Abstract

Accurate estimation of lithium-ion cell temperature within battery modules is essential for preventing thermal degradation, improving performance, and ensuring safe operation. This work proposes a Physics-Informed Neural Network (PINN) framework for estimating the surface temperature of battery cells by integrating a fractal-based thermal model as a physical constraint during network training. Three approaches are evaluated and compared: a conventional neural network (NN), a physics-informed neural network (PINN), and an inverse PINN (iPINN). Results show that all models achieve high temperature estimation accuracy, with a mean squared error below 0.05 degrees C, while using measurements from only a limited number of cells within the module. Among the evaluated approaches, the iPINN demonstrates superior robustness and adaptability, maintaining high prediction accuracy even under reduced sampling rates. These results demonstrate that combining physics-based constraints with data-driven models enables accurate and scalable temperature monitoring of battery modules, providing a promising approach for advanced battery thermal management systems and anomaly detection in lithium-ion batteries.