Optimization of knowledge graph search methods enabled by large language models with applications to fault diagnosis of electromechanical systems

Under the trend of industrial intelligence upgrading, electromechanical systems serve as a crucial component for ensuring the normal operation of equipment. The accuracy of fault diagnosis directly affects production efficiency and operation maintenance costs. Although traditional knowledge graphs can structure fault diagnosis knowledge for electromechanical systems, their reliance on keyword matching or specialized query languages for search leads to issues such as semantic ambiguity and low retrieval efficiency, limiting their application in the field of fault diagnosis. This paper proposes a knowledge graph search optimization algorithm empowered by large language models. Firstly, a general pre-trained large language model is fine-tuned to simulate user queries and generate answers, enabling the extraction of more relevant entities. Secondly, we design an evaluation metric for query complexity, utilizing the large language model to assess the complexity of user queries and dynamically determine the hops and search scope in the knowledge graph. Finally, the generated set of reasoning paths is scored for relevance, and the large language model outputs responses to fault-related queries based on the extracted path information, thereby completing the fault diagnosis. Experimental results on electromechanical system fault diagnosis datasets demonstrate that the proposed algorithm significantly improves fault diagnosis accuracy and the interpretability of the model's responses, offering a new technical pathway for the application of large language models in the vertical domain of electromechanical system fault diagnosis.