导航

Acta Aeronautica et Astronautica Sinica ›› 2026, Vol. 47 ›› Issue (19): 233284.doi: 10.7527/S1000-6893.2026.33284

• Solid Mechanics and Vehicle Conceptual Design • Previous Articles    

Knowledge graph- and large language model-driven fault diagnosis for avionics equipment

Yujie CHENG1,2,3,4(), Zheng HU1,3,4, Yongmei GAO5, Jiyan ZENG1,3,4, An ZHOU1,3,4, Bo SUN1,3,4   

  1. 1.Institute of Reliability Engineering,Beihang University,Beijing  100191,China
    2.Hangzhou Institute for International Innovation (International Innovation College),Beihang University,Hangzhou  310020,China
    3.School of Reliability and Systems Engineering,Beihang University,Beijing  100191,China
    4.National Defense Science and Technology Key Laboratory of Reliability and Environmental Engineering;Technology,Beijing  100191,China
    5.China Machinery Standard Technology Research Institute (Beijing) Co. ,Ltd. ,Beijing  100190,China
  • Received:2025-12-25 Revised:2026-03-02 Accepted:2026-04-09 Online:2026-05-06 Published:2026-04-30
  • Contact: Yujie CHENG E-mail:chengyujie@buaa.edu.cn
  • Supported by:
    National Natural Science Foundation of China(61803013);Aeronautical Science Foundation of China(201933051001)

Abstract:

Knowledge graphs and large language models have emerged as cutting-edge technologies in recent years, and leveraging them for fault diagnosis of avionics equipment has become a current research hotspot. Knowledge-graph-based fault diagnosis methods offer strong interpretability but have limited semantic generalization capabilities. In contrast, large language models possess semantic understanding and generation abilities but suffer from poor controllability and are prone to hallucinations, which can reduce fault diagnosis accuracy. To improve the accuracy of avionics equipment fault diagnosis, this paper proposes a fault-diagnosis method driven by the integration of knowledge graphs and large models, centered around “fault elements”. First, at the data layer, a “expert-guided and large-model-assisted” fault-element extraction mechanism is developed to clean and structurally extract the raw data. Subsequently, a structured knowledge graph is constructed based on the extracted fault elements and various entities, with visualization implemented using Neo4j. Second, at the inference layer, the TuckER tensor decomposition model is employed for causal chain modeling and multi-hop reasoning, while a large language model is introduced to perform enhanced reasoning based on semantic subgraphs of the knowledge graph. Finally, at the decision layer, a consistency verification module is designed to cross-validate the subgraph reasoning results from the tensor decomposition model and the large language model. Conclusions are output only when they meet a preset threshold, balancing accuracy and consistency. The proposed method is validated on an avionics equipment dataset, and results show that it outperforms traditional methods in terms of diagnostic accuracy, information richness, and professional consistency, demonstrating good robustness and scalability. These findings provide a new technical approach for building high-reliability, high-trustworthiness intelligent diagnosis systems for avionics equipment.

Key words: fault diagnosis, knowledge graphs, knowledge reasoning, large language models, tensor decomposition model, knowledge-augmented

CLC Number: