Surface Texture Inspection of Machined Surfaces in Computer Integrated Manufacturing

Abstract

Engineering surfaces originated by a diverse manufacturing processes like casting, forging, and secondary processes like turning, milling, etc. Technological advances in the area of manufacturing have resulted in closer tolerances on the parts being produced by all these processes. This, in turn, has resulted in a need to evaluate and control the surface finish of the product which has a significant role to play in its function. Surface roughness evaluation helps in predicting the functionality of machine components. Surface roughness assessment techniques broadly classified into contact type & non-contact type. Many of the contact type techniques use a diamond stylus during measurement though these methods give accurate results they are very slow and hence they can be used only for sampling inspection of surfaces. To overcome these drawbacks, non-contact methods came into the era. Many of these methods are electro-optical in nature and have high measuring speeds, however, the use of these methods are restricted to laboratory inspection only. With the advent of low-cost microcomputers, researchers have started use of vision system based non-contact techniques for surface roughness evaluation. However many of the approaches developed so far do not give complete information for the complete characterization of the surface. In this context, the proposed methods assume special significance. The developed vision system consists of a personal computer (server) with a built in advanced image processing card and charge-coupled device (CCD) camera to capture the image of a properly illuminated machined surface. The surface image is stored in the server. The integrated software, installed in the server, helps in processing the surface image for noise removal and for assessing the surface texture. The proposed method is amenable to automation. Near to one hundred percent inspection of components is possible. As the method is non-contact in nature, high measuring speeds are possible. The identification of surface texture images from machined surfaces using the image-processing technique has been a prominent research area in the past few decades. The aim of this study is to identify various machined surface texture images using machine learning techniques. The charge-coupled device is used to grab machined components images. Based on captured images twelve statistical features are extracted to construct feature vector. Algorithm based on Grey Level Co-occurrence Matrix (GLCM) utilized in statistical features extraction from the machined surface images. Four machine learning algorithms such as Random Forest, Support Vector Machine, Artificial Neural Network and J48 are utilized to characterize machined surfaces. Training and Ten-fold cross-validation processes are utilized to identify machining processes. It is found that Artificial Neural Network and Random forest gives 100 % training accuracy and 99 % cross validation accuracy. Results demonstrate the efficiency of proposed methodology which is useful for identifying texture images. In this work, Image texture features of machined surfaces are extracted through machine vision system. The image texture features are extracted using GLCM algorithm with regard to correlate with discint roughness parameters recorded by contact type surface profilometer. The image acquisition is done at different roughness levels for feature extraction. The variation between each feature and roughness parameter is investigated. Multiple regression models are developed subjected to estimation of surface roughness parameter (Ra) and the qualitative detection of the degree of surface roughness. It is observed that the linear detection model shows better performance characteristics in comparison with non-linear detection model. The comparison graph of actual results and measured results show that linear detection model has a maximum relative error of 2.01 % which is found to be more reliable than non-linear detection model of -9.60 %, indicating better surface detection capability over non-linear detection model. The results demonstrate that the correlation between texture features and surface roughness parameters is an effective means of surface roughness estimation in a non-contact manner. In this study, surface roughness parameters are measured after machining on a shaper machine using a machine vision system and compared with that obtained through stylus method. Machining operation involves complexity and produces different surface finish with different cutting conditions, therefore in the present study correlation between surface roughness parameters (viz. arithmetic average height (Ra); maximum height of peaks (Rp); root mean square roughness (Rq); maximum height of the profile (Rt) and ten-point height (Rz)) and optical surface finish parameters (i.e. mean, standard deviation, skewness, and kurtosis) are developed for varied values of cutting parameters (i.e. depth of cut and RPM of pulley drive). The linear relation model with optical parameters and surface roughness parameters is developed. It is observed that, all the roughness parameters can be estimated with a fair degree of accuracy (R2 > 0.92), using optical statistical parameter kurtosis, while mean, skewness and standard deviation obtained through same image processing data fails to estimate roughness parameters. In the present art of work, surface roughness measurement technique is proposed based on optical method using the statistical properties of binary images. Grounded metal surfaces are used to develop binary digitized speckle patterns by beam of He Ne laser (633 nm) on the machined surface. Parameters such as W (white pixels count) and B (black pixels count) and W/B pixels ratios are computed from the binary images. The obtained results assert the relationship between image parameters and degree of surface roughness. A linear relationship is observed between parameters obtained from proposed model and measured value of surface roughness using surface profilometer. The proposed method with simple setup has a great potential for in-process measurements of surface roughness mainly based on statistical modeling with the binary images. Statistical analysis shows that performance of maximum relative error in prediction of surface roughness is 9%, resulting in better performance characteristics of presented linear detection model. The simplicity of the required optical system and simple method rationalize a great potential for automation and it can be used for in-process measurement. Keywords: Feature extraction, Grey level co-occurrence matrix, Machine Vision System, Roughness parameters, Surface texture, Surface profilometer