Evaluating Physics-Informed Neural Networks for Food Kinetic Modeling

Kinetic modeling is essential for predicting changes in food quality during processing and storage. This study evaluates the application of physics-informed neural networks (PINN) for food kinetic modeling, integrating kinetic insights into neural network frameworks. Based on three case studies, namely seed drying, bread baking, and kiwi softening, PINN models outperform empirical models in terms of extrapolation and interpolation tasks and match predictive power of physics-based models in interpolation tasks. PINNs also enable transfer learning, allowing efficient model adaptation to new datasets with comparable prediction accuracy to empirical kinetic models. However, limitations of PINN models exist in handling discontinuities found in multistage food processing and storage. Despite this, the generalization and data efficiency of PINN models position them as powerful tools for modern food kinetic modeling.