ALGORITHMS: Collaborative Research: Parallel Reduced Order Modeling with In-Situ Error Correction and Performance Optimization

计算流体动力学（CFD）已被广泛应用于空气动力学、气象学、海洋学、燃烧和生物医学流动等领域的各种现象的模拟。然而，计算流体动力学的有用性受到无法及时进行系统模拟的限制。这是由于执行高保真非定常CFD模拟的复杂性和计算费用。本研究的根本目标是开发一种快速、鲁棒和高效的降阶模拟技术，以加速CFD模拟。降阶模拟环境必须适用于流体流动问题的研究，跨越广泛的应用领域。改善流体模拟性能的现有技术有：(a)计算并行化和(b)采用降阶建模。这些技术中的每一种都证明了将模拟时间减少了10倍到100倍。然而，适当的正交分解（POD）的实际应用尚未实现，因为在没有与相同场景的完整模拟进行比较的情况下，无法量化降阶模型的准确性。这项工作将调查和证明POD仿真方法的可行性，以及使用残差跟踪和相关方法来验证和纠正降阶仿真的能力。此外，我们将演示三维并行CFD求解器的自适应性能控制。

Computational fluid dynamics (CFD) has been employed in the simulation of a wide variety of phenomenon in areas such as aerodynamics, meteorology, oceanography, combustion and biomedical flows. However the usefulness of computational fluid dynamics is limited by an inability to perform system simulations in a timely fashion. This is due to the complexity and computational expense of performing high fidelity unsteady CFD simulations. The fundamental goal of this research is the development of a rapid, robust and efficient reduced order simulation technique for accelerating CFD simulation. The reduced order simulation environment must beapplicable to the study of fluid flow problems across a wide spectrum of applications domains. Existing techniques for improving the performance of fluids simulations are: (a) parallelization of the computations and (b) employment of reduced order modeling. Each of these techniques alone have demonstrated reductions of ten-fold to a hundred-fold in simulation time. However, the practical application of proper orthogonal decomposition (POD) has not been realized due to the inability to quantify the accuracy of the reduced order model without a comparison to a full simulation of the same scenario. This work will investigate and demonstrate feasibility of POD simulation methods and the ability to use residual tracking and related methods to validate and correct reduced order simulations. In addition we will demonstrate self-adaptive performance control of a 3D parallel CFD solver.