Component Classification And Defect Identification Based On Multifractal

Mechanical flexible manufacturing technology can meet the diversity and individual needs of the product.The necessary detection link in the production process is to accurately classify the parts and identify the defective parts.The general method of parts classification and identification based on visual features is simply to extract the geometric features of the objects to realize the classification and identification.There is no analysis of the fine structure of parts from the perspective of different scales.Personalized mechanical parts in the context of flexible manufacturing have some characteristics of irregularity and self-similarity.Fractal theory is an effective tool for describing the distribution of singular nonlinear structures.It is suitable for the study of complex mechanical parts structure images that are difficult to model.In essence,multifractal can fully consider the different levels of local and global aspects of parts images,and describe the parts of the image structure information in a comprehensive and effective way,so as to make up for the fact that a single fractal dimension cannot be from different scales at the local and global levels.Analyze the disadvantages of the fractal structure of complex parts images.In recent years,combined with multifractal theory and pattern recognition,the accurate identification and classification of mechanical parts have been continuously applied in the automatic production process.Based on multifractal theory,this paper analyzes several different types of mechanical parts classification and defect identification.The main research content of the paper is as follows:(1)Based on the box method,an improved triangular box method for calculating multifractal spectrum is proposed.The multi-fractal spectrum is obtained by calculating the standard library image by two algorithms.The multi-fractal spectrum calculated by the improved triangular box method is compared.The characteristic values are more distinct,the distinguishability is higher,and the comparison after feature extraction is favorable.(2)For different types of mechanical parts(four types of mechanical parts such as ring gears,gears,screws,and nuts),use the triangular box method to calculate the multifractal spectrum of the images of the four types of parts to describe four different types of parts images.Features,extracting the seven kinds of multifractal spectral eigenvalues of the four types of parts images.Based on this,the principal component analysis method is used to obtain the normalized weight distribution model,new identification eigenvalues and their distribution,as a basis for differentiating four different types of parts images: gear ring,gear,screw,and nut.In order to achieve the classification of four types of parts to identify.(3)For images of normal and defective mechanical parts of the same type(using tooth-ring parts as an example),calculate the multifractal spectrum information of normal parts and defective parts and extract the 7 kinds of eigenvalues of multifractal spectrum.Singular values and their corresponding multifractal spectral values are used as identification feature quantities.The design Bayesian classifier was chosen as the pattern recognition classification algorithm.After the training of the a priori data,the normal and defective tooth ring parts could be identified more accurately.