Battery capacity assessment is a crucial research direction in the field of lithium-ion battery applications. In the previous research, a novel data-driven state of health (SOH) estimation method based on the voltage relaxation curve at full charging is developed. The experimental results have shown the evidence of the superiority of accurate battery SOH estimation based on physical features derived from equivalent circuit models (ECMs). However, the earlier research has limitations in estimating battery capacity with a diversity of battery charging states of charge. This study represents an extension of the previous work, aiming to investigate the feasibility of this technology for battery degradation evaluation under various charging states so that the application capability in practice is enhanced. In this study, six ECM features are extracted from 10-min voltage relaxation data across varying charging states to characterize the battery degradation evolution. Gaussian process regression (GPR) is employed to learn the relationship between the physical features and battery SOH. Experimental results under 10 different state of charge (SOC) ranges show that the developed methodology predicts accurate battery SOH, with a root mean square error being 0.9%.

To improve the safety of electric vehicles and battery energy storage systems, early prediction of thermal runaway (TR) is of great significance. This work proposes a novel method for early warning and short-term prediction of the TR. To give warning of TR long time in advance, a variety of battery models are established to extract key features, such as Pauta feature and Shannon entropy of voltage deviation, and then local outlier factor algorithm is used for feature fusion to detect abnormal cells. For the short-term prediction, the predefined threshold and variation rates are used. By measuring the real-time signals, such as voltage and temperature, their variation rates are calculated, based on which TR can be predicted exactly. The real data including TR from an electric vehicle are used to verify the method that it can give a warning on TR long time before it happens up to 74 days. This is remarkable for providing replacement recommendations for abnormal cells. It can also predict the occurrence of TR 33 seconds in advance to ensure the safe use of batteries.

The safety of lithium-ion batteries in electric vehicles (EVs) is attracting more attention. To ensure battery safety, early detection is necessary of a soft short circuit (SC) which may evolve into severe SC faults, leading to fire or thermal runaway. This paper proposes a soft SC fault diagnosis method based on the extended Kalman filter (EKF) for on-board applications in EVs. In the proposed method, the EKF is used to estimate the state of charge (SOC) of the faulty cell by adjusting a gain matrix based on real-time measured voltages. The SOC difference between the estimated SOC and the calculated SOC through coulomb counting for the faulty cell is employed to detect soft SC faults, and the soft SC resistance values are further identified to indicate the degree of fault severity. Soft SC experiments are developed to investigate the characteristics of a series-connected battery pack under different working conditions when one battery cell in the pack is short-circuited with different resistance values. The experimental data are acquired to validate the proposed soft SC fault diagnosis method. The results show that the proposed method is effective and robust in quickly detecting a soft SC fault and accurately estimating soft SC resistance.