AI Driven Open Source Framework for Next Generation Heat Exchangers

选定的项目是为数据驱动的湍流建模创建一个开源策划数据集。该数据集将围绕印刷电路热交换器（PCHE）和冷板冷却系统建立。PCHEs和冷板被许多解决电气化和NetZero的行业大量使用。例如，在燃气轮机、电动汽车和核反应堆中都可以找到它们。设计更高效的热交换器不仅可以降低运行这些冷却系统的能耗，而且对于在高温环境中以安全和经济有效的方式安装组件也至关重要。然而，为了设计这些冷却系统并获得高性能的冷却系统，需要精确的计算流体动力学软件（CFD）来模拟流动行为。湍流模型被用作计算机辅助工程（CAE）软件包的一部分，几乎在每一个科学或工程行业。其中包括能源生产（来自化石能源、核能和可再生能源）、暖通空调、航空航天、汽车、工业加工等。虽然高分辨率技术，如大涡模拟（LES）和直接数值模拟（DNS）正变得越来越普遍，但与当前能力相比，计算需求使得这些技术无法承受许多工业模拟。出于这个原因，在未来几十年里，雷诺平均纳维-斯托克斯（RANS）模拟预计仍将是预测与工程和工业问题实际相关的流动的主要工具。然而，具有强逆压梯度、分离、流线曲率和反应化学的流动通常无法通过RANS方法进行预测。开发提高RANS模拟精度的方法将有助于弥合RANS和LES之间的这一关键能力差距。在这个项目中，我们的目标正是通过训练一个人工智能模型来提高RANS模拟的准确性，而几乎没有额外的计算成本。该数据集将包括各种直接数值模拟（DNS）和大涡模拟（LES）数据。它将立即用于机器学习增强校正湍流闭合模型。

The project selected is the creation of an open-source curated dataset for data driven turbulence modelling.The dataset will be built around printed circuit heat exchanger (PCHE) and cold plate cooling systems. PCHEs and cold plates are heavily used by many industries tackling electrification and NetZero. They are found for example in gas turbines, electric cars, and nuclear reactors. Designing more efficient heat exchangers not only reduces energy consumption to run these cooling systems but is also essential to allow installation components in high temperature environments in a safe and cost-effective manner. However, to design those cooling systems and obtain highly performant ones, accurate computational fluid dynamics software (CFD) to model flow behaviour becomes necessary. Turbulence models are used as part of computer aided engineering (CAE) software packages in almost every single scientific or engineering industry. These include energy generation (from fossil, nuclear, and renewable sources), HVAC, aerospace, automotive, industrial processing, and many others.While higher resolution techniques such as large-eddy simulation (LES) and direct numerical simulation (DNS) are becoming more widespread, the computational demands compared to current capabilities make these techniques unaffordable for many industrial simulations. For this reason, Reynolds-averaged Navier-Stokes (RANS) simulations are expected to remain the dominant tool for predicting flows of practical relevance to engineering and industrial problems over the next few decades. However, flows with strong adverse pressure gradients, separation, streamline curvature, and reacting chemistry are often poorly predicted by RANS approaches. Developing methods to improve the accuracy of RANS simulations will help bridge this critical capability gap between RANS and LES. In this project, we aim to do exactly that by training an AI model which can be used to improve the accuracy of RANS simulations at almost no extra computational cost. The dataset will feature a variety of direct numerical simulation (DNS) and large-eddy simulation (LES) data. It will be for immediate use in machine learning augmented corrective turbulence closure modelling.