Charging phase health indicators for battery state-of-health estimation: a systematic comparison of CC, CV, and combined approaches under cross-battery validation

This study investigates the effectiveness of various charging-phase health indicators for estimating the State-of-Health (SOH) of lithium-ion batteries. While most research focuses on the constant-current (CC) phase, this paper systematically evaluates indicators derived from the constant-voltage (CV) phase, such as charge throughput, time duration, and current decay time constants. The researchers used the NASA battery aging dataset to compare CC-only, CV-only, and combined CC+CV indicator sets. To ensure practical relevance, the study employed Leave-One-Battery-Out (LOBO) validation, which tests the model on a battery not included in the training set, providing a more realistic assessment of how these models perform in real-world deployment compared to standard, less rigorous cross-validation techniques.

The results confirm that the CC and CV phases contain complementary information regarding battery degradation. A combined model using both CC and CV indicators achieved the highest predictive accuracy (R² = 0.874). A critical finding of this work is the significant discrepancy between evaluation methods: standard 5-fold cross-validation yielded much higher accuracy than the LOBO approach, with a 119% increase in Root Mean Square Error (RMSE) under the more rigorous validation. This highlights that conventional evaluation metrics often provide an overly optimistic view of model performance for real-world battery management systems.

The study provides actionable guidelines for selecting health indicators under computational constraints. By using SHAP (SHapley Additive exPlanations) analysis, the authors identified the CV-to-CC time ratio as the most influential indicator for SOH estimation. Because this ratio is easy to calculate and does not require complex numerical differentiation, it serves as a robust, lightweight feature for embedded battery management systems where computational resources are limited.