Research On Key Construction Technology Of Knowledge Graph For Aero-engine Design

Intelligent manufacturing is the key to enhancing the core competitiveness of manufacturing enterprises,and it is the driving force to promote the transformation and upgrading of the manufacturing industry and accelerate the high-quality development of the manufacturing industry.Promoting the implementation of aero-engine intelligent manufacturing is an important way to accelerate the aero-engine development process and narrow the gap with the international advanced level.The design of modern aero-engines is based on knowledge-based digital design.However,the knowledge involved in the development of aero-engines is mostly recorded in the form of unstructured data,which makes it impossible to dig and utilize the objectively valuable relationships between knowledge.Knowledge graph technology has powerful semantic processing and interconnected organization capabilities.The construction of aero-engine domain knowledge graphs will provide the possibility to realize large-scale knowledge mining in the aero-engine field,and is the technical basis to realize intelligent manufacturing design of aero-engines.Based on this,the key construction technologies of aero-engine knowledge graph to aero-engine design are explored in this paper.The foundation for aero-engine knowledge mining and application is also constructed.The main contents are as follows:To solve the problem of the lack of aero-engine knowledge graph corpus,the aero-engine entity data set and the relational data set are established by sorting out the text data accumulated in the current aero-engine development process.The qualities of these data sets are evaluated.The experimental results on entity data sets show that the F values of lattice LSTM and LR-CNN models proposed by other researchers on entity data sets reach 91.98% and 92.85% respectively,which are basically equivalent to the scores of the two models on MSRA data sets.The experimental results of relational datasets show that the scores of CNN,BLSTM and Bert-BGRU-Attention models on ours dataset are basically the same as those of the two high-quality annotated Chinese relational datasets.The comparison experiments show that the two both data sets can satisfy the needs to construct aero-engine knowledge graph.To perform the task of entity extraction in the construction progress of aero-engine knowledge graph,the Lattice-Transformer model that introduced vocabulary information into the character-based entity extraction structure to improve the recognition performance index is proposed.On the Weibo,MSRA public datasets and ours entity data sets,the F value of lattice transformer model achieved 65.98%,95.09% and 95.21%respectively,which were higher than the model scores of other researchers,and the p value and R value also achieved good scores.For the task of relational extraction to construct aero-engine knowledge graph,a bidirectional gated recurrent network model based on pre-training language model and attention mechanism improvement—Bert-BGRU-Attention model is proposed.The experimental results show that the F value of the model is slightly higher than Self-ATT-CNN,Att-BLSTM,TRE,ITEM,et.al models proposed by others,reaching88.02% and 68.33% respectively on two foreign public datasets.The F values of the models reached 88.02% and 87.96% respectively on the two Chinese relational datasets.On ours relational data set,the P value,R value and F value of the model are 88.24%,89.47% and 88.85% respectively.Finally,Neo4 j database is used to store the extracted results in the paper.The aero-engine knowledge graph system platform is established to explore the application of the knowledge graph.Taking the automatic construction of fault tree for one engine "engine fulcrum bearing working abnormally" as an example,this paper explores the application of knowledge graph in the aided Aero-engine Design and development.