Mechanistic-probabilistic learning fusion approach for state of health estimation in LiFePO4 batteries under high-rate discharge cycling

Pure electric construction vehicles represent a transformative development in sustainable heavy machinery. However, their unique operational profiles, characterized by persistent high-rate discharge cycling, accelerate battery degradation beyond typical electric vehicle scenarios. This study presents a novel mechanistic-probabilistic learning framework for accurate state-of-health (SOH) estimation under high-rate discharge cycling conditions. To explore the health behaviour of LiFePO4 batteries subjected to high-rate discharge cycling, an integrated framework combining mechanistic analysis and Gaussian mixture regression (GMR) is proposed. Specifically, the incremental capacity and post-mortem analyses are introduced to identify aging mechanism. Moreover, the comprehensive correlation between chemical mechanisms and incremental capacity curves is established. The results reveal that the loss of irreversible lithium-ion is the primary cause of failure under high-rate discharge cycling. Incremental capacity features and voltage are extracted as potential health indicators. A stacked auto-encoder neural network is proposed to obtain the fusion health indicator. Furthermore, the GMR is built for accurate and reliable SOH estimation. Compared to existing methods, the proposed approach demonstrates significantly enhanced reliability and precision in SOH estimation, with a maximum relative error of less than 2 %.