Research On Assistant Assembly Technology Based On Digital Twins

With the development of science and technology,the structure of industrial equipment has become complex.In the production activities of manufacturing enterprises,assembly is an important and complex link.In recent years,artificial intelligence and image recognition have been rapidly developed and applied,and began to be used in the auxiliary assembly work.Most of the current auxiliary assembly technologies are combined with augmented reality technology.The camera is used to take pictures of real objects and collect images.Through in-depth learning and training models,parts are accurately identified to achieve auxiliary assembly.In view of the difficulties in early sampling and the lack of sample data in industrial auxiliary assembly engineering,this paper proposes a method to use its digital twins to sample and enhance samples,and forms a sample set with the actual sample data collected in the real environment,trains the recognition model and applies it to the physical inspection,finally supports the identification of auxiliary assembly parts,and completes the purpose of auxiliary assembly.The main research contents of this paper:(1)In this paper,the requirements of the aided assembly system are studied in depth,and the technical route of the system development is given.This paper analyzes from four aspects: image acquisition,image annotation,model training and object recognition.Determine the development platform and development tools,select the convolution neural network of deep learning training image model,and design the system function module.(2)In this paper,the digital twins are established to use 3Ds Max software to carry out 3D modeling of the target object,import it into the unity3 D virtual space,build a simulation environment,put the digital twins into the simulation environment,simulate natural light through the unity3 D lighting system,and build a complete simulation environment in the virtual space.This paper uses the virtual camera in the virtual space to complete the sample sampling of the digital twin,controls the position,attitude and angle of the virtual camera through the algorithm,and generates multiple samples through the random range control of the camera coordinates to form a sample set.(3)This paper takes the tire of a certain type of vehicle as an example,designs a comparative experiment to verify the feasibility,builds a virtual scene in unity3 D,puts its 3D model,collects the first sample of the 3D model,and collects the photos of the same type of vehicle on the network to form the second sample,trains the two recognition models for physical inspection through fast r-cnn,and compares their intersection and comparison,It is verified that the digital twin space sampling can be applied to physical detection.