Online identification of knee point in conventional and accelerated aging lithium-ion batteries using linear regression and Bayesian inference methods

Abstract

Accurate online knee point identification is crucial for predictive maintenance and secondary utilization of batteries. The “knee point” refers to the point in the battery capacity degradation curve where the degradation rate changes from linear to non-linear, marking a critical transition indicating the onset of accelerated capacity loss. However, challenges such as incomplete monitoring data, prevalent noise, difficulty in extracting characteristic parameters, and capacity regeneration phenomena hinder precise, real-time knee point detection. This study integrates physical mechanism modeling, signal processing techniques, and statistical inference to propose a robust, efficient solution for knee point identification. The proposed method employs feature extraction based on capacity loss mechanism models, denoising using variational mode decomposition (VMD), and a hybrid framework that combines linear regression with Bayesian inference. This dynamic model updates boundary limits in real-time, enabling highly accurate knee point identification across two positive materials, lithium cobalt oxide (LCO) and lithium iron phosphate (LFP), under various operating conditions. Comprehensive evaluations show that the proposed method achieves accuracies exceeding 94 % for conventional aging batteries and 92 % for accelerated aging batteries, surpassing existing methods. Additionally, the method demonstrates resilience to noise interference and capacity regeneration phenomena, maintaining high accuracy even under complex conditions. These results suggest that the proposed method has broad adaptability, making it a valuable tool for real-time battery health monitoring and providing a solid foundation for future research on battery aging diagnostics.