an introduction to turbulence modeling for...

An Introduction to

Turbulence Modeling for CFD

Gerald Recktenwald⇤

February 19, 2020

⇤Mechanical and Materials Engineering Department Portland State University, Portland, Oregon,[email protected]

Turbulence is a Hard Problem

• Unsteady

• Many length scales

• Energy transfer between scales:! Large eddies break up into small eddies

• Steep gradients near the wall

ME 4/548: Turbulence Modeling page 1

Engineering Model: Flow is “Steady-in-the-Mean” (1)

0 0.5 1 1.5 2t

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

2

u



Reality:

Turbulent flows are unsteady: fluctuations at a point are caused byconvection of eddies of many sizes. As eddies move through the flowthe velocity field changes in complex ways at a fixed point in space.

Turbulent flows have structures – blobs of fluid that move and thenbreak up.



Engineering Model:

When measured with a “slow” sensor (e.g. Pitot tube) the velocityat a point is apparently steady. For basic engineering analysis wetreat flow variables (velocity components, pressure, temperature) astime averages (or ensemble averages). These averages are steady(ignorning ensemble averaging of periodic flows).


Engineering Model: Enhanced Transport (1)

Turbulent eddies enhance mixing.

• Transport in turbulent flow is much greater than in laminar flow: e.g.pollutants spread more rapidly in a turbulent flow than a laminar flow

• As a result of enhanced local transport, mean profiles tend to be moreuniform except near walls.

• Near walls, gradients of velocity, temperature (and other scalars (e.g.,chemical concentration) are steep.

Visualize two-layer: high viscosity in core of pipe and low viscositynear the wall – Cartoon view only!


Engineering Model: Enhanced Transport (2)

Turbulence models are a way to account for enhanced mixing whiletreating the flow as steady-in-the-mean

• Apparent e↵ect of turbulence is to increase the e↵ective viscosity,thermal conductivity, and di↵usivity.

• Enhanced transport coe�cients are properties of the flow, not realthermophysical transport coe�cients.

• Most commonly used turbulence models provide a way to compute thee↵ective transport coe�cients, e.g. the turbulence viscosity.


Computational Models (1)

1. Direct Numerical Simulation (DNS)

• No modeling of turbulence• Resolve all time scales and spatial scales• Useful for fundamental research• Don’t scale to highest Re and hard to implement in complex

geometries

Rodi1 cites a DNS simulation in 2015

With increasing computer power, channel-flow simulations with ever-increasing Rwere carried out over the years, and at the time of writing, the largest calculationis that of Lee and Moser (2015) at R = 125, 000. This employed 242⇥ 109 gridpoints and ran for several months on the Peta-FLOP/S computer Mira.

1Journal of Hydraulic Engineering, 2017, 143(5)


World’s Fastest computers as of November 2018

See https://www.top500.org/lists/2018/11/. Mira is now #21 with Rmax = 0.06Rmax(Summit)

Note: the average US home uses ⇡1 kW.


Trend in the Top 500

https://www.top500.org/lists/2018/11/



2. Large Eddy Simulation (LES)

• Model the small scale (sub-grid) turbulence• Larger, energy-containing eddies are resolved in unsteady flow• Transient data needs to be averaged to obtain engineering quantities• Useful for rigorous engineering analysis• StarCCM+ can do LES and DES (Detached Eddy Simulation)



3. Reynolds-Averaged Navier Stokes (RANS)

• Turbulence e↵ects are replaced by enhanced mixing – eddy viscosity• Standard approach for basic engineering computations• Extremely limited ability to capture complex structure


Reynolds Averaged Navier Stokes

RANS


Reynolds Averaged Navier-Stokes (RANS)

• Average the Navier-Stokes equations in attempt to solve equations thatare steady in the mean.

• Averaging creates additional unknowns: the Reynolds stresses

• Use a turbulence model to compute the Reynolds stresses.

. Simplest models are based on the Boussinesq eddy viscosity

. Additional equations are necessary to compute the eddy viscosity

. The k � " model is the standard engineering model


Reynolds Decomposition (1)

Instantaneous velocity is

u(x, y, z, t) = exu(x, y, z, t) + eyv(x, y, z, t) + ezw(x, y, z, t)

Decompose each component into a mean and fluctuating component

u = U + u0 v = V + v0 w = W + w0



Reynolds averaging rules

1. f + g = f + g

2. af = af (a is a constant)

3.@f

@s=

@f

@s

4. f g = f g



Substitute into the Navier-Stokes equations and average. Consider one ofthe convection terms in the x-direction momentum equation

@

@x(⇢uu) =

@

@x

h⇢(U + u0)(U + u0)

irule 3

=@

@x

⇥⇢�UU + 2Uu0 + u0u0

�⇤rule 2, with ⇢ = constant

=@

@x

⇣⇢UU + ⇢u02

⌘by definition of U , and Uu0 = Uu0 = 0



The averaging process has the important property that although u0 = 0,

u0u0 6= 0

Terms like u0iu

0j are called Reynolds stresses, and arise from the process of

Reynolds averaging of the momentum equations.



