What is Turbulence?

Turbulence Modelling in CFD ¾ Unsteady, aperiodic (irregular) motion in which transported quantities
(mass, momentum, scalar species) fluctuate in time and space
– Identifiable swirling patterns characterize turbulent eddies.
by Danie de Kock
– Enhanced mixing (matter, momentum, energy, etc.) results
¾ Contains a wide range of turbulent eddy sizes (scales spectrum).
– The size/velocity of large eddies is on the order of mean flow.
z Large eddies derive energy from the mean flow
– Energy is transferred from larger eddies to smaller eddies
z In the smallest eddies, turbulent energy is converted to internal energy
by viscous dissipation.
¾ Results in mixing matter, momentum and energy
¾ Characteristics:
– Fluctuations
Sources: 1. FM White, “Viscous Fluid Flow”, 2nd Edition, McGraw-Hill International Editions, 1991. – Eddies
2. FLUENT Training Notes, Fluent Inc., Lebanon, New Hampshire, USA, www.fluent.com, 2005
– Random
– Self-sustaining
– Enhanced mixing

MSM732 October 2005 MSM732 3 of 57 October 2005

Agenda Examples of Turbulent Flow

¾ What is Turbulence? ¾ Numerical (DNS)
¾ Is the Flow Turbulent?
¾ Why Model Turbulence?
¾ Turbulence Modelling Approaches
¾ Turbulence Models
– Zero-equation Models
– One Equation
– Two Equation Center for Turbulence Research,Stanford University, Stanford, CA,USA, 2002 (ctr-sgi1.stanford.edu/CTR)
– Reynolds Stress Model (RSM) ¾ Experimental
– Large Eddy Simulation
– Detached Eddy Simulation
¾ Wall Function Approach
¾ Examples
¾ Turbulence Model Guidelines

MSM732 2 of 57 October 2005 MSM732 4 of 57 October 2005

Turbulence Modelling 1 Turbulence Modelling 2

 Is the Flow Turbulent? Overview of Computational Approaches
¾ Direct Numerical Simulation (DNS)
– Theoretically all turbulent flows can be simulated by numerically solving the
External Flows ρUL
where Re L ≡ full Navier-Stokes equations.
µ
Rex ≥ 5×10 5 along a surface – Resolves the whole spectrum of scales. No modeling is required.
L = x, D, Dh, etc. – But the cost is too prohibitive! Not practical for industrial flows
ReD ≥ 20,000 around an obstacle ¾ Large Eddy Simulation (LES)
Other factors such as free-stream – Solves the spatially averaged N-S equations. Large eddies are directly
turbulence, surface conditions, and resolved, but eddies smaller than the mesh sizes are modeled.
Internal Flows disturbances may cause earlier
– Less expensive than DNS, but the amount of computational resources and
transition to turbulent flow.
ReDh ≥ 2,300 efforts are still too large for most practical applications.
¾ Reynolds-Averaged Navier-Stokes (RANS) Equations Models
– Solve ensemble-averaged Navier-Stokes equations
Natural Convection
gβ∆TL3ρ – All turbulence scales are modeled in RANS.
Ra ≥ 108 − 1010 where Ra ≡
µα – The most widely used approach for calculating industrial flows.
¾ There is not yet a single turbulence model that can reliably predict all
turbulent flows found in industrial applications with sufficient accuracy.

MSM732 5 of 57 October 2005 MSM732 7 of 57 October 2005

Why Model Turbulence? Turbulence Scales and Prediction Methods

¾ Direct Numerical Simulation (DNS) is the solution of the time-dependent
Navier-Stokes equations without recourse to modeling
– Mesh must be fine enough to resolve smallest eddies
– Only practical for simple low-Re flows.
¾ The need to resolve the full spectrum of scales is not necessary for most
engineering applications
– Mean flow properties are generally sufficient
¾ Turbulence modelling provide:
– Less cell count
– Steady state solutions possible

– Less CPU time

– Less memory

MSM732 6 of 57 October 2005 MSM732 8 of 57 October 2005

Turbulence Modelling 3 Turbulence Modelling 4

 RANS Modelling Approach RANS Modelling Approach
¾ Decompose velocity into mean and fluctuating parts: ¾ The RANS based turbulence models calculate the Reynolds
r r r Stresses by one of two methods (also called turbulent closure):
ui ( x , t ) = U i ( x , t ) + ui′( x , t ) (1) Eddy-Viscosity Models (EVM): Using the Boussinesq assumption, the
Reynolds stresses are related to the mean flow by a turbulent viscosity,
µt:
r
ui′( x , t ) ⎛ ∂U i ∂U j ⎞ 2 ∂U k 2
Boussinesq Hypothesis: Rij = − ρ ui′u′j = µ t ⎜⎜ + ⎟ − µt δ ij − ρkδ ij
u i (t) (isotropic stresses) ⎝ ∂x j ∂xi ⎟⎠ 3 ∂xk 3

ui µt = eddy viscosity (not a fluid property)

- Strain rate tensor, Sij, described in terms of mean flow.

