【摘要】 现有装备保障性分析工作还存在着一些挑战，例如对客观数据和历史数据利用不足、数据知识有效表征化程度不高以及无法进行知识推送等问题。而知识图谱是利用数据表示现实世界实体与关系的信息网络，是解决上述问题的有效方案。为了实现数据驱动的装备以可靠性为中心的维修分析(RCMA),提出一种可以支撑保障性分析的装备RCMA知识图谱构建方法。首先，梳理装备RCMA流程，分析可以用于知识图谱的装备RCMA实体与关系，实现知识图谱的模式层构建。其次，通过单步骤-单模型的实体关系联合抽取方法，使用细粒度三分类模型OneRel从装备RCMA相关文本数据中抽取出三元组，实现知识图谱的数据层构建。最后，选用Neo4j图数据库进行存储，完成了装备RCMA知识图谱的构建。针对装备RCMA相关文本数据进行知识抽取实验，实验结果表明，使用实体关系联合模型的知识抽取在精确率上达到91%,比传统流水线方法用到的知识抽取模型精确率更高，且在构建流程上实现了优化。更多还原

【Abstract】 The existing work in equipment supportability analysis is confronted with a number of challenges, including insufficient use of objective and historical data, a low level of effective data knowledge representation, and the inability to perform knowledge dissemination. Knowledge graphs, which utilize data to depict the entities and relationships in the real world, offer an effective solution to these issues. To achieve data-driven equipment maintenance analysis centered on reliability(RCMA), this paper proposes a method for constructing an equipment RCMA knowledge graph that can support supportability analysis. Firstly, the equipment RCMA process is combed, and the entities and relationships of equipment RCMA that can be used for the knowledge graph are analyzed to achieve the construction of the schema layer of the knowledge graph. Secondly, a single-step-single-model entity-relationship joint extraction method is used, and the fine-grained three-category model OneRel is used to extract triples from equipment RCMA-related text data, achieving the construction of the data layer of the knowledge graph. Finally, the Neo4j graph database is selected for storage, and the construction of the equipment RCMA knowledge graph is completed. Knowledge extraction experiments were conducted on equipment RCMA-related text data, and the results show that the knowledge extraction using the entity-relationship joint model has an accuracy rate of 91%, which is higher than the accuracy rate of the knowledge extraction model used in the traditional pipeline method, and the construction process has been optimized.更多还原