This study evaluates the capabilities of Multimodal Large Language Models in assessing weld quality in specialized industrial contexts.

• Models perform better on online images vs. real-world welding examples
• Performance varies across industrial contexts (RV/Marine, Aeronautical, Farming)
• Results suggest potential for AI in quality control but highlight challenges with real-world applications

This research demonstrates both the promise and limitations of deploying MLLMs in engineering quality assessment, providing insights for implementing AI in manufacturing inspection processes.

Do Multimodal Large Language Models Understand Welding?