航天器再入各流域复杂绕流问题超大规模可扩展并行算法高效实现与验证

With the development of modern billion-flops high-performance computer system, it has become a very important and key basic problem in orbit-control development of the manned space flight project how to establish a large-scale scalable parallel algorithm for efficient and stable operation of hundreds of thousands or more millions of processor cores and to solve the reentry aerodynamics problems around spacecraft covering flow regimes, which it has been being difficult for traditional research methods to compute in the aerospace field. The unified Boltzmann model equation in describing reentry hypersonic flow transport phenomena around large-scale and complex structure spacecrafts will be presented covering the full spectrum of flow regimes from high rarefied free-molecular flow to continuum in this project, and the mathematical model for parallel computation will be set up for the numerical schemes on the Boltzmann model equation. The large-scale parallel algorithm with multilayer domain decomposition for the Boltzmann model equation of multiphase space will be presented for 100PF level heterogeneous supercomputer. The high performance parallel computing platform with high efficient data communication, floating-point operations and automatic fault tolerance will be developed for multi-core heterogeneous CPU/GPU, and the large-scale scalable parallel computing software and application environment will be presented to solve the complex hypersonic aerothermodynamics problems covering the whole of flight regimes from outer space reentry for spacecrafts such as the re-entry module and TianGong craft. The proving and evaluating platform of the low-density wind-tunnel experiment, DSMC and NS coupled parallel algorithm, and implicit GKS and UGKS methods will be established for validation of the extreme-scale parallel computation in solving the Boltzmann model equation. The aerothermodynamics characteristics and complex flow mechanism of the re-entry module and large-scale spacecrafts will be revealed with massively parallel computation. The researching results of this project will form the high-performance computing core competitiveness in solving the Boltzmann model equation of multiphase space relying on our domestic 100P heterogeneous extreme-scale computer with independent innovation research characteristics in the international field.

随着现代亿亿次计算机发展，建立高效稳定运行于海量处理器核超大规模并行算法，解决航天飞行器离轨再入空气动力学难题，已成为载人航天工程轨控研制迫切希望解决的关键基础问题。项目研究提出描述大型复杂结构航天器再入飞行各流域高超声速流动输运现象统一的Boltzmann模型方程及数值求解格式并行计算数学模型，建立求解多相空间Boltzmann模型方程多层区域分解超大规模并行算法，发展具有高效率数据通信、浮点运算与自动容错功能的众核异构高性能并行计算平台，研制适于数万数十万以上CPU核求解大型航天器再入各流域复杂高超声速绕流问题可扩展并行计算软件及使用环境，结合发展DSMC及其与NS耦合并行算法、隐式GKS与UGKS方法及低密度风洞实验验证评估，研究揭示飞船返回舱再入、大型航天器寿命末期离轨陨落跨流域空气动力特性与复杂流动机理，形成依托国产100PF超级计算机多相空间Boltzmann模型方程高性能计算应用研究体系

随着大内存、高速度超级巨型计算机研制使用，高性能并行计算逐渐成为复杂科学计算领域的主宰，高性能计算环境对解决航天飞行器从外层空间再入大气层各流域复杂空气动力学问题面临机遇挑战，项目针对一类共性关键基础问题：超级计算机处理器节点数量超过一定规模，作业运行过程处理器出现问题而中断运行几率急剧增大，极大限制了超级计算运行效率。本项目研究提出描述复杂结构航天器再入飞行各流域复杂高超声速流动输运现象统一Boltzmann模型方程及数值求解格式并行计算数学模型，建立面向100PF超级计算机求解多相空间Boltzmann模型方程多层区域分解超大规模并行算法，发展具有高效率数据通信、浮点运算与自动容错功能的众核异构高性能并行计算平台，研制适于数千数万更大规模处理器核求解航天器再入尤其是项目执行期间正值我国天宫一号目标飞行器超期服役两年半于2016年3月16日突然失灵无控陨落再入过程，通过将研究重心集中在天宫一号陨落再入近空间跨流域多次解体形成非规则解体物气动力热问题，实施大规模精细计算沿弹(轨)道修正当地化关联参数快速预测解体物气动力热耦合飞行力学数值预报。建立了求解Boltzmann模型方程离散速度空间区域分解MPI+Open ACC众核二级并行程序开发架构与基于硬件加速技术的超大规模异构并行算法，形成了求解 Boltzmann 模型方程统一算法、DSMC、N-S 及其耦合算法、隐式气体动理学格式直角网格法、低密度风洞实验验证分析平台，首次实现了天宫一号目标飞行器陨落再入过程多次解体形成非规则解体物气动力热问题大规模并行计算，研制形成了近空间非规则解体物高性能计算验证建设平台。利用此平台软件系统，开展天宫一号无控陨落再入大气层多次解体过程模拟，证实本项目数理建模、并行算法程序软件研制准确可靠性与工程应用价值及进一步研究前景，有力推动新型流体力学高性能计算基础算法与求解Boltzmann模型方程多相空间数值模拟超大规模并行计算技术在解决载人航天工程重大战略需求应用发展。