Battery state of health estimation for partial charge segments using incremental capacity analysis

Abstract

Accurately estimating the state of health of a lithium-ion battery is essential for ensuring safe and efficient operation. However, performing this estimation under operational conditions poses significant challenges, as only limited measurement data — current, voltage, and temperature — are typically available. This work presents a data-driven framework for state of health estimation based on automatic feature extraction. To capture battery degradation patterns, incremental capacity curves derived from constant-current charge phases are analysed. Through proper orthogonal decomposition, compact features are extracted from these curves by reducing the dimensionality of the feature space while preserving essential information. These low-dimensional features are then used to train a machine learning model for accurate state of health estimation. The framework can process partial charge segments from different state of charge regions, eliminating the need for complete charge cycles. Validation is performed using data from large-scale ageing experiments. The approach utilises structured reference tests conducted at 25 °C alongside constant-current phases extracted from dynamic load-cycle data within a 20–30 °C range. The model consistently captures degradation trends, confirming the robustness of the proposed framework across the conditions examined in this study.