Workpiece Surface Roughness Detection Based On Machine Vision

Surface roughness is an important indicator of surface quality and has a direct influence on the service performances,lives and safety of machine parts,especially for special parts(seal,relative movement).Therefore,detecting the surface roughness of machine parts fastly,accurately and nondestructively is very important for the application of the parts and the safety of the system.Based on the theory of machine vision the non-contact and nondestructive detection for face milling surface roughness is achieved by the image texture analysis for visual inspection of the face milling surface roughness.The main contents of this thesis are as follows:(1)The hardware platform for visual inspection of the face milling surface roughness was built which consisted of VHX-1000 type super depth 3D microscopy system,CCD camera and computer.The surface images of face milling could be captured by this system.(2)Original images of the surfaces of machine parts were gathered preprocessed,then grayed,denoised and rotated.Denoise processing is achieved by median filtering.Image was rotated quickly and accurately by Hough transform and bilinear interpolation.(3)The texture statistical analysis and feature extraction of the face milling image in a specific direction were conducted by Hough Transform and GLCM.The optimal eigenvalues for GLCM were confirmed.The 14 statistical parameters based on GLCM,including Angular Second Moment,Contrast,Correlation,Differential Moment,Inverse Difference Moment,Sum Average,Sum Variance,Sum Entropy,Entropy,Difference Variance,Difference Entropy,Correlation Information Measurement I,Correlation Information Measurement ? and Maximum Correlation Coefficient were extracted and analyzed according to surface textures.The database model is established between statistical characteristic parameters of GLCM and surface roughness.(4)The relationship between the 14 GLCM based statistical parameters and the roughness texture direction and roughness level was analyzed.The GLCM eigenvalues of image were analyzed,and the result showed that GLCM eigenvalues of the image along the texture direction can get good result.The four eigenvalues,contrast,difference moment,sum variance and difference variance were selected to characterize the level of surface roughness.Hough transform was applied to detect surface information and obtain the value of ?.The vertical texture image was obtained by the image rotation.Four eigenvalues of GLCM were calculated along the vertical direction,then the characteristics of the image were extracted.(5)The detection model for the face milling surface roughness Ra is established by BP neural network algorithm,which was tested by taking the texture parameters of face milling surface as input and the Ra as output.The result showed that the absolute average error between test results and actual value was not more than 0.10?m,and the average of relative error was less than 3.8%.Experiments showed that the detection error for the detection model was with the allowance and meet precision requirement,and the detection model could detect surface roughness fastly,accurately and nondestructively.It has a great significance on the study of fast non-contact with face milling surface roughness.