• CN: 11-2187/TH
  • ISSN: 0577-6686

Journal of Mechanical Engineering ›› 2026, Vol. 62 ›› Issue (18): 241-255.doi: 10.3901/JME.260762

Previous Articles    

Intelligent Maintenance Decision-making Method for Aero-engine Bearings Integrating Knowledge Graph and Large Language Model

WANG Shuo1, WANG Qinghua1, HAN Jingxuan1, LEI Yaguo1, WU Tonghai1, LI Yanjun2   

  1. 1. Key Laboratory of Education Ministry for Modern Design and Rotor-Bearing System, Xi'an Jiaotong University, Xi'an 710049;
    2. AECC Sichuan Gas Turbine Research Establishment, Chengdu 610500
  • Received:2026-02-01 Revised:2026-06-10 Published:2026-11-20

Abstract: Wear surface carries valuable wear mechanism information for wear failure, thus serves as critical support for the maintenance decision-making of aero-engine bearings. However, bearing maintenance decision analysis contains multi-attribute knowledge with complex interrelationships, and wear surface descriptions cover multi-source wear mechanisms, which have limited the accuracy and interpretability of maintenance decision-making for aero-engine bearings. To address this issue, an intelligent maintenance decision-making method is proposed for aero-engine bearings by integrating knowledge graph and large language model (LLM), with a focus on the structured representation of wear analysis knowledge and hierarchical knowledge retrieval. First, an aero-engine wear analysis knowledge graph with a “Cause-Mechanism-Severity-Maintenance/Prevention” schema is established based on ontology definition and domain knowledge extraction, which can transform textual wear analysis knowledge into machine-readable and queryable structured information. Second, a hierarchical knowledge subgraph retrieval and refinement strategy is designed based on bearing wear description characteristics, improving the overall consistency between the retrieved knowledge and the query and then enabling knowledge-graph-enhanced intelligent maintenance decision-making with the LLM. Finally, validation experiments are conducted using bearing wear surfaces under different wear mechanisms and severity levels, and the results indicate that the proposed method improves the decision accuracy from a range of 44.67%-70.67% achieved by comparative models to 95.46%. Moreover, the proposed method is applied to evaluate the wear surfaces of 56 bearing rings from 12 types across two actual aero-engine models, involving wear mechanisms such as scratches, spalling, and pitting, and achieves a maintenance decision accuracy of 95.33%.

Key words: bearing wear analysis, knowledge graph, large language model, maintenance decision-making

CLC Number: