Inverse aerodynamic design of turbo components for Carnot batteries by means of physics informed networks enhanced by generative learning

我们开发了基于深度神经网络的新模拟方法，大大缩短了传统CFD的计算时间。特别是，我们应用物理学通知神经网络的流体流动的粗略表示在卡诺电池中操作的涡轮机械组件。这种粗糙的解决方案通过条件生成对抗网络（GAN）进行改进，以创建湍流的真实精细结构。特别是，我们研究了这种基于深度学习的解决方案的物理和泛化特性，以改变边界条件和几何形状。这使我们能够快速评估设计强烈变化的边界条件下，因为它们是典型的卡诺电池的放电循环。通过这种方式，我们提供了一个宝贵的工具，逆卡诺电池的涡轮部件的设计。

We develop new simulation methods based on deep neural networks that shorten computation times with respect to traditional CFD considerably. In particular, we apply physics informed neural networks for a rough representation of fluid flows in turbo-machinery components operating in Carnot batteries. Such rough solutions are refined by conditional generative adversarial networks (GAN) in order to create realistic fine structure of turbulent flows. in particular, we study the physical and the generalization properties of such deep learning based solutions to fluid flow with respect to changing boundary conditions and geometry. This enables us to rapidly evaluate designs under strongly changing boundary conditions, as they are typical for the discharging cycle of a Carnot battery. In this way, we provide an invaluable tool for the inverse design of turbocomponents for Carnot batteries.