Time - Isotropic viscosity assumed
(2) Reynolds-Stress Models (RSM): Solves individual transport equations
¾ Similar fluctuations for pressure, temperature, and species for the Reynolds stresses.
concentration values - Turbulent viscosity is not employed, no assumption of isotropy
- Contains more “physics”
- More complex and computationally expensive than (1)

MSM732 9 of 57 October 2005 MSM732 11 of 57 October 2005

RANS Modelling Approach Turbulence Modelling Approach

r r r
¾ Substitute ui ( x , t ) = U i ( x , t ) + ui′( x , t ) into the Navier-Stokes ¾ Calculating µt for Boussinesq Formula
equations and average – Based on dimensional arguments, µt can be determined from the
– Eg. Steady incompressible flow without body forces: conservative variables k, ε or ω.
k ≡ ui′ui′ / 2 is the turbulent kinetic energy
∂U i (t ) ∂p(t ) ∂ 2U i (t )
z
ρU k (t ) =− +µ z ε ≡ ν (∂ui′ ∂x j )(∂ui′ ∂x j + ∂u′j ∂xi ) is the dissipation rate of k (to thermal energy)
∂xk ∂xi ∂x j ∂x j
z ω ~ ε / k is the specific dissipation rate
⎛ ∂ (U i + ui′ ) ∂ (U i + ui′ ) ⎞ ∂ ( p + p′) ∂ ⎛⎜ ∂ (U i + ui′ ) ⎞⎟ – µt is calculated differently depending upon the turbulence model.
ρ ⎜⎜ + (U k + uk′ ) ⎟⎟ = − + µ
⎝ ∂t ∂xk ⎠ ∂xi ∂x j ⎜⎝ ∂x j ⎟
⎠ z Spalart-Allmaras
– This ‘single equation’ model solves one additional µt ≡ f (ν~)
– Results in the Reynolds Average Navier-Stokes (RANS) equations: transport equation for a modified viscosity.
z Standard k-ε, RNG k-ε, Realizable k-ε
∂U i ∂p ∂ 2U i ∂Rij
ρU k =− +µ + µt ≡ f (ρk ε )
2
– These ‘two equation’ models solve transport
∂xk ∂xi ∂x j ∂x j ∂x j equations for k and ε.
z Standard k-ω, SST k-ω
where Rij = − ρ ui u j (Reynolds stresses) – These ‘two equation’ models solve transport
µt ≡ f (ρ k ω)
equations for k and ω.

MSM732 10 of 57 October 2005 MSM732 12 of 57 October 2005

Turbulence Modelling 5 Turbulence Modelling 6

 Turbulence Modelling Approach Turbulence Models
¾ One Equation Model:
RANS-based
– Turbulent viscosity is indirectly solved for from single transport equation of
Zero-Equation Models modified viscosity
models
e.g. Mixing length theory – Eg. Spalart-Allmaras
One-Equation Models z Turbulent viscosity is determined from:
e.g. Spalart-Allmaras ⎡ (ν~ /ν )3 ⎤
µt = ρν~ ⎢ ~ 3 3⎥
Include Two-Equation Models Increase ⎣ (ν /ν ) + cν 1 ⎦
More e.g. Standard k-ε Computational z ν~ is determined from the modified viscosity transport equation:
Physics RNG k-ε Cost Dν~ 1 ⎡ ∂ ⎧⎪ ~ 2⎤
~ ∂ν ⎫⎪ + ρc ⎛⎜ ∂ν ⎞⎟ ⎥ − ρc f ν
~ ~
⎨(µ + ρν )
~ ⎢
k-ω Per Iteration ρ = ρcb1S ν~ + ⎬ b2 ⎜
σ ν~ ⎢ ∂x j ⎪⎩ ∂x j ⎪⎭ ⎟
⎝ ∂x j ⎠ ⎦⎥
w1 w
Dt d2
⎣
V2F Model z The additional variables are functions of the modified turbulent viscosity and
Reynolds-Stress Model velocity gradients.
Detached Eddy Simulation z Designed specifically for aerospace applications involving wall-bounded flows.
– Boundary layers with adverse pressure gradients
Large-Eddy Simulation
– turbomachinery
z Can use coarse or fine mesh at wall
Direct Numerical Simulation – Designed to be used with fine mesh as a “low-Re” model, i.e., throughout the viscous-
affected region.
– Sufficiently robust for relatively crude simulations on coarse meshes.

