Mechanical fault diagnosis of high voltage circuit breakers utilizing empirical wavelet transform and one-class support vector machine

According to the actual engineering requirement of high reliability in the fault diagnosis of circuit breakers, this paper proposes a new mechanical fault diagnosis method of high voltage circuit breakers utilizing empirical wavelet transform (EWT) and one-class support vector machine (OCSVM). First, the method uses EWT to separate the intrinsic mode functions (IMF) containing different physical significances in the vibration signals of circuit breakers accurately. Then, Hilbert spectrum analysis is used to obtain the time-frequency matrix and compute the time-frequency entropy, which constitute the feature vector used for classification. Afterwards, only the easily obtained normal vibration signals are used to train the OCSVM with the constant parameters optimized by particle swarm optimization (PSO); and the OCSVM is used to accurately judge if the circuit breaker mechanical fault occurs and improve the fault diagnosis reliability,. If OCSVM judges that a mechanical fault occurs, the support vector machine (SVM) is used to further recognize the specific fault type. Experiment on a SF₆ high voltage circuit breaker was conducted, the result proves that the proposed new method can differentiate the fault samples from normal samples more accurately. Thus, the new method can satisfy the requirement of high reliability for the mechanical fault diagnosis of high voltage circuit breakers.