You are on page 1of 50

19.

0 Release

Lecture 07: Large Eddy Simulation (LES)

Turbulence Modeling Using ANSYS CFD

1 © 2018 ANSYS, Inc.


Large Eddy Simulation - LES
Temporal Mixing Layer Turbulence

• In LES, not all scales are modeled but the


large turbulence scales are resolved by the
numerical method
• The smaller scales are not resolved due to
mesh resolution limits
• For the small scales, simple models are
introduced as it is assumed that these
scales have little contribution
• It is important to note that LES is much
more expensive than RANS and therefore
not a standard engneering tool for most
applications
• Costs are very high for wall-bounded flows

2 © 2018 ANSYS, Inc.


Motivation for Large Eddy Simulation (LES)

• Accuracy Improvements over 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

Courtesey Munsch Pumpen

3 © 2018 ANSYS, Inc.


Classical Derivation of LES
resolved modelled
• Flow is only resolved above given grid size
• Average Navier Stokes equations over grid cell
size (Vol) to “filter” out small scales
• Just like in RANS, averaging leads to additional
stress terms in NS equations
• Resolve for 𝑬𝑺 > ∆ and model for 𝑬𝑺 < ∆
(𝑬𝑺 – turbulence eddy size) Cell size D= 𝑉𝑜𝑙
3

• Sub-grid model is often an Eddy-Viscosity model


1 𝜕𝜌𝑢ො 𝑖 𝜕𝜌𝑢ො 𝑖 𝑢ො𝑗 𝜕 𝑝Ƹ 𝜕 𝑙𝑎𝑚 𝐿𝐸𝑆
𝑢ො 𝑖 = න 𝑢𝑖 𝑑𝑉𝑜𝑙 + =− + 𝜏Ƹ 𝑖𝑗 + 𝜏Ƹ 𝑖𝑗
𝑉𝑜𝑙 𝜕𝑡 𝜕𝑥𝑗 𝜕𝑥𝑖 𝜕𝑥𝑗
𝑉𝑜𝑙
෡𝑗
෡𝑖 𝜕 𝑈
𝜕𝑈
𝐿𝐸𝑆
𝜏Ƹ 𝑖𝑗 = 𝜌𝑢ො 𝑖 𝑢ො𝑗 − 𝜌𝑢ෟ
𝑢 𝑢𝑗 = 𝜇𝑡 + Eddy-Viscosity model
𝜕𝑥𝑗 𝜕𝑥𝑖

𝜕𝜌𝑢ො 𝑖 𝜕𝜌𝑢ො 𝑖 𝑢ො𝑗 𝜕 𝑝Ƹ 𝜕 𝜕𝑢ො 𝑖 𝜕𝑢ො𝑗 Formally same as RANS!


+ =− + 𝜇 + 𝜇𝑡 +
𝜕𝑡 𝜕𝑥𝑗 𝜕𝑥𝑖 𝜕𝑥𝑗 𝜕𝑥𝑗 𝜕𝑥𝑖

4 © 2018 ANSYS, Inc.


Large Eddy Simulation (LES) ∞
𝒌 = න 𝑬𝒅𝜿
𝟎
Log E
• Role of LES: Generation of largest eddies
− Turbulence kinetic energy spectrum cannot be
resolved down to the dissipative scales Energy transfer
(Kolmogorov scales)
− Energy has to be dissipated from the grid limit
spectrum at grid limit
− LES Eddy Viscosity provides required damping Viscous Dissipation

− LES does not model the small scales – it just LES Dissipation
dissipates them
− Everything of importance has to be resolved! 𝑑𝑘 = 𝐸𝑑𝜅 Log k

𝜕𝑢𝑖 𝜕𝑢𝑖 𝜕𝑢ො 𝑖 𝜕𝑢ො 𝑖


𝜀𝐿𝐸𝑆 = 𝜀𝐷𝑁𝑆 𝜀𝐷𝑁𝑆 =𝜈 𝜀𝐿𝐸𝑆 = 𝜈𝑡𝐿𝐸𝑆
𝜕𝑥𝑗 𝜕𝑥𝑗 𝜕𝑥𝑗 𝜕𝑥𝑗

LES – Smagorinsky Model 𝜈𝑡𝐿𝐸𝑆 = 𝑐∆ 2 𝑆

5 © 2018 ANSYS, Inc.


Test Case: Decaying Isotropic Turbulence (DIT)

• Standard LES testcase


• Artificial turbulence is generated as
initial condition (see picture)
• Turbulence is then allowed to
decay – integrating the Navier-
Stokes equations (LES form)
• Compare turbulent spectrum at
different times against experiment
of Comte-Bellot

6 © 2018 ANSYS, Inc.