MSM732 13 of 57 October 2005 MSM732 15 of 57 October 2005

Turbulence Models Turbulence Models

¾ Zero-equation models ¾ Two Equation Model:
– Turbulent viscosity(µt) solve from algebraic expression – Turbulent viscosity is indirectly solved for from two additional partial
– e.g. Mixing Length Model: differential equations
∂u – For Two-Equation models, turbulent viscosity correlated with turbulent
µ t ≈ ρl 2 kinetic energy (TKE) and the dissipation rate of TKE.
∂y
k2
z Where l is a mixing-length scale (from experimental data) Turbulent Viscosity: µt ≡ ρCµ
ε
– Transport equations for turbulent kinetic energy and dissipation rate are
solved so that turbulent viscosity can be computed for RANS
equations.
Turbulent
Kinetic Energy: k ≡ uiui / 2

Dissipation Rate of ∂u ⎛ ∂u ∂u ⎞
ε ≡ν i ⎜⎜ i + j ⎟⎟
Turbulent Kinetic Energy: ∂x j ⎝ ∂x j ∂xi ⎠

MSM732 14 of 57 October 2005 MSM732 16 of 57 October 2005

Turbulence Modelling 7 Turbulence Modelling 8

 Turbulence Models Turbulence Models
¾ Two Equation Model: Standard k-ε Model ¾ Two Equation Models: RNG k-ε
Turbulent Kinetic Energy Turbulent Kinetic Energy

⎛ ∂U j ∂U i ⎞ ∂U j ∂k ∂ ⎛ ∂k ⎞
∂k ∂ ⎧ ∂k ⎫ ρU i = µt S 2 + ⎜α k µ eff ⎟ − ρε
ρU i
∂xi
= µt ⎜
⎜ ∂x
+ ⎟ +
∂x ⎟⎠ ∂xi ∂xi ⎩
⎨( µt σ k ) ⎬ − ρε
∂xi ⎭ { ∂xi { ∂xi ⎜⎝ ∂xi ⎟⎠ { 1 ⎛ ∂U ∂U ⎞
1
424 3 1⎝44i 424j 4 1
4243 Generation 1442443 Dissipation S ≡ 2SijSij , Sij ≡ ⎜ j + i ⎟
43 144 42444 3
Convection Diffusion 2 ⎜⎝ ∂xi ∂xj ⎟⎠
Convection Generation Diffusion Dissipation
Dissipation Rate Dissipation Rate
∂ε ⎛ ε ⎞ ⎛ ∂U j ∂U i ⎞⎟ ∂U j ∂ ⎧ ∂ε ⎫ ⎛ε2 ⎞ ∂ε ⎛ε ⎞ ∂ ⎛ ∂ε ⎞ ⎛ε2 ⎞
= C1ε ⎜ ⎟ µt ⎜ ρU i = C1ε ⎜ ⎟ µ t S 2 +
ρU i
∂xi ⎝ k ⎠ ⎜⎝ ∂xi
+ + ⎨( µt σ ε )
∂x j ⎟⎠ ∂xi ∂xi ⎩
⎬ − C2ε ρ ⎜⎜ ⎟⎟
∂xi ⎭ ⎝4
k ⎠ ∂xi 142 ⎝ ⎠ ∂ ⎜⎜α ε µ eff ∂x ⎟⎟ − C2ε ρ ⎜⎜ k ⎟⎟ − {
R
1
424 3 144444244444 3 144 42444 3 142 3 1
4243
k
4 43 4 14 xi ⎝
42443 i ⎠
142 ⎝ 43⎠ Additional term
Convection Generation Diffusion Destruction Convection Generation Diffusion Destruction related to mean strain
& turbulence quantities
σ k ,σ ε , C1ε , C2ε are empirical constants αk,αε ,C1ε ,C2ε are derived using RNG theory

