Advances and Challenges in Deep Learning for Automated Welding Defect Detection: A Technical Survey

Abstract

Automated welding defect detection has emerged as a pivotal aspect of quality assurance in high-stakes industries such as aerospace, oil and gas, and construction. This paper presents a comprehensive review of state-of-the-art deep learning (DL) models tailored for welding defect detection, segmentation, and classification, emphasizing technical advancements and persistent challenges. A critical analysis of single-stage and two-stage architectures is conducted to evaluate their ability to address issues like small defect sizes, low image contrast, and diverse defect geometries. The study also highlights the integration of advanced preprocessing techniques, such as noise reduction and contrast enhancement, within DL workflows to improve feature extraction and detection accuracy. Persistent challenges, such as the scarcity of large, labeled datasets, lack of real-time applicability, and limited model interpretability, are explored in depth. To address these gaps, the survey proposes future directions, including the use of self-supervised learning, domain adaptation, generative adversarial networks (GANs), and explainable AI techniques to enhance the robustness, scalability, and transparency of welding defect detection systems. By synthesizing insights from more than a decade of research, this paper provides a detailed roadmap for advancing automated welding inspection technologies, enabling reliable deployment in real-world industrial environments.

Topics & Concepts