Engineering-adaptive electrochemical modeling for fault diagnosis in electric vehicle energy storage systems

Electrochemical models offer great potential for onboard monitoring of lithium-ion batteries, yet their complexity and dependence on high-quality data have limited their engineering deployment. To address this challenge, this paper proposed an engineering-adaptive modelling framework that enabled reduced-order electrochemical models to remain accurate and robust under temperature variations and battery aging in real-world vehicle operation. The framework was developed by applying a transfer-function-based reduction of the single particle model with electrolyte dynamics (SPMe) model and reformulating it into a state-space structure, supporting real-time iteration and internal state tracking. Furthermore, a practical parameter identification scheme was introduced, combining interpolation-enhanced preprocessing, long-term open-circuit voltage extraction, and polarization resistance estimation. This allowed one-shot particle swarm optimization (PSO)-based parameter identification using sparse onboard data, where PSO adaptively estimated internal parameters from voltage trends. The resulting lightweight process supported periodic updates and cross-platform deployment. While only three cells were used for initial parameter identification, the model was validated across 96 cells in the full battery pack, demonstrating scalability and long-term stability, with average residuals maintained around 12 mV throughout a full year of cross-seasonal operation. Based on residual trend analysis, a fault diagnosis method was further developed to detect and isolate subtle faults such as data loss and micro-overcharge. The results highlighted the framework’s diagnostic capability and engineering adaptability, providing a practical path for the large-scale deployment of physics-based battery models in real vehicles.