(equations written for steady, incompressible flow w/o body forces) (equations written for steady, incompressible flow w/o body forces)

MSM732 17 of 57 October 2005 MSM732 19 of 57 October 2005

Turbulence Models Turbulence Models

¾ Two Equation Models: Standard k-ε Model ¾ Two Equation Models: RNG k-ε
– “Baseline model” (Two-equation) – k-ε equations are derived from the application of a rigorous statistical
z Most widely used model in industry technique (Renormalization Group Method) to the instantaneous
z Strength and weaknesses well documented Navier-Stokes equations.
– Semi-empirical – Similar in form to the standard k-ε equations but includes:
z k equation derived by subtracting the instantaneous mechanical energy z additional term in ε equation that improves analysis of rapidly strained
equation from its time-averaged value flows
z ε equation formed from physical reasoning z the effect of swirl on turbulence
– Valid only for fully turbulent flows z analytical formula for turbulent Prandtl number
– Reasonable accuracy for wide range of turbulent flows z differential formula for effective viscosity
z industrial flows – Improved predictions for:
z heat transfer z high streamline curvature and strain rate
z transitional flows
z wall heat and mass transfer

MSM732 18 of 57 October 2005 MSM732 20 of 57 October 2005

Turbulence Modelling 9 Turbulence Modelling 10

 Turbulence Models Turbulence Models
¾ Two Equation Models: Realizable k-ε ¾ Two Equation Models: k-ω
k
– Distinctions from Standard k-ε model: µt = α * ρ specific dissipation rate: ω
ω
z Alternative formulation for turbulent viscosity Dk ∂U i ∂ ⎡⎛ µ t ⎞ ∂k ⎤
2 ρ = τ ij − ρ β * f β * kω + ⎢⎜⎜ µ + ⎟ ⎥
σ k ⎟⎠ ∂x j ⎦⎥
k 1
µ t ≡ ρC µ where C µ = is now variable Dt ∂x j ∂x j ⎣⎢⎝
ε
Ao + As
U *k ε 1
ε ⎡⎛ ⎞ ∂ω ⎤
ω ≈ ∝
ρ
Dω ω ∂U i
= α τ ij − ρ β fβ ω 2 +
∂
⎢⎜⎜ µ +
µt
⎟⎟ ⎥ k τ
– (A0, As, and U* are functions of velocity gradients) Dt k ∂x j ∂x j ⎣⎢⎝ σω ⎠ ∂x j ⎦⎥
– Ensures positivity of normal stresses; – Belongs to the general 2-equation EVM family. Two variations widely used:
– Ensures Schwarz’s inequality; standard k-ω model by Wilcox (1998), and Menter’s SST k-ω model (1994).
– k-ω models have gained popularity mainly because:
z New transport equation for dissipation rate, ε: z Can be integrated to the wall without using any damping functions
Dε ∂ ⎡⎛ µt ⎞ ∂ε ⎤ ε2 ε z Accurate and robust for a wide range of boundary layer flows with
ρ = ⎢⎜⎜ µ + ⎟⎟ ⎥ + ρc1Sε − ρc2 + c1ε c3ε Gb
Dt ∂x j ⎣⎢⎝ σ ε ⎠ ∂x j ⎦⎥ k + νε k pressure gradient
– Most widely adopted in the aerospace and turbo-machinery communities.
Diffusion Generation Destruction Buoyancy – Several sub-models/options of k-ω : compressibility effects, transitional flows
and shear-flow corrections.

MSM732 21 of 57 October 2005 MSM732 23 of 57 October 2005

Turbulence Models Turbulence Models

¾ Two Equation Models: Realizable k-ε ¾ Two Equation Models: SST k-ω
– Shares the same turbulent kinetic energy equation as Standard k-ε – k-ω model has many good attributes and perform much better than k-ε models
for boundary layer flows.
– Superior performance for flows involving:
– Wilcox’ original k-ω model is overly sensitive to the freestream value (BC) of
z planar and round jets ω, while k-ε model is not prone to such problem.
z boundary layers under strong adverse pressure gradients, separation – Zonal (blended) k-ω / k-ε equations
z rotation, recirculation
z strong streamline curvature
Outer layer
k-ω model transformed from std.
(wake and
k-ε model
outward) 3
k 2
ε=
Modified Wilcox’lk-
Inner layer ε ω model
(sub-layer, log-
layer) Wilcox’ original k-ω model

