GIS Circuit Breaker Mechanical Fault Diagnosis based on Voice-Print

In this work, a method is proposed to successfully diagnose mechanical faults in gasinsulated switchgear (GIS) circuit breakers prone to acoustic signal disturbance from significant background noise and challenging feature extraction. Initially, a multi-channel acoustic sensor array is employed to gather the original observation signal of the circuit breaker. The independent component analysis (ICA) method is employed to decompose the observed signal into multiple source signals, and the source signal with the lowest fuzzy entropy is chosen as the feature signal. Subsequently, the multiscale fuzzy entropy (MFE) of the feature signal is computed to generate the sound-print feature of the circuit breaker. Lastly, the fault of the circuit breaker is identified using the extreme learning machine (ELM) algorithm. The experimental results demonstrate that the utilization of acoustic signals as a detection method offers a novel approach for diagnosing mechanical faults in GIS circuit breakers. The suggested approach successfully extracts the sound-print information, resulting in a notable enhancement in fault diagnostic accuracy compared to the conventional method.