Design And Research Of Equipment Fault Intelligent Question Answering System Based On Knowledge Graph

In the process of tobacco production and processing,re-roasting of tobacco leaves is an essential step.During the production process,prolonged use of machinery and equipment can lead to various malfunctions.Often,these malfunctions require professional maintenance personnel to handle them.However,the actual maintenance process requires a lot of time,which can lead to production downtime and increased costs.The maintenance data stored afterwards is also not convenient for subsequent retrieval and use.Therefore,this thesis constructs a knowledge graph based on equipment failures based on these equipment maintenance data,combined with related technologies such as knowledge graphs.Named entity recognition technology is used to extract entities from equipment failure data,and polynomial naive Bayes is used for semantic understanding of question sentences,ultimately constructing an intelligent question-answering system based on equipment failures.The main research content of this thesis is as follows:(1)In the field of re-roasting equipment,a BERT-Bi-LSTM-Attention-CRF model is proposed for named entity recognition.The BERT-Bi-LSTM-CRF model is improved by introducing the Attention mechanism.By comparing commonly used models in the named entity recognition field and conducting experiments on different datasets,the effectiveness of the BERT-Bi-LSTM-Attention-CRF model in entity recognition of equipment failure datasets is verified.Moreover,experiments are conducted to compare the hyperparameters of the model,and the best experimental results of the model under this experiment are obtained,with P,R,and F1 values reaching 87.93%,91.79%,and89.81%,respectively.(2)A TF-IDF-based polynomial naive Bayes model is proposed for semantic classification of equipment failure query sentences.A dataset of equipment failure query sentences is constructed based on equipment failure entities and their relationships,as well as common query methods for equipment failures.TF-IDF is introduced to improve the polynomial naive Bayes model.The dataset of equipment failure query sentences is used to train the TF-IDF-based polynomial naive Bayes model,and the experimental results are compared with those trained under the polynomial naive Bayes model,verifying the effectiveness of the model.(3)A knowledge graph-based intelligent question-answering system for equipment failures is constructed.The front-end and back-end systems of the question-answering system are built using the Flask and Vue frameworks,with data exchange between the front-end and back-end systems using Axios.In the back-end system,the BERT-BiLSTM-Attention-CRF model is used for entity recognition of the question,and the TFIDF-based polynomial naive Bayes model is used to obtain the query intent of the question.The two are combined to construct Cypher statements and query the answer to the question in the Neo4 j graph database,which is then returned to the front-end page to achieve the entity recognition and intelligent question-answering functions of the system.A test set is randomly constructed to test the system’s functionality,with an entity recognition accuracy of 92% and an intelligent question-answering accuracy of 90%.