An adaptive data-driven approach for two-timescale dynamics prediction and remaining useful life estimation of Li-ion batteries

During the multi-cycle operation of a Li-ion battery, its process dynamics evolve in two distinct timescales: slow degradation dynamics over multiple cycles and fast cycling dynamics during each cycle. The slow inter-cyclic dynamics of capacity degradation describes remaining useful life (RUL), and the fast intra-cyclic dynamics of state of charge (SoC) and voltage provides an insight into available power, temperature change, and charge and discharge times. Hence, predicting both intra and inter-cyclic dynamics aids in understanding battery degradation and assessing its performance. To this end, we develop a data-driven approach to model both fast and slow degradation dynamics using operable adaptive sparse identification of systems (OASIS). Specifically, the developed method determines two battery models: inter-OASIS and intra-OASIS. The inter-OASIS model predicts capacity degradation and estimates RUL, and utilizing this prediction, the intra-OASIS model accurately predicts SoC and voltage dynamics. The developed method is demonstrated on a LiFePO4/graphite battery system.