Health status centered mechanical feature extraction for high voltage circuit breakers

Mechanical characteristics, including the displacement curves of the movable contacts and coil current curves are the most common routine monitoring objects of high voltage circuit breakers to evaluate the machines' condition. Generally, a high-performance mechanical characteristic tester has the ability to offer dozens of parameters consisting of stroke, speed, magnitude of current and so on. Besides, lots of new features have been proposed for specific needs. So choosing useful features from all the features above becomes an inevitable problem. However, most of the features extracted are focusing on fault diagnosis and rare attention has been paid to the health condition evaluation. Here, a new health status centered mechanical feature extraction framework is proposed. Firstly, a large-scale feature selection is carried out among 44 closing features based on monotonicity and consistency. Then the most sensitive ones are fed into a support vector regression(SVR) model for predicting the remaining useful life. Real data collected from several high voltage circuit breakers of full life circles were used in the experimental studies, with the results showing the superiority of the extracted features.