Fault identification of high voltage circuit breaker trip mechanism based on PSR and SVM

The trip mechanism is a weakness in circuit breakers. Traditional fault identification based on the coil current is difficult to report early mechanical defects such as coil-plunger jam. Here, the vibration signal during the trip process was extracted. Based on the coil current signal and vibration signal, the characteristics of the trip mechanism are analyzed. The phase space reconstruction (PSR) method is used to extract features from the vibration signal. Combined with the features from the coil current waveform, the feature set representing the health condition of the trip mechanism is proposed. The fault simulation tests are carried out and the variation of current vibration characteristics under fault conditions is studied. The fault identification model based on a support vector machine (SVM) is proposed and compared with the identification results when features are extracted from a single signal. When the power supply voltage is dispersed, the prediction accuracy of fault identification is 83.3% considering only the features of the current waveform or vibration signal. And the identification accuracy rises to 96.7% while using the feature set of current and vibration signals. On basis of the current signal, the method further combines the vibration signal so that the robustness of defect identification improves.

Automated summary

In this paper, a method for fault identification of trip mechanism based on vibration signal and coil current signal is proposed. Features are extracted from the vibration signal using phase space reconstruction (PSR) method, and combined with the features from the coil current waveform to form a feature set representing the health condition of the trip mechanism. Fault simulation tests are conducted to study the variation of current vibration characteristics under fault conditions, and a fault identification model based on support vector machine (SVM) is proposed. The identification accuracy of fault identification is 83.3% considering only the features of the current waveform or vibration signal, and the accuracy rises to 96.7% when using the feature set of current and vibration signals.