Enhancing battery health estimation using incomplete charging curves and knowledge-guided deep learning

Abstract

Accurately monitoring the health degradation of batteries is critical for ensuring the safety and reliability of electrochemical energy storage systems. While great progress has been made in this area by machine learning, a significant but overlooked challenge is the prohibitive demand for large high-quality degradation data for model training. Here, we reveal that although ubiquitous incomplete charging curves cannot derive state of health (SOH) labels, they provide valuable degradation knowledge that can effectively contribute to the development of SOH estimation models. We propose a knowledge-guided method that has high flexibility in using charging segments to estimate the SOH. More importantly, it takes advantage of incremental capacity indicators obtained from incomplete charging curves to guide the training of a deep neural network (DNN). The resulting pre-trained DNN can quickly adapt to SOH estimation with very few SOH labels. Validations show that using only 50 charging segments with SOH labels, our knowledge-guided DNN achieves an SOH estimation root mean square error of 21.08 mAh for 0.74 Ah batteries, which is 60 % less compared with conventional methods. Further validations on 3 datasets covering different battery chemistries and operating conditions confirm the superiority of the proposed method. Our work highlights the promise of employing domain knowledge to advance machine learning models for health monitoring purposes.