DIT Spectrum t=0.87

No LES Model With LES Model


1.00E-01
Exp T1

WALE C=0.5

DSM

SMAG C=0.18
1.00E-02
E(k)

E(k)
1.00E-03

1.00E-04
1.00E+00 1.00E+01 1.00E+02

k k

• Without LES model, the energy is accumulated at small scales (large wave number k)
• With LES models, energy is dissipated at grid resolution limit

7 © 2018 ANSYS, Inc.


SGS Models: Concept

• For dimensional reasons the eddy-viscosity scales ෡𝑗


෡𝑖 𝜕𝑈
𝐿𝐸𝑆 𝜕𝑈
like 𝜇𝑡 ~ 𝜌𝑉𝑡 𝐿𝑡 𝜏Ƹ 𝑖𝑗 = 𝜌𝑢ො 𝑖 𝑢ො𝑗 − 𝜌𝑢ෟ
𝑢 𝑢𝑗 = 𝜇𝑡 +
𝜕𝑥𝑗 𝜕𝑥𝑖
• An LES model only ‚models‘ turbulence scales below
the grid limit ∆. The relevant length scale in the 𝜇𝑡 ~ 𝜌𝑉𝑡 𝐿𝑡
model is therefore 𝐿𝑡 = ∆.
𝐿𝑡 ~∆
• In case of an algebraic model, the only velocity scale
is 𝑉𝑡 ~∆ 𝜕𝑈෡ൗ𝜕𝑥 (note that the flow velocity 𝑉𝑡 ~𝑈෡ is not ෡
𝜕𝑈
𝑉𝑡 ~
suitable as it is not Galilean invariant) 𝜕𝑥

• The simplest model formulation is therefore the


Smagorinsky model 𝜇𝑡 = 𝜌 𝑐𝑠 ∆ 2 𝑆መ

• Note that the eddy-viscosity of LES models decreases


Smagorinsky model
with grid size (less model in finer meshes)

8 © 2018 ANSYS, Inc.


Limits of Smagorinsky Model
• LES models should only provide damping in case under-resolved 𝜇𝑡 = 𝜌 𝑐𝑠 ∆ 2 𝑆መ
turbulence is present.
• In case of resolved flow they should provide 𝜇𝑡 = 0 For laminar shear flow:
• Most important, LES models should provide 𝜇𝑡 = 0 in case of ෡
laminar shear flow (laminar boundary or shear layers) 2
𝜕𝑈
𝜇𝑡 = 𝜌 𝑐𝑠 ∆
𝜕𝑦
• The LES model should also ‚turn-off‘ in the viscous sublayer
• The Smagorinsky model provides finite and fairly large eddy-
viscosity in such scenarios – alternative models are developed which
avoid this limitation
− Dynamic models – take information from neighbouring cells to determine state of
‚resolvedness‘. In case of resolved flow 𝑐𝑠 → 𝑐𝑠𝑑 = 0.
− WALE (Wall-Adapting Local Eddy-Viscosity Model) uses complicated tensor
formulation to achieve the same goal – advantage: locallity
− Implicit LES – uses/designs damping of numerical method to achieve proper LES

• There are many other models – difference for practical flows is


typically not large

9 © 2018 ANSYS, Inc.


SGS Models: Summary 𝐿𝐸𝑆
෡𝑗
෡𝑖 𝜕𝑈
𝜕𝑈
𝜏Ƹ 𝑖𝑗 = 𝜌𝑢ො 𝑖 𝑢ො𝑗 − 𝜌𝑢ෟ
𝑢 𝑢𝑗 = 𝜇𝑡 +
𝜕𝑥𝑗 𝜕𝑥𝑖

• Smagorinsky model (Smagorinsky, 𝜇𝑡 = 𝜌 𝑐𝑠 ∆ 2 𝑆መ


1963)
− Need ad-hoc near wall damping 3 Τ2
2
𝑆መ𝑖𝑗𝑑 𝑆መ𝑖𝑗𝑑
• WALE model (Nicoud & Ducros 1999) 𝜇𝑡 = 𝜌 𝑐∆ 5Τ2 5 Τ4
𝑆መ𝑖𝑗 𝑆መ𝑖𝑗 + 𝑆መ𝑖𝑗𝑑 𝑆መ𝑖𝑗𝑑
− Correct asymptotic near wall behaviour
• Dynamic model (Germano et al., 1991) 𝜇𝑡 = 𝜌 𝑐𝑠𝑑 ∆ 2 𝑆መ
− Local adaptation of the Smagorinsky
constant
• Dynamic sub-grid kinetic energy 0.5 ∆
𝜇𝑡 = 𝐶𝑘 𝜌𝑘𝑠𝑔𝑠
transport model (Kim & Menon 2001) 3Τ2
𝜕(𝜌𝑘𝑠𝑔𝑠 ) 𝜕 𝑘𝑠𝑔𝑠 𝜕 𝜇𝑡 𝜕𝑘𝑠𝑔𝑠
(Fluent only) + (𝜌𝑈𝑗 𝑘𝑠𝑔𝑠 ) = 𝑃𝑘 − 𝜌 + 𝜇+
𝜕𝑡 𝜕𝑥𝑗 ∆ 𝜕𝑥𝑗 𝜎𝑘 𝜕𝑥𝑗

10 © 2018 ANSYS, Inc.