Wall

MSM732 22 of 57 October 2005 MSM732 24 of 57 October 2005

Turbulence Modelling 11 Turbulence Modelling 12

 Turbulence Models Turbulence Models
¾ Reynolds Stress Model ¾ RANS Turbulence Models Descriptions
Model Description:
∂
∂t
(
ρ ui′u ′j + )∂
∂xk
( )
ρ U k ui′u ′j = Pij + Fij + DijT + Φ ij − ε ij Spalart-
Allmaras
A single transport equation model solving directly for a modified turbulent viscosity. Designed
specifically for aerospace applications involving wall-bounded flows on a fine, near-wall mesh.
Fluent’s implementation allows use of coarser meshes. •Option to include strain rate in k
production term improves predictions of vortical flows.

Dissipation The baseline two transport equation model solving for k and ε. This is the default k-ε model.
Stress-production Turbulent diffusion Standard k-ε Coefficients are empirically derived; valid for fully turbulent flows only. •Options to account for
Pressure strain viscous heating, buoyancy, and compressibility are shared with other k-ε models.
Rotation-production A variant of the standard k-ε model. Equations and coefficients are analytically derived.
RNG k-ε Significant changes in the ε equation improves the ability to model highly strained flows.
Modeling required for these terms •Additional options aid in predicting swirling and low Re flows.
A variant of the standard k-ε model. Its ‘realizability’ stems from changes that allow certain
Realizable k-ε mathematical constraints to be obeyed which ultimately improves the performance of this model.
– Attempts to address the deficiencies of the EVM. •Should not be used in conjunction with multiple rotating reference frames.
– RSM is the most ‘physically sound’ model: anisotropy, history effects and A two transport equation model solving for k and ω, the specific dissipation rate (ε/k) based on
Standard k-ω Wilcox (1998). This is the default k -ω model. Demonstrates superior performance for wall
transport of Reynolds stresses are directly accounted for. bounded and low Re flows. Shows potential for predicting transition. •Options account for
– RSM requires substantially more modeling for the governing equations (the transitional, free shear, and compressible flows.
A variant of the standard k-ω model. Combines the original Wilcox model (1988) for use near
pressure-strain is most critical and difficult one among them). SST k-ω walls and standard k-ε model away from walls using a blending function. Also limits turbulent
– But RSM is more costly and difficult to converge than the 2-equation models. viscosity to guarantee that τt ~ k. •The transition and shearing options borrowed from SKO. No
compressibility option.
– Most suitable for complex 3-D flows with strong streamline curvature, swirl Reynolds stresses are solved directly with transport equations avoiding isotropic viscosity
RSM assumption of other models. Use for highly swirling flows. •Quadratic pressure-strain option
and rotation.
improves performance for many basic shear flows.

MSM732 25 of 57 October 2005 MSM732 27 of 57 October 2005

Turbulence Models Turbulence Models

¾ Reynolds Stress Model ¾ RANS Turbulence Models Behavior and Usage
– RSM has high potential for accurately predicting complex flows.
Model Behavior and Usage
z Accounts for streamline curvature, swirl, rotation and high strain rates
Spalart- Economical for large meshes. Performs poorly for 3D flows, free shear flows, flows with strong
– Cyclone flows, swirling combustor flows separation. Suitable for mildly complex (quasi-2D) external/internal flows and b.l. flows under
– Rotating flow passages, secondary flows Allmaras pressure gradient (e.g. airfoils, wings, airplane fuselage, missiles, ship hulls).
Robust. Widely used despite the known limitations of the model. Performs poorly for complex
Standard k-ε flows involving severe ∇p, separation, strong stream line curvature. Suitable for initial iterations,
initial screening of alternative designs, and parametric studies.
Suitable for complex shear flows involving rapid strain, moderate swirl, vortices, and locally
RNG k-ε transitional flows (e.g., b.l. separation, massive separation and vortex-shedding behind bluff
bodies, stall in wide-angle diffusers, room ventilation)
Offers largely the same benefits and has similar applications as RNG. Unable to use with multiple
Realizable k-ε rotating reference frames. Possibly more accurate and easier to converge than RNG.

Superior performance for wall-bounded b.l., free shear, and low Re flows. Suitable for complex
Standard k-ω boundary layer flows under adverse pressure gradient and separation (external aerodynamics and
turbomachinery). Can be used for transitional flows (though tends to predict early transition).
Separation is typically predicted to be excessive and early.
Similar benefits as SKO. Dependency on wall distance makes this less suitable for free shear
SST k-ω flows.

