Data-driven artificial intelligence methods for real-time welding defect diagnosis: A critical review and future outlook

Методы искусственного интеллекта на основе данных для диагностики дефектов сварки в реальном времени: критический обзор и перспективы развития
Xiaoqi Chen, Xun William Xu, Mingming Zhang, Jan Polzer, Qunfeng Liu, XuFan Chen
2026-02-02

convolutional neural networksdigital twin technologiesin-process monitoringmultimodal data fusionreal-time welding defect diagnosis
This review provides a detailed analysis of real time welding defect detection systems focusing on the critical integration of advanced sensing technologies with artificial intelligence to enhance welding quality assurance. Traditional post process inspection methods are time consuming costly and fundamentally incompatible with the modern automated manufacturing requirement for real time quality control. This necessitates a shift toward in process monitoring systems that detect defects during the welding operation enabling immediate corrective action. The study evaluates the effectiveness of various sensor technologies including optical electrical acoustic thermal and radiographic sensors in identifying diverse welding defects. It then examines the application of advanced AI techniques for welding defect diagnosis covering specialized models such as convolutional and recurrent neural networks transformer and generative models transfer learning multimodal data fusion and hybrid approaches. The review also discusses key challenges such as data quality acquisition scarcity computational resource limitations and system integration complexity. Finally it highlights promising future research directions including lightweight AI models sophisticated multi sensor fusion strategies and digital twin technologies. These advancements have the potential to improve diagnosis accuracy and truly enable real time defect detection during the welding operation ultimately increasing manufacturing efficiency reducing waste and ensuring the production of safer and more reliable welded structures in critical industrial sectors. • Comprehensive review of sensor technologies for real time welding defect diagnosis. • Analysis of AI-based methods applied to welding anomaly diagnosis. • Discussion of current challenges. • Future directions identified such as lightweight AI models, and adaptive multi-sensor systems. • Industry 4.0 relevance to enhance welding quality and reduce waste.
1
AI-based diagnosis increasingly uses convolutional and recurrent neural networks, transformers, generative models, transfer learning, multimodal fusion, and hybrid architectures.
2
Lightweight AI, adaptive multi-sensor fusion, and digital twins are identified as promising directions for accurate real-time diagnosis, reduced waste, and safer welded structures.
3
Major barriers include insufficient data quality and availability, limited computational resources, and complex integration of sensing and AI systems.
4
Optical, electrical, acoustic, thermal, and radiographic sensors can identify diverse welding defects during the welding operation.
5
Real-time in-process monitoring is needed because traditional post-process welding inspections are time-consuming, costly, and incompatible with automated manufacturing.

Real-time welding operations and their in-process defect detection systems

The effectiveness, integration, challenges, and future development of sensor- and AI-based welding defect diagnosis for real-time quality assurance

Publication Details
Publication Date
2026-02-02
Journal
Publisher
ISSN
Cited by
18
Access Type
Author Information
Authors
Xiaoqi Chen
Xun William Xu
Mingming Zhang
Jan Polzer
Qunfeng Liu
XuFan Chen
Explore further
Open the scid.ai AI chat with a ready-made request: it will find papers on a similar topic and help build a literature review.
Find similar papers in the chat →
Make a presentation
100%