Against the backdrop of rapid advances in intelligent manufacturing, equipment fault diagnosis plays a crucial role in ensuring the stable operation of industrial systems and improving production efficiency. To address the limitations of traditional methods in handling unstructured, multi-source heterogeneous data and the inefficiency of knowledge updating, this paper proposes a method for constructing equipment fault diagnosis knowledge graph based on large language model. The proposed approach integrates prompt engineering with domain ontology constraints to automatically extract and structurally rep resent key knowledge from various sources such as maintenance documents and operation manuals. During knowledge fusion, semantic matching and conflict detection mechanisms are introduced to ensure the accuracy and consistency of the graph. The extracted knowledge triples are imported into the Neo4j graph database using Cypher statements. Experimental results demonstrate that the proposed method significantly improves the efficiency of knowledge construction, reduces manual labor, and enhances the semantic coherence of the knowledge graph. This provides a solid knowledge foundation for intelligent fault diagnosis systems and offers a new paradigm for the automated construction of domain-specific knowledge graph.