Building And Application Of A Metal Cutting Process Knowledge Graph

Metal cutting is an important forming method of metal processing technologies,and the design of metal cutting process is one key content of process planning."Machine tool-cutting tool-workpiece" constitutes a complete machining process system under which the interactions between cutting tools and workpieces form the cutting process.This paper classifies the data related to metal cutting into factual knowledge and process knowledge.The former describes all kinds of physical phenomena and their changing mechanisms of metal cutting,while the latter generally refers to all kinds of data stored in various application systems,which represent manufacturing enterprises to process workpieces with cutting methods.The sources of these knowledge are diverse and the data types are rich,including structured,semi-structured and unstructured.To solve the problem that these two types of knowledge are isolated and cannot be transformed into each other,this paper proposes to build a metal cutting process knowlege graph(MCPKG)to integrate them together in semantic level,which provides an novel approach to build an intelligent brain for metal cutting.Specific research contents are shown as follows:1).On the basis of analyzing the components and sources of factual knowledge and process knowledge,a complete ontology model correlating these two types of knowledge is built with OWL QL for the first time,which lays a foundation for the construction of the MCPKG.2).Build the basic approachs to generating,fusing and storing the data of the MCPKG.For factual knowledge,a "Bi-LSTM+CRF" knowledge extraction framework is established based on NLP(natural language processing).For process knowledge,based on OBDA(Ontology-Based Data Access)architecture,the mapping rules between relational data models and ontology models are designed to realize the transformation from relational data to triples.In addition,the data fusion method based on attribute similarity is established,and the data storage architecture "neo4j+Minio" is determined as well.3).Establish a new similarity computational model for process routes of workpieces with knowledge representation learning.First,a subgraph including the relationships among "workpiece-material-process-machine tool-cutting tool" is established over the MCPKG,and it connects the related entities with the relationships "material?is","in?machine","use?tool","has?process" and "next?process",so as to present the semantic meaning of these entities through the connectivity of graph structures in a direct way.Then,the corresponding triples are generated according to the structure of the subgraph,and Trans D algorithm is used to map the semantic relations in the subgraph to low-dimensional dense real value vectors.After that,K-Means algorithm is adopted to cluster these low-dimensional dense vectors representing the workpieces.The results show that the workpieces with a similar process sequence and processed by same cutting tools and machine tools can be better clustered into a same cluster.This method breaks through the singleness and limitation of the traditional classification according to the type of workpieces.Finally,typical process routes are extracted from the cluster results,which further verifies the expression ability of semantic meaning of lowdimensional dense vectors.This similarity computational model provides a novel,efficient and accurate way for process route resue.4).Set up a novel,personalized and accurate cutting tool selection approach.The core idea is that the internal relationships of the MCPKG is an important basis for tool selection,taking the material of workpieces,the structural characteristics of machining tasks and the correlations between tools as the primary factors for tool selection.First,a data model describing the relationship "structural feature-material-cutting tool" is designed to build a subgraph.Then,the personalized Page Rank(PPR)algorithm is employed with the data model to score and rank each tool,providing a precise basis for cutting tool selection,and the result of an illustrative example is discussed in detail to verify the algorithm.5).Develop an integrated application system of the MCPKG with B/S(Browser/Server)structure.The system is based on.NET MVC(Model View Control)framework,using C# language in the server side,j Query and easy UI in the client side,and d3.js is leveraged for visualization.The storage architecture of the system is designed as "Neo4j+Min IO+Oracle",in which the graph database Neo4 j stores triples,the document database Min IO stores various unstructured documents,and Oracle stores structured data like user and permission information.The system function modules include data management,knowledge graph visualization,process route planning,cutting tool selection and user & authority management.