Enhancing Printed Circuit Boards Operational Efficiency through Multimodal Retrieval-Augmented Generation Chatbots: Bridging Textual and Visual Documentation

This paper presents an enhanced Retrieval-Augmented Generation (RAG) chatbot framework for efficient retrieval and interpretation of complex technical documentation in the printed circuit board (PCB) manufacturing domain. Two RAG architectures, LongRAG and GraphRAG, are systematically evaluated using domain-specific equipment manuals. Quantitative results show that LongRAG achieves higher answer relevancy, while GraphRAG produces more faithful responses. Performance is further improved through multi-query expansion, cross-encoding, and reranking strategies. In addition, a multimodal chatbot integrating textual and visual information is qualitatively assessed by domain experts, demonstrating improved user comprehension and interaction quality. The results highlight effective design strategies for deploying RAG-based chatbots in industrial environments involving long-context and multimodal documentation. Impact Statement — The developed chatbot employs advanced LLMs, specifically OpenAI's o3-mini and Gemini-2.0-Flash, integrated with RAG techniques, to efficiently retrieve and deliver technical documentation within the printed circuit board (PCB) manufacturing sector. By providing accurate, contextually relevant responses and incorporating visual content, the chatbot has the potential to significantly reduce service response times and enhance operational efficiency. This advancement establishes an elevated performance benchmark for the industry, underscoring the transformative capacity of artificial intelligence in industrial contexts through quicker and simpler access to critical production information. Consequently, this technology holds considerable promise for the future of PCB manufacturing, promoting increasingly intelligent, efficient, and adaptable production systems.