Knitted Cloth Simulation Based On Meshgraphnet

With the continuous development of computer industry,computer simulation technology is being used more and more widely in computer animation,industrial design and manufacturing,and AR/VR.On the other hand,due to the rapid development of textile industry,fabric simulation technology plays an increasingly important role in the design of fabrics,simulation of mechanical properties of materials and intelligent manufacturing.This makes fabric simulation a hot spot for computer simulation.However,the simulation of knitted fabrics is a difficult point in fabric simulation research.Because the knitted fabrics are prepared in different ways and modeled in a complicated manner,resulting in different physical deformation characteristics of different knitting methods are not the same,and the simulation methods using the triangular mesh and spring-mass model commonly used in fabric and woven fabric simulation cannot fully reflect these detailed characteristics.Therefore,knitted fabrics are often simulated by the yarn model through the force analysis and constraint method at the nodes.Although this method can reflect the deformation characteristics of knitted fabrics,it also makes the method computationally intensive,hardware demanding,and slow due to the complex loop organization.In response to the above problems,the following work is carried out in this paper to address these issues.(1)Using the periodicity of the yarn model,the periodic model of the yarn is simulated using the coil structure,and the correspondence between the deformation of this model and the energy density is established.The overall model is simulated using the lattice model to establish the correspondence between the yarn and the lattice model,and the final hybrid model is obtained by mapping.(2)The distance field calculated based on geometric and physical factors is used to perform adaptive partitioning of the mesh edges,with subdivision at locations where deformation is obvious and appropriate coarsening operations at locations where little deformation occurs.This method can improve the quality of the generated model for low-resolution input meshes,while it has the effect of reducing the computational effort and improving the computational efficiency for high-resolution meshes.(3)The Young modulus and shear modulus inputs are averaged and sampled,and then the meshes of each frame of the motion process are obtained by physical simulation,and they are used as data sets to learn the deformation effects and the adaptive mesh partitioning algorithm by using the meshgraphnet graph neural network composed of codec architecture with the information of the points and edges of the mesh and the distance field as inputs.Experiments show that the method can generate a needle fabric model with close to real effects and predict deformation about five times more efficiently than physical methods.