Battery state-of-health estimation: An ultrasonic detection method with explainable AI

State of health (SOH) stands as a pivotal metric for evaluating the aging performance of batteries. Effective SOH estimation is imperative to maintain the performance and safety of battery-based energy systems. This paper integrates the benefits of non-destructive ultrasonic detection with explainable AI to propose a rapid and accurate SOH estimation method for lithium-ion batteries. Specifically, the method first utilizes a portable ultrasonic sensor to achieve real-time battery-based ultrasonic measurement. Then an explainable AI model named Generalized Additive Neural Decision Ensemble (GAN-DE) is derived to efficiently estimate battery SOH and quantify the influence of relevant solo ultrasonic features. To further consider the interaction effects of ultrasonic features, an improved model named GAN-DE with interaction (GAN-DEI) is also proposed. The results demonstrate that both GAN-DE and GAN-DEI can achieve satisfactory accuracy in SOH estimation, especially for the state of charge (SOC) ranges from 35 % to 65 %, with R2 reaching 0.971 and 0.991, respectively. Additionally, based upon the developed explainable AI models, the contributions of main effects and interaction terms derived from ultrasonic features can be quantified, while their dynamic effects are thoroughly explained. This could help engineers to quickly obtain reliable information about battery health, thus benefiting battery health management.