Research On Visual Detection Method For Surface Defect Of Ceramic Valve Core

During the batch production of ceramic valve core,cracks,small blobs,corner and edge breakage,insufficient polishing and other defects may appear on the surface of the valve core,and the defective rate can reach 20% in extreme cases.Compared with the artificial detection method,machine vision technology has the advantages of non-contact,fast speed and high stability.To diminish the influence of different defects scales on the system sensitivity,and reduce the impact of image grayscale uniformity,a detection method for ceramic valve core surface defect is proposed.The main research contents are as follows:(1)The development of ceramic products surface defects detection technology is reviewed.Characteristics and visual inspection difficulties of ceramic valve core surface defects are analyzed.The ceramic valve core surface defects are graded into first-class and second-class defects,defects detection system and grading-detection method are designed.During the image preprocessing,usual methods of ROI extraction are compared by experiments,and iteration is chose for the ROI extraction process.The peak signal to noise ratio and normalized mean square error being used as evaluation criteria of filtering algorithms,and the median filter with M=3 is selected to remove the image noise.(2)A SVM defect classifier for the detection of first-class defects is designed.Geometric and gray features of ROI are extracted,and dimension reduction of image features are conducted by PCA.The PSO algorithm is used for the optimization of classifier parameters.And the first-class defect detection process is established.(3)The second-class defect detection process is firstly established.Least squares method is applied to ROI fitting,and the fitting function is determined based on the balance of fit goodness and run duration.Then fitting result and the original image are processed by image difference technology to remove image grayscale uniformity caused by illumination and polishing.Experimental results of common edge extraction methods are compared,and Kirsch edge extraction algorithm is chose for the obtaining of defect candidate regions.Image features being selected,construction of the second-class defect SVM classifier and parameter optimization by PSO are conducted for the detection of second-class defects.(4)Experiments are conducted for the detection of first-class and second-class defects,the results show that the detection accuracy of PSO-SVM satisfies the demands well.To verify the stability of PSO-SVM classifier detection accuracy,three intensities of light are introduced to simulate the potential illumination fluctuations during on-line detection,and the experiments results indicate that the detection accuracy varies little under different light.And mixed defects detection experiment is performed to simulate defects random distribution during on-line detection,and the experiment results prove that the detection speed and accuracy meet the requirement of the system.The above experiments verify that the ceramic valve core surface defect detection algorithm is of good feasibility and effectiveness.The on-line application results show that the detection accuracy and speed of several Φ35 mm ceramics valve core types satisfy the expected requirements,and the designed method has the significance of application and popularization.