Generalizability of machine-learning (ML) based turbulence closures to accurately predict unseen practical flows remains an important challenge. It is well recognized that the neural network (NN) architecture and training protocol profoundly influence the generalizability characteristics. At the Reynolds-averaged Navier–Stokes level, NN–based turbulence closure modeling is rendered difficult due to two important reasons: inherent complexity of the constitutive relation arising from flow-dependent non-linearity and bifurcations; and, inordinate difficulty in obtaining high-fidelity data covering the entire parameter space of interest. Thus, a predictive turbulence model must be robust enough to perform reasonably outside the domain of training. In this context, the objective of the work is to investigate the approximation capabilities of standard moderate‐sized fully connected NNs. We seek to systematically investigate the effects of (i) intrinsic complexity of the solution manifold; (ii) sampling procedure (interpolation vs extrapolation); and (iii) optimization procedure. To overcome the data acquisition challenges, three proxy-physics turbulence surrogates of different degrees of complexity (yet significantly simpler than turbulence physics) are employed to generate the parameter-to-solution maps. Lacking a strong theoretical basis for finding the globally optimal NN architecture and hyperparameters in the presence of non-linearity and bifurcations, a “brute‐force” parameter-space sweep is performed to determine a locally optimal solution. Even for this simple proxy-physics system, it is demonstrated that feed-forward NNs require more degrees of freedom than the original proxy-physics model to accurately approximate the true model even when trained with data over the entire parameter space (interpolation). Additionally, if deep fully connected NNs are trained with data only from part of the parameter space (extrapolation), their approximation capability reduces considerably and it is not straightforward to find an optimal architecture. Overall, the findings provide a realistic perspective on the utility of ML turbulence closures for practical applications and identify areas for improvement.
Skip Nav Destination
Turbulence closure modeling with data-driven techniques: Investigation of generalizable deep neural networks
,
,
,
Article navigation
November 2021
Research Article|
November 29 2021
Turbulence closure modeling with data-driven techniques: Investigation of generalizable deep neural networks
Salar Taghizadeh
;
Salar Taghizadeh
a)
1
J. Mike Walker '66 Department of Mechanical Engineering, Texas A&M University
, College Station, Texas 77843, USA
a)Author to whom correspondence should be addressed: [email protected]
Search for other works by this author on:
Freddie D. Witherden;
Freddie D. Witherden
2
Department of Ocean Engineering, Texas A&M University
, College Station, Texas 77843, USA
Search for other works by this author on:
Yassin A. Hassan;
Yassin A. Hassan
1
J. Mike Walker '66 Department of Mechanical Engineering, Texas A&M University
, College Station, Texas 77843, USA
3
Department of Nuclear Engineering, Texas A&M University
, College Station, Texas 77843, USA
Search for other works by this author on:
Sharath S. Girimaji
Sharath S. Girimaji
1
J. Mike Walker '66 Department of Mechanical Engineering, Texas A&M University
, College Station, Texas 77843, USA
2
Department of Ocean Engineering, Texas A&M University
, College Station, Texas 77843, USA
Search for other works by this author on:
Salar Taghizadeh
1,a)
Freddie D. Witherden
2
Yassin A. Hassan
1,3
Sharath S. Girimaji
1,2
1
J. Mike Walker '66 Department of Mechanical Engineering, Texas A&M University
, College Station, Texas 77843, USA
2
Department of Ocean Engineering, Texas A&M University
, College Station, Texas 77843, USA
3
Department of Nuclear Engineering, Texas A&M University
, College Station, Texas 77843, USA
a)Author to whom correspondence should be addressed: [email protected]
Physics of Fluids 33, 115132 (2021)
Article history
Received:
September 09 2021
Accepted:
October 30 2021
Citation
Salar Taghizadeh, Freddie D. Witherden, Yassin A. Hassan, Sharath S. Girimaji; Turbulence closure modeling with data-driven techniques: Investigation of generalizable deep neural networks. Physics of Fluids 1 November 2021; 33 (11): 115132. https://doi.org/10.1063/5.0070890
Download citation file:
Pay-Per-View Access
$40.00
Sign In
You could not be signed in. Please check your credentials and make sure you have an active account and try again.
Citing articles via
Phase behavior of Cacio e Pepe sauce
G. Bartolucci, D. M. Busiello, et al.
How to cook pasta? Physicists view on suggestions for energy saving methods
Phillip Toultchinski, Thomas A. Vilgis
Pour-over coffee: Mixing by a water jet impinging on a granular bed with avalanche dynamics
Ernest Park, Margot Young, et al.
Related Content
Scale-resolving simulations of turbulent flows with coherent structures: Toward cut-off dependent data-driven closure modeling
Physics of Fluids (June 2024)
A posteriori study on wall modeling in large eddy simulation using a nonlocal data-driven approach
Physics of Fluids (June 2024)
AutoTurb: Using large language models for automatic algebraic turbulence model discovery
Physics of Fluids (January 2025)
Techniques for deriving explicit algebraic Reynolds stress models based on incomplete sets of basis tensors and predictions of fully developed rotating pipe flow
Physics of Fluids (November 2005)
Artificial neural network-substituted transition model for crossflow instability: Modeling strategy and application prospect
Physics of Fluids (April 2024)