类天宫飞行器轨道衰降过程空气动力特性一体化建模并行优化设计

Parallel Optimization Design of Unified Modeling for Aerodynamics During Orbital Declining Reentry of Tiangong Type Spacecraft

针对天宫一号目标飞行器无控飞行轨道衰降数值预报需要快速确定轨道积分高精度计算模型中的空气动力,在发展基于修正Boettcher/Legge非对称桥函数的天宫一号空气动力特性当地化算法基础上,对当地化算法的运算流程及对应程序代码进行了整体分析,根据原程序热点代码集中、数据独立性强及传输需求少等特点,发展了多核处理单元的并行优化方法。引入CUDA架构的GPU设备同时,开展了系统、算法以及语句三个层次的并行优化,设计了GPU内存对齐访问方案,使用数据传输函数,将算法求解部分内循环经过展开与合并,整理为整体移植入核函数的一个循环,利用GPU较强的并行计算能力提升运算效率,对函数、循环、指令等代码语句进行级别优化。使用设计的并行计算方案对类天宫飞行器空气动力特性当地化串行算法程序进行CPU+GPU移植优化,达到了近5倍的并行加速比,且使单次求解中GPU数据传输时间缩减为原来的23%,证实了并行方案和优化设计手段的高效实用性。在类天宫飞行器空气动力特性GPU并行算法程序验证基础上,使用GPU并行程序对天宫飞行器轨道衰降飞行340~120 km过程的气动特性进行了不同迎角、侧滑角等飞行姿态计算分析,提供了大量可供轨道飞行力学数值预报的空气动力计算数据。

Numerical prediction during the orbital declining reentry of the Tiangong-1 target spacecraft requires rapid determination of the aerodynamics in the high precision calculation model for orbital integration of uncontrolled flight.Tiangong-1 aerodynamics local algorithm was developed using the non-symmetric corrections of the modified Boettcher/Legg bridge functions,and the overall analysis of the operational process of the local engineering algorithm and the corresponding program code was conducted.Then,according to the characteristics of the original program,such as the concentrated hot code,strong data independence and less transmission demand,the parallel optimization method using multi-core processing unit was developed.The GPU equipment with CUDA architecture was introduced into the original program,and the parallel optimization of the system,the algorithm and the code were carried out at the same time,such as designing GPU memory aligned access scheme,using data transfer functions,leading the solution procedure of the algorithm into one kernel function by expanding and consolidating nest-loop,calling more GPU threads,and using other optimization aim at function,loop,instruction and so on.The local engineering algorithm program with CPU+GPU parallel implementation was optimized for the unified computation on reentry aerodynamics of the Tiangong-1 target spacecraft using the designed parallel computing scheme,and the obtained parallel speedup ratio was elevated nearly 5 times,and the data transmission time per step in GPU was reduced to 23%of the original runtime.The parallel scheme and optimized design technique were verified to be highly efficient and practical.Based on the verification of the GPU parallel algorithm program for the Tiangong-1 spacecraft aerodynamics,the aerodynamics during the orbital declining reentry of the spacecraft were computed and analyzed under different angles of attack and sideslip angles at 340~120 km flight altitude which provided a large number of aerodynamic data for the numerical forecast of orbital flight mechanics.