Low Frequency Vibration Signal based Low Memory Fault Classification Algorithm for C-GIS Switchgear Operator

This study proposes a vibration signal-based fault classification algorithm for the vacuum circuit breaker (VCB) of cubicle type gas-insulated switchgear (C-GIS), considering both sensor type and installation location. Vibration impacts generated during close and open operations are detected, transformed via FFT, and only the low-frequency components are used as input to a fully connected (FC) classifier. Compared with prior models such as 1D CNN, ResNet, and ResNet+DSC, the proposed FFT+FC approach achieved the highest classification accuracy when using an appropriate sensor type that does not suffer from overload and placing it at the optimal location determined by the Optimal Sensor Placement (OSP) algorithm. Experiments demonstrated that sensor type and placement have a substantial influence on the performance of FFT+FC, with overload, caused by the limited measurement range of certain sensor, leading to waveform distortion and accuracy degradation. These findings underscore the importance of both sensor selection and placement strategy in achieving reliable C-GIS fault diagnosis. Focusing on model type of fault classifier, using datasets collected from two structurally different C-GIS units, the results confirmed that the FFT+FC model not only provides competitive classification accuracy under optimal conditions but also requires significantly lower activation memory than convolution based models, making it suitable for resource limited industrial applications.