When Steady State Turbulence Models are

not good Enough - When and how to use

Unsteady Turbulence Models

Duhwan Lee
TAESUNG Software and Engineering, INC

2014 ANSYS, Inc.

May 15, 2014

ANSYS Confidential

Turbulence Models

RANS Models

1-eq.,
2-eq,
RSM,
EARSM
Transition

LES Models

Smagorinsky (reg.,

dyn)
WALE
k-equation (dyn.)
WMLES

Hybrid Models

DES/DDES/IDDES

Zonal/Embedded Models

2014 ANSYS, Inc.

May 15, 2014

ANSYS Confidential

SA-based
SST-based
k-e based
SAS
All w-equation
based models

Motivation for Scale-Resolving Simulations

(SRS)
Accuracy Improvements over RANS
Reduce uncertainty in CFD simulations relative to RANS
Flows with large separation zones (stalled airfoils/wings
, flow past buildings, flows with swirl instabilities, etc.)

Additional information required

Acoustics
Information on acoustic spectrum not reliable from
RANS
Vortex cavitation
low pressure inside vortex causes cavitation
resolution of vortex required
Fluid-Structure Interaction (FSI)
unsteady forces determine frequency response of solid.

2014 ANSYS, Inc.

May 15, 2014

ANSYS Confidential

Why not Unsteady RANS (URANS)

URANS

URANS means the application of existing

conventional RANS models in unsteady mode (k-e
SST, RSM, )
RANS models can provide unsteady results for flows
with strong flow instabilities
However, turbulence is still not resolved
- URANS typically provides unphysical large
scales no turbulence spectrum
- In acoustics that translates into a single mode
tonal noise
- In mixing process, results can be improved
relative to RANS but not reliably
However this is not a principal problem of URANS
and can be improved
4

2014 ANSYS, Inc.

May 15, 2014

ANSYS Confidential

Real Turbulence

Scale-Adaptive Simulation (SAS)

3/ 2
k U j k
t k
3/ 4 k

Pk c

t
x j
L x j k x j

U j
1

2

1Pk 2 2 L2 t U '' 3 k t

t
x j
k

y y

With:

URANS

kL

t c1/ 4

U i U i
2U i 2U i
U'
U'
; U ''
; LvK
x j x j
x j x j xk xk
U ''

v. Karman length-scale as natural length-scale:

U / y
L ~ 2
LvK
2
U / y
5

2014 ANSYS, Inc.

May 15, 2014

ANSYS Confidential

SAS

SAS Modell - 2D Periodic Hill

Scale-Adaptation based on t

4 higher t

t = 0.045 h/UB

2 higher t
6

2014 ANSYS, Inc.

May 15, 2014

ANSYS Confidential

SAS Modell - 2D Periodic Hill

Time averaged velocity profiles U

2014 ANSYS, Inc.

May 15, 2014

ANSYS Confidential

Detached Eddy Simulation (DES)

Hybrid Model:
RANS equations in boundary layer.
LES detached regions.
Switch of model:
Based on ratio of turbulent length-scale to grid size.
Different numerical treatment in RANS and LES regions.
RANS

Lt c

LES

Lt c

2014 ANSYS, Inc.

May 15, 2014

Overcomes threshold limit of LES

Explicit grid sensitivity in RANS
region
Open question concerning
transition region between RANS
and LES

ANSYS Confidential

Detached Eddy Simulation (DES) for SST

Strelets (2000)

k-equation RANS

( k ) ( U j k )
k 3/2

Pk

t
x j
Lt x j

t k
( )

t k
( )

x j

max( x, y, z)

k-equation DES

( k ) ( U j k )
k 3/2

Pk

t
x j
min Lt ; CDES x j

*w

k-equation LES

( k ) ( U j k )
k 3/2

Pk

t
x j
CDES x j

Lt

2014 ANSYS, Inc.

May 15, 2014

ANSYS Confidential

t k
( )

> RANS,
LES

Classical Derivation of LES

resolved

Average Navier Stokes

equations over grid cell size
Just like in RANS, averaging
leads to additional stress
terms in NS equations
Resolve for Lt> and model
for Lt<
1
ui
ui dVol

Vol Vol

Cell size

ui ui u j
p lam LES

ij ij
t
x j
xi x j

U i U j

ij ui u j ui u j t

x
xi
j

e.g. Smagorinsky model

ui u j
ui ui u j
p

t

t
x j
xi x j

x
i
j

10

2014 ANSYS, Inc.

May 15, 2014

modelled

ANSYS Confidential

2
t c S

Same as RANS!

