You are on page 1of 36

19.

0 Release

Lecture 06: Reynolds Stress Models


(RSM)
Turbulence Modeling Using ANSYS CFD

1 © 2018 ANSYS, Inc.


Contents

• In this lecture, we will cover:


− Theory of various Reynolds Stress Models
− The application of various Reynolds Stress Models

2 © 2018 ANSYS, Inc.


Motivation RSM
• Reynolds Stresses (−𝜌 𝑢𝑖 𝑢𝑗 ) appear directly in Reynolds averaged momentum equations
𝜕𝜌𝑈ሜ 𝑖 𝜕𝜌𝑈ሜ 𝑗 𝑈ሜ 𝑖 𝜕 𝑝lj 𝜕 𝜕 𝑈ሜ 𝑖 𝜕𝑈ሜ 𝑖 𝜕 −𝜌𝑢𝑖 𝑢𝑗
+ =− + 𝜇( + ) +
𝜕𝑡 𝜕𝑥𝑗 𝜕𝑥𝑖 𝜕𝑥𝑗 𝜕𝑥𝑗 𝜕𝑥𝑗 𝜕𝑥𝑗

• Eddy Viscosity Models: −𝜌 𝑢 𝑢 = 2𝜇 𝑆 − 2 𝜇 𝜕𝑈𝑘 𝛿 − 2 𝜌 𝑘𝛿 ; 𝑆 = 1 𝜕𝑈𝑖 + 𝜕𝑈𝑗


𝑖 𝑗 𝑡 𝑖𝑗 𝑡 𝑖𝑗 𝑖𝑗 𝑖𝑗
− Very simplistic model 3 𝜕 𝑥𝑘 3 2 𝜕 𝑥𝑗 𝜕 𝑥𝑖
− Linear relationship between stress and strain
− Reynolds stress and strain rate tensor are often not aligned
− Offer good predictions for shear flows (one dominant shear stress - boundary layers, Mixing layers,
jets, …) but have limitations for complex flow interactions
• When 3D non-equilibrium effects are important in flows, the eddy-viscosity assumption
loses its justification
− Flows with strong curvature or swirl (e.g. Hydrocyclone)
− Flows in corners with secondary motions
− Complex 3D interacting flows
• RSM Model solve directly the transport equations for the Reynolds Stresses
3 © 2018 ANSYS, Inc.
Methodology: Reynolds Stress Model (RSM)

• Derivation of exact transport equation for Reynolds Stresses (−𝜌 𝑢𝑖 𝑢𝑗 )


• Additional higher order terms appear in the exact equations → Modelling required
• Reynolds Stress Models solve an equation for the transport of each Reynolds
stresses in the fluid, together with an equation for e or w (7 equations)
• The Reynolds stresses are inserted directly into the momentum equations
• Formally, for each additional scalar – like, e.g., temperature – again a system of 3
flux equations would need to be added – Second Moment Closure on all levels
• RSM are computationally expensive:
− More equations to solve
− Less robust than Eddy-Viscosity Models
− Potentially higher generality and better accuracy

4 © 2018 ANSYS, Inc.


RSM Models - Derivation
𝜕𝜌𝑢′𝑖 𝜕𝜌𝑈ഥ𝑘 𝑢′𝑖 𝜕𝜌𝑈ഥ𝑖 𝑢′𝑘 𝜕𝜌𝑢′𝑖 𝑢′𝑘 𝜕𝑝′ 𝜕 2 𝑢′𝑖
• Transport equation for 𝒖′𝒊 and + + + =− +𝜇 × 𝒖′𝒋
𝜕𝑡 𝜕𝑥𝑘 𝜕𝑥𝑘 𝜕𝑥𝑘 𝜕𝑥𝑖 𝜕𝑥𝑘2
multiply by 𝒖′𝒋 → (1)
𝜕𝜌𝑢′𝑖 𝜕𝜌𝑈ഥ𝑘 𝑢′𝑖 𝜕𝜌𝑈ഥ𝑖 𝑢′𝑘 𝜕𝜌𝑢′𝑖 𝑢′𝑘 𝜕𝑝′ 𝜕 2 𝑢′𝑖
𝑢′𝑗 + 𝑢′𝑗 +𝑢′𝑗 + 𝑢′𝑗 = −𝑢′𝑗 + 𝜇𝑢′𝑗 (1)
𝜕𝑡 𝜕𝑥𝑘 𝜕𝑥𝑘 𝜕𝑥𝑘 𝜕𝑥𝑖 𝜕𝑥𝑘2
• Same for equation for 𝒖′𝒋 and 𝜕𝜌𝑢′𝑗 𝜕𝜌𝑈ഥ𝑘 𝑢′𝑗 𝜕𝜌𝑈ഥ𝑗 𝑢′𝑘 𝜕𝜌𝑢′𝑗 𝑢′𝑘 𝜕𝑝′ 𝜕 2 𝑢′𝑗
+ + + =− +𝜇 × 𝒖′𝒊
multiply by 𝒖′𝒊 → (2) 𝜕𝑡 𝜕𝑥𝑘 𝜕𝑥𝑘 𝜕𝑥𝑘 𝜕𝑥𝑗 𝜕𝑥𝑘2

