Research On Construction And Application Of Crane Maintenance Knowledge Graph

The crane is a common lifting equipment in modern industrial production.As a complex automated mechanical equipment containing multiple components,the crane is prone to breakdowns due to long and intensive work.Historical fault investigation records information such as fault history,cause of fault and solution measures.This information is often stored in a semi-structured or unstructured form,and it has characteristics such as non-standard content descriptions and uneven data quality,which makes later retrieval and utilisation difficult.At present,the crane maintenance still relies on manual experience,with low utilisation of historical maintenance data and low maintenance efficiency.In this thesis,based on the crane fault investigation records in a company,a knowledge graph for crane maintenance was constructed through knowledge extraction according to the characteristics of the corpus.The historical maintenance knowledge was structured in the form of a knowledge graph.An application system were developed based on the knowledge graph to assist equipment maintenance,so that the overall maintenance efficiency can be improved.The main contributions are as follows:(1)A bottom-up and top-down combined ontology construction method was proposed.The concept extraction algorithm of multi-feature fusion was used in a bottom-up manner to extract the core concepts in the domain,and the related strengths between concepts were calculated based on Dice measure.Then,a top-down method was used to inductively summarize the ontology of the overhead crane maintenance domain.Finally,a self-built ontology editing tool was used for ontology persistence.(2)Knowledge extraction was conducted on the crane fault investigation records.To address the entity nesting and long entity recognition issues in the corpus,a machine reading comprehension model fused with reinforcement learning was proposed,which performed entity recognition in a question-and-answer form and used pointer networks for decoding.This model solved the entity nesting and long entity recognition problems and further improved the recognition performance using reinforcement learning.To address the relationship overlap issue in the corpus,the relationship extraction was divided into two stages: recognizing the subject first and then the object.The recognition of the relationship between multiple entities was isolated to solve the sentence-level relationship overlap problem.(3)Based on the Django framework,the thesis designed and implemented a crane maintenance assistance system,which provided functions such as knowledge visualization,knowledge inquiry,and intelligent question-answering.It can assist equipment maintenance and improve work efficiency.This thesis presents a complete process for constructing the crane maintenance knowledge graph.According to the characteristics of the corpus,this thesis uses a machine reading comprehension model fused with reinforcement learning to extract knowledge from fault investigation records,and develops a maintenance assistant system based on the knowledge graph,which provides an intelligent solution for crane maintenance.The work in this thesis has good reference value.