Vol

SGS Models: Summary

Sub-grid stress : turbulent viscosity
Smagorinsky model (Smagorinsky, 1963)

Need ad-hoc near wall damping

1
ij kk ij 2 vt Sij
3
vt Cs S
2

t Cs

WALE model (Nicoud & Ducros 1999)

Correct asymptotic near wall behaviour

Sijd Sijd

Sij Sij

5/ 2

3/ 2

Sijd Sijd

5/ 4

vt CD S
2

Dynamic model (Germano et al., 1991)

Local adaptation of the Smagorinsky constant

1/ 2
vt Ck k sgs

Dynamic sub-grid kinetic energy transport model

(Kim & Menon 2001) (Fluent only)
3/ 2
k sgs u j k sgs
k sgs
ui
sgs k sgs

ij
Ce

Robust constant calculation procedure

t
x j
x j

x j k x j
Physical limitation of backscatter
11

2014 ANSYS, Inc.

May 15, 2014

ANSYS Confidential

Periodic Channel Flow

Near the wall turbulent streaks appear
They have to be resolved to obtain
correct wall shear stress and heat
transfer
Scaling of these structures is ~ Re2
Wall flows can only be computed with
LES for small domains and low Re
numbers
Nx

Ny

Nz

Lx

Ly

Lz

x+

y1+

z+

Re_T

12

180

Mesh 1

73

75

73

2h

15.7

0.254

7.85

590

Mesh 2

120

100

110

2h

30

15

2014 ANSYS, Inc.

May 15, 2014

ANSYS Confidential

Periodic Channel Flow

2,00E+01

U+ DNS

1,80E+01

1,00E+01

Re=180

U+ DSM

1,60E+01

U+ WALE

uu+ DNS

Re=180

9,00E+00

uu+ DSM
uu+ WALE

8,00E+00

1,40E+01
7,00E+00

1,20E+01
6,00E+00

1,00E+01

5,00E+00

8,00E+00

4,00E+00

6,00E+00

3,00E+00

4,00E+00

2,00E+00

2,00E+00

1,00E+00

0,00E+00
1,00E-01

3,00E+01
2,50E+01

1,00E+00

1,00E+01

1,00E+02

0,00E+00
1,00E+03
1,00E+0 2,10E+0 4,10E+0 6,10E+0 8,10E+0 1,01E+0 1,21E+0 1,41E+0 1,61E+0
0
1
1
1
1
2
2
2
2

9,00E+00
uu+ DNS

U+ DNS
8,00E+00

U+ DSM

7,00E+00

Re=590

2,00E+01

uu+ DSM

Re=590

6,00E+00
5,00E+00

1,50E+01

4,00E+00
3,00E+00

1,00E+01

2,00E+00

5,00E+00
0,00E+00
1,00E-01

13

1,00E+00

1,00E+00

2014 ANSYS, Inc.

1,00E+01

May 15, 2014

1,00E+02

0,00E+00
1,00E+00

1,00E+03

ANSYS Confidential

1,01E+02

2,01E+02

3,01E+02

4,01E+02

5,01E+02

NACA 0012 Airfoil Noise

NACA 0012: Rechord = 1.1106
Velocity inlet
Airfoil rotated by 7.3 degree
71.3 m/s

Pressure outlet

14

2014 ANSYS, Inc.

May 15, 2014

ANSYS Confidential

WB Unstructured Hex Mesh

Span: 0.05 chord; 80 nodes
In total ~ 11.4 Mio nodes
WALE LES model
Periodicity in spanwise
direction

Trailing edge

Leading edge

15

2014 ANSYS, Inc.

May 15, 2014

ANSYS Confidential

Q-criterion
Q-criterion (W2-S2): Q=109 , colored by z-velocity:
Leading edge

Trailing edge

Due to high Re number and moderate a, it looks still ok near trailing edge even
though span=0.05c
16

