A Noise-Robust xLSTM-Based Method for Fault Diagnosis in Gas-Insulated Switchgear Using Acoustic Signals Under Strong Noise Interference

Gas-insulated switchgear (GIS), a vital component in power systems, requires high operational reliability to ensure system safety. However, mechanical faults of GIS can trigger abnormal vibrations that accelerate insulation aging, cause dielectric breakdown, and lead to gas leakage—failures that may paralyze the power system and cause severe economic losses. Traditional contact-based diagnostic methods using accelerometers are often unsuitable due to GIS’s complex structure and may interfere with operations. In contrast, acoustic-based methods offer a non-contact, sensitive, and easily deployable alternative. However, acoustic signals are highly susceptible to noise, limiting the reliability of conventional detection. To address this issue, we propose a novel noise-robust fault detection framework structure based on the extended LSTM (xLSTM) for acoustic GIS fault diagnosis under strong noise interference. The framework consists of an encoder, TSC-mLSTM blocks, a magnitude mask decoder, a complex decoder, and a classifier. The encoder extracts shared representations for both denoising and classification tasks, while the TSC-mLSTM blocks capture time and frequency dependencies. The two decoders predict denoised spectrograms, and the classifier performs final fault detection. The effectiveness of the proposed method is validated by experimental data obtained from a real gas-insulated switchgear, achieving accuracies of 39.83%, 46.27%, 59.61%, 75.88%, 83.21%, and 89.09% under SNR levels ranging from–20 dB to 5 dB (in 5 dB intervals), and obtains an accuracy of 96.15% in the absence of noise, outperforming the compared methods. Additionally, ablation studies confirm the effectiveness of the model design.

Recommended citation: Dong, H., Zhang, Y., Zhang, B., Zhang, Z., Shao, Y., Yuan, G., & Liu, H. (2025). A Noise-Robust xLSTM-Based Method for Fault Diagnosis in Gas-Insulated Switchgear Using Acoustic Signals Under Strong Noise Interference. Expert Systems with Applications, 130827.
Download Paper | Download Slides | Download Bibtex