LES Setup

• Numerics of the CFD solver becomes very important for LES


• The numerical dissipation can be of the same order or larger than the dissipation
from the LES model
• It is therfore necessary to use numerical methods with low numerical dissipation –
this means typically central difference schemes (or variants)
• There are also very stringent grid requirements for LES. Under-resolved grids can
lead to very large errors
• LES in industrial flows requires a very careful planning and pre-studies to achive
accurate results
• It is easy to create results which are of lower quality with LES than a RANS solution
• There are ANSYS Best Practice Guidelines for Scale-Resolving Simulations (SRS)

11 © 2018 ANSYS, Inc.


Periodic Channel Flow
y z

• Near the wall turbulent streaks appear x

• 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
Reτ Nx Ny Nz Lx Ly Lz Dx+ Dy1+ Dyc+ Dz+

180 Mesh 1 73 75 73 2h 2 h 15.7 0.254 13.47 7.85

590 Mesh 2 120 100 110 2h 2 h 30 1 50 15

12 © 2018 ANSYS, Inc.


Periodic Channel Flow
DSM – Dynamic Smagorinsky Model
2,00E+01
U+ DNS 1,00E+01
1,80E+01 uu+ DNS

1,60E+01
U+ DSM
U+ WALE
Re=180 9,00E+00 Re=180 uu+ DSM
uu+ WALE
8,00E+00
1,40E+01
𝒖′ 𝒖′ 7,00E+00

U+ 1,20E+01
6,00E+00
1,00E+01
𝒗′ 𝒗′ 5,00E+00
8,00E+00
6,00E+00 𝒘′ 𝒘′ 4,00E+00

3,00E+00
4,00E+00
2,00E+00
2,00E+00
1,00E+00
0,00E+00
0,00E+00
1,00E-01 1,00E+00 1,00E+01 1,00E+02 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

Y+ Y+
3,00E+01 9,00E+00
U+ DNS uu+ DNS
U+ DSM 8,00E+00 uu+ DSM
2,50E+01

2,00E+01
Re=590 𝒖′ 𝒖′
7,00E+00

6,00E+00
Re=590
U+ 1,50E+01 𝒗′ 𝒗′ 5,00E+00
4,00E+00
1,00E+01 𝒘′ 𝒘′ 3,00E+00
2,00E+00
5,00E+00
1,00E+00
0,00E+00
0,00E+00
1,00E-01 1,00E+00 1,00E+01 1,00E+02 1,00E+03
1,00E+00 1,01E+02 2,01E+02 3,01E+02 4,01E+02 5,01E+02

Y+ Y+
13 © 2018 ANSYS, Inc.
LES - Limitations Free Shear Flow

• With LES all relevant scales have to be resolved


• Free Shear Flows
− Turbulent scales are large and can easily be resolved in
time and space Wall Boundary Layer
− Combustion chamber (if walls are not relevant)
• Wall Boundary Layers
− Turbulent structures near the wall are very small (much
smaller than BL thickness)
− Need to be resolved in time and space
− Excessive CPU costs even for small domain
𝜹∙𝜹∙𝜹
− Minimum resolution per boundary layer volume 𝜹 ∙ 𝜹 ∙ 𝜹
– 10x20x40=8000 (~104) cells even with wall model Courtesey Center for Turbulence
(WMLES) Research (CTR)
− There are many boundary layer volumes on an real wall Published in: Taraneh Sayadi; Curtis W. Hamman;
boundary layer Parviz Moin; Physics of Fluids 2012, 24,

14 © 2018 ANSYS, Inc.


Grid Scaling for Wall-Resolved LES of Wall-Bounded Flows
Near Wall Streaks
• Near the wall turbulent streaks require resolution
of ∆𝑥 + ≈ 40, ∆𝑦 + ≈ 1, ∆𝑧 + ≈ 20
• Since ∆𝑥 + 𝒂𝒏𝒅 ∆𝑧 + depend strongly on Reynolds y
number (𝐞. 𝐠. ∆𝑥 + = ∆𝑥𝑢𝜏 Τ𝜈)this leads to a
strong increase of resolution requirements with
Re number
• Example Channel flow Reτ LES Cells Number
− Cell numbers for wall-resolved LES increases drastically
with Re number 395 500 000
760 1 500 000
− Re number of most technical devices moderate to high – 1100 4 000 000
therefore wall-resolved LES not feasible 2400 19 000 000
18000 1 294 676 760

15 © 2018 ANSYS, Inc.


