Knowledge Graph Modeling And Causal Reasoning For Machining Quality Of Aerospace Thin-Walled Parts

Aerospace thin-walled parts are important components of missiles,rockets,and other weapon equipment,with a large aspect ratio,high hardness and high strength,weak stiffness,and extreme use conditions.Its production process involves various complex machining processes of cold and hot forming,and the molding process is dynamic and uncertain.High-quality accuracy requirements and various quality influencing factors cause the problem of large fluctuations in quality consistency easily.Also,the shell machining quality directly affects the operating performance of the missile in harsh and extreme conditions,which has been a difficult problem faced in production.Therefore,analyzing the causes of aerospace thin-walled parts machining quality abnormalities and implementing interventions is of great practical value.Aerospace thin-walled parts machining has a systematic relationship chain of "productequipment-process-inspection".It has accumulated a large amount of data resources with various types,multiple sources,and heterogeneous as well as time-varying characteristics.How to transform these massive discrete data into valuable knowledge and guide the cause analysis of machining quality based on knowledge will play a significant role in enhancing intelligent cognitive analysis and intervention control of shell quality problems.In this regard,this paper focuses on the fundamental problem of analyzing the factors affecting the machining quality of aerospace thin-walled parts.The research on semantic knowledge integration and reasoning of tracing causes over aerospace thin-walled parts machining quality is carried out,mainly including fine-grained knowledge graph modeling of multi-source quality data of thin-walled parts manufacturing events,quality knowledge graph generation,causality enhancement of quality knowledge,link prediction and quality reasoning of tracing causes based on causal knowledge.The main contributions and innovations of this paper are as follows.(1)A knowledge graph construction method for aerospace products with multiple sources of machining quality data is proposed.In view of the characteristics of multi-source heterogeneous quality data,low information integration,dynamic change of events,and low knowledge storage of thin-walled parts in complex long-process machining,a knowledge modeling based on machining task events is constructed to simulate the continuous evolution process of thin-walled parts machining events in the physical world.Then,a fine-grained quality knowledge description space with global and multi-level correlation is formed.Further,a knowledge graph generation method is designed for multiple knowledge sources,such as industrial documents with tables,records with time series data,and defect detection in images.Thus,a knowledge fusion model based on a neighbor sampling graph matching convolutional network is established to form a quality knowledge graph that integrates multiple process machining resources.The method provides a knowledge base for causal reasoning of thin-walled parts machining quality.(2)A causality enhancement learning method based on a quality knowledge graph is proposed.Based on the backdoor adjustment strategy of causal science theory,the structural causal graph model of quality knowledge graph is designed.Also,a causality enhancement learning method of quality knowledge graph is established with the help of the sum-product network mechanism to remove intervention factors with the "pseudo-correlation".Finally,a cause-effect enhanced knowledge graph for supporting the traceability analysis of thin-walled parts machining quality defects is formed.(3)A causality-enhanced knowledge graph-based machining quality reasoning of tracing causes method for thin-walled parts is proposed.On the one hand,the semantic links of quality knowledge are incomplete.On the other hand,it is complicated to effectively track machining quality problems and mine the major factors.Therefore,a thin-walled machining quality knowledge representation and link prediction model is designed to realize the semantic vectorization of quality knowledge.Also,the potential implicit relationships of quality knowledge are also predicted.Thus,the relationship links of the quality knowledge graph are complemented.Then the reasoning of the tracing causes model is proposed,transforming the quality knowledge graph into a vector embedding based on the knowledge representation and link prediction model.Further,the question and answer mechanism of the language model is combined,which makes full use of causal knowledge to enhance the reliability of the reasoning of tracing causes.In this way,this approach is realized to provide intelligent support for causal analysis decisions of quality problems.In conclusion,this paper focuses on analyzing knowledge evolution modeling,graph generation,causal knowledge relationship enhancement,and intelligent reasoning for multiprocess machining quality data of aerospace thin-walled parts.The scientific path of semantic information integration and quality factors analysis from quality data is gradually realized.It provides theoretical support for constructing machining quality knowledge graphs and reasoning of tracing causes.Thus,the effectiveness of the method in improving the machining quality management of aerospace thin-walled parts is verified by a typical application in an aerospace research institute in Shanghai.Moreover,it is proved that the method provides a significant theoretical and practical guiding value for the causal analysis of machining quality problems over aerospace thin-walled parts.