End-to-end Fault Diagnosis of High-Voltage GIS Devices Leveraging Masked Auto Encoder

The runtime fault diagnosis of High Voltage Gas Insulated Switches (HV-GIS) is essential for ensuring the safety and reliability of electrical systems. Traditional fault diagnostic algorithms predominantly rely on statistical feature extraction combined with machine learning classification methods. However, these approaches often underperform in complex conditions, such as various aging and working environments. To address this limitation, we propose a fault diagnosis framework for HV-GIS based on masked auto encoder (MAE) pretraining. Our framework renders images from sensing data and approaches the diagnosis task as an image classification problem and utilizes MAE pretraining to achieve generalized fault diagnosis. Experimental results indicate that the diagnosis model trained with our framework outperforms baseline models, offering a more effective and generalized solution for HV-GIS fault detection.