| Abstract Scope |
Artificial intelligence (AI) has been increasingly integrated into various manufacturing processes, including solid-state cold spray additive manufacturing (AM), and applications of LLMs are expanding across a range of tasks. However, despite rapid advances in both, little work has explored their intersection. Here, we introduce the first open-source benchmark dataset for assessing LLM knowledge of cold spray across key subtopics (mechanical properties, fatigue, fracture, corrosion, spray parameters, and fundamental materials science), requiring cold-spray–specific expertise across subtopics and advanced LLM testing methods. Questions are grouped by difficulty, including broad conceptual questions, detailed mechanistic questions probing advanced domain knowledge, and complex reasoning questions. We evaluate both open-source and commercial models for accuracy, consistency, depth of reasoning, and completeness to inform practical and responsible use-cases. Finally, we explore prompt engineering and RAG-based approaches to enhance performance. A well-informed LLM could enable accelerated cold-spray AM innovation, robust quality control, and optimized workflows. |