Research Papers

Applying Machine Learnt Explicit Algebraic Stress and Scalar Flux Models to a Fundamental Trailing Edge Slot

[+] Author and Article Information
R. D. Sandberg

Department of Mechanical Engineering,
University of Melbourne,
Parkville 3010, VIC, Australia
e-mail: richard.sandberg@unimelb.edu.au

R. Tan, J. Weatheritt, A. Ooi, A. Haghiri

Department of Mechanical Engineering,
University of Melbourne,
Parkville 3010, VIC, Australia

V. Michelassi

Baker Hughes, a GE Company,
Florence 50127, Italy
e-mail: vittorio.michelassi@bhge.com

G. Laskowski

General Electric Aviation,
Lynn, MA 01905
e-mail: laskowsk@ge.com

1Corresponding author.

Contributed by the International Gas Turbine Institute (IGTI) of ASME for publication in the JOURNAL OF TURBOMACHINERY. Manuscript received August 15, 2018; final manuscript received August 21, 2018; published online September 28, 2018. Editor: Kenneth Hall.

J. Turbomach 140(10), 101008 (Sep 28, 2018) (11 pages) Paper No: TURBO-18-1202; doi: 10.1115/1.4041268 History: Received August 15, 2018; Revised August 21, 2018

Machine learning was applied to large-eddy simulation (LES) data to develop nonlinear turbulence stress and heat flux closures with increased prediction accuracy for trailing-edge cooling slot cases. The LES data were generated for a thick and a thin trailing-edge slot and shown to agree well with experimental data, thus providing suitable training data for model development. A gene expression programming (GEP) based algorithm was used to symbolically regress novel nonlinear explicit algebraic stress models and heat-flux closures based on either the gradient diffusion or the generalized gradient diffusion approaches. Steady Reynolds-averaged Navier–Stokes (RANS) calculations were then conducted with the new explicit algebraic stress models. The best overall agreement with LES data was found when selecting the near wall region, where high levels of anisotropy exist, as training region, and using the mean squared error of the anisotropy tensor as cost function. For the thin lip geometry, the adiabatic wall effectiveness was predicted in good agreement with the LES and experimental data when combining the GEP-trained model with the standard eddy-diffusivity model. Crucially, the same model combination also produced significant improvement in the predictive accuracy of adiabatic wall effectiveness for different blowing ratios (BRs), despite not having seen those in the training process. For the thick lip case, the match with reference values deteriorated due to the presence of large-scale, relative to slot height, vortex shedding. A GEP-trained scalar flux model, in conjunction with a trained RANS model, was found to significantly improve the prediction of the adiabatic wall effectiveness.

Copyright © 2018 by ASME
Your Session has timed out. Please sign back in to continue.


Whitelaw, S. , and Kacker, J. , 1963, “ An Experimental Investigation of the Influence of Slot Lip Thickness of the Impervious Wall Effectiveness of the Uniform Density, Two Dimensional Wall Jet,” Int. J. Heat Mass Transfer, 542(9), pp. 1196–1201.
Medic, G. , and Durbin, P. A. , 2005, “ Unsteady Effects on Trailing Edge Cooling,” ASME J. Heat Transfer, 127(4), pp. 388–392. [CrossRef]
Joo, J. , and Durbin, P. A. , 2009, “ Simulation of Turbine Blade Trailing Edge Cooling,” ASME J. Fluids Eng., 131(2), pp. 211021–2110214. [CrossRef]
Ling, J. , Kurzawski, A. , and Templeton, J. , 2016, “ Reynolds Averaged Turbulence Modelling Using Deep Neural Networks With Embedded Invariance,” J. Fluid Mech., 807, pp. 155–166. [CrossRef]
Parish, E. J. , and Duraisamy, K. , 2016, “ A Paradigm for Data-Driven Predictive Modeling Using Field Inversion and Machine Learning,” J. Comput. Phys., 305, pp. 758–774. [CrossRef]
Ferreira, C. , 2001, “ Gene Expression Programming: A New Adaptive Algorithm for Solving Problems,” Complex Syst., 13(2), pp. 87–129. https://arxiv.org/ftp/cs/papers/0102/0102027.pdf
Weatheritt, J. , and Sandberg, R. D. , 2016, “ A Novel Evolutionary Algorithm Applied to Algebraic Modifications of the RANS Stress–Strain Relationship,” J. Comput. Phys., 325, pp. 22–37. [CrossRef]
Weatheritt, J. , Sandberg, R. D. , Ling, J. , Saez, G. , and Bodart, J. , 2017, “ A Comparative Study of Contrasting Machine Learning Frameworks Applied to RANS Modeling of Jets in Crossflow,” ASME Paper No. GT2017-63403.
Weatheritt, J. , Pichler, R. , Sandberg, R. D. , Laskowski, G. M. , and Michelassi, V. , 2017, “ Machine Learning for Turbulence Model Development Using a High Fidelity HPT Cascade Simulation,” ASME Paper No. GT2017-63497.
Duraisamy, K. , and Durbin, P. , 2014, “ Transition Modeling Using Data Driven Approaches,” CTR Summer Program, p. 427.
Zhang, Z. J. , and Duraisamy, K. , 2015, “ Machine Learning Methods for Data-Driven Turbulence Modeling,” 22nd AIAA Computational Fluid Dynamics Conference, p. 2460.
Milani, P. M. , Ling, J. , Saez-Mischlich, G. , Bodart, J. , and Eaton, J. K. , 2018, “ A Machine Learning Approach for Determining the Turbulent Diffusivity in Film Cooling Flows,” ASME J. Turbomach., 140(2), p. 021006. [CrossRef]
Sandberg, R. , 2015, “ Compressible-Flow DNS with Application to Airfoil Noise,” Flow, Turbul. Combust., 95(2–3), pp. 211–229. [CrossRef]
Ducros, F. , Nicoud, F. , and Poinsot, T. , 1998, “ Wall-Adapting Local Eddy-Viscosity Models for Simulations in Complex Geometries,” Conference on Numerical Methods in Fluid Dynamics, pp. 1–7.
Ivanova, E. , and Laskowski, G. M. , 2014, “ LES and Hybrid RANS/LES of a Fundamental Trailing Edge Slot,” ASME Paper No. GT2014-25906.
Sandberg, R. D. , and Sandham, N. D. , 2006, “ Nonreflecting Zonal Characteristic Boundary Condition for Direct Numerical Simulation of Aerodynamic Sound,” AIAA J., 44(2), pp. 402–405. [CrossRef]
Menter, F. , and Esch, T. , 2001, “ Elements of Industrial Heat Transfer Predictions,” 16th Brazilian Congress of Mechanical Engineering (COBEM), Vol. 109.
Leschziner, M. , 2015, Statistical Turbulence Modelling for Fluid Dynamics–Demystified: An Introductory Text for Graduate Engineering Students, World Scientific, Imperial College Press, London, U.K.
Pope, S. B. , 1975, “ A More General Effective-Viscosity Hypothesis,” J. Fluid Mech., 72(2), pp. 331–340. [CrossRef]
Spalart, P. , Shur, M. , Strelets, M. K. , and Travin, A. , 2015, “ Direct Simulation and RANS Modelling of a Vortex Generator Flow,” Flow, Turbul. Combust., 95(2–3), pp. 335–350. [CrossRef]
Koza, J. R. , 1992, Genetic Programming: On the Programming of Computers by Means of Natural Selection, 1st ed., MIT Press, Cambridge, MA.
Daly, B. , and Harlow, F. , 1970, “ The Generalized Gradient Diffusion Hypothesis,” Phys. Fluids, 13(11), p. 2634. [CrossRef]
Schmitt, F. G. , 2007, “ About Boussinesq's Turbulent Viscosity Hypothesis: Historical Remarks and a Direct Evaluation of Its Validity,” C. R. Mec., 335(9–10), pp. 617–627. [CrossRef]


