scale-adaptive simulation (sas) turbulence modeling...2013/10/18 · scale. • we use a...
TRANSCRIPT
© 2011 ANSYS, Inc. October 17, 2013 1
Scale-Adaptive Simulation (SAS) Turbulence Modeling
F.R. Menter, ANSYS Germany GmbH
© 2011 ANSYS, Inc. October 17, 2013 2
Unsteady RANS Based Models
• URANS (Unsteady Reynolds averaged Navier Stokes) Methods URANS gives unphysical single mode unsteady
behavior Some improvement relative to steady state (RANS)
but often not sufficient to capture main effects Reduction of time step and refinement of mesh do
not benefit the simulation
• SAS (Scale-Adaptive Simulation) Method Extends URANS to many technical flows Provides “LES”-content in unsteady regions Produces information on turbulent spectrum Can be used as basis for acoustics simulations
URANS
SAS-URANS
© 2011 ANSYS, Inc. October 17, 2013 3
Assumptions Two-Equation Models
• Largest eddies are most effective in mixing
• Two scales are minimum for statistical description of
large turbulence scales
• Two model equations of independent variables define
the two scales Equation for turbulent kinetic energy is representing the large scale
turbulent energy
Second equation (e , w, kL) to close the system
Each equation defines one independent scale
Both e- and w-equations describe the smallest (dissipate) eddies,
whereas two-equation models describe the largest scales
Rotta developed an exact transport equation for the large turbulent
length scales. This is a much better basis for a term-by-term modelling
approach
© 2011 ANSYS, Inc. October 17, 2013 4
Classical Derivation 2 Equation Models
jk
t
j
k
j
j
x
k
xkcP
x
kU
t
k
w
j
t
j
k
j
j
xxk
kP
kx
U
t
w
w
w
w
ww
w
w
kt
• The k-equation: Can be derived exactly from the
Navier-Stokes equations
Term-by-term modelling
• The e- (w-) equation: Exact equation for smallest
(dissipation) scales
Model for large scales not based
on exact equation
Modelled in analogy to k-
equation and dimensional
analysis
Danger that not all effects are
included
© 2011 ANSYS, Inc. October 17, 2013 5
Source Terms Equilibrium – k-w Model
Only one Scale in Sources (S~1/T)
Turbulence Model Input S
Output w
Output k
One input scale – two output scales?
Source terms do not contain information on two independent scales
2 2
2 2
1 2
( )( )
( )( )
j tt
j j k j
j t
j j j
U kk kS c
t x x x
Uc S c
t x x x
w w
w
w
w w w w
© 2011 ANSYS, Inc. October 17, 2013 6
Determination of L in k-w Model
k-equation:
y
kk
ycS
k
x
kU
t
k
k
k
ww
w )(
)()( 22
ww
w
kkccS
k 1)(0 22
2
2 ~0
kccS
• Diffusion term carries information on shear-layer thickness
• Turbulent length scale proportional to shear layer thickness
• Finite thickness layer required
• Computed length scale independent of details inside turbulent layer
• No scale-resolution, as Lt always large and dissipative
22~ Sk
S~w from w-equation
2 2
~ ~ ~t
k SL
S
w
© 2011 ANSYS, Inc. October 17, 2013 7
Rotta’s Length Scale Equation
• To avoid the problem that the e (w) equation is an equation
for the smallest scales, an equation for the large (integral)
scales is needed.
• This requires first a mathematical definition of an integral
length scale, L. In Rotta’s (1968) approach this definition is based on two-point
correlations
• Based on that definition of L, an exact transport equation can
be derived from the Navier-Stokes equations (the actual
equation is based on kL)
• This exact equation is then modelled term-by-term
Rotta, J.C.: Über eine Methode zur Berechnung turbulenter Scherströmungen, Aerodynamische Versuchsanstalt Göttingen, Rep. 69 A14, (1968).
© 2011 ANSYS, Inc. October 17, 2013 8
Two-Point Velocity Correlations
Measurement of velocity fluctuations with two probes at two different locations
For small r, all eddies contribute
For large r, only large scales contribute
For r > L, correlation goes to zero
Integral vs. r proportional to size of large eddies L
r
Fixed Probe
Shifted Probe
ijR~
1
r
L
Eddies
),('),('
),('),('~
txutxu
trxutxuR
ji
ji
ij
r
© 2011 ANSYS, Inc. October 17, 2013 9
Rotta’s k-kL Model
Integral Length Scale: ( , ) ( , , )iiL x t c R x t r dr
• The integral of the correlations
provides a quantity, L, with dimension ‘length’.
• L is based only on velocity fluctuations and can therefore be described by the Navier-Stokes equations.
• Exact equation for L (or kL, ..) can be derived.
