Data Mining Techniques for Weld Defect Detection by Image Classification

Abstract

Data mining techniques have become indispensable in automating defect detection and classification in industrial welding processes, particularly through the use of image-based data. This study investigates the application of data mining methodologies for detecting and classifying weld defects using photographic datasets processed with the Orange data mining software. By leveraging Orange's visual programming interface, the research demonstrates how image data can be analyzed and modeled to identify welding defects. Key machine learning techniques, including Artificial Convolutional Neural Networks, Logistic Regression, Random Forest, k-Nearest Neighbors for image classification, were applied to achieve high accuracy in defect recognition. Special emphasis was placed on image preprocessing and feature extraction to enhance model performance. The results confirm that Orange offers an intuitive platform for integrating image-based data into sophisticated machine learning workflows, enabling accurate and interpretable classification outcomes. This approach highlights the potential of combining image data with domain-specific software to optimize defect detection processes and improve manufacturing quality.