𝜕𝜌𝑢′𝑗 𝜕𝜌𝑈ഥ𝑘 𝑢′𝑗 𝜕𝜌𝑈ഥ𝑗 𝑢′𝑘 𝜕𝜌𝑢′𝑗 𝑢′𝑘 𝜕𝑝′ 𝜕 2 𝑢′𝑗


𝑢′𝑖 + 𝑢′𝑖 +𝑢′𝑖 + 𝑢′𝑖 = −𝑢′𝑖 + 𝜇𝑢′𝑖 (2)
𝜕𝑡 𝜕𝑥𝑘 𝜕𝑥𝑘 𝜕𝑥𝑘 𝜕𝑥𝑗 𝜕𝑥𝑘2

• Add (1)+(2) and time average (1) + (2)

𝜕𝜌𝑢′𝑖 𝑢′𝑗 𝜕𝜌𝑈 ഥ𝑘 𝑢′𝑖 𝑢′ 𝜕𝜌𝑈 ഥ𝑖 𝜕𝜌𝑈 ഥ𝑗 𝜕𝜌𝑢′𝑖 𝑢′𝑗 𝑢′𝑘


𝑗
+ + 𝑢′𝑘 𝑢′𝑗 +𝑢′𝑘 𝑢′𝑖 +
𝜕𝑡 𝜕𝑥𝑘 𝜕𝑥𝑘 𝜕𝑥𝑘 𝜕𝑥𝑘
Exact transport Equation for 𝜕 2 𝑢′ 𝑖 𝑢′ 𝑗
𝜕𝑢′ 𝑖 𝑝′ 𝜕𝑢′𝑗 𝑝′ 𝜕𝑢′ 𝑖 𝜕𝑢′𝑗 𝜕𝑢′ 𝑖 𝜕𝑢′𝑗
Reynolds Stress Tensor =−
𝜕𝑥𝑗

𝜕𝑥𝑖
+𝜇
𝜕𝑥𝑘2
− 2𝜇
𝜕𝑥𝑥 𝜕𝑥𝑥
+𝑝 ′
𝜕𝑥𝑗
+ 𝑝′
𝜕𝑥𝑖

5 © 2018 ANSYS, Inc.


RSM Models - Derivation

𝜕𝜌𝑢′𝑖 𝑢′𝑗 𝜕𝜌𝑈 ഥ𝑘 𝑢′𝑖 𝑢′


• Transport terms (Exact): +
𝑗
=
𝜕𝑡 𝜕𝑥𝑘
ഥ𝑗
𝜕𝜌𝑈 𝜕𝜌𝑈 ഥ𝑖
• Production terms (Exact): −𝑢′𝑖 𝑢′𝑘
𝜕𝑥𝑘
−𝑢′𝑗 𝑢′𝑘
𝜕𝑥𝑘
𝜕 𝜕 2 𝑢′ 𝑖 𝑢′ 𝑗
• Diffusion terms (Modeling): −
𝜕𝑥𝑘
𝜌𝑢′ 𝑖 𝑢′ 𝑗 𝑢′ + 𝑝′ 𝑢′𝑗 𝛿𝑖𝑘 + 𝑝𝑝′ 𝑢′ 𝑖 𝛿𝑗𝑘 + 𝜇
𝑘 𝜕𝑥𝑘2

