Physics-informed machine learning for accurate SOH estimation of lithium-ion batteries considering various temperatures and operating conditions

Accurate State of Health (SOH) estimation for lithium batteries (LIBs) is crucial for the safe operation of battery systems. However, the lack of physical properties and the varied operating conditions in real-world use further increase the difficulty of traditional SOH estimation, making it a significant challenge in current research. For this reason, this paper proposes a physics-informed machine learning (PIML) method for accurate SOH estimation of LIBs varied operating conditions. Considering the fully charged relaxation voltage data obtained easily in practical applications, firstly, this paper discussed the relaxation voltage data related to the battery's aging characteristics from the experimental tests. Secondly, the fractional-order equivalent circuit model (FOECM) is constructed and parameters characterizing battery degradation are identified for extracting the physical features. Ultimately, a novel PIML framework based FOECM of LIB is developed, then the datasets of three different battery types under different temperatures and discharge rates are used and validated for SOH estimation without considering any usage information. Experimental results show that the PIML method proposed in this paper can quickly achieve SOH estimation and keep the accuracy in 0.84 % for different types of batteries under varying experimental conditions. In addition, compared with other feature extraction methods, the PIML-based SOH estimation has obvious advantages with 16.2 %, which provides an important reference for the design and optimization of advanced battery management systems.