2014 ANSYS, Inc.

May 15, 2014

ANSYS Confidential

5%chord, 11M cells, t=1.5 s

Pressure and skin friction coefficients

Even on this grid cf is too low -> WMLES (see later)

0.020

Cf comparison: 2-D SST transition vs. 3-D ELES

2-D RANS
0.015

3-D ELES pressure side

3-D ELES suction side

Cf
0.010

0.005

0.000
0.0
17

2014 ANSYS, Inc.

May 15, 2014

ANSYS Confidential

0.2

0.4

x/chord

0.6

0.8

1.0

WMLES: Near Wall Scaling

Turbulent length scale is independent of Re
number
However thickness of viscous sub layer
decreases with increasing Re number
Turbulent structures inside sublayer are
damped out
Smaller turbulence structures near the wall
get exposed as Re increases
WMLES: models small near wall structures
with RANS and only resolve larger structures
less dependent on Re number
Some Re number dependence for boundary
layer remains as boundary layer thickness
decreases with Re number
RANS
18

2014 ANSYS, Inc.

May 15, 2014

ANSYS Confidential

Lt y
Low Re
y

High Re
y

High Re WMLES
y

Viscous sublayer

WMLES-Concept Algebraic Model

RANS/LES blending:
RANS

LES

t f d min dW , CSMAG S
dw
S
CSMAG

fd

Wall distance
Shear Strain Rate
Smagorinsky constant CSMAG=0.2
Modified grid spacing
damping function

The model has been modified and optimized for usage in ANSYSFluent and ANSYS-CFX
A hybrid RANS-LES approach with delayed-DES and wall-modelled LES capabilities
Mikhail L. Shur , Philippe R. Spalart , Mikhail K. Strelets , Andrey K. Travin , Int. J of Heat and Fluid Flow 29, 2008
19

2014 ANSYS, Inc.

May 15, 2014

ANSYS Confidential

WMLES Channel Flow Tests

Re
395
760
1100
2400
18000

Cells
Number
384 000
480 000
480 000
528 000
624 000

LES Cells
Number
384 000
1 500 000
4 000 000
19 000 000
1 294 676 760

Nodes
Number
818161
8110161
8110161
8111161
8113161

X+

Z+

40.0
76.9
111.4
243.0
1822.7

20.0
38.5
55.7
121.5
911.4

Very large savings between

WMLES and wall-resolved LES
Alternative is LES with wall
functions however x+ and z+
are a function of y+

20

2014 ANSYS, Inc.

May 15, 2014

ANSYS Confidential

WMLES Channel Flow at Different Re

Numbers
Solutions at very
different Re numbers
look essentially
identical

Re=395

Re=18000

Differences can only

be seen near the
wall.
Visible is higher
Eddy-Viscosity for
higher Re number
close to wall
RANS Eddy Viscosity
21

2014 ANSYS, Inc.

May 15, 2014

ANSYS Confidential

WMLES Flat Plate Grid

Geometry and Grid
Inlet boundary layer thickness
L x 0.4 L x 0.1 L (Streamwise,
Normal, Spanwise)
Approximately 3 spanwise
(0=0.032)
Grid ~ 1Million cells (see table)
Y+~0.05 (to allow for higher Re
numbers)
Expansion factor 1.15
For each boundary layer
thickness one needs
~10x40x20 cells

Re
1000
10000

22

2014 ANSYS, Inc.

May 15, 2014

Cells Num
ber
1 085 000
1 085 000

ANSYS Confidential

Nodes Numb X+
er
2517163
68
2517163
520

Y+

Z+

0.05 300 34
0.4 2300 307

WMLES Boundary Layer

Boundary layer simulation:
WMLES
Inlet: synthetic turbulence

Vortex Method
2 different Reynolds numbers

Re=1000

Re=1000

Re=10000

23

2014 ANSYS, Inc.

May 15, 2014

Re=10000

ANSYS Confidential

NACA0012 at 7.3 AoA, Re=1.1106

Skin friction coefficient on the suction side

24

2014 ANSYS, Inc.

May 15, 2014

ANSYS Confidential

Inflow Boundary Conditions

It is often important to specify a realistic turbulent inflow
velocity for accurate prediction of the (downstream) flow

ui x, t U i x

time averaged

uix, t

coherent random

Two methods are available for time-dependent inflow

boundary condition for LES:
Spectral Synthesizer (Fluent)
Vortex Method (Fluent)
Harmonic Turbulence Generator (CFX)
25

