wp6: cd airfoil test-case experimental and numerical data base

Property of Valeo – Duplication prohibitedFebruary 2008

1

WP6: CD Airfoil Test-caseExperimental and numerical data base

October 2008 S. Moreau

VEC Manager of the Fan System Core CompetenciesManager of the Group Simulation Competency Center


2

Airfoil chord length ~10 cm smUsm /40/16 0

Valeo CD and NACA12 airfoils, Flat Plate, V2 and V3 airfoils

Nozzle exit section 50 cm x 25 cm

RMP

Camber 12°Thickness 4%

RMP 11

26

Open-Jet Aeroacoustic Experiment in ECL Large Wind Tunnel

ECL Experimental Set-up, LMFA


3

CD Airfoil Experimental Data Base

Open-Jet Aeroacoustic Experiment in ECL Large Wind Tunnel

Moreau et al, AIAA J. 2004, 2005, JFE 2005

Valeo CD Airfoil: Re 1.5 105 ; M 0.05 ; several angles of attack (8° focus)Far field noise measurements: Noise spectra and directivities Remote Microphone Probe (RMP) measurements: Wall pressure statistics (Cp, frequency spectra, coherence, phase)Hot wire measurements: Velocity statistics (mean and RMS velocity components, Reynolds stress and frequency spectra)


4

CD airfoil in high loading conditions

Flat plate at zero angle of attack

Evidence of 2 mechanisms: vortex-shedding noise and trailing-edge noise

Experimental Far Field Noise

Spectrum

Directivity


5

70

60

50

40

30

Wal

l pre

ssur

e sp

ectr

a (d

B)

102

2 3 4 5 6 7 8

103

2 3 4 5 6 7 8

104

Frequency (Hz)

RMP 11 RMP 21 RMP 22 RMP 23 RMP 24 RMP 26

Wall pressure fluctuations must be statistically homogeneous

is deduced from coherence measurements

)(l y

d),()(l y

0

2

is deduced from the phase diagrams of streamwise cross spectra

cUGaussian model proposed (Roger & Moreau, AIAA 2002-2460).

Equivalent Corcos’ model

Experimental Wall Pressure Statistics

f-5


6

Hot Wire Measurements

Wake Zoom

Overview

Inlet surveyLES bc survey

Shear layer survey


7

leading edge

separation line

Reattachment of the

flow

Separation

bubble

Suction side streaklines

Direction of the flow

turbulent

laminar

Flow Visualization on CD Airfoil

« Oil » Flow Tuft Film

MVI_9612.avi

Evidence of laminar flow separation at the leading edge Possible flow separation at the trailing edge


8

Two families of results

k-, SST and V2F

k- TL and WL

Numerical Wall Pressure Coefficient

Moreau et al, AIAA J. 2003


9

Two families of results

k-, SST and V2F

k- TL and WL

Numerical Wall Friction Coefficient

Moreau et al, AIAA J. 2003


10

Comparison of Wall Pressure Coefficients

Good prediction of laminar flow separation at the leading edge No prediction of onset of trailing edge flow separation

SST model


11

Broadband Models: Generalized Amiet - 1

3 main mechanisms considered:

Tip and leakage flow are not considered yet

1 - Turbulence-interaction noise

2 - Trailing edge noise

3 - Vortex-shedding noise


12

Broadband Models: Generalized Amiet - 2

Turbulence Interaction Noise:

Trailing Edge Noise:

2

0

2

01

0

2

00

2

20

30 ,,),(22

),(

S

xk

UxL

S

xk

U

dU

S

xckxS wwpp

x

yz

)x,x,x(x 321

)y,x(S

0U

2

0

21

0

2

2

20

3 ,,,)(2

),(

S

xk

UxL

S

xkd

S

xckxS ypppp

wall-pressure spectrum

Inflow velocity statistics

Spanwise correlation length

Radiation integrals(including back-scattering

correction)


13

Stanford LES Set-up & Averaged Results

Choose the largest jet width (w = 50 cm) LES domain in the jet core, with velocity B.C.'s coming from

RANS (only mean values) Better prediction of leading edge flow (Cp) with LES


14

Grid for Stanford LES

• Single block-structured topology• Grid Size: 960 x 84 x 64 5.2 million nodes• Domain Size: (4 x 2.5 x 0.1) x chord (first LES able to resolve the spanwise coherence length)

Very regular, fine and orthogonal grid at LE

Stanford Reference LES grid in 2D slice (2003)


15

Grid Quality/Parameters of Stanford LES

• Almost a DNS resolution in the normal direction• Very regular and orthogonal grid near airfoil • Grid independence of the solution verified on pressure spectra• Energy-conserving hybrid finite-difference/spectral code• Dynamic sub-grid-scale model


16

Stanford LES Instantaneous Results

Leading-edge separation leading to transition on suction side Laminar boundary-layer on pressure side

