Research On Feature Extraction And Recognition Of Mechanical Vibration Signal Of High Voltage Circuit Breaker

High voltage circuit breaker is one of the key devices of power system,due to the influence of outdoor environment or the quality of the mechanical components,mechanical failure occurs constantly,and therefore economic will losse.The timefrequency characteristics of mechanical vibration signals can reflect the operation status of machinery.Therefore,domestic and foreign scholars have carried out a lot of researches on fault diagnosis of high voltage circuit breakers based on vibration signals.However,the effective extraction of signal features and the accurate identification of faults have always been difficult to study,and no significant breakthrough has been made.Therefore,in this paper,the vibration signal acquisition platform based on virtual instrument technology is designed to collect four types of vibration signals(Normal state,the screw of the base,delay fault,one of the energy storage spring lossing).The method of feature extraction and recognition of switching vibration signal of high voltage circuit breaker is studied.The main contents of the research include:(1)Signal denoising and time-frequency analysis.For the measured data,the improved threshold function is used for noise reduction,compared with the soft threshold and hard threshold method,the proposed method can effectively ensure the smoothness and mutation information of the estimated signal.Then the empirical wavelet transform and variational mode decomposition are selected to decompose the measured signals,and the selection of the eigenmode number of the measured signals is discussed in this paper,compared with the time-frequency resolution performance of empirical mode decomposition and ensemble empirical mode decomposition,the methods used in this paper can accurately obtain the true narrowband eigenmode component of the signal.(2)Feature extraction of signal.In order to quantify the time-frequency characteristics of the signal,two new feature extraction methods are proposed in this paper,an improved time-frequency entropy method and a new time segmentation energy entropy method are presented.Compared with traditional feature extraction methods,the proposed methods can better characterize the time-frequency characteristics of signals.(3)Classification of signal.It is difficult to obtain a large number of vibration samples due to the operate of circuit breakers infrequently,and the number of signal samples is uneven.This paper provides two feasible schemes for identifying mechanical vibration signals of circuit breakers,and the classification performance indicators of the recognition schemes are discussed in detail.Scheme 1 uses a multi-classification support vector machine to identify four types of vibration signals,and the classifier parameters are determined by random search optimization to reduce the recognition time overhead.The accuracy of scheme 1 is 95%,and the recall rate of all failure samples is 96.67%.In order to comprehensively consider the recall rate,accuracy performance indicators and the uneven distribution of sample size,scheme 2 creatively uses single-class support vector machine to detect fault samples to the maximum extent,then the probabilistic neural network optimized by cyclic traversal is used to identify the specific fault types,the accuracy of the model is 95%,and the recall rate of all fault samples is 96.77%,compared with the traditional fault diagnosis model,on the premise of guaranteeing accuracy,the proposed classification model can avoid the failure samples being mistaken as the normal samples and missing the optimal maintenance time of the equipment to the greatest extent.