𝜕𝑢′ 𝑖 𝜕𝑢′𝑗
• Dissipation term (e-equation): −2𝜇
𝜕𝑥𝑥 𝜕𝑥𝑥
𝜕𝑢′ 𝑖 𝜕𝑢′𝑗
• Pressure-strain term (Modeling): +𝑝′ + 𝑝′
𝜕𝑥𝑗 𝜕𝑥𝑖

The most critical term is the pressure-strain term. Models differ by how this term is formulated

6 © 2018 ANSYS, Inc.


𝜕𝑢′ 𝑖 𝜕𝑢′ 𝑗
The Pressure Strain Term 𝜙𝑖𝑗 = 𝑝 ′
𝜕𝑥𝑗
+
𝜕𝑥𝑖

• The pressure strain term can be re-formulated as two integral terms. The first of these terms
only contain fluctuating components and the second term depends also on the mean flow
velocity gradients
• These two terms are often called a “slow” and a “fast” term 𝝓𝒊𝒋 = 𝝓𝒊𝒋,𝟏 + 𝝓𝒊𝒋,𝟐 . The fast term
got its name from the fact that it explicitly depends on the mean flow velocity gradients and
therefore responds quickly to changes in these gradients, whereas the slow term does not
• There are many different forms and models for these two terms which largely define the
RSM. A compromise between simplicity and generality needs to be made:
– Overly simplistic models might miss some of the more important physics
– Overly complex formulations might be hard to calibrate and even harder to solve numerically
• The two more prominent models are (named by the authors):
– Launder-Reece and Rodi (LLR)
– Speziale-Sarkar-Gatski (SSG)

7 © 2018 ANSYS, Inc.


Pressure Strain Term – Slow Term

• The slow pressure strain correlations term tends to redistribute turbulence energy
between the Reynolds stress components towards isotropy. It is therefore often
formulated in terms of the anisotropy tensor:
𝑢𝑖′ 𝑢𝑗′ 1
𝑎𝑖𝑗 = − 𝛿𝑖𝑗
2𝑘 3

• In engineering models one typically uses for the slow term:


1
𝜙𝑖𝑗,1 = −𝐶𝑠1 𝜚𝜀𝑎𝑖𝑗 −𝐶𝑠2 𝜚𝜀 𝑎𝑖𝑘 𝑎𝑘𝑗 − 𝛿𝑖𝑗 𝑎𝑘𝑙 𝑎𝑘𝑙
3

• The simplest models like Launder-Reece-Rodi (LLR) use only the linear (𝐶𝑠1 ) term
• More complex models like Speziale-Sarkar-Gatski (SSG) include also the non-linear
term (𝐶𝑠2 )

8 © 2018 ANSYS, Inc.


Pressure Strain Term – Fast Term

• The fast pressure strain correlations term depends also on the mean flow velocity
gradients. It is often modeled as:

• In this equation 𝑃 = 0.5𝑃𝑖𝑗 where 𝑃𝑖𝑗 is the production term of the Reynolds stresses,
𝑆𝑖𝑗 is the mean strain rate tensor and Ω𝑖𝑗 the vorticity tensor

9 © 2018 ANSYS, Inc.


Turbulence Diffusion Term
𝜕
𝐷𝑖𝑗 = − 𝜌𝑢′ 𝑖 𝑢′ 𝑗 𝑢′ + 𝑝𝑢′𝑗 𝛿𝑖𝑘 + 𝑝𝑢′ 𝑖 𝛿𝑗𝑘
𝜕𝑥𝑘 𝑘

• “Diffusion” terms are characterized by the fact that the terms are under a
differential operator
• The simplest model for this term is given by:
𝜕 𝜇𝑡 𝜕
𝐷𝑖𝑗 = 𝑢′ 𝑖 𝑢′ 𝑗
𝜕𝑥𝑘 𝜎 𝜕𝑥𝑘

• This term transfers 𝑢′ 𝑖 𝑢′ 𝑗 from numerically high levels to low levels


(similar to a heat transfer operator)
• There are more complex forms but they tend to be unstable (non-robust)
under industrial conditiions

10 © 2018 ANSYS, Inc.


