Data-Driven State of Health and Functionality Estimation for Electric Vehicle Batteries Based on Partial Charge Health Indicators

Abstract

Effective Electric Vehicle (EV) operation relies on robust State of Health (SoH) estimation algorithms, key for informed battery management. Various algorithms have been proposed to estimate degradation, often depending on full charge segmentation or on conditions that deviate from real-world EV operation. Addressing this gap, this study introduces a cell-level SoH estimation algorithm based on partial charges. The proposed approach employs Health Indicators (HIs) derived from realistic laboratory testing, which contains a variety of voltage ranges during charge to replicate the complexity of real data. The study compares two commonly employed data-driven algorithms, Support Vector Regression (SVR) and Neural Networks (NN) and two estimation voltage ranges, which encompass the second and third Incremental Capacity (IC) peak. Along with the SoH, the battery functionality is estimated through the State of Function (SoF), leveraging degradation data and performance requirements for each tested cell. This enables the definition of an indicator quantifying the proximity of the battery to underperformance in specific applications. In general, the second IC peak shows higher correlation to the SoH. However, the NN SoH algorithm, when trained with high number of observations in the third IC peak, shows the lowest error with an average Root Mean Square Error (RMSE) of 0.00330. Moreover, the translation from SoH to SoF highlights the different performance requirements for each case and supports a functional definition of End of Life (EoL) beyond the fixed threshold.