CUDA기반 평면 구속 다물체와 대량 입자들과의 충돌해석

Abstract

GPU has been constantly developed by increasing cores in bulk, accordingly a graphic card has been able to handle more colorful and natural 3D images. To apply this powerful GPU performance to the general data processing, NVIDIA(a GPU manufacturer) has been making various graphic cards which can be used in general program and They published CUDA(compute united device architecture) to help general programmers can develop their program by using GPU easily. CUDA has been applied to various fields of study, but very little is known about the application to MBD(multi-body dynamics) in our country. In case of computational analysis of MBD that has many contact elements, it takes a considerable time and computer memory. However, if CUDA is employed in this MBD area, the analysis time should be significantly decreased.

In this study, We employ GPU and CUDA to carry out parallel computing of collision analysis of 2D multi-body and mass particles. In order to detect collisions between modeled multi-body and mass particles, Broad phase collision detection algorithm and DEM(discrete element method) are considered. In addition, Kelvin-Voigt contact model is used to calculate contact force.

A modeled example for this study has a multi-body system and many particles. The multi-body system has 20 circle shaped objects, 19 spring-damper force elements, and 2 revolute joints. We make serial and parallel analysis program and compare simulation time of two programs for this example increasing the number of particles. Consequently, it is noticed that the more particle is, the more computing time we can save. Expecially, for this example, when the number of particle is 900, the computing speed of parallel analysis program is about 5x faster than that of serial analysis program.