During the operation of the EMU, a large number of fault data will be generated, which record the fault overview, fault cause, treatment measures and other related fault information. Construction of knowledge graph of these data can provide knowledge for relevant staff in the maintenance process and assist them in decision-making, which is of great practical significance. Most of the existing EMU fault knowledge graph construction methods adopt supervised named entity recognition algorithms and relation extraction algorithms, which require a large amount of labeled data for training. To solve the above problems, a method for constructing EMU fault knowledge graph based on large language model is proposed. Firstly, the EMU fault ontology model is constructed based on the existing EMU fault data and drawing on the existing fault knowledge graph. Then the EMU fault triple is extracted by means of small samples using the in-context learning capability of the large language model. Finally, entity linking is performed based on the constructed fault ontology model to construct the EMU fault knowledge graph, and the graph is displayed in a visual way to guide the maintenance and diagnosis of EMU faults.