Fractional Domain Analysis And Defect Classification Method For Commutator Visual Inspection

The commutator is the core component of the motor,so quality inspection is a critical process in the commutator production process.At present,commutators are mainly based on manual testing in many production lines.With the increase of labor cost and manual testing,the detection standards are different,the detection efficiency is low,and the quality of inspection cannot be guaranteed.Therefore,the use of machine vision technology to develop automated commutator inspection production lines is an inevitable trend.The commutator is a three-dimensional cylindrical product whose metal surface defects are mainly stains,scratches and bumps hidden under the oxidized surface on the side of the commutator.In order to solve the problem of defect detection under the surface oxidation of commutator,a new method based on fractional domain commutator for metal oxide surface defect detection is proposed.The main work is as follows:1.For the image acquisition of the commutator side of the cylindrical shape,this paper uses the combination of linear array camera and coaxial line light source to control the motor to make uniform rotation motion by PLC,while the camera scans the commutator side in uniform rotation.,get a two-dimensional commutator side image.2.Extraction and segmentation of the area to be detected on the side of the commutator.First,median filtering is used to eliminate the noise generated in the acquisition image acquisition.The side metal region is then obtained by image segmentation.Finally,the edge fitting method is used to further extract the lateral detection area to completely eliminate the interference of the non-interest area.3.A fractional domain analysis method based on fractional Fourier transform is proposed for the background interference of metal surface of commutator.Firstly,the image is transformed from the spatial domain to the fractional order domain by using the fractional Fourier transform.The features of the different transform order images in the fractional order domain are also different.The transform order is determined according to the characteristics of the commutator image.coefficient.Finally,the ADF filter is used to set different f,g,and h parameter sizes to eliminate dissimilar image blocks,thereby achieving the purpose of weakening oxidative interference and highlighting the defect area of the commutator.4.Based on the Random Forest classification algorithm,the defect characteristics are analyzed based on the image preprocessing,and the features such as gray value,variance,tightness and circularity of the defect are selected as the basis for classification.Then through the simulation and actual test of Random Forest on the computer,the classification accuracy of the classifier can reach more than 92%,which is greatly improved compared with other image preprocessing methods in this paper.5.Finally,on the Win7 platform,simulations and related tests were performed in applications developed based on Microsoft Visual Studio 2015 and MATLAB 2012 a.The simulation experiment has achieved the expected effect,and the defects proposed above can be accurately detected in the experimental system to meet the system design requirements.