Turbulence Dissipation Term
𝜕𝑢′ 𝑖 𝜕𝑢′𝑗
𝐸𝑖𝑗 = −2𝜇
𝜕𝑥𝑥 𝜕𝑥𝑥
• This term is very similar to the dissipation term in the k-equation. In fact of
summed over its indices (contraction) the dissipation term in the k-equation
appears
• The tensorial form of this term is simplified by
2
𝐸𝑖𝑗 = − 𝜀𝛿𝑖𝑗
3

• This modeled term only acts on the diagonal elements of the Reynolds Stresses
and sums up to the dissipation 𝜀 in the k-equation
• Since the dissipation 𝜀 appears in several of the terms of a RSM, the equation
for 𝜀 needs to be solved as well (or alternatively the 𝜔–equation using the
relation 𝜀 = 𝐶𝜇 𝑘𝜔)

11 © 2018 ANSYS, Inc.


Standard Reynolds Stress Model Equation

• The standard Reynolds Stress model in Fluent is based on the e-equation


– The differential equation Reynolds stress transport with isotropic formulation is:
𝜕 𝜕 2 𝜕 𝜇𝑡 𝜕𝑢𝑖 𝑢𝑗
(𝜌𝑢𝑖 𝑢𝑗 ) + (𝑈𝑘 𝜌𝑢𝑖 𝑢𝑗 ) = 𝑃𝑖𝑗 + 𝜑𝑖𝑗 − 𝜌𝜀 𝛿𝑖𝑗 + 𝜇+
𝜕𝑡 𝜕𝑥𝑘 3 𝜕𝑥𝑘 𝜎𝑘 𝜕𝑥𝑘

𝜑𝑖𝑗 is the pressure-strain correlation and 𝑃𝑖𝑗 is production term


– As the turbulence dissipation appears in the individual stress equations, an equation for e is
required, the equation with isotropic formulation is:

𝜕 𝜕 𝜀 𝜕 𝜇𝑡 𝜕𝜀
(𝜌𝜀) + (𝜌𝑈𝑘 𝜀) = (𝑐𝜀1 𝑃 − 𝑐𝜀2 𝜌𝜀) + (𝜇 + )
𝜕𝑡 𝜕𝑥𝑘 𝑘 𝜕𝑥𝑘 𝜎𝜀 𝜕𝑥𝑘

12 © 2018 ANSYS, Inc.


Standard Reynolds Stress Model Equation
• LRR
• A simpler version, less accurate but more stable
• Select Linear Pressure-Strain in Viscous Model panel
• Can be used with EWT or wall functions
• SST
• More accurate than LRR, especially for swirling flow
• Select Quadratic Pressure-Strain in Viscous Model panel
• Not compatible with EWT, use Scalable Wall Functions

13 © 2018 ANSYS, Inc.


Reynolds Stress-w Models

• Reynolds Stress-w Models


– Stress-Omega model
▪ Based on the LRR model with the epsilon equation replaced by the omega
equation
– Stress-BSL model
▪ Solves the scale equation of the BSL k-w model, removing the freestream
sensitivity observed with the stress-omega model
▪ Uses same pressure-strain correlation as Stress-Omega, but without low
Reynolds number correction terms
• These two models are based on the w-equation, and related to each
other in the same way as the two-equation k-w and BSL models
• Advantage: More accurate near wall treatment

14 © 2018 ANSYS, Inc.


Reynolds Stress Models: When should be Used

• Reynolds Stress Models are recommended for following cases:


− Free shear flow with strong anisotropy, like a strong swirl component, including flows
in rotating fluids
− Flow with sudden changes in the mean strain rate
− Flows with strong streamline curvature
− Secondary flow
− Buoyant flow
− Flows where the strain fields are complex, and reproduce the anisotropic nature of
turbulence itself

15 © 2018 ANSYS, Inc.


