Knowledge graph reasoning: learning long-path fault prioritization from aircraft maintenance records

Kong Sun, Huanchun Peng, Yuxuan Zhang, Zhi Lv, Yongsheng Yang, Yuanxiang Li

Abstract: The integration of robust fault diagnosis technologies into complex civil aircraft systems plays a pivotal role in enhancing flight safety and ensuring system reliability. However, current data-driven and knowledge-driven approaches to aircraft fault diagnosis still struggle to effectively handle long-path reasoning in knowledge graph-based inference. To address this limitation, this paper proposes ​​Pathformer​​, a novel multi-path representation learning model that learns fault path priorities from maintenance records. First, we establish a maintenance record matching model to associate key entities in fault paths with their corresponding maintenance records. Next, we introduce a fault path confidence construction method that prioritizes troubleshooting paths by analyzing key entities within each path. This process is further refined through a training loss function that incorporates path confidence derived from historical maintenance records. Finally, we develop ​​Pathformer​​ to efficiently encode and rank long diagnostic paths. Experimental evaluations on a self-constructed fault knowledge graph covering four typical civil aircraft systems demonstrate the model’s superior path reasoning capabilities, achieving significant improvements over existing path reasoning approaches.

External IDs:doi:10.1007/s42401-025-00412-7