| As an enabling technology and means to practice the concept of intelligent manufacturing,digital twin technology can effectively solve the problem of cyber-physical integration of intelligent manufacturing.Currently,it has become the focus of academic circles and industrial circles all over the world.At present,in the field of industrial process monitoring,the utilization of data in the manufacturing process is mostly concentrated in the form of intuitive visualization for production management and control,and as the source of historical data,instead of being used to realize the virtual and real mapping and interactive fusion of physical space and information space.Manufacturers have an increasingly urgent need for real-time display and online monitoring of relevant status monitoring data such as equipment operating status and product production quality during the manufacturing process.Digital twin technology can accurately simulate and depict the behavior of physical entities in the real world.The establishment of digital twin model of the automated manufacturing process can represent and synchronously map the manufacturing process in real-time,which has important practical significance for enterprises to optimize the production process,reduce cost and increase efficiency,and improve quality.This thesis is based on the National Natural Science Foundation of China and the Guizhou Provincial Science and Technology Major Program.According to the actual engineering project requirements of an aerospace science and industry enterprise in Guizhou,the automated production line manufacturing process’ s digital twin model is established to realize the real-time perception and interaction of twin data.Data mining is performed on the data that cannot be intuitively used to represent the current state of the twin data,to solve the problem of real-time representation and synchronous mapping of the tool wear status of the CNC machining center on the production line and the product surface quality status,which is the status of the digital twin technology in the manufacturing process application in the monitoring field provides theoretical and technical support.Specifically include the following:(1)Aiming at the construction of the digital twin system of the automated production line manufacturing process,based on the analysis of the existing digital twin system architecture in the manufacturing workshop,the digital twin system architecture of the automated production line manufacturing process is proposed,and the relationship between the various levels in the system architecture is explained in detail;This thesis starts from the three dimensions of the critical elements of the twin model: entity modeling,digital twin virtual modeling,and virtual-real mapping association modeling.A unified logical data model has been established.A physical production line of an aerospace industry product has been built.The virtual model provides a virtual model for researching the subsequent production line manufacturing process’ s digital twin system.(2)Aiming at the problem of twin data perception and interaction,a unified architecture-standardized communication protocol is established based on OPC UA technology to solve the complex issue of data perception and interaction caused by the inconsistent communication interface protocol of different devices on the automation production line;based on the object Constructed the critical element data perception model with the node information modeling method,built additional software and hardware interfaces to solve the integration problem between twin heterogeneous data;A real-time mapping of the production line information space to the physical space is established to drive the continuous iterative update of the virtual model in the information space and to lay the foundation for subsequent twin data mining and the status monitoring of the production line manufacturing process.(3)Aiming at the problem of building a digital twin rule model for monitoring tool wear status of CNC machining centers,a lightweight status monitoring model based on deep-gated recurrent unit neural networks is proposed,which is then used to build a digital manufacturing process for automated production lines at a macro level.The twin model provides the basis.Aiming at the problem of tool wear assessment,this thesis proposes an algorithm of the CABGRUs network model.In the preprocessing stage,the time sequence signal collected by the acceleration sensor is denoised by wavelet threshold,and the lengthy signal generated by each tool feed is divided.Multiple training samples are used to filter out the noise and improve the robustness of the algorithm;use a convolutional neural network to extract features from time-series signal input adaptively,and construct a deep two-way gated recurrent unit neural network to learn time-series information between feature vectors;The idea of the attention mechanism is introduced into the improved network model,which effectively improves the recognition accuracy and generalization performance of the real-time monitoring of the network model.The experimental results show that the prediction accuracy of the proposed CABGRUs network model reaches 97.58%.The single test time is 8ms,which is better than traditional machine learning algorithms and can support the tool wear status digital twin rule model’s accurate construction.(4)Aiming at the problem of constructing the rule model of the digital twin of the surface quality inspection of products in production,and improved online inspection model of product surface quality based on model compression is proposed,which is deployed in edge computing equipment to identify the types of surface defects on the product and online mark the defect location,and then provide a basis for building a digital twin model of the automated production line manufacturing process at the macro level.Aiming at the problem of product surface quality evaluation,an algorithm to improve the YOLOV4 network model is proposed.That is,the data enhancement method using CutMix and the K-means++clustering algorithm is offered in the preprocessing stage to enhance the robustness and generalization ability of the model and so that the generated boundary candidate box can be screened out more advanced features earlier;with the help of CSP Darknet53 network and SPP module to extract the features of the original input image,through training to obtain the online detection model of the product surface quality;on this basis,the model is compressed,Under the condition that the accuracy of the model remains unchanged,the model volume and forward reasoning time are effectively reduced.The experimental results show that the proposed method based on model compression and improved deep convolutional neural network can compress the model size from 244 M to 4.19 M,with a prediction accuracy of97.41% and a detection speed of 47.8 f/s(AGX Xavier test results),both it is superior to similar deep learning algorithms and can support the precise construction of a digital twin rule model for product surface quality inspection.(5)In response to the actual needs of an automated production workshop for a product produced by an aerospace science and industry enterprise in Guizhou,this thesis proposes a technical framework for the development of a prototype system,developed a database application program and a prototype system for automated production line manufacturing process condition monitoring and applied it in the actual automated production line manufacturing scenario,the feasibility and effectiveness of the proposed related theories and technologies are verified. |
PDF Full Text Request