Why not LES in Engineering Flows? NACA 0012 Airfoil Test

Rechord = 1.1·106

Velocity inlet

Airfoil rotated by 7.3 degree

71.3 m/s

Pressure outlet

16 © 2018 ANSYS, Inc.


Why not LES? WB Unstructured Hex Mesh
• Span: 0.05
chord; 80 nodes
• Very specialized grid
• In total ~ 11.4
• LES needs 3D mesh – for Mio nodes
“2D” flow to resolve 3D
turbulence. This means • WALE LES model
resolution in spanwise
direction • Periodicity in
spanwise
• Minimum span for such
simulation is 5% of chord direction
length
• Periodic boundary
conditions in spanwise
direction
• If spanwise space is too Leading edge Trailing edge
small – danger of
“synchronizing” eddies
across boundary

17 © 2018 ANSYS, Inc.


Why not Wall-Resolved LES? Q-criterion
• Q-criterion (W2-S2): Q=109 , colored by z-velocity:
Leading edge Trailing edge

Due to high Re number and moderate a, turbulence structures look still uncorrelated near trailing
edge even though periodic conditions with only span=0.05c

18 © 2018 ANSYS, Inc.


Wall-Resolved LES? 5%chord, 11M cells, Dt=1.5 s

• Skin friction coefficient Cf shows that LES is lower by


0.020
30% against RANS (RANS is reliable for such flows) Cf comparison: 2-D SST transition vs. 3-D ELES

• On the current grid cf is too low -> near wall streaks 2-D RANS(SST)
2D RANS
cannot be properly resolved 0.015 3D LES – Suction side
3-D ELES- Pressure
3D LES pressureside
• It is estimated that the grid is too small by a factor Cf side
of 2 in both streamwise and spanwise direction 0.010

• Grid refinement would result in ~50 million cells for


this small domain of only 5% span 0.005

• Aspect Ratio =1 (span=chord) would already require


1x109 cells 0.000
0.0 0.2 0.4 0.6 0.8 1.0
• This is still a moderate Reynolds number! x/chord

19 © 2018 ANSYS, Inc.


Near Wall Scaling – What’s the Problem
Low Re 𝐿𝑡 = 𝜅𝑦
• Turbulent length, 𝐿𝑡 , scale is independent of Re
number y

• Turbulent structures inside sublayer (grey box) are


damped out and only structures above the sublayer
need to be resolved Viscous sublayer
High Re
• However thickness of viscous sublayer decreases
with increasing Re number y

• Smaller turbulence structures near the wall get


“exposed” as Re increases – these structures need
to be resolved in 3D space and time
• Strong Re number scaling for wall-resolved LES cost

20 © 2018 ANSYS, Inc.


Near Wall Scaling – What’s the Alternative?
Lt = k y
Low Re
• There are in principle two options on how to avoid
y
the resolution of the thin near wall structures
(meaning to resolve only the yellow scales):
− Wall Function
• Coarse near wall grid with cell centre in logarithmic High Re Wall Function Viscous sublayer
layer
• Use of Wall function to bridge inner layer (first cell at y
∆𝒚+𝟏−𝒍𝒐𝒈 )
∆𝒚+
𝟏−𝒍𝒐𝒈
− Wall Modeled LES (WMLES)
• Use RANS model in the inner part (pink in the figure). High Re WMLES
This allows to resolve the flow inside the near wall y
layer only in wall normal direction
RANS

21 © 2018 ANSYS, Inc.


WMLES-Concept – Algebraic Model
RANS LES
• RANS/LES blending: 𝜈𝑡 = 𝑓𝑑 min 𝜅𝑑𝑊 2 , 𝐶𝑆𝑀𝐴𝐺 ∆ 2
𝑆
*
− Blending of Prandtl‘s mixing length model
(dw – part) with Smagorinsky model (D Part)
− Near the wall, the RANS model (Prandtl) is 𝑑𝑊 - Wall distance
used, away from the wall the LES model is 𝑆 - Shear Strain Rate
selected 𝐶𝑆𝑀𝐴𝐺 - Smagorinsky constant 𝐶𝑆𝑀𝐴𝐺 =0.2
− There are now many WMLES formulations, ∆ - Modified grid spacing
some based on algebraic models, some based 𝑓𝑑 - damping function
on transport equations (like DES or SBES).
• The algebraic model has been modified
* A hybrid RANS-LES approach with delayed-DES and
and optimized for usage in ANSYS- wall-modelled LES capabilities, Mikhail L. Shur,
Fluent and ANSYS-CFX Philippe R. Spalart , Mikhail K. Strelets , Andrey K.
Travin , Int. J of Heat and Fluid Flow 29, 2008

22 © 2018 ANSYS, Inc.


