Wear Particles Identification Based On Cooperation Of Grey Relational Analysis And Principal Component Analysis

Abrasion of all kinds of mechanical facilities lead to invalidation, scrap and even the occurrenceof serious accident as well as induce exhaustion of energy and material, terrible waste and theproblem of personal safety. Thereby building the state monitoring of machine abrasion and improvingthe techniques of fault diagnosis have significant economy benefit and social efficiency.Ferrography is a kind of important means to judge and forecast the running state of mechanicalfacilities by wear particles analysis. On account of the abrasion state produced in mechanical facilities,it can efficiently improve the facility of fault monitoring and diagnosis as well as reduce theoccurrence probability of mechanical fault via wear particles intellectual identification. The thesis hasstudied the parameter extraction of wear particles and wears particles intellectual identification whichregards VC++6.0and OpenCv Function-Library as the platform and auxiliary tool respectively.In the ferroscopel image pre-processing process, the integration of image smoothing, medianfiltering, denoising, edge detection, mathematical morphology operations aim at providing a goodbasis for the following ferroscopel image segmentation. Ferroscopel wear particles imagesegmentation is an important step in ferroscopel image processing, image segmentation will directlyaffect the abrasive grit extraction and identification. This thesis contrast and analyze foursegmentation algorithms based on the carefully study and research on wear particles imagesegmentation methods. The thesis finally select the improved Watershed techniques as the imagesegmentation method by taking into account of the adjustment of parameters, the color distribution ofthe same region, the extent of the phenomenon of over-segmentation and the final segmentationresults in segmentation process.The extraction of wear particles characteristic parameters and the type identification of wearparticles are the key intention of this subject research. This thesis firstly studies the shape featureparameters as well as extracts feature parameters in view of large scale of samples. Analyzing andcomparing different feature parameters value of different kind wear particles, we summarize the keyfeature parameters of describing typical wear particles. The thesis studies the algorithms ofPCA(principal component analysis) and gray relative analysis on the basis of the techniques of pastwears particles intellectual identification as well as propose the identification method for oxide wearparticles hardly analyzed, severe sliding wear particles, fatigue wear particles. At the same time, themethod that takes advantage of the combination of PCA and Euclidean distance is also proposed toidentify the red oxide and black oxide wear particles as well as integration of PCA and correlationanalysis to identify serious slip of wear particles and fatigue wear particles. Finally, the thesis throughexamples of the above two methods to verify the identification method has been significantlyimproved compared to traditional identification methods and single intellectual identification method.This method has theoretical and practical significance in view of the promotion of machinery andequipment condition monitoring and fault diagnosis technology based on image information.