Robust Artificial Neural Network-Based Models for Accurate Surface Temperature Estimation of Batteries

The temperature of a battery cell is a major parameter that must be continuously monitored to ensure a safe operation. Most battery failures are linked to thermal runaway due to temperature rise in the battery, which if not detected early can result in battery destruction or fire hazard. This article proposes robust artificial neural network models with reduced complexity to estimate the surface temperature of different battery chemistries. The proposed models are accurate, reliable, and use no temperature sensor. Different neural network architectures are evaluated and optimized. Derivation of the models followed by experimental verification using commercial battery cells of different chemistries, specifications, and aging conditions is presented.