WMLES – Channel Flow Tests
Reτ WMLES LES Cells WMLES ΔX+ ΔZ+
Cells Number Nodes
Number Number
395 384 000 384 000 81×81×61 40.0 20.0
760 480 000 1 500 000 81×101×61 76.9 38.5
1100 480 000 4 000 000 81×101×61 111.4 55.7
2400 528 000 19 000 000 81×111×61 243.0 121.5
18000 624 000 1 294 676 760 81×131×61 1822.7 911.4

• Very large savings between WMLES and wall-


resolved LES
• Alternative is LES with wall functions – however Dx+
and Dz+ are a function of Dy+
• Good agreement in logarithmic layer
• With increasing Re number more of the inner layer
(in terms of y+) is covered by RANS
• Still requires minimum of ~10x40x20 cells for
boundary layer volume

23 © 2018 ANSYS, Inc.


WMLES – Channel Flow at Different Re Numbers
Ret=395 Ret=18000

• Solutions at very different


Re numbers look
essentially identical
• Differences can only be
seen near the wall.
• Visible is higher Eddy-
Viscosity for higher Re
number close to wall

RANS Eddy Viscosity

24 © 2018 ANSYS, Inc.


LES – Unsteady Inlet Conditions
RANS-LES-Interface with Vortex Method
(number of vortices = 1000)
• At an LES inlet one needs to specify resolved
unsteady turbulence structures
− Necessary especially when boundary layers/channel
flow enter the domain
• Numerous possibilities:
− Store-and-retrieve unsteady field (e.g. from pre-
cursor channel flow)
− Re-cycling from a downstream position (+re-scaling) 𝑢𝑖 𝐱, 𝑡 = 𝑈𝑖 𝐱 + 𝑢𝑖′ 𝐱, 𝑡
− Synthetic turbulence generator 𝑡𝑖𝑚𝑒−𝑎𝑣𝑒𝑟𝑎𝑔𝑒𝑑 𝑠𝑦𝑛𝑡ℎ𝑒𝑡𝑖𝑐 𝑡𝑢𝑟𝑏𝑢𝑙𝑒𝑛𝑐𝑒

• Synthetic turbulence generator is the most


user-friendly and generic way of creating Iso-surface of Q-criterion = 500
colored by eddy viscosity ratio
inlet conditions

25 © 2018 ANSYS, Inc.


Vortex Method (VM) - Fluent

• In essence, vorticity-transport is modeled by


distributing and tracking many point-vortices User provides RANS inlet profiles
on a plane (Sergent, Bertoglio)
𝑁

𝜔 𝐱, 𝑡 = ෍ Γ𝑘 𝑡 𝜂 𝐱 − 𝐱 𝑘 , 𝑡
𝑘=1

• Velocity field computed using the Biot-


Savart’s law
1 𝐱 − 𝐱 ′ × 𝜔 𝐱 ′ 𝐞𝑧
𝐮 𝐱, 𝑡 = − න ′ 2
𝑑𝐱 ′
2𝜋 𝐱−𝐱

26 © 2018 ANSYS, Inc.


Spectral Synthesizer

• The spectral synthesizer is based on harmonic functions:

෥ and 𝒕෤ include information


𝒙
from local turbulent length and time scales
present in the inflow RANS profiles

• The spectral synthesizer creates synthetic turbulence based on


inflow RANS profiles for k-w (or k-e)
− Profiles required, typically from auxiliary RANS precursor calculation
• Very unrealistic behavior possible when synthetic turbulence calculated
from uniform fields

27 © 2018 ANSYS, Inc.


Synthetic Turbulence Generator (STG) User provides RANS inlet profiles

• The STG generator is based on harmonic functions:


𝑁
𝑡
𝐮′ 𝐫, 𝑡 = 6 ෍ 𝑞 𝑛 𝐫, 𝑡 ⋅ 𝛔𝑛 ⋅ cos 𝑘 𝑛 𝐝𝑛 ⋅ 𝐫 + 𝜙 𝑛 + 𝑠 𝑛
𝜏
𝑛=1
• The details of this method are complex and not
given here (see publication)
• The STG generates synthetic turbulence based on
inflow RANS profiles for k-w (or k-e)
D. Y. Adamian and A. K. Travin, “Improved version of the
synthetic eddy method for setting nonstationary inflow
boundary conditions in calculating turbulent flows,” High
Temperature, vol. 49, no. 5, pp. 704–711, Oct. 2011.

28 © 2018 ANSYS, Inc.