2014 ANSYS, Inc.

May 15, 2014

ANSYS Confidential

Vortex Method
In essence, vorticity-transport is
modeled by distributing and tracking
many point-vortices on a plane
(Sergent, Bertoglio, Laurence, 2000)

w x, t

t x x , t
k

k 1

Velocity field computed using the

Biot-Savarts law
1
ux, t
2
26

2014 ANSYS, Inc.

May 15, 2014

x x w xe z
x x

dx

ANSYS Confidential

Vortex Method
3-D Wavy Channel (ReH = 10,600)
Mathey and Cokljat (2005)

Exp
Flow

xr
Periodic

Computational Domain

LES predictions of the

reattachment point

Random
number

xr
Exp.

4.7 h
Vortex
method

Periodic 5. H
VM

5.2 h

Random 7.7 h
27

2014 ANSYS, Inc.

May 15, 2014

ANSYS Confidential

Harmonic Turbulence Generator (HTG)

Input from
turbulence model
in form of Lt and w.
Produces Karman
spectrum
Co-operation with
Profs. Strelets and
Shur
(St. Petersburg)

28

2014 ANSYS, Inc.

May 15, 2014

ui 2 u n cos n, j x j w t n n ,i
n 1

k HTG k RANS k LES

slope k 5/3

u
n 1

ANSYS Confidential

Log scale

2
n

kRANS kLES ,

1
k
, w ~ w , Lt
Lt
c w

SRS - Model Summary

LES and WMLES assume that all wall boundary layers are
covered in unsteady Scale-Resolving mode (turbulence
structures resolved)
- LES costs excessive for moderate to high RE number wall bounded flows
- Even for WMLES, this is still very expensive, as one needs ~5000 cells to cover one
-

boundary layer volume ( x x )

On the other hand, LES of free shear flows is much less demanding

Hybrid RANS-LES models are therefore being developed which compute all
wall boundary layers in RANS mode and switch to LES only for the free shear
flow regions

- Simulations can be performed on grids much closer to the RANS grid

- Still - expensive time integration is needed

29

2014 ANSYS, Inc.

May 15, 2014

ANSYS Confidential

Embedded/Zonal Large Eddy Simulation

(ELES, ZFLES)
Suitable if zone with high
accuracy demands is
embedded into larger
domain which can be
covered properly by
RANS models

LES zone

Limited zone can then be

covered by LES or WallModelled WMLES model
LES zone needs to be
coupled to RANS zone
through interfaces

LES zone requires

suitable (WM)LES
resolution in time and
space
30

2014 ANSYS, Inc.

May 15, 2014

ANSYS Confidential

Rest: RANS zone

Coupled Zonal Modelling

In ELES/ZFLES e.g. MODEL2 can be LES turbulence model embedded
in a RANS or SAS model (MODEL1), or vice versa
Shadow face 2 acting as B.C. for
model2 in zone3

Shadow face 1 acting as B.C. for

model1 in zone2
wall
ZONE 1
RANS Model

ZONE 2

ZONE 3

LES Model

RANS Model

wall

There is STRONG need for model

interaction at this interface since
models are different in Zone 1
2 and Zone 2 3
31

2014 ANSYS, Inc.

May 15, 2014

ANSYS Confidential

Embedded LES and Zonal Forced LES

In many flows an area where
(WM)LES is required is
embedded in a larger RANS
region
In such cases, a zonal method is
advantageous

RANS and LES regions are

separately defined and use
different models
Synthetic turbulence is
generated at the interface to
convert RANS to LES turbulence
32

2014 ANSYS, Inc.

May 15, 2014

ANSYS Confidential

Flow Types: Globally Unstable Flows

Types of highly unstable flows:
Flows with strong swirl instabilities
Bluff body flows, jet in crossflow
Massively separated flows

Physics
Resolved turbulence is generated quickly by flow instability
Resolved turbulence is not dependent on details of turbulence in
upstream RANS region (the RANS model can determine the
separation point but from there new turbulence is generated)

