insideswg

Transcription

insideswg
WCRR, Sydney, November 22nd-25th, 2013
Comparison between CFD and Wind-Tunnel experiment for
slender bodies of aspect-ratio O(1)
in the presence/absence of cross-wind
Keith Weinman, Uwe Fey, Ralf Deiterding,
Moritz Fragner, Sigrid Loose, and Claus Wagner
DLR Göttingen
Institute of Aerodynamics and Flow Technology
WCRR, Sydney, 22nd-25th November, 2013, Keith Weinman, Slide 1
Contents
o Motivation.
o Preliminary objectives of the work.
o Overview of experimental and numerical facilities used.
o CFD, open geometry: comparisons at Re=[250,450]x103, β=0°.
o CFD, tunnel geometry: comparisons at Re=450x103, β=30°.
o What have we learned?
2
Motivation: Side-wind stability requirements for Trains
Wind effects can be correlated with train derailment
incidents (i.e. Simes,T, “A blow to train operations:
Can strong winds cause derailment”, Int. Rail Safety
Conference Melbourne, 2011).
Probability of wind-induced derailment increases with:
-- gusting side winds,
-- train speed,
-- area profile of vehicle normal to wind direction,
-- a reduction of vehicle mass.
Loongana,WA(2008)
New rail vehicle designs must pass
through a certification process,
described in various national
standards:
EN 14067-6, RIL 807.0401-0449
For risk assessment these standards require an understanding of the vehicle aerodynamics.
3
Estimating aerodynamic loads on vehicles
-Fz
Wind
Ril 807.0401-807.0449 - three pronged analysis
of vehicle kinematic behaviour to assess cross wind risk.
(Naumann et al,“Calculation of characteristic wind curves for
cross wind investigation“)
Fy
1.Determine vehicle cross-wind stability.
Q(t)
M A  M R fs
MA : aerodynamic moment.
MR : restoring moment due to vehicle mass on straight track.
CMx(Lee)
fs : calibration factor .. 0(1)
M R  mgbA
2. Evaluate vehicle response to external forcing neglecting
dynamic effects (springs, dampers etc). Wind loads
determined by aerodynamic coefficients and wind speed use wind tunnel with scaled model to derive characteristic
wind curves (CWC).
3. Multi-body simulation to calculate CWC including dynamic
effects of the vehicle.
4
Long term goals of the work.

Investigate flow topology (Oil flow, PIV, TSP).
Identification of flow features that influence vehicle aerodynamic loads
(i.e. attached vortices, laminar-turbulent transition).

Comparison of computed model aerodynamic loads against force balance
measurements in laboratory reference frame [RIL 807.0401] over
Reynolds number range [0.25, 0.35,0.45] Mio,
yaw angle range β=[0,…,30]o.

