Online Learning-Based Fault Detection for Nonstationary Processes with LSTM-AE-EWC

The detection of faults in dynamic systems, particularly under nonstationary conditions, presents significant challenges due to the complex nature of system behavior. Traditional fault detection methods often struggle to extract deep-level features and adapt to changes in operating conditions, resulting in reduced accuracy and reliability. To address this issue, we propose a novel nonstationary process monitoring framework that integrates long short-term memory (LSTM) networks, autoencoders (AEs), and elastic weight consolidation (EWC). The proposed method employs LSTM to capture long-term dependencies and dynamic characteristics of time-series data, while an AE is utilized to extract deeper features for fault detection. Furthermore, the method incorporates the online learning technique EWC to adaptively update the model, enabling it to accommodate changes in operating conditions. By leveraging the synergistic combination of LSTM, AE, and EWC, the proposed approach achieves effective feature extraction and fault detection in nonstationary processes. A case study on an insulated-gate bipolar transistor (IGBT) power system demonstrates that the proposed method can effectively detect faults in nonstationary processes.