Defect Recognition Of Ring Parts Based On Improved Multifractal Detrending Fluctuation Analysis

In industrial production,the feature extraction,classification and defect recognition of mechanical parts images are usually based on the geometric features of the parts in the images,and there is no deep exploration of the classification and defect recognition features of the parts.As an important branch of fractal,multifractal,especially the Multifractality Detrended Fluctuation Analysis(MF-DFA)algorithm,can start from both the global and local aspects to deeply explore the various characteristics of the target object.In addition,the calculation results of the MF-DFA algorithm have high accuracy and are suitable for processing non-linear signals,while the images of mechanical parts are mostly non-linear.Therefore,this thesis takes parts with good fractal characteristics(rings,bearings,nuts,gears)as an example,and uses the MF-DFA algorithm for feature extraction of part images to realize the classification and defect recognition of part images,which has practical significance and value.The main contents of this thesis are as follows:(1)In terms of feature extraction of parts images,an improved MF-DFA algorithm is proposed.The traditional MF-DFA algorithm is improved by using sliding window technology,Empirical Mode Decomposition(EMD)and triangular rotation overlay model.Compared with the traditional MF-DFA algorithm,the improved MF-DFA algorithm can more accurately describe the characteristics of the image and provide more accurate data for image recognition.(2)Research and analyze the multifractal characteristics of part images.The four types of parts images with prominent fractal characteristics such as gear ring,bearing,nut and gear are taken as the research object.Use the improved MF-DFA algorithm to calculate and analyze the generalized Hurst exponent h(q)of these four types of parts images.The results show that these four types of parts images have multifractal characteristics.The improved MF-DFA algorithm can describe these four types of parts well.The classification characteristics of the image.(3)Analyze and extract part image classification and defect feature values based on the improved MF-DFA algorithm.The improved MF-DFA algorithm is used to calculate the multifractal spectrum of these parts images,and seven multifractal spectrum characteristic values of four types of normal parts and defective gear parts images are extracted.Kernelized Principal Component Analysis(KPCA)is used to perform dimensionality reduction processing on these data to obtain accurate part image classification and defect feature values.(4)Based on the improved MF-DFA algorithm,classify the part image and identify the defect.Choose Least Squares Support Vector Machines(LS-SVM)as part image classification and defect recognition method,transform the part image classification and defect feature values into part image classification and defect training set and test set respectively,Obtain the results of part image classification and defect recognition.After prediction,it is found that the accuracy of part image classification and defect recognition is high,which proves that the improved MF-DFA algorithm can well express the classification and defect characteristics of part images.