Performance Evaluation of Digital Image Processing and Artificial Neural Networks for Weld Line Detection of Robotic Manipulator

Abstract

Robotic manipulators are used in industries for several purposes, with one of the purposes being welding. The objective of this paper is to automate the welding process by automating the task of weld line detection. A comparative evaluation between digital image processing technique and artificial neural networks to detect the center line of the weld piece has been analyzed. In the image processing technique, the image has been acquired and image pre-processing has been done in the form of RGB Thresholding, filtering, Binary thresholding, and morphological transformations to attain the center line. In utilizing artificial neural networks, a database of images for training and testing have been fed into the network with different training algorithms as well as different training ratios and learning rate, and the performance criteria of mean square error has been evaluated.