• L is a true measure of the size of the largest eddies
ry
1
L
Rii
r
© 2011 ANSYS, Inc. October 17, 2013 10
Exact Transport Equation Integral Length-Scale (Rotta)
Exact transport equations for F=kL (boundary layer form):
F
F
F
yvpvpR
y
drrr
RdrRR
r
drRy
rxUdrR
y
xU
x
U
t
ii
y
kk
iiyikiiik
k
y
y
y
k
k
''1
16
3
8
3
16
3
)(
16
3)(
16
3)()(
2
2
)(
1221
• Important term:
)(xkLFwith
y
yidrR
y
rxU12
)(
16
3
© 2011 ANSYS, Inc. October 17, 2013 11
Expansion of Gradient Function
Important term:
• Rotta:
22 3y y
y2 3
U( x r ) rU( x ) U( x ) U( x )r ...
y y y y 2
yyyyyy
ydrRr
y
xUdrRr
y
xUdrR
y
xUdrR
y
rxU12
2
3
3
122
2
1212
)(
2
1)()()(
0)(
122
2
yy drRr
y
xU
• Due to symmetry of Rij with respect to ry for homogeneous turbulence
ry
1 Rij
© 2011 ANSYS, Inc. October 17, 2013 12
• If z3=0 - No natural length scale
– No fundamental difference to other scale-equations
Transport Equation Integral Length-Scale (Rotta)
Transport equations for kL:
F
F
F
F yy
qcc
y
xUL
y
xULuv
x
U
t
tL
ii
k
k
zz
2/3
2
3
33
32
)()()()(
• Equation has a natural length scale:
33
3
2
/
/
yU
yUccL l
z
• Problem – 3rd derivative:
– Non-intuitive
– Numerically problematic 03 z
© 2011 ANSYS, Inc. October 17, 2013 13
Virtual Experiment 1D Flow
?
Logarithmic layer Lt=ky
2
1220y y
Ur R dr
y
)(~
)(~
1212 y
II
y
I rRrR
)(~
)(~
1212 y
I
y
III rRrR
)(~
)(~
1212 y
II
y
III rRrR
)(~
)(~
1212 y
III
y
III rRrR
wyconstxvxu
xvxu
rxvxuR
.)(
)(
)(~12
R12 asymmetric
012 yy drRr
© 2011 ANSYS, Inc. October 17, 2013 14
New 2-Equation Model (KSKL)
• With:
LkF
22
1 2 32
1''
j tk t
j
UP L U k
t x k y y
z z z
k F
F F F F
2 2'
' ; '' ;''
i i i ivK
j j j j k k
U U U U UU U L
x x x x x x Uk
vKLyU
yUL
22 /
/~ k
v. Karman length-scale as natural length-scale:
3/ 23/ 4j t
k
j j k j
U kk k kP c
t x L x x
1/ 4
t c F
© 2011 ANSYS, Inc. October 17, 2013 15
SAS Model Derivation
• Using the exact definition and transport equation of Rotta,
we re-formulated the equation for the second turbulence
scale.
• We use a term-by-term modelling approach based on the
exact equation.
• This results in the inclusion of the second velocity derivative
U’’ in the scale equation
• Based on U’’ the scale equation is able to adjust to resolved
scales in the flow.
• The KSKL model is one variant of the SAS modelling
concept, as these terms can also be transformed into other
equations (e- or w).
© 2011 ANSYS, Inc. October 17, 2013 16
Transformation of SAS Terms to SST Model
• Tranformation: w
k
c 4/1
1F
Dt
D
Dt
Dk
kDt
Dk
Dt
Dk
c
k
Dt
D
cDt
D F
F
F
F
F
F
www
24/14/1
111
2
2 2 2
22
2 1j t
j j j j j j j vK
U k k LS S
t x x x x x x x Lw
w w w w w w w z k
w wF
Wilcox Model BSL (SST) Model SAS
2 2
/
/vK
U yL
U yk
© 2011 ANSYS, Inc. October 17, 2013 17
2-D Stationary Flows: KSKL - RANS
NACA-4412 airfoil at 14°: trailing edge separation
KSKL
© 2011 ANSYS, Inc. October 17, 2013 18
Limitation of Growth by U’’
.
~
dUconst
dy
dU
dy
L
w
Homogenous Shear Inhomogeneous Shear
.
~
vK
dUconst
dy
dUy
dy
L L
w
Eddies grow to infinity
Eddy growth limited
by LvK.
© 2011 ANSYS, Inc. October 17, 2013 19
yUyU
2sin)( 0
SAS
RANS
One Model – Two Modes
RANS Model L~
SAS L~
© 2011 ANSYS, Inc. October 17, 2013 20
Scale-Adaptation based on t
SAS Modell - 2D Periodic Hill
t = 0.045 h/UB
4 higher t
2 higher t
© 2011 ANSYS, Inc. October 17, 2013 21
Time averaged velocity profiles U
SAS Modell - 2D Periodic Hill
© 2011 ANSYS, Inc. October 17, 2013 22
Fluent-SAS Model Volvo Bluff Body : Cold Case
DES-SST SAS-SST
Q = 1e6
© 2011 ANSYS, Inc. October 17, 2013 23
VOLVO Cold Case
Time-averaged
U-velocity
© 2011 ANSYS, Inc. October 17, 2013 24
Test case: Mirror Geometry
EU project DESIDER Testcase
Plate dimensions LW= 2.41.6
Cylinder Diameter : D = 0.2 m
Rear Face location : 0.9 m
Free stream Velocity: 140 km/h
ReD: 520 000
Mach: 0.11
2.4 m 1.6 m
140 km/h 0.9 m
© 2011 ANSYS, Inc. October 17, 2013 25
Test case: Mesh
Mesh: Box around the Plate & Cylinder Height of domain: 10 diameters (D=0.2m)
Coarse and fine meshes
wall-normal distance around 1-3 *10 -4 m
obstacle edges resolution: step sizes around 0.02*D (height) -0.03*D (circumf.)
