Accelerated DEM Simulation of Food Powder Mixing in a Planetary Mixer Using a Recurrent Neural Network with Stochastic Random Motion

Abstract

The uniform mixing of different food powder ingredient is a critical unit operation that significantly influences the quality and consistency of food products. However, their differences in particle properties often lead to non-uniform mixing. Discrete element method (DEM) plays important role in analysing the particle-scale mixing behaviour, but its application is often limited by the longer simulation times. Considering this limitation, this study proposes an integrated framework that integrates DEM with a recurrent neural network (RNN) to accelerate the prediction of powder mixing behaviour. Meanwhile, a short-term DEM simulation of mixing of different food powder ingredients was performed to generate particle trajectory data during planetary mixing. This data was used to trained the RNN model for capturing the periodic particle motion and stochastic component captures random particle diffusion resulting from particle-particle and particle–wall interactions. The trained model was further used to predict long-term mixing behaviour with substantially reduced computational cost. The predictive performance of the proposed framework was evaluated by comparing the mixing index, particle velocity, and spatial particle distribution with those obtained from conventional DEM simulations. The findings indicated that the developed integrated model accurately predicts powder mixing behaviour, with an error of less than 10%, while reducing the simulation time by approximately 60% compared. This study offers an effective alternative for computationally longer simulation times