Research On Automatic Construction Method Of Process Knowledge Graph

Machining process,as an important manufacturing production,has a large amount of potential knowledge,and the effective use of this knowledge can provide a data base for machining process optimization.Aiming at the problems of difficulty in extracting such knowledge and isolation from each other,we propose to automatically construct a knowledge graph of machining process to achieve semantic unification of diverse knowledge,so as to provide a new research idea for the digital twin of manufacturing industry.Aiming at the above research directions,the main research work is as follows:(1)Design and implementation of a process ontology model,including the theoretical knowledge required to develop the process and the process knowledge generated by the machining process.Under the guidance of domain experts,the ontology modeling language OWL Full is used for modeling.The model provides a unified model for knowledge graphs and realizes the organic integration of multiple knowledge.(2)A process information extraction method based on a pre-trained model is proposed.Firstly,it defines processing entities and relationship categories for processing process text language characteristics,formulates processing process domain biased entity annotation rules and defines relationship division methods,and then constructs processing process corpus;secondly,it establishes a domain adaptive multi-network coordinated Chinese named entity recognition method to effectively distinguish entity boundaries for processing entities,etc.;it establishes a pre-training and knowledge representation based relationship extraction architecture for processing process.Finally,a relationship extraction architecture based on pre-training and knowledge representation is established to realize automatic extraction of domain relationships.(3)A knowledge fusion method based on knowledge representation learning and clustering is proposed.For unstructured data,triads are generated and a similarity entity alignment method is established;for structured data,data-ontology mapping rules are formulated based on a virtual mapping architecture;finally,knowledge representation learning is used to generate text vectors and combined with clustering algorithms to achieve multi-source heterogeneous triad knowledge fusion and generate processing process knowledge maps for triad data.(4)Develop a B/S structured processing process knowledge graph application system.Determine the knowledge storage framework of Neo4j graph database for processing process triad,and finally realize the construction of processing process knowledge graph;adoptjQuery development framework to develop user interface and integrate and develop knowledge graph related applications.The system functions include information retrieval,process planning and information extraction.