Models
SAS: Most easy to use as it converts quickly into LES mode, and
automatically covers the boundary layers in RANS. Has RANS
fallback solution in regions not resolved by LES standards (t, x)
DDES: Similar to SAS, but requires LES resolution for all free shear
flows (t, x) (jets etc.)
ELES: Not really required as RANS model can cover boundary
layers. Often difficult to place interfaces for synthetic turbulence.
Green-recommended, Red=not recommended
33

2014 ANSYS, Inc.

May 15, 2014

ANSYS Confidential

Flow Types: Locally Unstable Flows

Types of moderately unstable flows:

BL Turbulence

Jet flows, Mixing layers

Physics

ML Turbulence
z

Flow instability is weak RANS/SAS models stay steady state.

Can typically be covered with reasonable accuracy by RANS
models.
DDES and LES models go unsteady due to the low eddy-viscosity
provided by the models. Only works on fine LES quality grids and
time steps. Otherwise undefined behavior.

Models
SAS: Stays in RANS mode. Covers upstream boundary layers in
RANS mode. Can be triggered into SRS mode by RANS-LES
interface.
DDES: Can be triggered to go into LES mode by fine grid and small
t. Careful grid generation required. Covers upstream boundary
layers in RANS mode.
ELES: LES mode on fine grid and small t. Careful grid generation
required. Upstream boundary layer (pipe flow) in expensive LES
mode. Alternative ELES with synthetic turbulence RANS-LES
interface.
34

2014 ANSYS, Inc.

May 15, 2014

ANSYS Confidential
Green-recommended,
Red=not recommended

Flow Types: Stable Flows

Types of marginally unstable flows:
Pipe flows, channel flows, boundary layers, ..

Physics
Transition process is slow and takes several boundary layer
thicknesses.
When switching from upstream RANS to SRS model, RANS-LES
interface with synthetic turbulence generation required.
RANS-LES interface needs to be placed in non-critical (equilibrium)
flow portion. Downstream of interface, full LES resolution
required.

Models
SAS: Stays in RANS mode. Typically good solution with RANS. Can
be triggered into SRS mode by RANS-LES interface.
DDES: Can be triggered to go into LES mode by fine grid and small
t. Careful grid generation required. Covers upstream boundary
layers in RANS mode.
ELES: LES mode on fine grid and small t. Careful grid generation
required. Upstream boundary layer (pipe flow) in RANS mode.
Synthetic turbulence RANS-LES interface.
35

2014 ANSYS, Inc.

May 15, 2014

ANSYS Confidential
Green-recommended,
Red=not recommended

Globally Unstable Flow Jets in Crossflow

Problem:
Hot air leaves engine nacelle and
heats wall
Heat shielding required
Experiments too expensive
RANS not accurate enough
Simulations ANSYS-Fluent

Confidential Toulouse
2014 ANSYS, Inc.
May 15, 2014
Courtesy:
Benjamin
Duda, ANSYS
Airbus

36

Generic Jet in Cross Flow Configuration

Infrared Thermography

Particle Image Velocimetry

Laser Doppler Anemometry

Hot and Cold Wire Measurements

Confidential Toulouse
2014 ANSYS, Inc.
May 15, 2014
Courtesy:
Benjamin
Duda, ANSYS
Airbus

37

Hexahedral Mesh
12,900,000 Elements
Min angle = 28.1
Max AR = 3,500
Max VC = 10

Confidential Toulouse
2014 ANSYS, Inc.
May 15, 2014
Courtesy:
Benjamin
Duda, ANSYS
Airbus

38

Scale Resolvability of Turbulence Models

Q-criterion:

SAS

DDES

ELES

URANS

Confidential Toulouse
2014 ANSYS, Inc.
May 15, 2014
Courtesy:
Benjamin
Duda, ANSYS
Airbus

39

Hybrid Tetrahedral Mesh

21,000,000 Elements
Min angle = 20.0
Max AR = 7,600
Max VC = 8

20 inflation layers

Confidential Toulouse
2014 ANSYS, Inc.
May 15, 2014
Courtesy:
Benjamin
Duda, ANSYS
Airbus