There are six Reynolds stresses, which can be represented by the followingmatrix. 2

4u0u0 u0v0 u0w0

u0v0 v0v0 v0w0

u0w0 v0w0 w0w0

3

5

The matrix is symmetric because u0v0 = v0u0 and v0w0 = w0v0. TheReynolds stresses are field variables, i.e., each term in the matrix is afunction of space in the flow.


Turbulence Kinetic Energy

Another important quantity is the turbulence kinetic energy

k =1

2u0ku

0k =

1

2

�u0u0 + v0v0 + w0w0

�(1)

If the turbulence is isotropic, then u0u0 = v0v0 = w0w0, and2

k =3

2u0u0 (2)

2Equality of normal stresses is a consequence of isotropy, not the definition of it.


Turbulence Intensity(1)

Turbulence Intensity is a measure of the velocity associated with thefluctuations in the flow relative to the mean flow. We start by defining theturbulence intensity in each coordinate direction

TIx =

pu0u0

Ux,0, T Iy =

pv0v0

Uy,0, T Iz =

pw0w0

Uz,0, (3)

where U0 is a representative length scale

The total turbulence intensity is

TI =

r1

3(u0u0 + v0v0 + w0w0)

U0(4)


Turbulence Intensity(2)

If the turbulence is isotropic, then from

TI =

r1

3(u0u0 + u0u0 + u0u0)

U0=

q(u0u0)

U0(5)

and from Equation (2),

TI =

r2

3k

U0(6)

thus, another interpretation of turbulence intensity is that it is a measureof (the square root of) local turbulence kinetic energy.


Turbulence Models


Kinds of turbulence models

1. Large eddy simulation

a. Resolve unsteady large scale motionsb. Model small scale (sub-grid scale) motions

2. Reynolds stress models

a. Treat flow as steady (no large scale motions)b. Solve model equations for the Reynolds stresses

3. Two-equation models

a. Treat flow as steady (no large scale motions)b. Use Boussinesq eddy viscosity to represent Reynolds stressesc. Solve model equations to compute eddy viscosity


Boussinesq model (1)

The viscous stress tensor is

⌧ij = µ

✓@Uj

@xi+

@Ui

@xi

◆� 2

3�ij

@Uk

@xk

�(7)

Assume (wish, hope) that the Reynolds stresses have the same form asthe viscous stresses.

�⇢u0iu

0j = µt

✓@Uj

@xi+

@Ui

@xi

◆� 2

3�ij

@Uk

@xk

�� 2

3⇢ �ijk (8)

where k is the turbulence kinetic energy


Boussinesq model (2)

In practice, the Boussinesq model involves replacing the viscosity in themomentum equations with an e↵ective viscosity

µe↵ = µ+ µt

where µ is the physical or moleclar viscosity, and µt is a turbulence

viscosity simulates the enhanced mixing due to turbulence.

General observations

• µt is a point value. µt has no direction, so therefore the Boussinesqmodel assumes that the turbulence is isotropic

• µt needs to be estimated by some other model


Prandtl’s Mixing Length Hypothesis

Prandtl made the direct analogy with the the molecular model of viscosityfrom the kinetic theory of gases.

molecular viscosity of gases turbulence viscosity

µ =1

3⇢`fVm µt = ⇢`mVt

`f = mean free path `m = mixing length

Vm = velocity of molecules Vt = velocity scale of turbulence

The mixing length is taken as an appropriate length scale for the flow.

The mixing length can vary in proportion to distance from a wall.


Estimate of Eddy Viscosity for Pipe Flow (1)

We can estimate the magnitude of the eddy viscosity in a pipe. Prandtl’smodel is

µt ⇠ ⇢Vt`m

Turbulent fluctuations are small compared to the mean velocity in thepipe, say,

0.01 u0

U 0.15

To estimate µt, take u0/U ⇠ 0.1 and Vt ⇠ u0 so that

Vt ⇠ 0.1U


Estimate of Eddy Viscosity for Pipe Flow (2)

To make a somewhat arbitrary choice of a single length scale thatcharacterizes the turbulent mixing, take

`m ⇠ D

4

Combining the preceding expressions gives the following estimate ofturbulence viscosity

µt = ⇢(0.1U)(0.25D) = 0.025⇢UD

From the estimate of µt we can compute an e↵ective Reynolds number

Ree↵ =⇢UD

µt=

⇢UD

0.025⇢UD= 40


The k � " model


How do we compute µt?

Short answer:

µt = Cµ⇢k2

"where k is the turbulence kinetic energy, " is the turbulence dissipationrate, and Cµ is a constant.

