A Machine Learning Approach for Turbulent Scalar Mixing with Applications in Film Cooling - PowerPoint Presentation

Slide1

A Machine Learning Approach for Turbulent Scalar Mixing with Applications in Film Cooling

1

Sandia National Laboratories is a multi-mission laboratory managed and operated by National Technology and Engineering Solutions of  Sandia, LLC, a wholly owned subsidiary of Honeywell International,  Inc., for the U.S. Department of Energy's National Nuclear Security Administration under contract DE-NA0003525.  SAND2017-6417C

Pedro M. Milani*, Julia Ling†, John K. Eaton*

* Department of Mechanical Engineering, Stanford University† Citrine Informatics

Ann Arbor, MI - Advances in Turbulence Modeling Symposium

July 12 2017Slide2

Motivation

Film cooling is a technique commonly used on modern gas turbine blades.LES is too expensive for full blade simulation, so Reynolds-averaged Navier-Stokes (RANS) methods required in industry.

2Gas turbine blade (Image from: http://www.me.umn.edu/labs/tcht/measurements/what.html)Slide3

Goal

Improve temperature field predictions in RANS simulations of film cooling configurations.

3Schematic of an inclined jet in crossflow.Slide4

RANS Modeling

4Slide5

Turbulent Scalar Flux

5

Average transport of scalar due to turbulent fluctuations.Gradient diffusion hypothesis (GDH) is simplest model:

Need to specify

Typical RANS models use fixed Extract from high-fidelity simulationUse machine learning to predict

 Slide6

Machine Learning Approach

6

Machine Learning: a broad class of algorithms to process large amounts of data and extract patterns from it. Slide7

Datasets

7

1) Baseline Jet (ReD = 3,000)2) Skewed Jet(ReD = 5,800)3) Cube(ReH = 5,000)Slide8

Diffusivity Results

8

Center plane Z/D = 0Plane X/D = 2LESRANS

MLSlide9

Forward Propagation

9

Reynolds-averaged Advection Diffusion (RAAD) equationCalculate the scalar field that results from the diffusivity field.Slide10

Temperature Results

10

LES fieldCenter plane Z/D = 0X/D = 2X/D = 2RANS diffML diffX/D = 2

X/D = 2Slide11

Temperature Results

11

TotalInjectionWallX/D = 4Y/D = 0.1Slide12

Conclusions and Future Work

12

We used machine learning approaches to incorporate high-fidelity simulation knowledge into turbulence models, with good results to a film cooling simulation.Test robustness for other film cooling flows and generalize to different classes of turbulent flows.Use data driven models to learn about the underlying physics.Slide13

13Slide14

(Some) References

Bodart, J., Coletti

, F., Bermejo-Moreno, I., and Eaton, J., 2013. High-Fidelity Simulation of a Turbulent Inclined Jet in a Crossflow. CTR Annual Research Briefs, Center for Turbulence Research, Stanford University, Stanford, CA.Folkersma, M., 2015. “Large Eddy Simulation of an Asymmetric Jet in Crossflow”. Master of Science thesis, Tampere University of Technology, Tampere, Finland, October.Rossi, R., Philips, D., and Iaccarino, G., 2010. “A Numerical Study of Scalar Dispersion Downstream of a Wall-Mounted Cube Using Direct Simulations and Algebraic Flux Models”. International Journal of Heat and Fluid Flow, 31, pp. 805–819.Ling, J., Jones, R., and Templeton, J., 2016. “Machine Learning Strategies for Systems with Invariance Properties”. Journal of Computational Physics, 318, pp. 22–35.Ling, J., and Templeton, J., 2015. “Evaluation of Machine Learning Algorithms for Prediction of Regions of High Reynolds Averaged Navier Stokes Uncertainty”. Physics of Fluids, 27.Ling, J., Ryan, K., Bodart, J., and Eaton, J., 2016. “Analysis of Turbulent Scalar Flux Models for a Discrete Hole Film Cooling Flow”. Journal of Turbomachinery, 138(1).14