The Research For Fault Diagnosis Of High-Voltage Circuit Breakers Based On Vibration Signal Processing

High-voltage circuit breakers are essential control and protection equipment for power plants and substation power distribution devices.They can protect generators,transmission systems,power distribution systems and their electrical equipment.High reliability is the basic requirement for high-voltage circuit breakers.Thus it is necessary to monitor the status of the circuit breakers.The vibration signal generated during the operation of the high-voltage circuit breakers contains very rich mechanical state information.Therefore,this paper is mainly in relation to faults diagnosis of high-voltage circuit breakers based on mechanical vibration signal processing.The research object is ABB 12 kV iVD4 intelligent high-voltage vacuum circuit breaker,and the research is focused on feature extraction and fault identification of the collected vibration signal.Firstly,the existing after-sales data is investigated to determine the actual operation of the relatively high incidence of several typical mechanical failures—unsmooth transmission chain,lack of lubricant and relax of opening spring.Under unload condition of the circuit breakers,the fault combination of the above three kinds of faults is simulated and the fault implantation test is carried out to obtain the vibration signal.The obtained signal samples are pretreated and manually annotated for subsequent feature extraction and fault identification.Secondly,two novel methods of vibration signal feature extraction for highvoltage circuit breaker based on vibration signal processing are studied.One is to use singular value decomposition to determine the number of signal modes,and thus to do feature extraction by using variational modal decomposition.The first-order modal center frequency,the first-order modal energy and the peak-to-peak distance of the vibration signal are used to form the eigenvector.The other is to assume that each modal corresponds to a spectrum of Gaussian function,and the Gaussian mixture model(GMM)and the expected maximization algorithm are proposed to do spectrum clustering and thus to determine the variance,mean value and the weight of each modal,which are used to form the eigenvalues of the vibration signal.Nonlinear State Estimate Technology and Support Vector Machine are applied to classify eigenvalues extracted by the above two methods.The classification results are compared.Finally,aiming at the difficulty of feature extraction,the short-time Fourier transform is used to convert the vibration signals into time-frequency maps,and the convolution neural network is used for image training and classification.The experimental results show that all the above novel methods can realize the effective diagnosis of common mechanical faults such as unsmooth transmission chain,lack of lubricant and relax of opening spring.Therefore,this study has a certain practical value.