Qualitative agreement with experimental observation


17

2D vs 3D LES Instantaneous Results

NACA0012 Re ~ 5-9 105 and M~0.2

Drastic change of flow topology after transition

3D 2D


18

#26

Experiment

LES

Stanford Broadband Noise (BBN) Sources

Excellent qualitative and quantitative agreement


19

Stanford BBN Prediction

Good agreement of both analogies with experimental data Effect of finite-chord up to 2 kHz

Discrepancy between the two analogies at high frequencies

Acoustic Analogies based on wall pressure statistics (Amiet) and on velocity statistics near the trailing edge (Ffowcs-Williams

and Hall)


20

Evaluation of Other Unsteady Methods

Starting from the same “LES domain” with the same RANS boundary conditions and, if possible, the same LES grid:

Unsteady RANS: no unsteadiness was observed

Detached Eddy Simulations (DES-SA within Fluent 6.1)

Lattice Boltzmann (RANS/DNS) > Powerflow (EXA)

LES with Immersed Boundary Technique

Moreau et al, CTR Summer Program 2004


21

Grid and Simulation Parameters of SP-2004


LBM: Powerflow 6.1.22 (3ddp) Grid: 73551 voxels and 1491 surfels in a 2D slice (~1.2 M in 3D)

Smallest cell at LE is similar to body-fitted LESModel: RANS k- in 2D - No model in 3DSimulation parameters: (Re=1.5 105)

Time step: t =2.0e-7

CPU time for 100 time steps: 5 minutes on SGI Octane (1 CPU)

LES-IB: Structured Cartesian (PhD: S. Kang)Grid: 4.8M in 3D (0.15M in a 2D slice)

Smallest cell at LE is ~2.5 larger than body-fitted LESModel: LES + Dynamic ProcedureSimulation parameters: (Re=1.5 105)

Time step: t =1.0e-4

CPU time for 100 time steps: 20 minutes on Linux cluster (8 CPU)


22

SP-2004 (Stanford) Grids

Reference LES DES-SA

LBM-DNS LES-IB

1-2 coarsening


23

SP-2004 (Stanford) Flow field Results

Reference LES DES-SA

LBM-DNS LES-IB

Instantaneous Velocity Field


24

SP-2004 (Stanford) Wall Pressure Coefficient


2.5

2.0

1.5

1.0

0.5

0.0

-0.5

-1.0

Pre

ssur

e C

oeff

icie

nt

1.00.80.60.40.20.0

Normalized distance

Experiment RANS (v2f) LES DES IB-LES

2.5

2.0

1.5

1.0

0.5

0.0

-0.5

-1.0

Pre

ssur

e C

oeff

icie

nt

1.00.80.60.40.20.0

Normalized distance

Experiment RANS (v2f) LES LB-RANS LB-DNS

Only IB-LES provided a complete flow field close to reference LES


25

SP-2004 (Stanford) Wake Velocity Profiles


Two coarse grids in any of the new simulations to yield good wake

X-wire


26

SP-2004 (Stanford) Flow field Results

DES gives unrealistic flow field (over production of k at LE) Grids for IB-LES need to re-visited for better TE prediction

All over estimate the pressure fluctuations at low frequencies


27

STAR-CD LES Simulations (2005-2006)Moreau et al, AIAA 2005-2916

PISO algorithm for time discretization

Central differencing and upwind MARS scheme (no stable solution could be obtained even in 2D with CDS for all grid topologies tested).

To keep a CFL number below 1 throughout the computational domain, a maximum allowable time step t = 1.5 10-5 s is used.

5 to 10 time units run to eliminate the transient and collect reliable statistics (based on the free stream velocity of 16 m/s).

Smagorinsky sub-grid scale (SGS) model together with a van Driest near-wall damping and WALE SGS are used.

k- based DES is selected.

Evaluation of different numerical Schemes Evaluation of different SGS models

Re-assessment of the DES model with a different code


28

Final MARS LES Grid Topology (2005)

Zoom LE Zoom TE

1,115,000 cells• Similar grid 1-2 coarsening as DES with Fluent 6.1• Only 5% chord span• Good near-surface resolution: x+≤ 20 ; y+≤ 1 ; z+≤ 10


29

MARS LES Instantaneous Velocity Field

Stanford Reference LES STAR-CD MARS LES

Similar small structures created after separation convected downstream towards the trailing edge

Larger flow separation at the leading edge in MARS LES.

More coherent structures at the trailing edge in MARS LES.


30

MARS LES Wall Pressure Coefficient

Qualitative agreement on the laminar separation bubble (good level of pressure plateau but too large extent of the bubble, 11.2% instead of 3.7%)

First simulation to predict the positive pressure gradient up to mid-chord.


31

MARS LES Wake Velocity Profiles

Excellent agreement in the near wake for the MARS LES

Too large diffusion and deflection of the wake in the DES.

