Fault Diagnosis Of Reciprocating Machinery Based On Image Recognition

Reciprocating compressor structure with more excitation source is complex,the vibration signal performs stronger non-linear and non-stationary,its fault feature extraction is much more difficult,currently,there is a lack of mature diagnosis method,in view of this situation,this paper presents a intelligent diagnosis method based on image recognition,this method uses reciprocating compressor vibration parameter graphs as the study object,bypassing the complex fault information of the image,but converts to the distinction between the fault images,by extracting the texture feature parameters in the image,uses fuzzy kernel clustering method to identify the fault samples with texture features,in order to realize fault intelligent classification.Around the key technology:signal processing method,image feature extraction method,artificial intelligence classification algorithm are studied,including the following aspects:1.In view of the reciprocating compressor vibration signals are non-linear and non-stationary,using the methods of time domain analysis,frequency domain analysis and time frequency analysis to obtain time-domain waveform,power spectrum,smoothing pseudo Wignal distribution time-frequency graph,after contrasting analysis of the treatment,smoothing pseudo Wignal distribution time-frequency graph is ultimately selected as the further study object of image feature extraction.2.This paper discusses the image texture and the method of feature extraction,and studies the gray histogram,gray co-occurrence matrix and the method of gray level-gradient co-occurrence matrix texture feature extraction,also proposes the method for fault diagnosis of reciprocating compressor based on gray level-gradient co-occurrence matrix.This method accurately describes gray distribution probability and statistical correlation of reciprocating compressor vibration parameter image,and gradient distribution of each pixel and texture features of the image,effectively extracts the image texture feature information.3.From the angle of image recognition used BP neural network to diagnose reciprocating compressor vibration signal,this paper introduces the basic theory of neural network,studies the structure and design of BP network and the BP learning algorithm,finally,the BP neural network is applied to fault diagnosis of reciprocating compressor valves,in order to realize valve intelligent identification.4.The paper proposes a method for fault diagnosis of reciprocating compressor based onfuzzy kernel clustering,uses vibration time-frequency graphs as the research object,inputs the texture feature quantities as samples to Gauss feature space through nonlinear mapping,clustering in the high dimensional feature space,in order to obtain higher accuracy results.