WMLES – Flat Plate Grid

• Geometry and Grid


− Inlet boundary layer thickness d
− L x 0.4 L x 0.1 L (Streamwise, Normal,
Spanwise)
− Approximately 3 d spanwise (d0=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 d
one needs ~10x40x20 cells ReΘ Cells Nodes ΔX+ ΔY+ ΔZ+
Number Number
1000 1 085 000 251×71×63 68 0.05 ÷ 300 34
10000 1 085 000 251×71×63 520 0.4 ÷ 2300 307

29 © 2018 ANSYS, Inc.


WMLES with Unsteady Inlet Conditions

• Two inlet Reynolds number using ReΘ=1000


ReQ =1000 and ReQ =10000
• Grid stays the same as for lower
Reynolds number
• Inlet synthetics computed with
Fluent‘s Vortex Method
• Quality measure is how ”realistic” is ReΘ=10000
synthetic turbulence
− Is turbulence maintained?
− Are there large jumps in wall shear at
inlet?

30 © 2018 ANSYS, Inc.


Skin Friction Coefficient for Different Reynolds Numbers

• The wall shear stress recovers quickly


from the synthetic turbulence imposed
for ReΘ =1000 ReΘ=1000

• For the high Reynolds ReΘ =10000


number, the wall shear stress shows
even less variation near the inlet than
for the low Reynolds number
• The Vortex Method produces
ReΘ=10000
turbulence which is reasonably close to
‘real’ turbulence to be accepted by the
WMLES method
• Similar behavior for STG

31 © 2018 ANSYS, Inc.


Backward Facing Step

VM synthetics
• Back Step :
– Vogel Eaton
– Re=U*H/n=28000
– Domain span 4H
– Grid 2.8x106 cells
– Dxmax+~200, Dzmax~100,
Dy+<1
– CFL<1
– Vortex Method at Inlet
– (IDDES, WMLES and SAS
can all act as WMLES)

32 © 2018 ANSYS, Inc.


Summary: Grid Resolution Boundary Layers
x-streamwise, y-wall normal, z-spanwise

• Wall-Resolved LES ∆𝑥 + < 40, ∆𝑧 + < 20, 𝑁𝑦 ≈ 30 − 40

• Wall-Function LES ∆𝒚+


𝟏−𝒍𝒐𝒈
+
∆𝑥 + < 2∆𝑧 + , ∆𝑧 + ≈ ∆𝑦1−log , 𝑁𝑦 > 20
in log-layer
𝛿 𝛿
• Wall-Modelled LES 𝑁𝑥 ≈ ≈ 10; 𝑁𝑦 ≈ 20 − 30; 𝑁𝑧 ≈ ≈ 20
∆𝑥 ∆𝑧

Boundary Layer Volume with dxdxd

33 © 2018 ANSYS, Inc.


Grid Resolution for Mixing Layers

• Turbulence length scale from RANS is 𝑘 3/2


𝑅𝐴𝑁𝑆
𝑘1/2
𝑅𝐴𝑁𝑆
𝐿𝑅𝐴𝑁𝑆
𝑡 = =
proportional to mixing layer thickness 𝜀 𝐶𝜇 𝜔

• One needs at least 10 cells across mixing


𝐿𝑅𝐴𝑁𝑆 ≈ 0.7 ⋅ 𝛿 𝑚𝑖𝑥𝑖𝑛𝑔
layer (absolute minimum) 𝑡

• Similar spacing in other two directions 𝛿 𝑚𝑖𝑥𝑖𝑛𝑔 = Thickness of mixing layer


• More easily achieved than boundary

max
layer resolution as no near wall 𝑅𝐿 = 𝐿𝑅𝐴𝑁𝑆 ≤ 0.1
𝑡
refinement required
∆max = Max. edge length of cell

34 © 2018 ANSYS, Inc.


Selection of LES Model
• Smagorinsky model:
− Gives non-zero eddy-viscosity for plane shear (laminar flow)
− Cannot be used for transitional flows
− Needs near wall damping
− Not recommended
• Dynamic models:
− Avoid deficiencies of Smagorinsky model
− Require test filter
− Require averaging in time
− Relatively high model complexity
• WALE model:
− Avoids problems of Smagorinsky model
− Good compromise between simplicity and accuracy
− Note that WALE is not a WMLES formulation!
• WMLES
− Use for higher Re numbers to avoid unfavorable Re scaling of classical LES models

35 © 2018 ANSYS, Inc.


Time - Sampling of Statistically Steady Problems

In case of : f Start-up
• LES requires “start-up-time“ to reach correct
solution. Do not start time-average during start-
up phase
• Monitor quantities of interest (f =global
variables like lift, drag, and/or local quantities
like maximum heat transfer coefficients etc.)
• Start averaging only after monitors reach
“constant levels“ – where their mean value does t
no longer change
• For safety – use multiple averaging time
windows and compare solutions
• Starting the averaging too early can lead to large
errors – which average out only very slowly

36 © 2018 ANSYS, Inc.


