如今航天器内部结构功能越来越复杂。电缆作为其"生命线"，在人工检测过程中需要耗费大量人力物力，如何快速定位质量问题并找出解决方案仍然具有挑战性。针对此问题，该文采用基于知识图谱的方法来抽取多源异构的电缆知识实体，该模型通过基于transformer的双向编码表征(bidirectional encoder representations from transformers，BERT)网络对输入文本进行词向量嵌入，然后通过双向长短时记忆(bidirectional long short-term memory，BiLSTM)网络对输入序列的上下文特征进行提取，最终输入随机条件场(conditional random field，CRF)网络中预测实体类别，同时以此模型提取出的实体作为数据层，构建知识图谱。所搭建的实体抽取模型的准确率、召回率、F1值等指标均比常见模型有所提升。最终以某航天精密机械加工所的线缆测试数据构建了该领域的知识图谱模型，实现了可视化查询及对质量问题的追溯定位。
Knowledge Graph Based Method for Tracing Quality of Aerospace Products
Nowadays,the internal structure of spacecraft has been increasingly complex.As its"lifeline",cables require extensive manpower and resources for manual testing,and it is challenging to quickly and accurately locate quality problems and find solutions.To address this problem,a knowledge graph based method is employed to extract multi-source heterogeneous cable knowledge entities.The method utilizes the bidirectional encoder representations from transformers(BERT)network to embed word vectors into the input text,then extracts the contextual features of the input sequence through the bidirectional long short-term memory(BiLSTM)network,and finally inputs them into the conditional random field(CRF)network to predict entity categories.Simultaneously,by using the entities extracted by this model as the data layer,a knowledge graph based method has been constructed.Compared to other traditional extraction methods,the entity extraction method used in this study demonstrates significant improvements in metrics such as precision,recall and an F1 score.Ultimately,employing cable test data from a particular aerospace precision machining company,the study has constructed the knowledge graph based method in the field to achieve visualized queries and the traceability and localization of quality problems.
万方数据
