Machine Learning Approach for Surface Defects Identification of Shielded Metal Arc Welding

Shielded Metal Arc Welding (SMAW) is one of the most employed welding technique in fabricationprocess. Spatter and incomplete fusion are the surface defects that are frequently found in SMAW welds.The images of the weldments are captured through a high-resolution digital camera and subsequently processed for further analysis by Machine Learning (ML)algorithms such as Artificial Neural Network (ANN), Convolutional Neural Network (CNN), ResNet50 and Recurrent Neural Network (RNN)on training images to develop the models. These ML models are evaluated on test images and the model exhibiting the highest prediction accuracy is implemented as a classifier for classification of the defects on the weldments. In this work, the classifier identifies thetype of surface defect(s) considering the image of weldment as an input by the learner and associates the most probable causes of the surface defects detected,providing a valuable feedback to improve the performance