Research On Online Monitoring And Fault Diagnosis Of Mechanical Characteristics Of 10kV Vacuum Circuit Breaker

Circuit breakers are the most important switching devices in power systems,10 k V vacuum circuit breakers are widely used in the system,with a large number of installations and a relatively large number of operations.At the same time,mechanical failure is the most important type of failure.In this paper,the on-line monitoring and fault diagnosis research are carried out for the mechanical characteristics of a springoperated 10 k V vacuum circuit breaker.An on-line monitoring system for circuit breakers was designed to realize automatic control of circuit breakers and automatic collection of information.Opening(closing)coil current signals,travel signals of the spindle,vibration signals,voltage and current signals of energy storage motor,micro switch signals and auxiliary contact signals were collected,and normal working state,opening(closing)coil iron core jam fault,spindle rotation jam fault,opening spring missing fault,these five mechanical states of the circuit breaker were simulated to provide the measured data for the fault diagnosis research.In the hardware design of the lower computer,a separate and modular design was adopted,which can improve the system integration and compatibility,and enhance the interface protection performance of the hardware.The upper computer program was developed based on the Lab VIEW platform,which realized the multi-threaded concurrent execution of the online monitoring software.Due to the existence of impulse noise and small fluctuation noise in the measured signal,moving least squares method was improved to be a filtering method of moving robust regression,which reduced the impact of impulse noise on the noise reduction effect.And the wavelet analysis method and the combination method of moving median filter and moving mean filter were used to reduce the noise of the measured waveform.In order to extract the characteristics of one-dimensional DC signal in time domain,a time-domain feature extraction method based on cumulative derivatives was proposed to extract special points of the waveform.And the short-term energy entropy ratio method was used for the vibration signal to extract the initial moment of the vibration event.The time-domain feature information in various signals was effectively extracted.In order to achieve a more comprehensive reflection of the state of the circuit breaker,multi-source information has been used for diagnosis.In one cycle of circuit breaker closing-opening-energy storage action,74-dimensional features were extracted from the signals.The feature dimension was reduced to 11 dimensions by principal component analysis(PCA),which reduced the linear correlation between multidimensional features.Then the BP neural network was used for classification,and finally the five states of the circuit breaker were distinguished 100% accurately.The successful diagnosis of the faults proves the correctness and feasibility of the work in this paper from hardware design,online monitoring,data noise reduction to feature extraction.