A hybrid temperature distribution monitoring method for Lithium-ion battery module by integrating multi-physics with machine learning

Abstract

The temperature distribution monitoring of the battery system is essential to prevent thermal runaway and ensure operational safety. In real applications, there is an intricate thermal dynamics inside the battery pack which causes heat transfer and heat dissipation inhomogeneities. It is not easy to describe with the control-based lumped thermal model. The accuracy of temperature monitoring will be affected if the pack-level thermal dynamics are not captured. Motivated by this, a hybrid lumped multi-physics coupled neural network (MPNN) model is proposed. The hybrid MPNN model combines the mechanism-driven multi-physics coupled model and the data-driven machine learning model for thermal non-uniformity compensation. A hybrid MPNN-based close-loop observer is further proposed to achieve a real-time estimation of the internal temperature of each cell in the battery pack. The model parameters of the multi-physics coupling model are identified based on the recursive least square algorithm and genetic optimization algorithm by the experimental data. The computational fluid dynamic is applied to simulate the thermal behavior and validate the multi-physics coupling model at the system level. Results indicate that the proposed hybrid MPNN model can capture the complex thermal distribution non-uniformity in the battery system more accurately compared with the traditional model. The hybrid MPNN model combined with the unscented Kalman filter method can accurately monitor the temperature distribution to prevent thermal runaway.