Research Of Mechanical Vibration Signal Feature Extraction And Fault Diagnosis Of High Voltage Circuit Breaker

In recent years,the power generation and capacity of China continue growing although they have been the first one position in the world.Moreover,the rising of capacity must lead the increasing of power devices,such as high voltage switches.High voltage circuit breaker(HVCB)with the function of protection and isolatio is one of the most important switching equipments in regional power grid.So it is the key to guarantee the safty and reliability in electric power system because it undertakes the tasks of controlling and protecting.The mechanical vibration signal of HVCB contains abundant information and status of equipments.And the types and states of the mechanical faults of HVCB can be recognized by monitoring the mechanical vibration signal.Because the low temperature of winter in Heilongjiang province,the working condition of HVCB is very bad and the individual faults frequently take place.And the operation frequency is usually very low to ensure the safty of electricity.Therefore the maintenance data of HVCB is not enough to establish the diagnosis model.In addition,the analytical model based on the fixed HVCB is not generalized because of the differences among HVCBs.In this thesis,the vibration signal acquisition and fault diagnosis are researched based on the vacuum circuit breaker whose type is titled as ZW32-12.A platform for signal acquisition is set up and the feature extraction methods of HVCB based on wavelet packet-correlation dimension and EEMD(Ensemble Empirical Mode Decomposition)-characteristic entropy are respectively proposed and simulated.Neural network and support vector machine are also used to classify different kinds of discharge,and the validity and generalization of the feature quantity are comprehensively studied.The main works and results of this thesis are as follows:(1)A platform for mechanical vibration signal acquisition is set up,and the selected sensor with good temperature characteristic can adapt to the cold winter of Heilongjiang region.The platform not only acquire the mechanical vibration signals of the operation mechanism,but also the signals of displacement/velocity,opening/closing coil current and opening(closing)current of ZW32-12.The management system software is jointly developed by C++ Builder and Matlab,and the hardware circuit is composed of DSP(Digital Signal Process)and its peripheral circuits.The CAN(Controller Area Network)bus serves to transmit the data between the management unit and the circuit.(2)The model of feature extraction of HVCB is proposed by combining the methods of wavelet-packet and correlation-dimension.Firstly,the vibration signals with noise-reduction are decomposed by three layers of wavelet-packet and the coefficients of the decomposed signals are reconstructed.Then,the phase space is also reconstructed based on the time-delay method.And the delay time is determined by the C.C algorithm,and the embedding dimension is gained by the improved False Nearest Neighbors approach.The correlation dimension is calculated by the G-P(Garsesrger-Procacai)algorithm and it is used as the feature vector.Additionally,the simulation and testing of three conditions,i.e.,the insufficient self lubrication of HVCB,the loose screw and the invalid spring of the buffer,are carried out and analyzed.The experiment results indicate that the proposed method can effectively acquire the characteristics of the vibration signal of HVCB.(3)The model of extracting feature vector is proposed by combining EEMD and information entropy theory.Firstly,the signal with noise-reduction is decomposed based on the method of EEMD,and the main four intrinsic mode functions(IMF)are selected and processed by Hilbert transform.The analytic signals are then obtained and the corresponding envelopes are solved,respectively.Then the analytic signal envelope is divided into m segments along the time axis.Eventually the EEMD-feature entropy vector is obtained by calculating and normalizing the energy of each segment.In addition,the methods of EMD(Empirical Mode Decomposition)and EEMD are respectively adopted to extract the features of the vibration signals.The simulation results indicate that EEMD can effectively solve the mode mixing phenomenon when EMD is merely used.(4)The optimal classification-method is analyzed to solve the problems from the small action samples and the poor generalization rate of the recognition model.Based on the samples from wavelet packet-correlation dimension and EEMD-feature entropy,the classification and identification of signals are respectively conducted by BP(Back Propagation)neural network,particle swarm optimization of BP neural network,RBF(Radial Basis Function)neural network,wavelet neural network and support vector machine(SVM).The obtained classification results show that the highest recognition rate of wavelet packet-correlation dimension features is gained by SVM model,and the values in some states reach 100%.Comparatively,the recognition rate of EEMD-feature entropy is similar under the above classification methods,up to 90%.The testing identification results show that,the fault prones of HVCB,such as insufficient lubrication and invalid spring,are resulted from fact that the very low temperature of Heilongjiang region in winter.