Streamline Curvature

• Standard eddy-viscosity models are insensitive to streamline


curvature and system rotation (however – see Curvature Correction)
• Enhancements:
− Reynolds Stress models
− Non-linear algebraic stress models
− Curvature correction models
• Test cases:
− Cyclone/Hydrocyclone flow

16 © 2018 ANSYS, Inc.


Flow in Cyclone

• Example: flow in cyclone


• Barrel diameter: 0.29 m
Uin=16.1m/s
• Re D = 310,000
• Fluid: air
• Mixed element mesh, but mainly hexahedral h1
− 250127 nodes, 237790 elements

Measurement h2
lines

17 © 2018 ANSYS, Inc.


Flow in Cyclone
Tangential velocity Axial velocity
z=0.978 RS_tran_x z=0.978 RS_tran_x
RS_tran_y RS_tran_y
2.5
data
SSG model 1.5
data 1.0

uaxial /Uin
utang /Uin
0.5
0.5

-0.5 0.0
-1.5
-0.5
-2.5 -1.0 -0.5 0.0 0.5 1.0
-1.0 -0.5 0.0 0.5 1.0
r/R
r/R

z=0.978 RS_tran_x z=0.978 RS_tran_x


RS_tran_y RS_tran_y
2.0
data 1.0 data
1.5
LRR model 1.0

uaxial/Uin
0.5
utang/Uin

0.5
0.0
-0.5 0.0
-1.0
-1.5
-0.5
-2.0
-1.0 -0.5 0.0 0.5 1.0
-1.0 -0.5 0.0 0.5 1.0
r/R
r/R

18 © 2018 ANSYS, Inc.


Hydrocyclone

Z=20mm

Z=32mm

Z=53mm

Z=73mm

Z=117mm

19 © 2018 ANSYS, Inc.


Hydrocyclone Flow

Tangential Velocity

20 © 2018 ANSYS, Inc.


Hydrocyclone Flow
Tangential Velocity

21 © 2018 ANSYS, Inc.


Rotating Channel: Effect of EARSM Curvature Correction

22 © 2018 ANSYS, Inc.


NACA 0012 Wing Tip Vortex

• Turbulence intensity (Tu) in the tip vortex:

SST SST with Curvature Correction (CC)

EARSM EARSM with Curvature Correction (CC)

23 © 2018 ANSYS, Inc.


Comments on Reynolds Stress Models

• RSM model develops partial differential equations for each of the six
Reynolds’ stress terms. Allow for anisotropy.
• Advantage:
− Most general of classical turbulence models
− Only need initial/boundary values
− Very accurate calculate of mean flow properties and all Reynolds stresses for many
flows
• Disadvantages:
− Robustness impaired relative to eddy-viscosity models
− Large computing costs (7 extra equations)
− Not as widely validated as k-epsilon and mixing length
− Performs just as poorly as k-epsilon in some flows (axisymmetric jets, unconfined
recirculating flows)
24 © 2018 ANSYS, Inc.
Explicit Algebraic Reynolds Stress Model (Beta)

• Explicit Algebraic Reynolds Stress Models (EARSM) allow an extension of the


current BSL model to additional flow effects:
− Flows with streamline curvature and system rotation
− Secondary flows
− Flows with buoyancy
− Beta feature in Fluent
• Basis:
− Wallin and Johansson (2000)

25 © 2018 ANSYS, Inc.


EARSM – Stress Strain Relation –Wallin-Johansson
2
𝜏𝑖𝑗 = 𝑢𝑖′ 𝑢′ 𝑗 = 𝑘 𝑎𝑖𝑗 + 𝛿𝑖𝑗
3
Linear part of Stress-Strain relation

