Automatic Detection and Characterization of Weld Defects Using CNN Algorithm in Machine Learning

Abstract

Conventional radiographic technique uses visual inspection of scanned output for defect detection. This makes the inline testing of products time consuming and hectic. Convolutional Neural Network (CNN) algorithm in machine learning can be used for the automation of defect detection in radiography thereby reducing human intervention and associated delays. By the use of robotics the welding parameters can be adjusted and the issue of welding defects can be resolved. By combining the two, the defect detection process can be modified into a digital manufacturing process. A dataset created from radiography test data is used for training the algorithm and for writing a program to train this dataset which can be used for defect detection and its characterization.