With the increasing complexity of industrial systems, efficiently extracting and modelling knowledge from massive technical standards documentation has become a major challenge. This paper presents GraphRAG-KM, an automated framework for accurately converting unstructured industrial documents into structured ontology (OWL) and conceptual (UML) models. The framework integrates multi-layered technologies, including MinerU, retrieval-augmented generation (RAG), large language models (LLMs), and K-Means clustering, to precisely extract entities, relationships, and hierarchical structures. Specifically, MinerU and OCR preprocess PDF documents into structured Markdown texts, which are then semantically enriched by GraphRAG via targeted retrieval and indexing. Subsequently, LLMs infer implicit attributes and relationships, thereby enhancing semantic completeness. Meanwhile, K-Means clustering identifies latent associative relations, which significantly improves model completeness and semantic depth. Experimental evaluations on the MIL-STD-6016 standard demonstrate that GraphRAG-KM achieves a knowledge extraction accuracy of 94.7%, significantly outperforming traditional LLMs methods (approximately 30%). Reading efficiency improved by 58.66% in readability assessments, and comprehension accuracy increased by 29.2% compared to traditional reading methods. Furthermore, automated modelling efficiency improved by 6.75 times compared to manual methods. GraphRAG-KM thus represents an effective solution for intelligent, scalable, and efficient knowledge management in industrial engineering contexts.