关键词: 盾构, 液压推进系统, 故障诊断, 知识图谱, 本体建模, 关系路径推理, 知识驱动

Abstract: When shield tunneling under complex working environments, its hydraulic propulsion system often encounters various challenges, such as concealed fault modes, scarce fault samples, limited generalization ability of data-driven models, and the difficulty of integrating and reusing empirical knowledge. To address these challenges, this paper proposes a knowledge-driven auxiliary fault diagnosis method based on a knowledge graph. It aims to construct a structured knowledge system integrating equipment structure, monitoring points, fault phenomena, fault causes, and maintenance suggestions, thereby enabling multi-path reasoning diagnosis from either observed fault phenomena or preliminary diagnostic results, and improving diagnostic completeness and interpretability under complex conditions. Methodologically, an AMESim-based simulation model of the shield hydraulic propulsion system is first established. Based on actual engineering parameters, typical faults—including speed control valve spring failure, directional valve internal leakage, relief valve internal leakage, and hydraulic cylinder internal leakage—are parameterized and simulated to provide a foundation for fault mechanism analysis and knowledge extraction. An ontology-based knowledge representation approach is then adopted to classify domain knowledge into five ontology categories: equipment structure, monitoring points, fault phenomena, fault causes, and maintenance suggestions. A knowledge ontology model containing concepts, relations, and attributes is constructed, with semantic relations such as hasPhe, cause, located, monitor, and advice defined and expressed in the form of triples. Neo4j graph database is employed for knowledge storage and visualization, resulting in a fault diagnosis knowledge graph comprising 183 entities and 205 triples. In terms of reasoning mechanism, a multilevel semantic reasoning method based on relationship paths is proposed, including a phenomenondriven reasoning process for simple faults and a hybrid diagnostic process integrating data-driven model outputs for complex faults. Entity matching is used to establish the mapping between model outputs and knowledge graph entities. The results demonstrate that the constructed knowledge graph can infer fault causes, locations, and maintenance suggestions from a single fault phenomenon, forming a complete diagnostic chain. For complex faults, it supplements mechanism interpretation and corrective strategies based on preliminary classification results, enabling extended and structured diagnostic outputs. Engineering case validation shows that the proposed method generates multi-branch reasoning paths for typical faults and provides clear fault localization and maintenance recommendations, establishing a hybrid knowledge-driven and data-driven diagnostic framework for the hydraulic propulsion system of shield machines and offering a systematic technical approach for intelligent maintenance of large-scale engineering equipment.

Key words: shield, hydraulic propulsion system, fault diagnosis, knowledge graph, ontology modeling, relationship path reasoning, knowledge driven