RSM Physically the most sound RANS model. Avoids isotropic eddy viscosity assumption. More CPU
time and memory required. Tougher to converge due to close coupling of equations. Suitable for
complex 3D flows with strong streamline curvature, strong swirl/rotation (e.g. curved duct,
rotating flow passages, swirl combustors with very large inlet swirl, cyclones).

MSM732 26 of 57 October 2005 MSM732 28 of 57 October 2005

Turbulence Modelling 13 Turbulence Modelling 14

 Turbulence Models Detached Eddy Simulation (DES)
¾ Large Eddy Simulation ¾ Motivation
– Large eddies: – For high-Re wall bounded flows, LES becomes prohibitively
expensive to resolve the near-wall region
z Mainly responsible for transport of momentum, energy, and other scalars,
directly affecting the mean fields. – Using RANS in near-wall regions would significantly mitigate the
mesh resolution requirement
z Anisotropic, subjected to history effects, and flow-dependent, i.e., strongly
dependent on flow configuration, boundary conditions, and flow ¾ RANS/LES hybrid model based on the Spalart-Allmaras
parameters. turbulence model:
– Small eddies: ⎡ ∂ ⎤
Dν~ ⎛ ν~ ⎞
2
⎧⎪ ~ ∂ν ⎫⎪ + ...⎥
~
⎨(µ + ρν )
~ 1
z Tend to be more isotropic and less flow-dependent = Cb1S ν~ − Cw1 f w ⎜ ⎟ + ⎢ ⎬
Dt ⎝ d ⎠ σ ν~ ⎢⎣ ∂x j ⎪⎩ ∂x j ⎪⎭ ⎥⎦
z More likely to be easier to model than large eddies.
– LES directly computes (resolves) large eddies and models only small d = min (d w , C DES ∆ )
eddies (Subgrid-Scale Modeling).
– Large computational effort – One-equation SGS turbulence model
z Number of grid points, NLES ∝ Reu2τ ¾ DES is a practical alternative to LES for high-Reynolds
z Unsteady calculation number flows in external aerodynamic applications

MSM732 29 of 57 October 2005 MSM732 31 of 57 October 2005

Large Eddy Simulation (LES) Wall Function Approach

∂ui ∂ui u j 1 ∂p ∂ ⎛ ∂ui ⎞ ¾ Accurate near-wall modeling is important:
N-S equation + =− + ⎜ν ⎟
∂t ∂x j ρ ∂xi ∂x j ⎜ ∂x ⎟ – Successful prediction of frictional drag, pressure drop, separation, etc.,
⎝ j ⎠ Filter; ∆
depends on the fidelity of local wall shear predictions.
u(x, t ) = u (x, t ) + u′(x, t )
123 123 – Near-wall modeling is used to supply boundary conditions for turbulent
resolved scale subgrid scale
Sub-grid scale (SGS) flows.
turbulent stress
∂ui ∂ui u j 1 ∂p ∂ ⎛⎜ ∂ui ⎞⎟ ∂τ ij ¾ Most k-ε and RSM turbulence models are not valid in the near-
Filtered N-S + =− + ν − τ ij ≡ ui u j − ui u j
equation ∂t ∂x j ρ ∂xi ∂x j ⎜⎝ ∂x j ⎟⎠ ∂x j wall region:
– Special near-wall treatment is
¾ Spectrum of turbulent eddies in the Navier-Stokes required to provide proper BC’s
equations is filtered:
– The filter is a function of grid size
– Eddies smaller than the grid size are removed and modeled by a
sub-grid scale (SGS) model
– Larger eddies are directly solved numerically by the filtered
transient N-S equation y+≈ 200