We need to estimate k and "

In the k � " model, we solve two transport equations: one for the k fieldand one for the " field.

Then, the e↵ective viscosity in the Boussinesq model is

µe↵ = µ+ µt


k � " model

Standard equations. See [3].

Transport equation for k

Dk

Dt=

@

@xi

µe↵

�k

@k

@xi

�

| {z }di↵usion

+

µt

✓@Ui

@xj+

@Uj

@xi

◆� 2

3⇢ �ijk

�@Uj

@xi| {z }production

� cD⇢ k3/2

`m| {z }dissipation

Transport equation for "

D"

Dt=

@

@xi

µe↵

�"

@"

@xi

�

| {z }di↵usion

+ c",1

µt

✓@Ui

@xj+

@Uj

@xi

◆� 2

3⇢ �ijk

�@Uj

@xi| {z }production

� c",1⇢ "2

k| {z }dissipation


What is "?

Use a simple scaling argument to set an upper bound on the dissipationrate (see Panton [1, § 22.10]).

velocity scale of largest eddy ⇠ u0

kinetic energy of the eddy ⇠ u20

length scale of the eddy ⇠ L

turn-over time of the eddy ⇠ L

u0

An upper estimate of the dissipation rate, " is

" ⇠ kinetic energy of an eddy

time for one rotation=

u20

L/u0=

u30

L(9)


What is "?

But . . . dissipation happens on the smallest scale.

Thus, " ⇠ u30

Lgives an upper bound, but not a good scale

At the smallest scales the turbulence is (tends to be) isotropic. Theturbulence kinetic energy of isotropic turbulence is

k =3

2u0u0 isotropic turbulence

Thus, we can estimate the velocity scale as

|u0| ⇠pk so that " ⇠ k3/2

L


What is "?

Introduce a dissipation length scale, `" such that the dissipation rate is

" ⇠ CDk3/2

`"(10)

where CD corrects (,) any error in the scale estimate.

But. . . now we need to estimate `"

Turn Equation (10) around

`" ⇠ CDk3/2

"(11)

Use Prandtl’s mixing length hypothesis to estimate " from the mixinglength


Prandtl’s mixing length hypothesis

`m is the length scale for the eddies responsible for mixing, i.e., the mostenergetic eddies. `m > `".

`m = C 0`"

Prandtl’s estimate of the turbulence viscosity (eddy viscosity) is

µt ⇠ ⇢Vt`m

where Vt is the velocity scale

At the dissipation scale, which is isotropic

Vt ⇠pk


Prandtl’s mixing length hypothesis

Combine estimates of Vt, `" in Prandtl’s expression for eddy viscosity

µt = C⇢

✓CD

k3/2

"

◆⇣k1/2

⌘

or

µt = Cµ⇢k2

"(12)

where Cµ is a constant adjusted from experimental values


Calculation of the eddy viscosity, µt, in a CFD code

1. Solve equations for velocities and pressure, using current guess at µe↵.

2. Solve the k and " equations

3. At each point in the flow compute

µt = Cµ⇢k2

"

4. Update µe↵ = µ+ µt.

5. Return to step 1 until convergence.


Wall Functions


Wall Functions

In many engineering flows, we don’t have the computational power toresolve the velocity profile near the wall. One solution is to fudge it withwall functions

Expandthe y scale


Wall variables

Scaling arguments lead to the inner (near-wall) coordinates

y+ =yu⇤⌫

(13)

u⇤ =

r⌧w⇢

friction velocity (14)

u+ =u

u⇤(15)

In the viscous sublayeru+ = y+

In the log-law (inertial) sublayer

u+ = c1 ln y+ + c2 (16)


Wall treatment

k-ε model is used to compute μeff

in the central region of the flow

Wall functions are used to compute μeff

near solid surfaces.

Remember the prism layer cells in StarCCM+?


Wall treatment

Adjust the near-wall mesh spacing so that near-boundary cells have goodvalues of y+, where good depends on the wall treatment.

In StarCCM+, the value of y+I is called the “wall y+”.

I

B

yI

ures


Wall treatment models in StarCCM+

The main wall treatment models in StarCCM+ are

• High y+: Classic wall treatment

• Low y+: Extend cells into the viscous sublayer

• All y+: Recommended model that blends low y+ and high y+

Wall treatment is considered in another slide deck.


References

[1] Ronald L. Panton. Incompressible Flow. Wiley, New York, secondedition, 1996.

[2] Stephen B. Pope. Turbulent Flows. Cambridge University Press,Cambridge, UK, 2000.

[3] John C. Tannehill, Dale A. Anderson, and Richard A. Pletcher.Computational Fluid Mechanics and Heat Transfer. Taylor andFrancis, Washington, D.C., second edition, 1997.


an introduction to turbulence modeling for...

Documents