A Parallel, High-Fidelity Coupled Machine Learning/ CFD Solver Method for Solving Exascale CFD Problems with Turbulence Damping at the Liquid-Gas Inte

模拟气液界面湍流阻尼的性质对于准确预测部件功率损失、润滑剂分布和传热起着至关重要的作用。目前，人们对这种界面湍流阻尼以及如何在计算模型中最好地表示它的理解有限。在EPSRC Cornerstone项目期间，一种概念验证方法得到了验证，该方法使用高保真计算流体动力学（CFD）来模拟平面上矩形通道内的气液流动。本研究旨在将这一概念扩展到发动机的代表性几何形状和条件，以及其他多相流情况。这些高保真的百亿亿次测试用例将使用前面提到的CFD概念验证方法进行模拟，并在高性能计算机（hpc）上使用并行领域特定语言OPS来适应百亿亿次测试用例。这些测试用例形成了一个模拟数据库，从中可以训练机器学习模型。与传统计算模型相比，将使用耦合的机器学习/CFD方法，并分析其效率、可扩展性和准确性。

Modelling the nature of gas-liquid interface turbulence damping plays a crucial role in making accurate predictions of component power losses, lubricant distribution and heat transfer. Currently, there is a limited understanding of this interface turbulence damping and how to best represent it in computational models. During the EPSRC Cornerstone project, a proof-of-concept approach was demonstrated that uses high-fidelity computational fluid dynamics (CFD) to simulate air-liquid flow in a rectangular channel on a flat surface. This study seeks to expand this concept to engine-representative geometries and conditions, as well as other multi-phase flow cases. These high-fidelity, exascale test cases will be simulated using the aforementioned CFD proof-of-concept approach, adapted for exascale cases using the parallel domain specific language OPS on high performance computers (HPCs). These test cases form a database of simulations, from which a machine learning model can be trained. A coupled machine learning/CFD approach will be used and analysed for its efficiency, scalability, and accuracy compared to traditional computational models.