MSM732 30 of 57 October 2005 MSM732 32 of 57 October 2005

Turbulence Modelling 15 Turbulence Modelling 16

 Wall Function Approach Wall Function Approach
¾ Standard Wall Functions ¾ Two-Layer Zonal Model
– Used for low-Re flows or
Mean Velocity
flows with complex near-wall
phenomena.
UP C1µ/ 4k 1P/ 2 ρ C 1µ/ 4k 1P/ 2 yP
U ∗ = 1 ln(Ey∗ ) ∗ – Zones distinguished by a
where U ≡ y∗ ≡
κ τw / ρ µ wall-distance-based turbulent
Reynolds number
Temperature ρ ky
Re y ≡ µ
⎧ Pr y ∗ ( y * < yT* )
⎪ – High-Re k-e models are used in the turbulent core region.
T∗ = ⎨ ⎡ 1
⎪ Prt ⎢ ln (Ey ) + P ⎥⎦
∗ ⎤
( y * > yT* ) thermal sublayer thickness
⎩ ⎣κ – Only k equation is solved in the viscosity-affected region.
– ε is computed from the correlation for length scale.
(Tw − TP ) ρ c pC 1µ/ 4k 1P/ 2
T* ≡ and P is a function of the fluid
q&′′ and turbulent Prandtl numbers.
– Zoning is dynamic and solution adaptive.

MSM732 33 of 57 October 2005 MSM732 35 of 57 October 2005

Wall Function Approach Wall Function Approach

¾ Non-equilibrium Wall Functions ¾ Computational Grid Guidelines
– Log-law is sensitized to pressure gradient for better prediction of Wall Function Two-Layer Zonal
adverse pressure gradient flows and separation. Approach Model Approach
– Relaxed local equilibrium assumptions for TKE in wall-neighboring
cells.
– Thermal law-of-wall unchanged
~
U C1µ/ 4 k1/ 2 1 ⎛ ρ C1µ/ 4 k1/ 2 y ⎞
= ln ⎜⎜ E ⎟⎟
τw / ρ κ ⎝ µ ⎠

dp ⎡ yv y − yv y ⎤
2
~ ⎛ y⎞ z First grid point in log-law region z First grid point at y+ ≈ 1.
where U = U − 1 ⎢ ln ⎜ ⎟ + + v⎥
2 dx ⎣ ρκ k
∗ 1/ 2
⎝ yv ⎠ ρκ k
∗ 1/ 2
µ⎦ z At least ten grid points within
z At least ten points in the BL. buffer & sublayers.

z Better to use stretched quad/hex z Better to use stretched quad/hex

cells for economy. cells for economy.

MSM732 34 of 57 October 2005 MSM732 36 of 57 October 2005

Turbulence Modelling 17 Turbulence Modelling 18

 Wall Function Approach Examples
¾ Estimating Placement of First Grid Point ¾ 2-D Backstep – cont.:
– Estimate the skin friction coefficient based on correlations either Predicted reattachment lengths
approximate or empirical: Std. k-ε Real. k-ε SST k-ω Wilcox k-ω Measured
− 0 .2
z Flat Plate- c f / 2 ≈ 0.0359 Re L xr/H 5.8 6.6 6.6 7.3 6.4

−0.2
z Pipe Flow- c f / 2 ≈ 0.039 Re D

– Compute the friction velocity: uτ ≡ τ w / ρ = U e c f / 2

– Back out required distance from wall:

z Wall functions • Two-layer model
y1 = 50ν/uτ y1 = ν/ uτ

– Use post-processing to confirm near-wall mesh resolution

MSM732 37 of 57 October 2005 MSM732 39 of 57 October 2005

Examples Examples
¾ 2-D Backstep: ¾ 2-D Backstep – cont.:
– Experiments conducted at NASA Ames (Driver and Seegmiller, 1985); – Five different near-wall mesh resolutions with the standard wall
ReH= 3.74 x 104, α = 0 deg. functions (SWF) and the enhanced wall treatment (EWT).
– The flow features re-circulation, reattachment, and re-developing BL.
– Computed using SKE, RNG, RKE, and k-ω models on a fine mesh.

MSM732 38 of 57 October 2005 MSM732 40 of 57 October 2005

Turbulence Modelling 19 Turbulence Modelling 20

 Examples Examples
¾ Impinging Flow Over a Blunt Plate: ¾ Impinging Flow Over a Blunt Plate –cont.:

Predicted separation bubble

Standard k-ε

Realizable k-ε

Experimentally observed
reattachment point is at x/d = 4.7

MSM732 41 of 57 October 2005 MSM732 43 of 57 October 2005

Examples Examples
¾ Impinging Flow Over a Blunt Plate –cont.: ¾ 2D U-Bend:
– The standard k-ε model gives spuriously large turbulent kinetic energy – Comparison with experimental data of Monson et al. (1990)
on the front face, underpredicting the size of the recirculation.

