Explainability-driven model improvement for SOH estimation of lithium-ion battery

Deep neural networks have been widely used in battery health management, including state-of-health (SOH) estimation and remaining useful life (RUL) prediction, with great success. However, traditional neural networks still lack transparency in terms of explainability due to their “black-box” nature. Although a number of explanation methods have been reported, there is still a gap in research efforts towards improving the model benefiting from explanations. To bridge this gap, we propose an explainability-driven model improvement framework for lithium-ion battery SOH estimation. To be specific, the post-hoc explanation technique is used to explain the model. Beyond explaining, we feed the insights back to model to guide model training. Thus, the trained model is inherently explainable, and the performance of the model can be improved. The superiority and effectiveness of the proposed framework are validated on different datasets and different models. The experimental results show that the proposed framework can not only explain the decision of the model, but also improve the model's performance. Our code is open source and available at: https://github.com/wang-fujin/Explainability-driven_SOH.