-0.2

-0.1

0.0

0.1

0.2

y/c

1.00.90.80.70.6

x-velocity

Experiment present LES


32

MARS LES Wall Pressure Spectra

Too high levels everywhere

No homogeneous statistics close to TE

P2

P8

P14P60

P60: mid-chord (-60 mm)P14: -14 mmP8: -8 mmP2: -2 mm

Origin: TE

• Too large structures and coherence at the trailing edge• Fluctuations are getting damped towards TE (MARS upwinding)


33

MARS Fine Scale Structures

Turbulent re-attach.Turbulent T.E.Separation

Laminar L.E. separation

Iso-values of normalized Q colored by the streamwise vorticity

Larger structures than reference LES


34

DES Simulation Issue

Velocity Field Sub-grid Turbulent Viscosity

Well attached flow field all the way to the trailing edge as in RANS simulations

Short laminar separation bubble is not captured

Transition occurs at the stagnation point (local turbulent kinetic energy overproduction in k- model).


35

Conclusions Summer 2005

The STAR-CD MARS LES reproduces all qualitative features of the flow encountered in the ECL experiment and simulated in the reference LES (Wang et al, Stanford 2004).

A short laminar separation bubble is formed, reattaches and sheds small vortices that are convected towards the trailing edge.

Evolution of the boundary layer seems to be well captured, especially the experimental positive pressure gradient up to mid-chord.

Yet the laminar separation bubble is too wide and the wall pressure fluctuations are damped at the trailing edge (most likely due to MARS upwind scheme). No use for self-noise prediction.

All DES found inadequate for this attached flow (transitional airfoil with a short laminar bubble).

Further grid optimization required to remove instabilities with Central Differencing Scheme (CDS).


36

Final CDS LES Grid Topology (2006)

690,000 cells

• 3-4 grid coarsening in all directions• 10% chord span• Good near-surface resolution: x+~ 2.5 ; y+~ 2 ; z+~ 3.6

3-4 coarsening

Stability of CD scheme without significant oscillationsOnly a small jump in the turbulent viscosity

D. Laurence, VKI Lecture Series 2005Y. Addad, PhD UMIST


37

CDS LES Wall Pressure Coefficient

Same qualitative agreement on the laminar separation bubble

No significant differences between the two SGS models


38

CDS LES Wake Velocity Profiles

Same excellent agreement in the near wake for the CDS-LES

No significant differences between the two SGS models


39

CDS LES Wall Pressure Spectra

Good overall predicition with the CD SGS model has only a moderate effect on spectra


40

Iso-values of normalized Q colored by the streamwise vorticity

t = 0.07 s

t = 0.25 s

CDS Fine Scale Structures

Smaller structures than with MARS scheme


41

SP-2006 (Stanford) Unstructured LES

Excellent agreement with both structured LES and experimental data

Comparison of Urms

0.000

0.020

0.040

0.060

0.080

0.100

0.120

0.140

0.160

0.180

-0.100 -0.080 -0.060 -0.040 -0.020 0.000 0.020 0.040 0.060 0.080 0.100 0.120

y/C

Urm

s/U

ref

2005 Wake Data (18 Jan) @ x/C = 0.057356

2003 LES (M. Wang) Wake Data @ x/C = 0.057356

2006 LES (CDP 2.3) Wake Data @ x/C = 0.05736

103

104

-110

-100

-90

-80

-70

-60

-50

-40

f (Hz)

Pow

er S

pect

ral D

ensi

ty (

dB)

Wake Spectra at x/c = 0.057356

EXP (y/H = 0.00038238)

LES (y/H = -0.00045381)

x/C = 0.057356

0,0

0,2

0,4

0,6

0,8

1,0

1,2

-0,100 -0,080 -0,060 -0,040 -0,020 0,000 0,020 0,040 0,060 0,080 0,100

y/C

Vm

ag/U

ref

2005 ECL Experiment

2003 RANS-V2F

2004 LES (M. Wang)

2006 LES (CDP 2.3)

<4% difference


42

Conclusions Summer 2006

The STAR-CD CDS LES improves the MARS LES significantly (over dissipative in the TE region) and compares favorably with the ECL experiment and the reference LES (Wang et al, Stanford 2004).

A regular grid with smooth jumps is required if coarsening is to be used to yield numerical stability and limited oscillations. (3-4) coarsening as suggested by Laurence seems to provide the best compromise and still yield reasonable grid sizes (< 1 Million nodes).

Different SGS models do not yield significant differences

Unstructured LES solver (CDP, Stanford) with the same numerical schemes and SGS as the reference structured LES yields similar results .

But all LES still show weaknesses in the laminar flow recirculation and especially in the transition process (we have as many bubble sizes as LES)

Yet in the trailing edge region, similar statistics are achieved and consequently same broadband noise prediction

wp6: cd airfoil test-case experimental and numerical data base

Documents