Research on Fault State Evaluation of Fast Vacuum Switch Based on Vibration Signal

Abstract

The fast vacuum switch is the core component of the mechanical HVDC circuit breaker. In order to accurately evaluate the operation state of the fast vacuum switch with the bidirectional coil-metal disk structure, an experimental test platform was built in this paper to simulate the failures such as screw loosening and jamming. The vibration signals during opening under different conditions are decomposed based on the wavelet packet transform (WPD) and the Hilbert-Huang transform (HHT), and four different eigenvectors were established. Then, three state evaluation models of BP neural network, support vector machine (SVM) and random forest (RF) were built, and the influence of different training and validation sets on the model was analyzed. Finally, three models are trained based on the same training set and validation set, whose features and prediction accuracy are compared and analyzed. The results of the identification studies on two different mechanical faults show that the composite eigenvector construction strategy has high prediction accuracy in fast vacuum switch state evaluation. In addition, the prediction accuracy and stability of the fast vacuum switch state evaluation model based on RF are high.