Identification of most appropriate numerical modeling approach (in terms of
agreement with experiment, numerical stability on typical meshes, and cost).
5
Side Wind facility at Göttingen (SWG)
Technical Details
Closed Loop facility
Cross-section: 2.4 m x 1.6 m
0.5 Mw Fan with Umax= 65 m/s
TI < 0.22 %
Ubulk continuously adapted to match target Reynolds
number due to variations in temperature.
dTt/dt ≈ 7° C/min @ Re = 450000
dTo/dt ≈ 1° C/min @ Re = 450000
Force/moment measurement with 6 component
balance.
Splitter plate to avoid contamination from tunnel
Floor boundary layer.
Atmospheric pressure/temperature
1:25 scale (Remax=0.5 Mio)
6
Estimation of aerodynamic loads: Coordinate System
Xo=(0,0,0)
7
Side Wind facility at Göttingen (SWG) : Visualization Methods
Particle Image Velocimetry[1][2] Temperature Sensitive Paint[3][4]
Velocity field computed
by cross-correlation of
time-separated particle
images.
Luminescent molecular
probe which measures surface
temperature differences
produced by convective heat
transfer.
Oil Flow Visualisation
Flow patterns created by
viscous shear forces acting
on model surface.
References
[1] Loose et al., „Optical measurement techniques for high Reynolds number train investigations“, Experiments in Fluids, 40, 643-653, 2006
[2] Haff et al., „Wind Tunnel Experiments with a high speed train model subject to cross-wind conditions“, Proccedings of the first international conference on Railway
Technology, J.Pombo (Editor), Civil-Comp press, Stirlingshire, Scotland
[3] Tropea et al. [Ed.], Springer Handbook of Experimental Fluid Mechanics, Chap 7.4: „Transition detection by Temperature- Sensitive Paint“, Springer, Heidelberg, 2007
[4] Fey et al.,“Investigation of Reynolds number effects in high-speed train wind tunnel testing using temperature-sensitive paint“, „2nd symposium on Rail-Aerodynamics
2013, Berlin; 15-17, May 2013
8
Numerics: DLR Triangular Adaptive Upwind Code (TAU)
 Compressible NS solver
 Hybrid mesh (hexa, prisms, tetras)
 Dual-grid/cell centered metric
 Explicit time integration
 Multi Grid (domain decomposition)
 Low Mach Number Preconditioning
 RANS/LES/Hybrids with transition
Note : Conversion from TAU C.S. to WT
C.S. required in post-processing integral loads.
9
RANS (SWG): Feature identification using CFD result at Re=450,000, β=30o
Isosurface:
Contours :
Grid
:
total pressure
M∞
D2.R4
10
NGT2-Model: Topology of near-wall flow @ Re=0.45 Mio, β=30°
PIV cut
Topology of CFD solutions agrees broadly with flow visualization....
11
Comments on Flow visualization
but some important differences
can be seem!
 laminar-separation
 turbulent transition
which can impact on the vehicle
aerodynamic loading.
12
Flow topology at β=30°. Comparison with PIV data (X=-0.25)
View from
rear
View from
front
URANS, SA, D2.4)
Qualitative agreement between CFD and PIV is reasonable.
13
Integral force and moments: Re=250,000, β=20o
Grid
N (Mio)
O1
16
Predicted integral force coefficients
appears reasonable but …..
RANS solutions may not satisfy
standards such as RIL 807.0401, even
for low β and Re values.
20000 iterations (96 domains) requires ~ 24
hours - the computational cost equilvalent
across all calculations shown.
14
N .bl
11-20
X/H
17
Y/H (Mio)
17
Z/H
17/3
Y+max
0.6
Quantification of Integral Force and Moment Coefficients
Define :
subscript c - value from a sampled set (i.e. CFD, experiment)
subscript e – value from a sampled set (i.e. CFD, experiment)

