Adaptive Extended Kalman Filter-Based Multi-Fault Detection and Diagnostics for High-Voltage Battery Management Systems in Electric Vehicles

Abstract

This paper proposes an adaptive Extended Kalman Filter (EKF) based fault detection and diagnosis (FDD) system for 800 V NMC lithium-ion battery pack (192S×2P) used in battery electric vehicles (BEVs). The voltage, current and thermal innovation residuals per cell are generated by a two-RC equivalent circuit model (ECM) parameterised by hybrid pulse power characterisation (HPPC) tests. Adaptive ±3σ detection thresholds, which are proportional to the EKF innovation covariance S_k, cover the entire 806.4 V operating range and take into consideration model uncertainty due to SOC and temperature. It is able to identify four fault classes that are all critical for safety: internal short circuit (ISC), voltage sensor bias, thermal runaway precursor, and overcharge/over-discharge. A hardware-in-the-loop (HIL) bench is used to validate simulation results in MATLAB/Simulink using a drive profile from the WLTP Class 3 driving cycle. The proposed method achieves the overall detection accuracy of 98.9%, weighted F1-score of 0.987, false positive rate (FPR) of 0.7% with a mean detection latency of 19.2 ms which meets the ISO 26262 ASIL-D reaction time constraint of 50 ms.