Numerical Computing Method Of Large-scale Engineering Structure Based On Distributed Heterogeneous Platform

With the numerical analysis model of engineering structure gradually becoming more refined,the fine-grained structural analysis models have a large degree of freedom,which places higher demands on the efficiency of numerical calculations.At present,most of the numerical analysis platforms at home and abroad are based on traditional CPU computing platforms,and the computing efficiency is generally low,Therefore,how to improve the numerical computing efficiency of large-scale engineering structures has become an urgent problem to be solved.This paper develops FEM(Finite Element Method)and DEM(Discrete Element Method)geometric nonlinear parallel algorithms based on CPU-GPU heterogeneous platform,The high-precision and high-efficiency of geometric nonlinear numerical calculation of large-scale engineering structures are realized.Based on the vector mechanics,based on the beam element constitutive,the deformation description mechanism is introduced on the basis of the CR formula computing theory,and the deformation increment of the element is obtained by subtracting the rigid body displacement through the virtual reverse motion,and the element is obtained by superimposing the increment of the forward motion of the element,the total deformation is combined with the Newton-Raphson method to establish the finite element geometric nonlinear static computing method.On the basis of static computing method,based on the basic principle of Newmark-? method,the finite element geometric nonlinear dynamic computing method is established.The accuracy of the static and dynamic algorithms was verified by using hexagonal space frame and K6 reticulated shell as the computing models.Based on the CPU-GPU heterogeneous platform,and based on the potential parallelism of CR columnar calculation theory,the computing methods of unit parallel,node parallel and degree of freedom parallel are proposed.Using the CUDA architecture's "fine-grained" programming design idea,a finite element geometric nonlinear static and dynamic parallel algorithm is implemented on a CPU-GPU heterogeneous platform.Hexagonal space rigid frame and K6 reticulated shell are used as the calculation models to verify the accuracy of static and dynamic parallel algorithms respectively.The computing efficiency and accuracy of the parallel algorithm are verified with the large frame structure as the calculation model,the maximum acceleration ratios of static and dynamic parallel algorithms reached 17.95 and 17.25 respectively.The DEM of the rod system is completely independent in solving the equation of motion and calculating the internal force of the element,with natural parallelism.In this paper,based on the inherent parallelism of DEM computing theory,this paper proposes the computing method of unit parallel and node parallel,and establishes the corresponding DEM parallel computing framework and corresponding geometric nonlinearity static and dynamic computing programs based on the CPU-GPU heterogeneous platform.The computing accuracy of DEM static and dynamic parallel algorithms was verified by using hexagonal space frame and frame structure respectively.A large spherical shell model was used to test the dynamic parallel algorithm,the test results show that the DEM parallel algorithm has a good acceleration effect compared to the conventional serial computing program,the maximum acceleration ratio of the dynamic parallel algorithm reached 12.70.