Research On Diesel Engine Fault Diagnosis Based On Domain Knowledge Graph

Diesel engine,as a common power device,is widely used in many fields such as vehicle,ship,military industry and so on.Once the diesel engine fails,it will not only lead to production interruption,property loss,but also serious safety accidents.Therefore,it is necessary to diagnose the cause of diesel engine fault timely and accurately and ensure its safe and reliable operation.However,the knowledge distribution of diesel engine fault is sparse and lack of correlation,which makes it difficult to use the knowledge efficiently and brings great difficulties to diesel engine fault diagnosis.Based on this,this paper studies a new way of diesel engine fault diagnosis based on knowledge graph supported by the national key R & D plan.It describes fragmented diesel engine fault knowledge by using knowledge graph,and uses Bayesian network to infer the cause of failure.The named entity recognition algorithm,relation extraction algorithm and knowledge graph complement algorithm proposed in this paper can also provide reference for the construction and improvement of knowledge graph in other fields.This paper mainly studies following aspects:In order to obtain the entities needed to construct knowledge graph from the text of diesel engine fault diagnosis,this paper proposes word-set attention guided Chinese named entity recognition method.On the basis of introducing external vocabulary information,the word set vector is divided according to the position of the character in the word.By using the attention mechanism at word set level,we can focus attention on the appropriate word set,ignore the unreliable part,overcome the disadvantage of traditional methods ignoring the importance of each word set,so as to improve the recognition effect of Chinese named entity.This paper verifies the method by means of public data set and named entity identification data set in diesel engine fault diagnosis field.In order to obtain the relationship needed to construct knowledge graph from diesel engine fault diagnosis text,this paper studies the relationship extraction algorithm,and proposes relation extraction based on multi-scale attention and BERT method.The method uses special symbols to mark the entity to reflect the boundary and position of the entity,and solves the problem of BERT using entity information in this way;In order to reduce the influence of redundant information in sentences,attention mechanism is used to improve the accuracy of relation extraction in word scale and segment scale.This paper verifies the method by using the open data set and the data set of the established diesel engine fault diagnosis domain relationship extraction data.In order to solve the problem that the fault knowledge graph of diesel engine constructed by the above steps is incomplete,this paper studies the knowledge graph completion technology and proposes non-local convolution embedding method.Based on the traditional method,this method introduces non-local operation,establishes long-distance dependency,increases the interaction between entities and relationships,and improves the effect of knowledge graph completion.The method is verified by open data sets,and an example is analyzed on the constructed diesel engine fault knowledge graph.This paper combines the knowledge graph of diesel engine fault with Bayesian network to diagnose the fault.The Bayesian network is constructed based on the fault reason query subgraph of the knowledge graph,and the most likely fault reason is deduced by Bayesian network,and the fault diagnosis is completed.This method solves the problem that the results of knowledge graph retrieval have little priority,and the problem that the construction of Bayesian network structure needs the participation of experts.This paper verifies the effectiveness of the method by a fault example.