Original scientific paper

https://doi.org/10.17559/TV-20250312002462

Low-Dose X-Ray Visual Weld Defect Feature Extraction and Classification

Pengyu Gao ; Guangdong provincial Welding Engineering Technology Research Center, Guangdong University of Technology, Guangzhou 510006, Guangdong, China
Yali Huang ; Guangdong provincial Welding Engineering Technology Research Center, Guangdong University of Technology, Guangzhou 510006, Guangdong, China
Jingguo She ; Guangdong provincial Welding Engineering Technology Research Center, Guangdong University of Technology, Guangzhou 510006, Guangdong, China
Yangshuo Tian ; Guangdong provincial Welding Engineering Technology Research Center, Guangdong University of Technology, Guangzhou 510006, Guangdong, China
Xindu Chen ; School of Electromechanical Engineering, Guangdong University of Technology, Guangzhou 510006, Guangdong, China
Xiangdong Gao ; (Corresponding author) 1) Guangdong provincial Welding Engineering Technology Research Center, Guangdong University of Technology, Guangzhou 510006, Guangdong, China 2) Guangzhou Zhengtian Technology Co., Ltd, Guangzhou 510006, China *

* Corresponding author.

Abstract

This study proposes a low-dose X-ray-based nondestructive testing method for weld defect automatic detection, focusing on resistance spot welds and butt joint laser welds. By exploiting the high penetration capability and resolution of X-ray imaging, the proposed method detects internal defects such as porosity, cracks and lack of fusion through contrast analysis. Image preprocessing techniques including median filtering, Fourier and wavelet transforms, adaptive histogram equalization and Hough Transform were applied to suppress noises and enhance defect features. Also, Contrast Limited Adaptive Histogram Equalization effectively improves image contrast and reveals subtle defect patterns. For classification, the Random Forest algorithm was adopted to extract relevant features and perform defect recognition. Experimental results show that the proposed method achieves high accuracy in classifying weld defects, with strong generalization ability and minimal misclassification. The classification accuracy reached for resistance spot welds and laser welds confirms the method robustness. This approach enhances the automation and reliability of welding quality control by providing an efficient and accurate solution for internal defect identification. The contributions of this work lie in integrating advanced image processing with machine learning techniques for precise defect detection in complex weld structures using low-dose imaging, thus supporting intelligent inspection systems in industrial applications.

Keywords

feature extraction; non-destructive testing; random forest; weld defect detection; X-ray imaging

Hrčak ID:

342636

URI

https://hrcak.srce.hr/342636

Publication date:

31.12.2025.

Visits: 435 *