A Fast High-Order CFD for Turbulent Flow Simulation in Cardio-Devices

项目摘要 计算流体力学(CFD)在复杂医疗设备流动分析与设计中的应用 目前，人工心脏瓣膜和心脏辅助装置等装置已成为 医疗器械研发社区。然而，FDA最近的一项对照研究表明 论证了传统CFD在预测与以下因素相关的层流-过渡-湍流流动方面的局限性 心血管设备。特别是，在医疗器械领域没有使用统计湍流模型 根据实验数据统一成功地进行了基准测试。大涡模拟(LES)是 在这项研究中推荐用于未来的模拟。 为了解决FDA小组在未来的模拟中使用LES的建议，我们建议开发一种 与NIH任务相关的低雷诺数湍流的先进新一代CFD， 使用(1)边界层区大涡的高阶欧拉涡度输送方法，并直接 边界附近的数值模拟；(2)现有的无网格 拉格朗日旋涡方法(LVM)用于大尺度远离边界层流动的大涡模拟。速度 评估大约占计算成本的80%，将在多核CPU上并行化 和多图形处理器。该项目的具体目标是： 具体目标1：发展一种紧凑的高阶有限体积方法，用于层流流动的模拟 涡量输运方程(VTE)；加速多核CPU和多GPU上的速度计算； 并使用FDA“关键路径”问题#1等严格验证层流代码 (喷嘴)，以及稳定和脉动狭窄血流的DNS值。 具体目标2：在VTE的背景下开发动态亚网格尺度(SGS)模型，并为 过渡流；以及验证用于湍流的高阶有限体积代码，其中使用 FDA“关键路径”问题#1(喷嘴)，以及稳定和脉动的狭窄血流。 具体目标3：完成为低雷诺数大涡模拟所建议的混合代码的开发 耦合高阶欧拉和拉格朗日涡元解算器的湍流 稳定；并使用一系列基准来验证最终产品，包括FDA的“关键路径” 问题#1(喷嘴)和#2(泵)，以及实际的血泵；例如，HeartMate II。 具体目标4：实施一个系统，以便成功地记录、传播和维护 软件，并适应协作研究；并开发与主流CFD的接口，以确保 互操作性和无缝迁移到所提议的技术。 长期影响：目前，传统的CFD和统计湍流模型的应用仅限于 根据相关趋势研究设备性能。也就是说，CFD还不是一个真正的预测性设计 分析工具，至少在心脏装置设计的情况下，它涉及高度复杂的非定常流动 多种共存的层流、过渡流和湍流区。提出的高阶混合域名系统-大涡模拟算法 方法被设计为一种预测工具，它避免特别的模型常量，尤其是在边界内 这是切应力和血液损伤的关键来源。拟议中的技术将从根本上 改变医疗器械社区未来使用CFD的方式。这将带来长期而重大的影响 研究和技术成熟项目是一个CFD软件，(1)非常容易学习和使用，因为它 避免了经常繁琐和容易出错的体积网格过程；(2)可以可靠地用作预测 工具，这要归功于没有带有特殊模糊因子的湍流模型，这必须始终是 针对每个新的流程问题进行了“校准”和“验证”；以及(3)可以在台式机上以数量级运行 更快的周转时间，将长达一个月的产品设计周期减少到几天，这要归功于先进的 商用多核CPU和多GPU上的算法和加速计算。

Project Summary Application of Computational Fluid Dynamics (CFD) to the flow analysis and design of complex medical devices such as prosthetic heart valves and ventricular assist devices is by now standard practice in the medical devices research and development community. However, a recent controlled study by the FDA has demonstrated the limitations of traditional CFD in predicting laminar-transitional-turbulent flows of relevance to cardiovascular devices. In particular, no statistical turbulence model used in the medical devices community benchmarked uniformly successfully against experimental data. Large Eddy Simulation (LES) was recommended in this study for future simulations. To address the recommendation of the FDA panel to use LES in future simulations we propose to develop an advanced new-generation CFD for low-Reynolds-number turbulent flows of relevance to the NIH mission, using (1) a high-order Eulerian vorticity transport method for LES in the boundary layer region, and Direct Numerical Simulation (DNS) in the immediate vicinity of the boundary; and (2) an existing meshless Lagrangian Vortex Method (LVM) for LES of the large scale flow away from the boundary layer. The velocity evaluations, which constitute roughly 80% of the computational cost, will be parallelized on multicore CPUs and multi-GPUs. The Specific Aims of the project are: Specific Aim 1: To develop a compact high-order finite volume method for laminar flow simulation via the vorticity transport equation (VTE); to accelerate the velocity evaluations on multicore CPUs and multi-GPUs; and to rigorously validate the laminar flow code using, among others, the FDA "Critical Path" problem #1 (nozzle), as well as DNS of steady and pulsatile stenotic flow. Specific Aim 2: To develop a dynamic Subgrid-Scale (SGS) model in the context of VTE and tailored for transitional flow; and to validate the high-order finite volume code for turbulent flow using, among others, the FDA "Critical Path" problem #1 (nozzle), as well as steady and pulsatile stenotic flow. Specific Aim 3: To finalize the development of the proposed hybrid code for LES of low-Reynolds-number turbulent flow by coupling the high-order Eulerian and Lagrangian vortex element solvers, accurately and stably; and to validate the final product using a series of benchmarks, including the FDA "Critical Path" problems #1 (nozzle) and #2 (pump), as well as an actual blood pump; e.g., the HeartMate II. Specific Aim 4: To implement a system for successful documentation, dissemination, and maintenance of the software, and to accommodate collaborative research; and to develop interfaces to mainstream CFD to ensure interoperability and seamless migration to the propsed technology. Long-Term Impact: At present, application of traditional CFD and statistical turbulence models is limited to the study of device performance in terms of relative trends. That is, CFD is not yet a truly predictive design and analysis tool, at least in the case of cardio-device design, which involves highly complex unsteady flow with multiple coexisting laminar, transitional, and turbulent flow regimes. The proposed high-order hybrid DNS-LES method is designed to be a predictive tool that avoids ad hoc model constants especially within the boundary layer, which is a key source of shear stress and blood damage. The proposed technology will fundamentally alter how the medical devices community will use CFD in future. The long-term significant impact of this research and technology maturation project is a CFD software that (1) is incredibly easy to learn and use, as it obviates the often tedious and error prone volumetric meshing process; (2) can be used reliably as a predictive tool thanks to the absence of turbulence models with ad hoc fudge factors, which must invariably be "calibrated" and "validated" for each new flow problem; and (3) can be run on a desktop at order-of-magnitude faster turn-around times, reducing month-long product design cycles to just days, thanks to advanced algorithms and accelerated computing on commodity multicore CPUs and multi-GPUs.