Development of ultra-fast computing method for powder mixing process
Powder mixing is an important operation in many industries. Numerical simulations using the discrete element method (DEM) have been widely used to analyze powder-mixing processes. However, one of the current limitations of the DEM simulation is its high computational cost. Recently, approaches that combine machine learning models and numerical simulations have attracted considerable attention for high-speed computing. However, there has been no research on high-speed computing methods for powder mixing that account for individual particle motions. Here, we propose an original machine learning model, namely, a recurrent neural network with stochastically calculated random motion (RNNSR), which enables a long-time-scale powder mixing simulation with low computational cost and high accuracy. The RNNSR is designed to learn individual particle dynamics with periodicity from short-term DEM simulation results and predict powder mixing for a longer period. The RNNSR combines a recurrent neural network and a stochastic model to predict both convective and diffusive mixing. The simulation results obtained using the RNNSR were quite similar to those obtained using the DEM in terms of the degree of powder mixing, particle velocity, and granular temperature. It was also demonstrated that the RNNSR has the capability of ultrafast computing in powder-mixing simulations. In conclusion, we demonstrated the effectiveness of the RNNSR for ultrafast computation of the powder mixing process.