Simulation Of Particle Reinforced Composite Materials Damage Evolution In Macro-and Meso-Scales Based On GPU

As a new type of composite material, particle reinforced composite materials, which has good mechanical properties and secondary machining, have been widely used in mechanical, biological, aerospace, military, motor and other important industrial areas. With the development and progress of science and technology, lots of numerical simulation have been carry on in home and aboard. For the accurate simulation of some complex models, the researchers must solve two questions that are long computation time and big occupied computer storage.According to the programmable ability of GPU and the ability of high speed parallel processing, this article makes CUDA in both macro scale and micro scale’s numerical simulation of particle reinforced composite materials and this method can speed the parallel computing better. The mixed programming of CUDA Fortran and Fortran program is achieved. Through a simple example, the difference between CUDA parallel program and Fortran serial program briefly is analyzed. At the same time, the rationality, feasibility and superiority of GPU acceleration is highlighted. The article focuses on analyzing CUDA programming essentials, and it is the important to make full use of Share memory, so not only can it hidden the delay of the data transmission better, but it also saves the storage of space and time, and improves the efficiency.The article improves the algorithm of the original application process through CUDA. The original program need calculate the displacement of macro finite element and the whole components is divided into discrete micro model scale and macro model scale at first. Next, call the CUDA program to instead the program code of corresponding Voronoi unit stiffness matrix. This article adds two subroutines, that are using the program of H matrix calculation and H_C matrix calculation instead of the original subroutine, The stress and stain response of the whole components can be got by the coupling of the final computing result and the computing result of macro element at last.Through the establishment of related different calculation model, the process of the damage evolution process of simulating particle is reinforced.The article analyzes the computing speed of modals under the different experimental conditions, the different gauss point number and the different stress parameters, and proves the method through contrasting the calculation results by the original program sufficiently. It is the main conclusion that when the data size of calculation is small, the acceleration of GPU is not obvious, and when the data size of calculation is bigger, the celeration of GPU is more obvious.lt is also turned out that the more core of CUDA, the better effect of acceleration.