Do Multimodal Large Language Models Understand Welding?
This addresses the problem of applying multimodal LLMs to high-stakes technical domains like welding for industry practitioners, but is incremental as it evaluates existing models with new data and a prompting strategy.
This paper evaluated two state-of-the-art multimodal LLMs on assessing weld acceptability across three industrial contexts, finding they performed relatively well on unseen real-world images but better on online images, and introduced WeldPrompt prompting strategy that improved recall in some contexts but showed inconsistent performance.
This paper examines the performance of Multimodal LLMs (MLLMs) in skilled production work, with a focus on welding. Using a novel data set of real-world and online weld images, annotated by a domain expert, we evaluate the performance of two state-of-the-art MLLMs in assessing weld acceptability across three contexts: RV \& Marine, Aeronautical, and Farming. While both models perform better on online images, likely due to prior exposure or memorization, they also perform relatively well on unseen, real-world weld images. Additionally, we introduce WeldPrompt, a prompting strategy that combines Chain-of-Thought generation with in-context learning to mitigate hallucinations and improve reasoning. WeldPrompt improves model recall in certain contexts but exhibits inconsistent performance across others. These results underscore the limitations and potentials of MLLMs in high-stakes technical domains and highlight the importance of fine-tuning, domain-specific data, and more sophisticated prompting strategies to improve model reliability. The study opens avenues for further research into multimodal learning in industry applications.