Consider a vector space spanning the range of test results F such that an error norm E
can be constructed using the difference of sets on the space as a difference:
En ( F )  Fc  Fe
n
where n is the order of the vector norm e.g. (0,1,2,∞)
 The definition allows construction of quality quantification measures based on RIL
807 0401, for example …
En ( Fi )
M 1n ( F )  i
 0.5%
Fi ,e
15
M 2n ( F )  i En ( Fi )  0.1
Measured force and moment components: Re=250,000,β=0o
Force Components (3)
*
*
Moment Components (3)
*
Mx b
*
i
Balance CS-> Model CS
M xi 
M xi   ijk Fjb xk
b
U  2 Ao d o
*
*
16
Integral force and moments: Re=250,000, β=0o
11
Grid
N (Mio)
N .bl
X/H
Y/H (Mio)
Z/H
Y*max
O1
16
11/20
17
17
17/3
0.6
Fa ,b  C f a  Cb
M a ,b  C M a  CM b
2
2
Grid
Cfx
Cfy
Cfz
M2(Cf) Cmx
Cmy Cmz
M2(CM)
Exp
D1
-0.18
-0.003
-0.11
0
0.015
0.07
0.04
0
SA
D1
-0.26
0.04
-0.078
0.04
0.032
0.9
0.3
0.5
SA(RC)
D1
-0.27
0.05
-0.07
0.04
0.03
0.6
0.23
0.3
MSST
D1
-0.26
0.03
-0.08
0.04
0.031
0.8
0.29
0.3
17
Integral force and moments: Re=450,000, β=0o
Grid
N (Mio)
N .bl
X/H
Y/H (Mio)
Z/H
Y+max
O1
16
11/20
17
17
17/3
2.0
M 2 (C f )  C f c  C f e
2
Grid CFx±0.01
CFy±003 CFz±0.09
M2(Cf)
Exp
D1
-0.22
0.003
0.07
0
SA
D1
-0.27
0.04
0.078
0.05
SA+RC D1
-0.26
0.05
0.075
MSST
-0.29
0.03
0.07
D1
M 2 (CM )  CM c  CM e
CMx
2
CMy
CMz
M2(CM)
-0.04
0.02
0
-0.032
-0.9
0.3
1.2
0.045
-0.03
-0.647
0.23
0.9
0.058
-0.03
-0.86
0.9
1.2
0.003
• Variation between experiment and CFD is considerably larger than variation across CFD
for both integral and moment coefficients.
•
Turbulence model influence is not marked for integral forces, but appears more signficant
for the integral moments.
18
Simulating in SWG: Base Grids
D2
D1
Base grid dimensions
Grid
β(odegs)
N (Mio)
Nlayers
Nprisms (Mio)
Npyras (Mio)
Ntetra (Mio)
Y+max
D1/D2
0
16
11/20
17
0.02
39
0.9
D2.1
30
19
11/20
21
0.02
49
0.9
D2.2
30
22
11/20
26
0.02
56
0.9
D2.3
30
29
11/20
32
0.02
75
0.9
D2.4
30
36
11/20
43
0.02
86
0.9
19
RANS (SWG): Estimation of Rolling moment (CMx): Re=450,000, β=30o
 Loads computed on D1 (β=0°) are comparible to the results computed using mesh O1.
 CFD analysis of the 30° case is not yet complete.
 Adapatation improves the quality index, but no RANS calculation satisfies the quality
requirement for all moments and forces.
 Transition specifications return minor improvements here (under investigation).
Grid
Exp
Cfy
Cfz
M2(CF)
Cmx
M2(CMx)
-4.7
-1.8
0
-3.07
0
SA+RC
D2.R0
-3.1
-5.3
3.8
-2.8
0.27
SA+RC
D2.R1
-3.1
-5.2
3.7
-2.9
0.17
SA+RC
D2.R2
-4.0
-3.9
2.2
-3.1
0.03
SA+RC
D2.R3
-4.3
-3.3
1.5
-3.2
0.13
SA+RC)
D2.R4
-4.5
-2.6
0.9
-3.2
0.13
SA+RC *
D2.R4
-4.5
-2.6
0.8
-3.2
0.13
* Transition line co-ordinates determined from TSP image
20
OFV -transition
What have we learned?
 The flow about the NGT2 is highly complex and provides a
significant challenge for RANS methods.
 RANS methods are tuned for lower yaw angle ranges.
 RANS mesh refinement studies suggest that the solution
quality improves with refinement: prediction of the attached
vortex systems and associated induced load components due
to these vortices is improved. Note that attached vortices
appear to be stationary.
 Strong turbulence model influences are not demonstrated,
suggesting that other modeling error sources dominate. All
meshes are near-wall resolving … suggesting that unsteady
modeling may be more appropriate.
21
What have we learned?
 Quality
indices for integral forces are easier to achieve
in comparison to integral moments
∆FxMAX= 5N
∆FyMAX= 10N
∆FzMAX= 30N
22
What have we learned?
 Flow features returned from CFD agrees broadly
with flow visualization .
 TSP is an effective method to assess turbulent transition
as well as deduce footprints of attached vortices on
complex geometries and detect laminar separation bubbles.
 Observation and CFD results inside SWG show the
existence of complex vortex systems attached to the
vehicle. Laminar separation, APG effects and transition
effects are present, each of which on an individual basis
pose a significant challenge to conventional RANS
modeling-
23
What have we learned???
Work to date supports the findings in the literature that
suggests unsteady computational methods may be more
appropriate. At the present time both URANS and DDES
solutions are being computed for validation against the
present RANS and experimental data.
 HOWEVER, THERE IS A NEED TO EVALUATE SYSTEM
COMPLEXITY IN A CONSITENT MANNER. To this end we
are in the process of devising numerical experiments of
increasing complexity that are relevant to train simulations
Experimental program is being extended so that unsteady
force and flow field data can be collected for use in validation
of flow-resolving calculations.
24
25
RANS (SWG): integral force and moments: Comments on adaptation
• Mesh refinement considerably enhances the ability of the RANS method
to resolve the vortical flow attached to the vehicle.
N2.R1 (19 Mio)
N2.R4 (36 Mio)
26
Numerical convergence
27
Plate surface pressure distributions: Re=250,000 β=20o
Re=250,000, β=20°
Influence of turbulent production evident.
28
29
Influence of Reynolds number and Yaw angle
l
0.75 Mio yaw = -30°
1.05 Mio yaw = -30°
30
Flow topology: Shear stress distribution along plate (CFD)
Skin friction estimates demonstrate stationary
vortical footprints which are „attached“ to the
model. These footprints are only weakly
influenced by the modeling approach used
(SA/SST models demonstrated).
Re=450K, β=30o
Re=450K, β=0o
31
Vortex Footprints (VF)
At β=30° two VFs are
observed downwind of the
model.
Initialization of Tunnel BC's
Iterate inflow boundary pressure until pressure
at sensor location is achieved.00
i.e.: at Re = 450K Pdyn (target) = 2055
Pa
32
WCRR, Sydney, 22-15th November, 2013, Keith W
Visualization of the flow field: Temperature Sensitive Paint (TSP)
33

