complex capillary interactions formed between particles at intermediate moisture contents. Considering this challenge, this study developed a machine learning-based capillary force model to accurately predict liquid bridge forces in wet food powders. A comprehensive database of capillary forces was generated for particle-particle and particle-wall interactions under varying particle sizes, moisture contents, and contact angles using numerical modelling of the liquid bridge model. The generated database was subsequently used to train CatBoost regression models for predicting capillary forces with high accuracy. Finding suggest that, the developed models showed excellent predictive capability, with relative prediction errors below 10% over the selected conditions. The trained capillary force model was integrated into the discrete element method to calculate inter-particle capillary interactions and simulate the mechanical behavior of food powders. Simulation results showed excellent agreement with experimental observations of particle column collapse run-out length. This study will provide an accurate and efficient numerical approach for simulating wet food powder systems.