DNS and Visual Analysis of Superstructures in Turbulent Channels with Mixing by Parallel Injection

为了分析高雷诺数平行喷射通道中上部结构的发生及其对湍流混合的影响，提出了直接数值模拟（DNS）、涡定义和识别以及基于特征的可视化相结合的方法。标准的、现成的解决方案无法用于此目的。关于DNS，核心问题是在高性能计算系统上以优异的效率访问高雷诺数。此外，必须包括合适的模型来数值描述与混合有关的所有流体特性。涡旋提取的主要挑战有三个方面。首先，高强度湍流排除了基于流动导数局部分析的标准涡定义。相反，全局的、拉格朗日的或分层的涡定义是必要的，它们基于流图上的过滤操作，而不是速度场。其次，涡定义和参数调整必须适应，这样它就不会集中在靠近注入的上游涡，而是处理下游不太明显、噪音更大、更不稳定的涡结构。第三，在可视化分析方面，主要的挑战与数据集的大小有关：DNS通常提供的数据集在模拟期间不能完全存储。因此，为可视化分析提供实时解决方案是必要的。为了分析这些现象，必须将DNS、漩涡提取和可视化结合到一个计算导向意义上的反馈循环中。在实时进行多尺度POD和自动涡提取的同时，随后以交互式方式对所得到的涡进行可视化分析，从而允许对可视化参数和进一步的仿真参数进行调整。DNS、POD和可视化分析的有效结合将有助于识别上层建筑，并帮助解释它们对运输过程的影响。

In order to analyze the occurrence and the impact of superstructures onturbulent mixing in channels at high Reynolds numbers with a parallel injection,a combination of Direct Numerical Simulation (DNS), vortexdefinition and identification, and feature-based visualization, isproposed. Standard, off-the-shelf solutions are not available for thispurpose.Concerning DNS, the central issue is to access high Reynolds numberswith excellent efficiency on HPC systems. Additionally, suitablemodels must be included to describe numerically all fluidproperties relevant for mixing.The main challenge in vortex extraction is three-fold. Firstly,high-intensity turbulence excludes standard vortex definitions that arebased on a local analysis of the flow derivatives. Instead, global,Lagrangian, or hierarchical vortex definitions are necessary that arebased on filtering operations on the flow map instead of the velocityfield. Secondly, vortex definitions and parameter tuning has to beadapted such that it does not focus on upstream vorticesclose to injection but tackles the less obvious, noisierand more unsteady vortex structures downstream. Thirdly, in terms ofvisual analysis, the main challenge is associated with the sheer size ofthe data sets: DNS typically delivers data sets that cannot becompletely stored during the simulation. Hence, on-the-fly solutions forthe visual analysis are necessary.To analyze the phenomena, DNS, vortex extraction and visualization haveto be combined into a feedback cycle in a computational steering sense.While a multi-scale POD along with an automatic vortex extraction is carried outon-the-fly, the resulting vortices are later visually analyzed in an interactive manner,allowing adaptation of both the visualization parameters and further simulationparameters. This efficient combination of DNS, POD, and visual analysis shall allow the identification of superstructures and help explain their impact on transport processes.