| With the proposal of "Made in China 2025",the production mode of manufacturing industry is transforming to intelligence.As one of the important ways to realize intelligent manufacturing,digital twin has attracted extensive attention from researchers at home and abroad.As a common manufacturing means,welding is widely used in machinery manufacturing,aerospace and other fields.Quality monitoring of welding process is an important means to improve welding quality.Aiming at the problems of low intelligence and poor precision in real-time detection and analysis technology of welding quality in current automobile parts manufacturing,this paper takes a certain body-in-white welding production line as the research object,and researches on welding quality monitoring system of automobile parts based on digital twin.The main research contents are as follows:(1)Firstly,according to the actual demand of quality control system of a certain body-in-white welding production line,the design requirements and design principles of the system are proposed.Then,the function modules of the system are designed.Finally,the network architecture and function architecture of the system are determined to establish the overall framework of the auto parts welding quality monitoring system based on digital twin.(2)The digital twin models of welding production line including geometric models,motion models and physical behavior models are constructed.Firstly,the geometric models of welding production line are constructed by UG,3ds Max and Unity3 D.Secondly,two kinds of motion models driven by real-time data are controlled based on motion transformation matrix,and the motion errors are detected based on forward kinematics of robot,so as to realize high-precision dynamic motion simulation of the models.Then,the physical behavior models are built based on the mapping of physical rules in the machining process to simulate the physical processes such as collision detection and welds generation.Finally,model grid optimization,occlusion elimination algorithm,LOD technology and model batch processing are used to optimize the virtual scene,which provided technical support for welding process monitoring and process optimization.(3)A Dense Net(Dense Convolutional Network)weld defect recognition model based on transfer learning is proposed.The Dense Net is improved by transfer learning technology,so as to optimize initial network parameters,reduce training time and improve the model accuracy.The result of trained model shows that the accuracy up to 99.8%.The effectiveness and superiority of the algorithm is verified further by the control experiment.(4)Based on the analysis of the characteristic parameters affecting welding spatter during the welding process,an optimized BP neural network prediction model based on Hyperband algorithm is established to predict welding spatter,which takes the main characteristic parameters as the input and the number of spatter as the output.The experimental results show that the prediction accuracy of spatter is 88%.Furthermore,the adaptive control strategy of welding process parameters is proposed to improve welding quality by controlling welding current and time,so as to realize real-time detection,analysis,prediction and control of welding quality.(5)The visualization module,welding defect recognition module and process adaptive module of the welding quality monitoring system are developed and integrated by C# and python.The overall performance,module performance and display performance of the system are tested.The results show that the system can keep a good running state. |
