Research On Fault Diagnosis Of Compressor Valve Based On HHT And BP Neural Network

With the advancement of science and technology and the development of productivity,the complexity of the operation of machinery and equipment has increase and fault diagnosis technology has also received the attention of everyone.For gas valve fault diagnosis of compressors,Hilbert-Huang Transform,as a newly developed signal processing method with adaptive time-frequency decomposition capability can better extract fault features.BP neural network has strong pattern recognition ability and has unique advantages in solving fault classification.However,traditional BP algorithm has the disadvantages of slow optimization speed and easy to fall into local minimum.For this purpose,this paper introduces Principal Component Analysis and a Genetic Algorithm Particle Swarm Optimization(GA-PSO).The algorithm respectively improves the structure of the BP neural network and the optimization of the algorithm.Based on the collected valve acceleration and vibration signals.The signal features were extracted using Hilbert-Huang Transform and the Principal Component Analysis was used to reduce the dimension.The GA-PSO algorithm was used to optimize the parameters of the BP neural network for the status of the compressor valve.Identification,this paper can be divided into the following sections:(1)Briefly introduced the research background and significance of the subject,analyzed the common fault diagnosis methods and problems of the compressor valve;(2)Introduce two common and typical signal processing methods: Short-time Fourier Transform and Wavelet Transform.Combining with the simulation analysis of the compressor valve,the limitations of the two methods are summarized.(3)Based on Hilbert-Huang Transform,a compressor valve feature extraction method is proposed: Firstly,the noise reduction step is performed and then the energy characteristics of the Intrinsic Modal Function component modal signal and the characteristics of the marginal spectrum between the cells are extracted by the Hilbert-Huang processing method.The characteristics are enough to be eigenvectors.Finally,Principal Component Analysis is used to reduce the dimension of the eigenvectors.(4)The artificial neural network is briefly introduced,and the principles,algorithms,etc of the BP neural network are introduced.Use the feature extraction method based onHilbert-Huang Transform to obtain the feature vectors of four working conditions,and use the standard BP neural network to identify and classify all kinds of faults of the compressor valve.Finally,the recognition results are analyzed and the limitations of the standard BP algorithm are summarized.(5)The GA-PSO algorithm is proposed to optimize the BP neural network.The GA-PSO algorithm is obtained by integrating Genetic Algorithm into Particle Swarm Optimization.The new algorithm combines the advantages of the Particle Swarm Optimization and the Genetic Algorithm.Parameter optimization process for BP neural networks.Finally,by using the three networks are standard BP neural network,PSO-BP neural network,GA-PSO-BP neural network for fault diagnosis of the compressor valve.The test verified that the GA-PSO-BP neural network is superior to the compressor valve fault diagnosis.In the other two networks,it is feasible.(6)Summary.