40

Hybrid Cartesian Mesh

13,100,000 Elements
Min angle = 6.0
30 Elements < 15
Max AR = 6,000
Max VC = 16

20 inflation layers

Confidential Toulouse
2014 ANSYS, Inc.
May 15, 2014
Courtesy:
Benjamin
Duda, ANSYS
Airbus

41

Time Averaged Temperature Distribution

Turbulence models (on hex)

Lateral spreading of temperature in good a

greement with all three SRS

Poor URANS prediction due to unphysical
damping of lateral jet wake
II movement

Meshing strategies (with SAS)

Generally very good agreement

Small underestimation for hex mesh at cen
ter

T T

T j T

Confidential Toulouse
2014 ANSYS, Inc.
May 15, 2014
Courtesy:
Benjamin
Duda, ANSYS
Airbus

42

Mean Thermal Efficiency on Wing Surface

URANS

SAS

EXP

Confidential Toulouse
2014 ANSYS, Inc.
May 15, 2014
Courtesy:
Benjamin
Duda, ANSYS
Airbus

43

Geometry of the Cavity

D = 4 in

Pressure far field

L = 5 D, W = D
Lx x Ly x Lz = 18 D x 17 D x 9 D

Outlet

M = 0.85
Massflow
Inlet

P = 62100 Pa
T = 266.53 K
Re = 13.47 E 6

Symmetry

44

2014 ANSYS, Inc.

May 15, 2014

ANSYS Confidential

Mesh: 5.8 e 6 Cv double O-grid

Inlet
Side

Bottom
45

2014 ANSYS, Inc.

May 15, 2014

ANSYS Confidential

Turbulent structure by q-criterion. SAS Model

Eddy viscosity ratio @ q = -500000 (q = 1/2 (S:S W:W))

46

2014 ANSYS, Inc.

May 15, 2014

ANSYS Confidential

Wave propagation by Fluctuating Density,

SAS Model
Eddy viscosity ratio @ q = -500000 (q = 1/2 (S:S W:W))

47

2014 ANSYS, Inc.

May 15, 2014

ANSYS Confidential

SAS: k20 k25

k29
k20

48

2014 ANSYS, Inc.

May 15, 2014

ANSYS Confidential

SAS: k26 k29

k29
k20

49

2014 ANSYS, Inc.

May 15, 2014

ANSYS Confidential

Stable Flow Example

Flow configuration:

Simulation: baseline (no flow control)

Testcase of EU Project ATAAC

http://cfd.mace.manchester.ac.uk/twiki/bin/view/ATAAC/WebHome

50

2014 ANSYS, Inc.

May 15, 2014

ANSYS Confidential

Flow over a wall mounted hump, Geometry

and Grid
Geometry:
Spanwise extent:
3.16 H (bump height)
5.6 dinterface (d boundary layer thick
ness).

Grid:
RANS grid with only 5 cells in span
wise direction
LES grid: 200x100x100 (2 million)
Grid resolution per inlet boundary
layer (Dx/d=10, Dz/d~20, NY~40).
EU project ATAAC
51

2014 ANSYS, Inc.

May 15, 2014

ANSYS Confidential

Flow over a wall mounted hump

Q criterion:

VM_WMLES_CD

52

2014 ANSYS, Inc.

May 15, 2014

ANSYS Confidential

Flow over a wall mounted hump Wall

Shear Stress and Wall Pressure
The Re number at the
RANS-LES interface is
ReQ=7000
If the simulation in the
LES region is carried out
with a standard LES
model (WALE) the
solution is lost
immediately after the
interface
The WMLES formulation
is able to carry the
solution smoothly across
and provide a good
agreement with the data
for two different time
steps (CFL~0.5 and
CFL~0.12)

53

2014 ANSYS, Inc.

May 15, 2014

RANS-LES Interface

ELES with WMLES ANSYS-Fluent

ANSYS Confidential

Overall Summary
SRS is making its way into industrial CFD

Different types of model recommended for different types of applications

Currently favored methods within ANSYS software:
SAS globally unstable flows
DDES globally and locally unstable flows
ELES/WMLES stable flows
Ongoing work:
Synthetic turbulence is good but far from perfect
WMLES requires special grid resolution not as generic as other model
s (inherited from LES)
User guidelines
More Zonal formulations
54

2014 ANSYS, Inc.

May 15, 2014

ANSYS Confidential