Integration of Machine Learning and Computational Fluid Dynamics to Develop Turbulence Models for Improved Turbine Wake Mixing Prediction

TitleIntegration of Machine Learning and Computational Fluid Dynamics to Develop Turbulence Models for Improved Turbine Wake Mixing Prediction
Publication TypeConference Paper
Year of Publication2020
AuthorsAkolekar H, Zhao Y, Sandberg R, Pacciani R
Conference NameASME Turbo Expo 2020 Turbomachinery Technical Conference and Exposition
Conference LocationVirtual Event, September 21-25, 2020
This paper presents development of accurate turbulence closures for wake mixing prediction by integrating a machine-learning
approach with Reynolds Averaged Navier-Stokes (RANS)-based computational fluid dynamics (CFD). The data-driven modeling
framework is based on the gene expression programming (GEP) approach previously shown to generate non-linear RANS models with
good accuracy. To further improve the performance and robustness of the data-driven closures, here we exploit that GEP produces
tangible models to integrate RANS in the closure development process.
Specifically, rather than using as cost function a comparison of the GEP-based closure terms with a frozen high-fidelity dataset,
each GEP model is instead automatically implemented into a RANS solver and the subsequent calculation results compared with reference
data. By first using a canonical turbine wake with inlet conditions prescribed based on high-fidelity data, we demonstrate that
the CFD-driven machine-learning approach produces non-linear turbulence closures that are physically correct, i.e. predict the
right downstream wake development and maintain an accurate peak wake loss throughout the domain. We then extend our analysis to
full turbine-blade cases and show that the model development is sensitive to the training region due to the presence of deterministic
unsteadiness in the near wake region. Models developed including this region have artificially large diffusion coefficients to overcompensate
for the vortex shedding steady RANS cannot capture.
In contrast, excluding the near wake region in the model development produces the correct physical model behavior, but predictive
accuracy in the near-wake remains unsatisfactory. We show that this can be remedied by using the physically consistent models in
unsteady RANS, implying that the non-linear closure producing the best predictive accuracy depends on whether it will be deployed in
RANS or unsteady RANS calculations. Overall, the models developed with the CFD-assisted machine learning approach were found to be robust and capture the correct physical behavior across different operating conditions.

paper GT2020-14732

Refereed DesignationRefereed