CC BY 4.0Zhang, HansiHansiZhangSchmidt, WilmaWilmaSchmidtShen, XiaozhiXiaozhiShenCao, QiushiQiushiCaoMonka, SebastianSebastianMonkaPaschke, AdrianAdrianPaschke2025-11-112025-11-112025-09https://publica.fraunhofer.de/handle/publica/499007https://doi.org/10.24406/publica-618210.24406/publica-61822-s2.0-105019640306In the context of Industry 4.0, effective maintenance is critical for minimizing manufacturing downtime and ensuring production reliability. While first Graph Retrieval-Augmented Generation (RAG) frameworks enhance contextual understanding and accuracy in maintenance chatbots, Knowledge Graph (KG) construction in manufacturing remains tedious and error-prone. To address this, we propose a semi-automated KG construction pipeline that integrates rule-based methods, Small Language Models (SLMs), and Large Language Models (LLMs), significantly reducing manual efforts in KG construction. We evaluate the constructed KG in a Graph RAG setting on real-world maintenance scenarios in a production line. Our results highlight the potential to significantly enhance the efficiency and intelligence of manufacturing maintenance workflows. Our work aims to spark discussions on efficient Graph RAG frameworks for maintenance scenarios in manufacturing.enfalseKnowledge Graph ConstructionGraph RAGLLMManufacturingconference paper