ARMAX-based capacity fade estimation for lithium-ion cells across state-of-charge and temperature variations

Abstract

This paper presents an ARMAX-based framework for State-of-Health (SoH) estimation of lithium-ion batteries using short-duration impulse-response characterization. Battery dynamics are identified through a third-order equivalent circuit model whose parameters are estimated under different SoC, SoH, temperature, C-rate, and Depth of Discharge conditions. The identified parameters are organized into a multidimensional look-up table (LUT) and subsequently interpolated through polynomial hypersurfaces, enabling continuous parameter representation and optimization-based estimation. Two Status Estimation Process (SEP) strategies are investigated: an exhaustive LUT-based search and a Goodness-of-Fit (GoF)-based local optimization over the interpolated parameter space. The proposed methodology is experimentally validated using independent NMC and LFP datasets, including different aging trajectories, low-temperature conditions, high C-rate operation, and representative driving-cycle current profiles. Furthermore, the algorithm is implemented and benchmarked on a Texas Instruments F28379D embedded platform to assess its computational requirements for real-time Battery Management System (BMS) applications. Experimental results show SoH estimation errors generally below 5% over the considered operating range while requiring significantly less training data than data-driven approaches. The proposed framework provides an effective trade-off between estimation accuracy, computational complexity, and embedded implementation suitability, making it a promising solution for practical battery health monitoring.