Turning Surface Roughness Measurement Based On Image Texture Analysis

Surface roughness is an important index for the surface quality of parts,affect the use of parts performance,appearance and longevity directly,especially for special functions parts(seal,relative movement).Therefore,we must measure the surface roughness accurately.Turning workpiece surface roughness non-contact nondestructive detection was realized through the image texture analysis method based on the theory of computer vision.The main contents of this paper are as follows:(1)The measurement system hardware platform was built with VHX-1000 type super depth 3D microscopy system as the core,the turning surface clear microscopic images were captured.(2)Turning surface texture statistical analysis and feature extraction was implemented based on gray level co-occurrence matrix(GLCM).Firstly,according to the cutting surface image characteristics,determine the optimal structure parameters of GLCM,and provide accurate data for the subsequent feature extraction.Secondly,extract the 14 surface texture based on gray level co-occurrence matrix feature parameters by the principle of turning and turning surface image feature.The feature parameters: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 II and Maximum Correlation Coefficient.To establish the database model between GLCM matrix statistical characteristic parameter and the corresponding surface roughness.(3)The test model of turning workpiece surface roughness Ra was built by multiple regression analysis method,the quantitative calculation of the turning workpiece surface roughness was realized.The multivariate linear regression testing model and multivariate nonlinear regression model were established respectively,fitting the workpiece surface texture characteristic parameters and the mathematic expression of the workpiece surface roughness Ra mapping relation.Both multiple regression analysis methods have high precision and the accuracy of measurement meets the requirement by the samples tested.Experiments show that,the nonlinear multiple regression detection model for detecting precision is better than the linear multiple regression model.(4)The turning surface roughness detection model of Ra based was establish on BP neural network algorithm..The parameter turning surface image texture features were as the input and the corresponding surface roughness value of Ra as the expectations output,the neural network detection model was built.Experiments showed that the BP neural network detection model has higher accuracy,and the accuracy was higher than the multiple regression model.(5)The software design of turning surface roughness detection system was completed based on MARLAB software development platform,and also the development of the graphical user interface.Finally,the application of the detection system for detection model was investigated experimentally,and then the contrast analysis were carried between the detection results and the traditional probe type measuring results,the results show that the average error rate of all detection models are in the allowed range,meet the measurement requirements,the turning surface roughness can be quickly accurately and non-destructive detected.