Grahic Jump Location
Fig. 1

(a) Computational domain (not to scale) and (b) LES blocks

Grahic Jump Location
Fig. 2

Instantaneous temperature fields obtained from LES and unsteady RANS using the baseline kω-SST model, dashed lines denote time-averaged temperature at Tslot + 2 K and Tfs – 2 K. (a) Instantaneous LES temperature field for t/s = 1.14. (b) Instantaneous LES temperature field for t/s = 0.126. (c) Instantaneous URANS temperature field for t/s = 1.14. (d) Instantaneous URANS temperature field for t/s = 0.126.

Grahic Jump Location
Fig. 3

Comparison of ηwall obtained from LES, experiments [1] and baseline RANS for different t/s

Grahic Jump Location
Fig. 4

Thermal diffusivity obtained from LES using Eq. (11)

Grahic Jump Location
Fig. 5

Thermal diffusivity obtained using GGDH, Eq. (12)

Grahic Jump Location
Fig. 6

Alignment of strain and anisotropy tensor using baseline kω-SST model: (a) t/s = 1.14 and (b) t/s = 0.126

Grahic Jump Location
Fig. 7

Training regions for RANS (top) and heat flux (bottom) models

Grahic Jump Location
Fig. 8

Alignment of strain and anisotropy tensor using different trained nonlinear RANS models for t/s = 1.14 case. Note that arrows are pointing toward the areas of highest interest (close to the lower wall) where significant improvement in the alignment is observed compared to Fig. 6.

Grahic Jump Location
Fig. 9

Rxx (a) and Rxy (b) from trained models compared with baseline RANS and LES profiles at x/s = 30 for t/s = 1.14

Grahic Jump Location
Fig. 10

Rxx (left) and Rxy (right) from trained models compared with baseline RANS and LES profiles at x/s = 30 for t/s = 0.126 at different blowing ratios: (a) BR = 1.26, (b) BR = 1.07, and (c) BR = 0.86

Grahic Jump Location
Fig. 11

Comparison of streamwise heat flux obtained from different flux models for case t/s = 1.14: (a) LES streamwise heat flux, (b) HF.1 streamwise heat flux, (c) HF.2 GGDH streamwise heat flux, and (d) HF.3 GGDH streamwise heat flux.

Grahic Jump Location
Fig. 12

Comparison of wall-normal heat flux obtained from different flux models for case t/s = 1.14: (a) LES wall-normal heat flux, (b) HF.1 wall-normal heat flux, (c) HF.2 GGDH wall-normal heat flux, and (d) HF.3 GGDH wall-normal heat flux.

Grahic Jump Location
Fig. 13

Comparison of ηwall for different RANS models for t/s = 0.126 at different blowing ratios: (a) BR = 1.26, (b) BR = 1.07, and (c) BR = 0.86.

Grahic Jump Location
Fig. 14

Comparison of ηwall for different RANS and heat flux models for t/s = 1.14



Some tools below are only available to our subscribers or users with an online account.

Related Content

Customize your page view by dragging and repositioning the boxes below.

Related Journal Articles
Related eBook Content
Topic Collections

Sorry! You do not have access to this content. For assistance or to subscribe, please contact us:

  • TELEPHONE: 1-800-843-2763 (Toll-free in the USA)
  • EMAIL: asmedigitalcollection@asme.org
Sign In