Artificial intelligence techniques for precise current sensor fault diagnosis in battery systems using enhanced equivalent circuit models

Abstract

Accurate diagnosis of current sensor faults is a critical requirement for improving the reliability and safety of lithium-ion battery systems, especially in electric vehicles and renewable energy applications. This paper proposes a comprehensive diagnostic framework that combines enhanced equivalent circuit modeling, adaptive filtering, and artificial intelligence techniques. A five-phase permanent magnet synchronous motor (PMSM) Simulink model is developed to collect operational data, which is pre-processed using an Adaptive Smooth Variable Structure Filter with a time-varying boundary layer (ASVSF-VBL) to suppress noise measurement. State estimation is carried out using a hybrid Kalman Filter-Cuckoo Algorithm (KFCA), while a CNN-LSTM architecture captures spatial and temporal features for effective fault identification. Open-circuit transistor faults are monitored through a pulse-width modulation voltage source inverter, and model parameters are optimized using a Self-adaptive Bonobo Optimizer combined with Least Mean Squares (SaBO-LMS). Simulation studies demonstrate that the proposed method achieves a fault diagnosis accuracy of 94%, significantly outperforming VICO (70%) and RLS-UKFJEM (65%). It also delivers smoother current tracking, higher voltage stability up to 550 V, improved state of charge estimation, and reduced computational complexity. These results confirm the suitability of the proposed approach for real-time intelligent battery management systems.