Hyper Flux

流体流动的计算机模拟在包括飞机、汽车、船舶和风力涡轮机在内的众多复杂系统的气动设计中发挥着越来越重要的作用。然而，越来越明显的是，对于广泛的流动问题，用于气动设计的当前生成软件包不适合用于该目的。具体地说，对于非定常流动(高度分离流动、涡流主导流动、声学问题等)，当前生成软件包缺乏所需的精度；因为它们普遍基于‘低阶’(一阶或二阶)精确的数值方法。为了解决具有挑战性的非定常流动问题，并消除对额外的物理原型的需要，需要基于先进的高阶精度数值方法的更新软件。此外，该软件必须能够在所谓的非结构化网格上实现高阶精度-用于对复杂的工程几何形状进行网格划分，并且它必须能够有效地利用下一代的多核计算硬件(如NVIDIA Tesla GPU、Intel Xeon Phi协处理器和AMD FirePro GPU)，这些硬件可能是未来HPC平台的基础。先进的高阶通量重建(FR)方法，结合多核加速器，可以提供一种“游戏改变”技术，能够在复杂的工程几何形状附近执行目前难以处理的非定常湍流模拟。然而，在使用上述技术来解决现实世界的流动问题之前，仍需要解决各种技术问题，这些问题往往涉及“滑动平面”(两个计算网格以不一致的方式彼此滑动的情况)。该建议的主要目标(学术部分)是开发一种在许多核心加速器上使用FR方法有效地处理滑行飞机的方法，并在金融(CFMS和Zenotech)和非金融(空中客车、EADS、BAE、劳斯莱斯、ARA、英国空气动力学中心)项目合作伙伴提出的一系列行业领先测试案例中展示FR方法在多核加速器上的性能。建议的学术部分将由Peter Vincent博士(帝国理工学院航空系讲师)领导，并将在现有工作的基础上由3个EPSRC DTA，1个Airbus/SEPRC iCASE DTA资助和EPSRC早期职业奖学金(EP/K027379/1)。

Computer simulations of fluid flow are playing an increasingly important role in aerodynamic design of numerous complexsystems, including aircraft, cars, ships and wind turbines. It is becoming apparent, however, that for a wide range of flowproblems current generation software packages used for aerodynamic design are not fit for purpose.Specifically, for scenarios where flow is unsteady (highly separated flows, vortex dominated flows, acoustics problems etc.)current generation software packages lack the required accuracy; since they are ubiquitously based on 'low-order' (first- orsecond-order) accurate numerical methods. To solve challenging unsteady flow problems, and remove the need forexpensive physical prototyping, newer software based on advanced 'high-order' accurate numerical methods is required.Additionally, this software must be able to achieve high-order accuracy on so-called 'unstructured grids' - used to meshcomplex engineering geometries, and it must be able to make effective use of next-generation 'many-core' computinghardware (such as Nvidia Tesla GPUs, Intel Xeon Phi Co-Processors, and AMD FirePro GPUs), which will likely underpinfuture HPC platforms.Advanced high-order Flux Reconstruction (FR) methods, combined with many-core accelerators, could provide a `gamechanging'technology capable of performing currently intractable unsteady turbulent flow simulations within the vicinity ofcomplex engineering geometries. However, various technical issues still need to be addressed before the above technology can be used `in anger' to solve real-world flow problems, which often involve `sliding planes' (situations whentwo computational meshes slide across one another in a non-conforming fashion). The key objectives (of the academiccomponent) of the proposal are to develop a treatment for sliding planes that works effectively with FR methods on manycoreaccelerators, and demonstrate the performance of FR methods on many-core accelerators for a range of industry ledtest cases proposed by the financial (CFMS and Zenotech) and non-financial (Airbus, EADS, BAE, Rolls-Royce, ARA, UKAerodynamics Centre) project partners.The academic component of the proposal will be lead by Dr. Peter Vincent (a Lecturer in the department of Aeronautics atImperial College London), and will build upon current work funded by 3 x EPSRC DTAs, 1 x Airbus/EPSRC iCASE DTA,and an EPSRC Early Career Fellowship (EP/K027379/1).