Mechanical Fault Diagnosis of High Voltage Circuit Breaker Based on Autoencoder and Metric Learning

In the field of mechanical fault diagnosis for high-voltage circuit breakers, the application of metric learning has proven effective in analyzing datasets characterized by a limited number of samples. However, its efficacy tends to diminish when dealing with high-dimensional data. This study presents an autoencoder framework tailored for efficient feature extraction and data dimensionality reduction, thus harmonizing the strengths of autoencoders with metric learning to enhance the classification accuracy of high-dimensional datasets with scarce samples. Additionally, the integration of the K-nearest neighbor algorithm as a metric learning tool facilitates the effective matching and classification of sample data. Experimental findings indicate that the methodology introduced in this research yields a 6.6% improvement in classification accuracy compared to conventional metric learning techniques in the context of fault diagnosis for high-voltage circuit breakers. This research not only aids in enhancing the accuracy of high-voltage circuit breaker fault diagnosis but also offers a new perspective for machine learning applications involving small samples and high-dimensional data.