Research On The Key Technology Of Fault Knowledge Modeling And Fault Reasoning Of CNC Machine Tools

The competitiveness of future manufacturing depends not only on the accuracy of the equipment,but also on the ability to improve the reliability and availability of the equipment and a full life-cycle maintenance strategy for the equipment with long-term impact.According to The Freedonia Group(USA),in general mechanical engineering,machine tools are responsible for 40% to 60% of the total engineering workload.The demand for CNC machine tools in China is growing steadily at an annual rate of 9.2%.To make full use of CNC machine tools,CNC machine tool fault diagnosis as an important topic has received a lot of attention from academia and industry,and a lot of research has been conducted.The current main work in the field of CNC machine tool fault diagnosis includes fault diagnosis and repair of CNC machine tools,maintenance strategy support of CNC machine tools,etc.However,the current number of personnel engaged in the field of machine tool fault diagnosis and maintenance in China is insufficient,especially the lack of experts and scholars in the field with rich practical experience and theoretical knowledge.The skills and experience of these experts are also often extremely limited because of the geographical,time and professional differences that cannot be shared through communication.Sometimes,due to the complexity of faults,multiple experts from different fields are required to collaborate to solve them.In addition,the traditional "business trip" type of machine tool fault diagnosis and after-sales strategy obviously cannot meet the modern manufacturing industry’s demand for high availability of CNC machine tools.This paper investigates a prominent issue in the health management of CNC machine tools: how to learn from historical cases the implicit correlation mechanism of failure causes,phenomena and solutions,so as to realize remote fault diagnosis and health management of CNC machine tools.Specifically,this paper extracts the semantic features in the historical cases from the perspective of semantic analysis and builds a fault diagnosis system for CNC machine tools by doing so.This system uses the knowledge extraction method to organize the historical fault cases of CNC machine tools into a fault knowledge base and designs a knowledge inference model to realize fault inference diagnosis.This system can assist maintenance engineers to quickly locate the cause of faults and realize remote diagnosis.The main research contents of this paper are as follows.(1)Semantic pre-processing research of fault knowledge.For the problem that the original fault case data and user input fault phenomena are not standardized and unified,natural language reasoning and interpretation models are applied to realize word separation,remove deactivated words and mine the semantic/syntactic association of key words by analyzing fault descriptions.This paper also applies data visualization methods to show the correlation of each word group in the fault case database after interpretation.(2)Knowledge graph construction of fault cases.The traditional relational knowledge base is extended to a fault knowledge graph,and the elements in the case base are associated with predicates based on the statistical information in the fault case base,such as the number of times a fault cause causes a fault phenomenon,the number of times multiple fault phenomena are accompanied,etc.,and the strength of the association is described by assigning weights.On this basis,data visualization and graph data analysis methods are used to mine the potential failure patterns and features in the historical cases.(3)Research on fault inference based on knowledge graphs.A fault inference mechanism based on dynamic graphs is constructed to extend the traditional multilayer inference networks into graph neural networks(GNNs)and realize the variational inference of Markov logic.The Markov logic-based graph neural network integrates the inherent semantic information of fault cases and the connection between cases.(4)Construction of fault mode prediction model based on graph data mining.The temporal evolution of the fault knowledge graph is investigated to make the fault knowledge graph conform to the evolution law of different life cycles of CNC machine tools and to predict the interaction between complex nonlinear fault entities through knowledge evolution modeling.(5)Planning and development of remote fault diagnosis system for CNC machine tools.The basic requirements of the fault diagnosis system and its development objectives are discussed,and the design of fault case knowledge base,workshop machine tool information knowledge base and fault diagnosis system are realized.The PC-based and mobile-based applications are also developed to provide the theoretical basis and experimental platform for the application of remote fault diagnosis.The experiments show that the remote fault inference model constructed in this study can correctly identify 89% of the fault types based on the current case library and provide corresponding solutions;the fault knowledge mapping time series prediction model proposed in this paper can simulate the fault occurrence pattern of CNC machine tools throughout their life cycle.The work in this study can provide a theoretical basis and application feasibility for further customized fault diagnosis and failure mode prediction.Future research includes extending the fault diagnosis mechanism of single device to swarm-based diagnosis based on migration learning and constructing separate user portraits for different machine tools to achieve personalized diagnosis.