Non-contact surface roughness evaluation of milled Al and Cu specimens by 1D and 2D wavelet transformation using histogram based linear regression model

A contactless system networked with contactless surface roughness using convolutional neural network (CNN) algorithms helps to sort out all the issues in offline surface roughness measurement. In this study, aluminium and copper specimens were milled to create different roughness textures. Then the roughness of the surface was measured using a contact stylus with a 2 μm radius tip and also using a noncontact 3D surface profiler. The CMOS camera is used to capture the image of the specimen’s surface in white light illumination. The extracted signal vectors from the image pixel intensities and image histogram details were processed using MATLAB with a wavelet toolbox to obtain transformed mean image intensity. These collected data were correlated and formed a linear graph to build the non-contact surface roughness measurement system in which the knowledge is automatically defined by the CNN algorithm. 1D and 2D biorthogonal wavelets are examined for high-grounded training to aid the CNN algorithm. The milled aluminium specimens generated the curve correlation graph with an error of 2 to 7% with the stylus parameter and 4 % with a 3D surface profiler. The usage of image histogram based linear regression models also produces only 8% of error which shows the reliability in the quantification procedure of surface roughness evaluation.