Research And Application Of Aircraft Maintenance Entity Relationship Construction And Fault Reasoning Algorithm

With the continuous improvement of the global aviation maintenance,repair ＆operations（MRO）market influence and increasing market share,China’s aviation MRO industry has also faced huge challenges while achieving rapid development.my country’s civil aviation maintenance industry gives full play to its technological advantages,researches civil aircraft maintenance plans based on big data,and seeks more reasonable plans to enhance aircraft performance and reduce maintenance costs,which is of very practical significance.The big data application technology of processing linked data represented by knowledge graph is one of the paths to better solve the aircraft maintenance plan.However,in the field of aircraft maintenance,knowledge graph research work is relatively lacking,and relevant data sets are lacking.Therefore,based on the text information of aircraft maintenance logs of domestic large airlines,this thesis analyzes the classical algorithm of entity naming and relationship extraction and optimizes and improves it,and obtains the knowledge graph in the field of aircraft maintenance.For problems with relatively low and large sparsity,use knowledge inference algorithm to complete them.The main work of this thesis is as follows:1.Aiming at the problems of fuzzy entity boundary,long entity length,large and complex number of entities,and small sample size in this field,this thesis deeply studies the BERT-Bi LSTM-CRF algorithm for adversarial training and existing entity naming based on annotated maintenance data.The third chapter of this thesis improves the Tokenizier word segmentation in the BERT embedding layer,and then uses adversarial training to add perturbed samples to the embedding layer,and finally enters Bi LSTMCRF to finally realize entity recognition in the field of aircraft maintenance;use sentencelevel and word-level two-layer Bi LSTM-Att algorithm conducts research on relation extraction tasks and generates a knowledge graph in the field of aircraft maintenance.Relevant experiments show that the improved method can achieve good results on the dataset in the field of aircraft maintenance.2.In view of the low integrity of the knowledge graph in the field of aircraft maintenance and the large sparsity problem,the path-based knowledge inference algorithm generates and selects the path feature set by random walk,which will introduce bad paths,and this kind of algorithm handles Low-connectivity graphs（data sparse cases）do not work well;and reinforcement learning knowledge inference methods such as Deep Path and MINERVA’s reward functions require manual tuning to achieve good performance,which is not only inefficient and laborious,but also difficult to adapt to realworld KGs In this thesis,we propose a reinforcement learning knowledge graph reasoning method based on generative adversarial imitation learning.Relevant experiments show that the knowledge reasoning method in this thesis achieves good results in complementing the knowledge graph of aircraft maintenance.3.In terms of practical application,this thesis uses related algorithms to extract the entities and relationships of airborne sensors in the field of aircraft maintenance from a large number of unstructured text data,and uses the Neo4 j graph database to store the relevant data,forming machine knowledge in the field of aircraft maintenance.Atlas.And based on the graph data,combined with the knowledge graph reasoning algorithm proposed in this thesis,the application of the adaptive maintenance plan in the field of aircraft maintenance,the application guide of the maintenance manual,the identification and guidance of the airborne sensor with high failure rate,and the detection of the potential coupling fault of the airborne sensor are carried out.Practice shows that related applications can play an active role in the daily maintenance process,which has strong practical significance.