Physics-informed Neural Network for Estimation of Lithium-Ion Battery State-of-health

Currently, lithium-ion batteries are becoming the most promising power source for a variety of portable electronics as well as electric vehicles. Some of the advantages that promote their widespread usage include their long battery cycle life, high durability, low self-discharge rate, and fast charge rate. However, despite their superiority in comparison with other power sources, there exists a lack of understanding regarding their battery lifetime owing to their sophisticated electrochemical actions, which cannot be sufficiently modeled and predicted using traditional physics-based models. This limitation has motivated the development of numerous data-driven approaches. However, data-driven methods also have certain limitations, such as low interpretability and inability to extrapolate well. This necessitates an alternative method that can leverage the strengths of both models while complementing their drawbacks. In this study, the state-of-health of lithium-ion batteries is estimated using a physics-informed neural network with the integration of physics in the deep learning pipeline. The results of this study indicate that the proposed model outperforms the conventional data-driven methods in RMSE and physical inconsistency.