This study explores the application of a machine vision system integrated with convolutional neural network (CNN) for detecting and classifying welding defects. By leveraging the power of deep learning approaches, the proposed approach aims to enhance the efficiency and reliability of defect classification. This method not only reduces human dependency, but also establishes a framework for automated welding quality control systems. A CNN-based machine vision system has been developed to classify welding defects in radiographic images. Particularly, two transfer learning algorithms, mainly, ResNet-18 and ResNet-50, have been applied and evaluated in order to determine the most effective method in detecting and classifying weld defects. The dataset covered three classes of weld defects: cracks, lack of penetration, and porosity. The performance of each ResNet-based CNN model was assessed using performance evaluation metrics and visualization techniques. ResNet-50 emerged as the best performing model and had a strongest response in the weld defects regions, achieving an average accuracy of 96.061%. This model proved effective in detecting and classifying defects, demonstrating its potential to significantly enhance the reliability and automation of detection and recognition.
Loading....