A Novel Health Indicator-Based RBF-FGSPF Framework for Health Prognosis of Lithium-Ion Batteries in Dynamic Operating Conditions

The onboard measurement of lithium-ion battery (LIB) health indicators (HIs) such as capacity and internal resistance (IR) are hardly achievable during dynamic discharge conditions, thereby making them unsuitable for real-time health prognosis. Therefore, we propose a remaining useful life (RUL) estimation framework for LIBs using a novel HI "discharge pulse power (DPP)"that is suitable for both complete and partial discharge conditions. The main contribution consists of virtually predicting the end of the discharge voltage using the radial basis function-based finite Gaussian sum particle filter (RBF-FGSPF) model and estimating RUL as a function of DPP in variation with the cycle number. The novelty also lies in the model architecture where the radial basis function neural network (RBFN) is adaptively trained online, that is, its parameters are identified by the FGSPF upon the availability of new observations. The robustness of the proposed framework is verified on the Centre for Advanced Life Cycle Engineering (CALCE) lifetime degradation dataset based on three stressors: charge cut-off current, discharge rate, and ambient temperature and also validated experimentally on realistic drive profiles having variable C-rates and temperatures. The results verify that the proposed framework shows an improvement of 27.89% in absolute error (AE) for all operating conditions when compared to the benchmark RUL prediction methods for LIBs.