developed in cooperatrion with Universität
Hohenheim

Sensitivity 8%/°C @40°C
(=luminosity pro 1° change in temp.)
Luminescent molecular
probe which measures surface
temperature differences produced
by convective heat transfer.
(reference).
Visualization of the flow field: Temperature Sensitive Paint (TSP)
Luminescent molecular
probe which measures surface
temperature differences produced
by convective heat transfer.
(reference).

Developed in cooperatrion with Universität
Hohenheim

Sensitivity 8%/°C @40°C
(=luminosity pro 1° change in temp.)
34
OV322
Overview of Numerics: Modified JST Dissipation Model
Central Scheme Flux Model
r → ∞ G → H(0)
aP
1
1
Fˆ f   FˆL  FˆR   (G (...) D (2)  G (...) D (4) )
aPr→ 0 G → H(1)
2
2
a = 4.5
D(2)=ε2 ( fj+1 - fj )
SHOCK REGIONS:
(4)
SHOCK FREE REGIONS: D =ε4 (L( fj+1 )-L( fj ))
Description of G and unsteadiness indicators (P)
Unsteadiness characterised by 1/P
G(a,r,P)
G ( a, r , P )  tanh aP r 
H(1)
H(0)
r is set like 1/ M local (e.g. shock sensor)
PDucros (u ) 
PMach 
(.u )
2
((.u )     )
M local
2
2
P
M ref
WCRR, Sydney, 22nd-25th November, 2013,
WCRR, Sydney, 22nd-25th November, 2013, Keith Weinman, Slide 35
Experimental Overview: OFV issues
WCRR, Sydney, 22nd-25th November, 2013,
WCRR, Sydney, 22nd-25th November, 2013, Keith Weinman, Slide 36
Experimental Overview: TSP Sensitivity
SWG @ 40o C:
Sensitivity (TSP) = 8%/oC
8% change in light intensity
per oC
WCRR, Sydney, 22nd-25th November, 2013,
WCRR, Sydney, 22nd-25th November, 2013, Keith Weinman, Slide 37
Numerics: DLR Triangular Adaptive Upwind Code (TAU)
Compressible NS solver
Hybrid mesh (hexa, prisms, tetras),
Central Scheme + matrix dissipation form
of JST stabilization + Skew-symmetric
inviscid flux form [Ducros (2005), Kok (2009)].
Dual-grid/cell centered metric,
Implicit/Explicit time integration,
H based on Ducros et al. (2005)
Jameson Dual + Global Time Stepping,
Multi Grid (domain decomposition),
Low Mach Number Preconditioning,
RANS/LES/Hybrids with transition.
2nd Diff.
𝐺→0 𝑎𝑠 𝐴→0
𝐺→1 𝑎𝑠 𝐴→∞
4th Diff.
𝐻→0 𝑎𝑠 s ≪ w
𝐻→1 𝑎𝑠 s ≫ w
S  0.5ui , j  u j ,i 
W  0.5ui , j  u j ,i 
38
Scalar Dissipation +
JST Model
Scalar Dissipation +
JST Model (CB/32)
Results with Modified JST Dissipation
Matrix Dissipation +
Mod. JST Model
Matrix Dissipation +
Mod. JST Model (CB/32)
39
WCRR, Sydney, 22-15th November, 2013, Keith W
Visualization of the flow field: Particle Image Velocimetry (PIV)
Velocity field computed
by cross-correlation of
time-separated particle
images.
40