𝑎𝑖𝑗 = 𝛽1 𝑇1,𝑖𝑗 + 𝛽2 𝑇2,𝑖𝑗 + 𝛽3 𝑇3,𝑖𝑗 + 𝛽4 𝑇4,𝑖𝑗 + 𝛽6 𝑇6,𝑖𝑗

1 1
𝑇1,𝑖𝑗 = 𝑆𝑖𝑗 ; 𝑇2,𝑖𝑗 = 𝑆𝑖𝑘 𝑆𝑘𝑗 − 𝐼𝐼𝑆 𝛿𝑖𝑗 ; 𝑇3,𝑖𝑗 = Ω𝑖𝑘 Ω𝑘𝑗 − 𝐼𝐼Ω 𝛿𝑖𝑗 ; 𝑇4,𝑖𝑗 = 𝑆𝑖𝑘 Ω𝑘𝑗 − Ω𝑖𝑘 𝑆𝑘𝑗
3 3
2
𝑇6,𝑖𝑗 = 𝑆𝑖𝑘 Ω𝑘𝑙 Ω𝑙𝑗 + Ω𝑖𝑘 Ω𝑘𝑙 𝑆𝑙𝑗 − 𝐼𝑉𝛿𝑖𝑗 − 𝐼𝐼Ω 𝑆𝑖𝑗 ;
3
𝑁 2𝐼𝑉 1 𝑁
𝛽1 = − , 𝛽2 = 0, 𝛽3 = − , 𝛽4 = − , 𝛽6 = − ,
𝑄 𝑁𝑄1 𝑄 𝑄1

9 𝑃𝑘 Non-linearity due to Pk
𝑁= 𝐶1′ +
4 𝜀
9
𝐴1 = 1.2; 𝐶1′ = 𝐶 −1 , 𝐶1 = 1.8
4 1

26 © 2018 ANSYS, Inc.


Corner Flows

• Early separation of linear Eddy Viscosity


Models in corners observed ¼ of cross section of square duct,
secondary flow into corner
• Can be caused by lack of anisotropy in the
stress formulation(differences in normal
stresses near wall)
• Anisotropy is cause of secondary flows
into the corner
• Reynolds Stress Model (RSM) could
account for this – but is often not robust
enough for complex flows
• Explicit Algebraic RSM (EARSM) offer an
attractive alternative with reduced
numerical effort and increased robustness

27 © 2018 ANSYS, Inc.


Flat Plate

• Calibrated for log layer


• Results very close to SST model
• Combination of (S-) WJ-EARSM and BSL (A1=1.245)

28 © 2018 ANSYS, Inc.


Fully Developed Square Duct Flow

• Secondary Flow
requires anisotropies
• The results in
secondary velocity
driving fluid into the
corner
• Higher velocity of axial
flow near corner
• Less prone to
separation?

DNS of Huser and Biringen, (1993)


29 © 2018 ANSYS, Inc.
Stanford Diffuser – Numerical set-up

• Computational domain dimensions:


− -3H < X < 45H H
− 0 < Y < 4H
− 0 < Z < 4H
• Mesh Inlet Outlet

− 145 x 91 x 121 nodes


− 1,596,595 nodes totally
− <Y+> = 0.22
• Boundary conditions
− Developed flow at inlet
− Zero gauge pressure at outlet
30 © 2018 ANSYS, Inc.
Stanford Diffuser – Flow Topology
X/H=16

• Flow topology depends


strongly on turbulence
model
• Stress anisotropy
necessary to obtain
correct behaviour

31 © 2018 ANSYS, Inc.


Stanford Diffuser – Distribution of Pressure Coefficient

Data for Diffuser 1

32 © 2018 ANSYS, Inc.


Wing-Body Separation

SST
• Corner separation
overpredicted by SST
and “linear“ BSL-EARSM
(only b1 coefficient
active)
• Non-linear EARSM
stress-strain relation S-BSL-EARSM,
isotropic term only
S-BSL-EARSM,
full model
gives reduced
separation
• Consistent with
expectations

33 © 2018 ANSYS, Inc.


Setting up RSM Turbulence Models in ANSYS
Fluent

Turbulence Modeling Using ANSYS CFD

34 © 2018 ANSYS, Inc.


SMC Models in Fluent

• Epsilon based models:


− Linear Pressure-Strain Model
− Quadratic Pressure-Strain Model
• SSG Reynolds Stress Model (Speziale-Sarkar-Gatski)
• Stress-Omega Model
• Stress-BSL Model

35 © 2018 ANSYS, Inc.


Summary

In this lecture, we covered:


• Theory of Reynolds stress models
• How to apply Reynolds stress models
• CFX and Fluent RSM results compared with experimental data

36 © 2018 ANSYS, Inc.

You might also like