Research On Mechanical Fault Diagnosis System Of High-Voltage Circuit Breaker Based On Vibration Signal

High voltage circuit breaker is one of the most important devices in the operation of power system.In this context,this paper performs fault diagnosis of the measured circuit breaker vibration signal by machine learning algorithm and builds a fault diagnosis system.The main research content includes four parts: signal acquisition and decomposition,signal feature extraction,signal classification,and the research of fault diagnosis system.Firstly,a high-voltage circuit breaker vibration signal acquisition platform was built to simulate and acquire four types of fault signals(energy storage spring dislodged,base screw loose,joystick paddle twisted and deformed,and operation mechanism stuck).The original vibration signals are noise reduced by using improved wavelet threshold denoising.A method based on the combination of spectral analysis of the eigenmodal function and center frequency is proposed for determining the optimal number of decomposition layers K in the variational modal decomposition,and the simulated signals are compared with the empirical modal decomposition and the ensemble empirical modal decomposition for analysis.Secondly,two improved methods for extracting the features of high-voltage circuit breaker vibration signals are proposed.One is the method of extracting the envelope feature entropy by combining the variational modal decomposition and Hilbert transform.The second is the improved method of time-segmented energy entropy,which calculates the energy entropy of the signal in the time and frequency domains respectively on the basis of the variational modal decomposition.And the two methods are compared.Then,two improved fault diagnosis models are proposed.Model 1: For the problem that the classification results of support vector machine are greatly affected by the kernel parameters and penalty factors,an optimized support vector machine fault diagnosis model based on the gray wolf algorithm is proposed.The optimal combination of parameters is found by the gray wolf algorithm and input into the support vector machine to achieve accurate identification of vibration signals.The final diagnosis rate is 94%.Model 2: For the problem that the sample distribution of high-voltage circuit breaker vibration signal is not uniform enough,a learning vectorized neural network fault diagnosis model based on single classification support vector machine and sparrow algorithm optimization is proposed.The single classification support vector machine is used to identify the fault samples,and the initial weights of the learning vectorized neural network are iteratively optimized by the sparrow search algorithm to find the best initial weights and input them into the fault diagnosis model for subsequent training and identification.The final diagnosis rate is 96%,which further improves the accuracy of fault diagnosis.Finally,a mechanical fault diagnosis system of high-voltage circuit breaker based on vibration signal was built using Lab VIEW software.The system includes a signal acquisition module,an improved wavelet threshold denoising module,a feature extraction module and a fault diagnosis module.Through the interworking of these modules,the mobile computer online monitoring and diagnosis is realized.