LES Set-Up Consideration
• Read “Best Practice Guidelines SRS“
• Get LES-experience with simple test cases (e.g. channel
flow etc.) before proceeding to more complex application
• Determine if application/project is suitable for LES:
− Estimate domain size and boundary conditions (e.g. periodic, …)
− Estimate grid size and number of time steps → CPU/Storage cost.
Compare to available CPU power and project duration
• Define/select numerical scheme
− Model selection
− Low dissipation (e.g. BCD)
− Fast transient with low/no sub-iterations
• Define run-time strategy
− Start-up time – averaging time etc. Plan Carefully – LES can cost you many project
− Post-processing (write out required quantities during run) months and a large amount of CPU power

37 © 2018 ANSYS, Inc.


Setting up LES Models
in ANSYS Fluent
Turbulence Modeling Using ANSYS CFD

38 © 2018 ANSYS, Inc.


Setting Up LES in Fluent: Preparation
• Generate inlet profiles (velocity and turbulence)
− This is most often done with a separate, steady RANS calculation
in a small periodic domain immediately upstream of the inlet
− Realistic, non-uniform, inlet profiles allow the establishment of a
resolved turbulent flow in the proximity of the inlet in the LES
calculation
− Turbulence profiles are needed for realistic synthetic turbulence
generation at the inlet in the LES calculation Velocity contours from
• For any fluctuating velocity algorithm steady RANS solution
• Run a steady state precursor RANS calculation in the LES
flow domain
− Unless the initial condition is a quiescent flow field
− Use generated inlet profiles (velocity and turbulence) for
boundary conditions for the precursor calculation
− Check mesh resolution (next slide) and re-mesh if necessary
• Generate initial turbulence fluctuations
− TUI Command: Velocity contours after
/solve/initialize/init-instantaneous-velocity init-instantaneous-velocity
− The TUI command needs to be executed before switching from
steady to transient

39 © 2018 ANSYS, Inc.


Setting Up LES in Fluent: Check Mesh with Precursor RANS

• An estimate of the turbulence integral length 𝑘 3/2


𝑅𝐴𝑁𝑆
𝑘1/2
𝑅𝐴𝑁𝑆
𝐿𝑅𝐴𝑁𝑆 = =
scale, Lo, can be taken from the RANS results 𝑜
𝜀 𝐶𝜇 𝜔

• Normally, it is preferred to have at least 10 grid


cells per integral length scale
• The grid cell size can be estimated using
1ൗ
𝐿𝑔𝑟𝑖𝑑 = 𝑉 3
• A custom field function for Lo/Lgrid can be defined
and plotted to examine local mesh resolution
− In the picture to the right higher values imply more grid
points per integral length scale, meaning higher
resolution
Contour plot showing ratio of
− The lower limit of the contour scale is 10, so areas integral length scale to cell size
without any contours may require higher resolution
during mesh generation for LES

40 © 2018 ANSYS, Inc.


Setting Up LES in Fluent: Subgrid-Scale Models in Fluent

• After init-instantaneous-velocity, change analysis type


from steady to transient
• Go the Viscous panel and choose Subgrid-Scale Model
− Smagorinsky-Lilly
• Dynamic Stress option will be exposed after this is selected
– Enables the dynamic model (slide 10)
− WALE
• Recommended model for most cases
− WMLES
• Developed for wall bounded flows at increasing Re
– Review mesh guidelines in the SRS Best Practices document
− WMLES S-Omega
• Enhances standard WMLES formulation by computing the LES
portion of the model using 𝒂𝒃𝒔 𝑺 − 𝛀 instead of S to compute
the LES portion of the model
– S = strain rate, W = vorticity magnitude

41 © 2018 ANSYS, Inc.


Setting Up LES in Fluent: Inlet Boundaries

• At inlet boundaries specify profiles for


velocity components
− If a precursor RANS calculation was performed,
these would already be set
• Select the Fluctuating Velocity Algorithm
− Vortex Method, Spectral Synthesizer, or Synthetic
Turbulence Generator (STG)
• All Fluctuating Velocity Algorithms require
profiles for turbulence
− Method = “K and Epsilon”
− Accurate profiles ensure realistic turbulent
fluctuations

42 © 2018 ANSYS, Inc.


