Predicting Transitional and Turbulent Flow Around a Turbine Blade With a Physics-Informed Neural Network

Abstract

Abstract Despite the demonstrated usefulness of RANS for many industrially-relevant problems, it can be challenging to accurately simulate certain flow features with the method. Due to the Reynolds-averaging process, the Reynolds-averaged Navier-Stokes equations require a turbulence model to close the equations, and the simple physical arguments and approximations used in many turbulence models can cause erroneous results when applied to flows featuring separation or strong pressure gradients. Physics-informed neural networks (PINNs) offer a way to model aerodynamic problems without explicitly requiring a closure. The network can use sparse training data and the unclosed RANS equations to reconstruct the flow without a turbulence model. In this work, PINNs are applied to two problems of relevance in the tubomachinery community, a variable area channel known as the periodic hills, and the T106C low-pressure turbine blade with two different levels of inlet turbulent intensity. These turbulent flows feature shear layers, separation bubbles, as well as favourable and adverse-pressure-gradients. We demonstrate that PINNs are capable of modelling wall-bounded quantities such as Cf and Cp in such complex flows, capturing sensitive features such as the change in separation length when the turbulent inlet conditions are altered. Excellent predictions of wake mixing are also achieved with sparse training data required. Based on these results, future potential work should focus on reconstructing the wake loss region of a linear cascade with sparse experimental data.