A Comprehensive Review of YOLO-Based Object Detection for Intelligent Welding Process

The advancement of high-performance computing and neural network technologies has continuously improved the performance of the object detection algorithm You Only Look Once (YOLO), enabling its application in a wide range of industries. In the welding industry, YOLO is actively utilized in the development of automated monitoring systems, leading to a surge in related research. This review paper aims to comprehensively analyze the application of YOLO in diverse welding environments by examining recent studies. Welding monitoring is categorized into defect inspection, process monitoring, and safety management, and the specific tasks achievable within each domain are elucidated. In addition, this paper analyzes YOLO-based algorithms optimized for welding; their network architectures and performance are evaluated and the potential of YOLO in diverse welding applications is explored. Furthermore, the advantages and limitations of YOLO in welding work environments are discussed, and the key challenges and future research directions are identified. Specifically, this review emphasizes the necessity for constructing robust datasets, optimizing models for real-world industrial applications, tracking network development trends, and developing enhanced modules. By addressing these aspects, the essential elements required for YOLO to evolve into a more advanced and intelligent welding monitoring system can be incorporated.