Physics-informed neural network for increasing prediction accuracy of microscale variations of single plant cell during drying

Bulk level variations of plant foods during drying are mainly governed by microscale characteristic variations. Investigating such microscale variations have been challenging with physics-based models due to heterogeneity of microstructures, largely unknown property data, and limitations of numerical simulations. On the other hand, the development of data-driven machine learning (ML) models for predicting microscale variations has not yet been succeeded due to the inability of having a sufficient dataset for extracting an interpretable solution. Therefore, in this work, the Physics-Informed Neural Network (PINN) capabilities are explored to improve the prediction accuracy of moisture concentration variations of a single plant cell during drying with low dimensional input data. In particular, additional information using relevant physics conditions is provided into the feedforward neural network by altering the loss function. The performance of PINN is investigated and compared against pure data-driven ML model predictions for benchmark cases. It can be highlighted that PINN with additional physics information is significantly improved the prediction accuracy even if the training data is very low, indicating the possibilities of integrating PINN for accurately investigating microscale characteristic variations of plant foods during drying.