Setting Up LES in Fluent: Solver Setup

• In the Solver tab, select Methods


− Recommended Settings
• Pressure-Velocity Coupling: SIMPLEC or PISO
– See the best practices document (slide 37) for details
– All underrelaxation factors (under Controls) should be
1, or very close to 1 (0.9 – 0.7)
• Pressure: Second Order
– 2nd order should be good for most cases
– Avoid PRESTO for LES
• Momentum & Energy: Bounded Central
Differencing
• Transient Formulation: Bounded Second
Order Implicit
− Non-Iterative Time Advancement (NITA)
• If the solution converges well, consider using
Non-Iterative Time Advancement
– With Fractional Step for Pressure-Velocity coupling
– More details on NITA and LES available in best
practices document (Slide 37)
43 © 2018 ANSYS, Inc.
Setting Up LES in Fluent: Time Step Size
• Temporal resolution should match or exceed the spatial
resolution for LES
− Let’s say we have a cell with largest edge length of ∆ x and local
average velocity, U
− It takes a time interval ∆ t for the flow to travel across the cell
− The time step should be small enough to provide an adequate 𝛥𝑥
temporal resolution of the flow as it passes through the cell 𝛥𝑡 ≤
− The real velocity can be higher than the averaged velocity
𝑈
− Pre-cursor RANS simulation is used for the assessment of ∆ t
• Good practice to account for differences between
instantaneous and averaged velocities as well as for the
∆𝑥 𝑈∆𝑡
∆𝑡 ≈ ≈ 0.5
errors introduced from RANS by using a safety factor of 2 2𝑈 ∆𝑥
• RANS precursor calculation can be used to estimate Dt
• Use Cell Convective Courant Number in post-processing to
confirm time step is small enough
• Smaller time steps can be used during startup, if convergence
problems arise, or any time during the calculation
44 © 2018 ANSYS, Inc.
Setting Up LES in Fluent: Solution Procedure
1. Set the monitors with relevant global (e.g., 1. 2.
forces/moments) and local quantities (e.g., velocity,
pressure)
2. Set the autosave of the data files (e.g., every a few
hundreds time-steps)
− In case it is needed to restart, it will be possible to do so using
the most recent autosave files
3. Start the transient run and continue until a
statistically steady state is reached (see slide 36)
4. Start sampling the data (to compute mean and r.m.s.
values) 4. 4.
− GUI: Run Calculation/Advanced … (click on the “Data Sampling
for Time Statistics” button)
− GUI: Initialization/Reset Statistics to reset the data sampling
5. Continue sampling for a sufficiently long period of 4.
time
− Several flow-through times ( L / U0 )
− Mean fields need to recover homogeneity
6. Post-process the results (mean, r.m.s., power spectra,
etc.)

45 © 2018 ANSYS, Inc.


Post-Processing Variables Fluent

Variable Definition
Velocity (u, v, w) Instantaneous Resolved Velocity
Velocity (Mean X Velocity (& Y, Z)) Time Averaged Resolved Velocity
𝑢ത , 𝑣ҧ , 𝑤

Velocity (RMSE X Velocity (& Y, Z)) Fluctuating Velocity (u’,v’,w’)

Reynolds Stress (Resolved UV Reynolds = 𝑢′ 𝑣 ′ , 𝑢′ 𝑤 ′ , 𝑣 ′ 𝑤 ′ time averages of


Stress (& UW, VW) products of fluctuating velocity
components
Reynolds Stress Use custom field function = rmse-x-
(Resolved UU Reynolds Stress (& VV, velocity ^ 2 (& rmse-y-velocity ^ 2 ,
WW) rmse-z-velocity ^ 2)

Resolved Turbulent Kinetic Energy:


Use Custom field function = (1 / 2) * (rmse-x-velocity ^ 2 + rmse-y-velocity ^ 2 + rmse-z-
velocity ^ 2)

46 © 2018 ANSYS, Inc.


LES Summary

• LES allows resolution of large turbulent scales


• Numerous models are available in ANSYS-CFD
− Suite of LES models in both codes
− Unsteady inlet conditions with high quality synthetic turbulence
− Numerical schemes optimized for LES
− Post-processing for LES
• LES can be applied to free shear flow but is very expensive for wall bounded flows
• Re number scaling of LES can be reduced by WMLES – still expensive
• Both LES and WMLES can be used in Embedded/Zonal formulation to reduce LES
domain size
• Often 104-105 time steps required
• Still much more expensive than RANS

47 © 2018 ANSYS, Inc.


Appendix

48 © 2018 ANSYS, Inc.


Embedded/Zonal Large Eddy Simulation (ELES)
LES zone

• Suitable if zone with high accuracy


demands is embedded into larger Synthetic
domain which can be covered Turbulence
properly by RANS models RANS zone
• Limited zone can then be covered by
LES or Wall-Modelled WMLES model
• LES zone needs to be coupled to
RANS zone through interfaces
• LES zone requires suitable (WM)LES
resolution in time and space

49 © 2018 ANSYS, Inc.


Zonal LES: Test cases

• DIT-x: decay rate validation


− Modelled and resolved k

50 © 2018 ANSYS, Inc.

You might also like