Automated Weld Defect Detection Using Gated Attention

Posted: 2025-12-1

Source: Nature.com

https://www.nat...598-025-24406-9

Welding, a widely employed technology for metal joining, is paramount in ensuring the quality and safety of critical infrastructure. However, inspecting weld defects is crucial to preventing disasters, and traditional manual methods are prone to errors, particularly in areas with limited accessibility. To address these drawbacks, researchers seek intelligent systems that replace tedious manual inspections and reduce human errors. Although Structural Health Monitoring (SHM) methods using sensor technology have been implemented, concerns about sensor faults and limitations in monitoring smaller and sensitive areas persist. Vision-based SHM methods offer an alternative, yet they require skilled personnel and face challenges in monitoring specific defects in steel structures. In defect detection, Convolutional Neural Networks (CNNs) have transformed computer vision applications in SHM and deep learning. However, existing studies often focus on single-vision tasks, lacking a unified framework for comprehensive weld visual inspection. Segmenting defects or abnormalities in industrial images demands a higher level of accuracy than what is desired in natural images. At the same time, a precise segmentation mask may not be critical in natural images; even marginal segmentation errors can cause fatal accidents in field settings.

Additionally, reliance on sensor-based SHM methods poses reliability issues, while vision-based methods encounter difficulties in monitoring weld defects in intricate structures. The automated detection of weld defects based on weld images is still an evolving research discipline. Performing this involves several stages, including evaluation of the quality of the weld images, segmentation, and identification of defects. Deep learning models have shown considerable potential to automate weld image segmentation, allowing accurate segmentation of weld areas, which is crucial to determining the severity of defects.

Welding is a fundamental process in manufacturing and construction, playing a pivotal role in joining materials by melting and fusing them. This versatile technique finds widespread application across diverse industries, creating robust and structurally sound components. Regular inspection and continual oversight of the welding process are imperative. Manual inspection methods, however, prove inadequate in addressing the ever-growing demand for safety and precision. In recent years, integrating deep learning and computer vision technologies has revolutionized the welding field by enhancing and automating inspection and safety protocols.

This paper proposes DeepFuse WeldNet, an automated Weld detection Framework that employs a novel Gated Attention Squeeze and Excitation Fusion U-net (GASEUNet) for weld defect segmentation. Our research focuses on enhancing the accuracy of the model for more precise identification and measurement of weld porosity within images. The model provides a significant boost in segmentation accuracy that is better than existing methods. Our experimental results demonstrate the effectiveness of GASEUNet, with an impressive Jaccard Coefficient (JC) of 98.12. The proposed DeepFuse WeldNet also consists of a fully automated Weld Inspection Framework that classifies the weld defects as safe or unsafe by comparing the measurements to the standards provided by the American Welding Society (AWS)

Read the full article at Nature.com