A physics-informed neural network coupling framework for predicting heat and mass transfer characteristics of grains

Grain drying is a critical process in agricultural engineering, where optimizing drying efficiency and ensuring product quality depend on accurately understanding the internal temperature and moisture distribution. To obtain the dynamic evolution of internal microscopic information, this study developed a Physics-Informed Neural Network (PINN) coupling framework. By integrating the interaction between temperature and moisture, along with transfer learning, this framework can predict the spatiotemporal evolution of grain temperature and moisture under various drying conditions. The result shows that compared with purely data-driven neural networks (DNN), the PINN coupling framework with added physical constraints achieves lower prediction errors when applied to unknown data. Under the drying condition of 333.15 K, the prediction accuracy of particle temperature and humidity by PINN are improved by 62.4% and 55.3%, respectively, compared with DNN. Furthermore, incorporating transfer learning into the PINN framework significantly improves computational efficiency, enabling the model to adapt more effectively to varying drying conditions. The proposed computational framework offers an innovative approach to explore and predict the dynamic evolution of temperature and moisture in agricultural materials, providing a valuable tool for optimizing drying processes in agriculture and related applications.