Standard k-ε model Reynolds-Stress model

(exact)

Contours of TKE production

MSM732 42 of 57 October 2005 MSM732 44 of 57 October 2005

Turbulence Modelling 21 Turbulence Modelling 22

 Examples Examples
¾ 2D U-Bend – cont.: ¾ 2D U-Bend – cont.:
– Streamwise Velocity Comparisons – Stream Function Contours

Spalart-Allmaras Standard k-ε

r*
r*
θ = 180
θ=0

U/Uref r* U/Uref

θ = 90 RNG k-ε RSM

U/Uref

MSM732 45 of 57 October 2005 MSM732 47 of 57 October 2005

Examples Examples
¾ 2D U-Bend – cont.: ¾ Flow in a Rotating Channel:
– Pressure Coefficients – Represents flows through
rotating internal passages Flow configuration:
(e.g. turbomachinery
applications) Johnston et al. (1972)
– Rotation affects mean axial
momentum equation through
turbulent stresses.
– Rotation makes mean axial
velocity asymmetrical.
ReH = 11,500
– Computations are carried out
using SKE, RNG, RKE and Ro = 0.21
RSM models are with the
standard wall functions.

MSM732 46 of 57 October 2005 MSM732 48 of 57 October 2005

Turbulence Modelling 23 Turbulence Modelling 24

 Examples Examples
¾ Flow in a Rotating Channel –cont.: ¾ Cyclone – cont.:
– Tangential velocity profile at 0.41 m below the vortex finder
Predicted axial velocity profiles (ReH = 11.500, Ro = 0.21)

Symmetric
profiles

MSM732 49 of 57 October 2005 MSM732 51 of 57 October 2005

Examples Examples
¾ Cyclone:
0.1 m • LES of the Flow Past a Square Cylinder (ReH = 22,000)
– 40,000 cell hexahedral mesh
– High-order upwind scheme was used
CD St
– Computed using SKE, RNG, RKE
0.12 m Dynamic Smag. 2.28 0.130
and RSM models with the standard
Dynamic TKE 2.22 0.134
wall functions
Exp.(Lyn et al., 1992) 2.1 – 2.2 0.130
– Represents highly swirling flows Uin = 20 m/s
(Wmax = 1.8 Uin)
0.97 m

Time-averaged streamwise
velocity along the wake centerline
0.2 m Iso-contours of instantaneous vorticity
magnitude
CL spectrum
MSM732 50 of 57 October 2005 MSM732 52 of 57 October 2005

Turbulence Modelling 25 Turbulence Modelling 26

 Examples Turbulence Models Guidelines
¾ Choices to be Made:
• LES of the Flow Past a Square Cylinder (ReH = 22,000)

Flow Computational
Physics Resources

Turbulence Model
& Computational
Near-Wall Treatment Grid

Turnaround
Streamwise mean velocity Streamwise normal stress Accuracy Time
along the wake centerline along the wake centerline Required Constraints

MSM732 53 of 57 October 2005 MSM732 55 of 57 October 2005

Turbulence Models Guidelines Summary

¾ How Complex is the Flow? ¾ Successful turbulence modeling requires engineering
– Extra strain rates judgement of:
z Streamline curvature – Flow physics
z Lateral divergence – Computer resources available
z Acceleration or deceleration – Project requirements
z Swirl z Accuracy
z Recirculation (or separation) z Turnaround time
z Secondary flow – Turbulence models & near-wall treatments that are available in
– 3D perturbations commercial solver
– Transpiration (blowing/suction)
– Free-stream turbulence
– Interacting shear layers

MSM732 54 of 57 October 2005 MSM732 56 of 57 October 2005

Turbulence Modelling 27 Turbulence Modelling 28

 Where to Get More Information
¾ On the web:
– www.ionasoftware.com/gross/fluids/turbulence.html
– www.cfd-online.com
¾ Books:
– David C. Wilcox, “Turbulence Modelling for CFD”, 2nd Edition, DCW
Industries, 1998
z A comprehensive introduction to turbulence modelling for CFD. Covers
everything from simple algebraic models to advanced RSM and LES
methods. The main focus is on two-equation eddy-viscosity models.
– P. Sagaut, “Large Eddy Simulation for Incompressible Flows - An
Introduction”, Springer Verlag, 2001
z Presents a comprehensive account of the available knowledge in the field
of Large Eddy Simulation, and also the first unified view of the various
existing approaches.

MSM732 57 of 57 October 2005

Turbulence Modelling 29