Machine Learning Application for Prediction of Surface Roughness of Milled Surface

The traditional methods of measuring surface roughness make use of a stylus-based instrument. Accuracy of this instrument depends upon the roughness in the surface and is an indirect method that involves down time. Machine vision techniques have attracted researchers in the area of machining including analysis of surface quality. Grey level co-occurrence matrix (GLCM) is the most widely used statistical technique for feature extraction of machined surfaces. As surface roughness is a widely accepted measure of quality of the machined component, this work aims at prediction of surface roughness value of milled surface using a regression model. The surface roughness values of milled images were found by mechanical means. These images were then subjected to feature extraction by GLCM and discrete wavelet transform (DWT). Based on the available data of images, a multiple linear regression model was developed to predict the surface roughness value of the machined surface without actual measurement. The proposed model is tested on various milled surface images and it predicts an accuracy of 84.18% for freshly milled surfaces.