Mechanical fault diagnosis of high-voltage circuit breakers using multi-segment permutation entropy and a density-weighted one-class extreme learning machine

Abstract

Condition monitoring for high-voltage circuit breakers (HVCBs) is of great significance for the safety of power grids. Based on machine-learning methods, most relevant studies have contributed significantly to improving the classification accuracy of known states. However, these studies have neglected the detection of unknown faults. In this study, a new one-class classifier, called a density-weighted one-class extreme learning machine (DW-OCELM), was proposed to detect unknown faults of HVCBs. The DW-OCELM determines the classification boundary considering data distribution by introducing the notion of density weight, such that samples located in low-density regions are more likely to be separated, improving detection performance. On this basis, a multi-class classifier was developed based on the homogeneous combination of multiple DW-OCELMs to classify known states. In addition, the proposed classifiers were trained based on multi-segment permutation entropy calculated from vibration signals. Experiments on a 35 kV HVCB demonstrated that the proposed methods outperformed other state-of-the-art techniques.