Development of high-speed aerodynamic simulator of aerobraking spacecraft

研究了大气制动航天器气动特性模拟的大规模并行计算系统。模拟方案是直接模拟蒙特卡罗方法，这是广泛用于稀薄气体流动。对大规模并行计算系统的单元进行了概念设计。发现大规模并行计算系统的单元是用流水线方法实现的，不是由中央处理单元组成，而是由分布式处理单元组成。该装置的试制成功，并显示出预期的计算性能。该单元的运算精度与系统时钟速度成正比，在20MHz系统时钟下，运算精度可达0.3GFLOOPS。通过系统的适当制造可以获得更多的计算性能。此外，通过将该单元实现为集成电路，将显著增强计算性能。目前计算性能的瓶颈是上位机之间的通信速度，可以通过PCI等更高速的通信基础系统来提高，并发现本计算系统的思想适用于各种流体运动的模拟方案。特别是LB（格子玻尔兹曼）方法是适合这样的应用。对基于LB法的流体模拟计算系统进行了概念设计。

The massively parallel computing system for simulating the aerodynamic characteristics of the aerobraking spacecraft was investigated. The simulation scheme is the direct simulation Monte Carlo method, which is widely used for the rarefied gas flow. the conceptural design for the unit of the massively parallel computing system was carried out. It is found that the unit of the massively parallel computing system is realized by the pipeline method, and is composed not of the central procession unit but of the distributed processing units. The trail manufacturing of the unit was successfully conducted and demonstrated the computation performance as was expected. The computation preformance of the unit is proportional to the system clock sped and reached about 0.3 GFLOOPS for the 20 MHz system clock. The more computational performance can be attained by a proper manufacturing of the system. Furthermore the computational performance will be significantly enhanced by realizing the unit as an integrated circuit. The bottle neck of the present computational performance is the communication speed between the host computer, which can be improved by means of more high-speed communication bas system like PCI.It is also found the idea for the present computing system is applicable to various simulation schemes for the fluid motion. Especially the LB (Lattice Boltzmann) method is suitable to such an application. The conceptual disign for the computing system for the fluid simulation based on LB method was carried out.