Flow: Air as ideal gas
10D
© 2011 ANSYS, Inc. October 17, 2013 26
Validation: Near field SPL
Sensors downstream the mirror
10 100 1000Frequency [Hz]
0
10
20
30
40
50
60
70
80
90
100
110
120
130
140
SP
L [
dB
]
Freestream Velocity = 140 km/h
Experimental data
SAS model
Sensor 121
10 100 1000Frequency [Hz]
0
10
20
30
40
50
60
70
80
90
100
110
120
130
140
SP
L [
dB
]
Freestream Velocity = 140 km/h
Experimental data
SAS model
Sensor 122
10 100 1000Frequency [Hz]
0
10
20
30
40
50
60
70
80
90
100
110
120
130
140
SP
L [
dB
]
Freestream Velocity = 140 km/h
Experimental data
SAS model
Sensor 123
Grid ~ 3 million nodes
© 2011 ANSYS, Inc. October 17, 2013 27
Blow-Down Simulation – SAS (SST)
Mesh – 1x107 control
volumes hybrid
unstructured
Scale resolving results: SAS and DES show
similar flow pattern
SAS model does not
rely on grid spacing
SAS can be applied to moving
meshes with more confidence
Courtesy VW AG Wolfsburg: O. Imberdis, M. Hartmann, H. Bensler, L. Kapitza
VOLKSWAGEN AG, Research and Development, Wolfsburg, Germany
D. Thevenin University of Magdeburg
© 2011 ANSYS, Inc. October 17, 2013 28
Flow Topology and Mass Flow
Intake
Valve Exp. RANS DES SAS
3 mm 1 0.95 0.985 0.996
9 mm 1 0.988 - 0.99
Mass flow Rates
Courtesy VW AG Wolfsburg: O. Imberdis, M. Hartmann, H. Bensler, L. Kapitza
VOLKSWAGEN AG, Research and Development, Wolfsburg, Germany
D. Thevenin University of Magdeburg
© 2011 ANSYS, Inc. October 17, 2013 29
Geometry of the Cavity
• D = 4 in
• L = 5 D, W = D
• Lx x Ly x Lz = 18 D x 17 D x 9 D
• M = 0.85
• P = 62100 Pa
• T = 266.53 K
• Re = 13.47 E 6
Massflow
Inlet
Pressure far field
Symmetry
Outlet
© 2011 ANSYS, Inc. October 17, 2013 30
Mesh: 5.8 e 6 Cv – double O-grid
Inlet Side
Bottom
© 2011 ANSYS, Inc. October 17, 2013 31
Eddy viscosity ratio @ q = -500000 (q = 1/2 (S:S –W:W))
Turbulent structure by q-criterion
© 2011 ANSYS, Inc. October 17, 2013 32
Eddy viscosity ratio @ q = -500000 (q = 1/2 (S:S –W:W))
Wave propagation by Fluctuating Density
© 2011 ANSYS, Inc. October 17, 2013 33
k20 – k25 k20
k29
© 2011 ANSYS, Inc. October 17, 2013 34
k26 – k29 k20
k29
© 2011 ANSYS, Inc. October 17, 2013 35
Testcase Description – Experimental Test Facility and Data
• The experimental data is provided by the Institute of
Aerodynamics and Fluid Mechanics from TUM (not
yet released)
• Experiments are performed in a wind tunnel
including a moving belt
Courtesy by TU Munich, Inst. of Aerodynamics
© 2011 ANSYS, Inc. October 17, 2013 36
Computational Mesh 2
• 108,034,893 Cells
• Four Refinement Boxes
• MRF-Zones
Number
of
Inflations
First layer
height
Car 20 0.02 mm
Road 20 0.02 mm
Courtesy by TU Munich, Inst. of Aerodynamics
© 2011 ANSYS, Inc. October 17, 2013 37
DrivAir Generic Car Model
• Courtesy Tu Munich
• Currently studied
with ANSYS CFD
(Fluent and CFX)
• Data not yet public
© 2011 ANSYS, Inc. October 17, 2013 38
Overall Summary
• SAS is a second generation URANS model It is derived on URANS arguments
It can resolve turbulence structures with LES quality
A strong flow instability is required to generate new – resolved
turbulence
• Examples Flows past bluff bodies
Strongly swirling flows (combustion chamber)
Strongly interacting flows (mixing of two jets etc.)
• SAS Model is first and relatively save step into Scale-
Resolving Simulations (SRS) modeling Worthwhile to try
Alternative Detached Eddy Simulation (DES)