Artificial Intelligence enhanced smart welding islands: Foundation models revolutionizing manufacturing

Welding is a pivotal process in equipment manufacturing, yet it faces significant challenges including skilled labor shortage, quality inconsistency, and operational inefficiency due to manual operations. While automation mitigates reliance on human expertise, existing industrial welding robots are typically optimized for specific and repeatable tasks in mass production, lacking the adaptability required for Engineering-to-Order manufacturing with high workpiece variability like shipbuilding. Therefore, this study proposes a framework of smart welding island service system (SWISS) that leverages large language models (LLMs) to achieve responsive and intelligent decision-making for agile manufacturing. This system comprises four modules: TrajectoryGPT, which automatically generates weld seam trajectory from design models by interpreting geometric features and process constraints. WeldGPT, which produces tailored welding procedures by integrating general standards with specific factory requirements. RobotGPT, which synthesizes trajectory, process plans, and real-time perception for workpiece recognition and positioning to generate G-code and execute adaptive and quality-assured welding. OperationGPT, which orchestrates out-of-order execution and synchronization of welding operations to reduce idle time and improve throughput. Domain-specific foundation models are developed through fine-tuning and retrieval-augmented generation, combining general world knowledge with high-precision industrial data to enhance welding flexibility and system scalability. The module-level experiments on the four LLM-based components and system-level case studies are conducted to validate the feasibility and rationality of the proposed system. The results highlight a pathway for smart welding islands and practical applications of LLMs in industry. This research is expected to offer guidance for practitioners facing similar challenges and inspiration for scholars exploring next-generation intelligent manufacturing.