Machine Vision System Based Monitoring Approach For Friction Stir Welding Process

Development of intelligent systems for the monitoring of manufacturing processes is always in great demand. In this study, an approach based on image processing through 2D wavelet transform is proposed for monitoring of weld quality in friction stir welding process. Welding experiments are conducted over three ranges of two most influencing process parameters; tool rotational speed and welding speed. Top surface images of the welds are captured and converted to binary images for the analysis using wavelet transformation. The technique of peak signal to noise ratio is executed to find the best mother wavelet function out of 53 available functions for optimum decomposition. All the images are decomposed to second level and the coefficients from the approximations of the images are computed. In this work, a novel indicator based on double logarithmic values of average root mean square of the approximation coefficients of the image is introduced to study the behaviour of weld joint strength. It is observed that, with the increase in the proposed indicator, ultimate tensile strength of the joints is found to follow a decreasing trend. The proposed methodology can be implemented as a tool based on machine vision system for the monitoring of the process with less human intervention.