Simulations of LAGOON landing-gear noise using

AIAA 2015-2993 AIAA Aviation 22-26 June 2015, Dallas, TX 21st AIAA/CEAS Aeroacoustics Conference

Simulations of LAGOON landing-gear noise using Lattice-Boltzmann Solver Alois Sengissen∗ AIRBUS Operations SAS, 316 route de bayonne, 31060 Toulouse cedex 03, France

Jean-Christophe Giret†

Downloaded by PURDUE UNIVERSITY on August 3, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.2015-2993

CS, Systèmes d’ information, 22 avenue Galilée, 92350 Le Plessis Robinson, France.

Christophe Coreixas‡ and Jean-Fran¸cois Boussuge§ CERFACS, 42 Avenue G. Coriolis, 31057 Toulouse cedex, France.

This paper aims at investigating and analyzing numerical simulations of landing-gear configurations of increasing complexity using the Lattice-Boltzmann solver “LaBS”. The LAGOON (LAnding-Gear nOise database for CAA validatiON) project, supported by Airbus,1, 2 provides an accurate experimental database on simplified landing-gear configurations perfectly suitable for this purpose. First, an assessment of the numerical approach accuracy is carried out on LAGOON1 configuration by comparing both aerodynamic and near-field acoustic results with the LAGOON database disclosed in the frame of the NASA BANC workshop. Then, further investigations are focused on the influence of mesh refinement, subgrid scale model and wall law parameters. Finally, the best practices obtained are applied on LAGOON2 & 3 configurations and allow to capture the impact of some geometrical components added onto LAGOON1 baseline.

I.

Introduction

The understanding and the prediction of aircraft airframe noise is important nowadays to meet future aircraft noise requirements (ACARE 2020 goals3 ). As engine noise has been greatly reduced through the use of high bypass ratio engines, the design of new aircraft involves the prediction and if needed the reduction of airframe noise coming from the deployed landing-gear and high-lift devices. In particular, a deployed landing-gear contributes roughly to one third of the total acoustic energy emitted by the aircraft in approach conditions. In addition to wind tunnel experiments, computational aeroacoustics provides a promising path to better understand sound generation mechanisms and to speed up the design process of a quieter aircraft. In this perspective, the LAGOON (LAnding-Gear nOise database for CAA validatiON) project, supported by Airbus,1, 2 provides an accurate experimental database on simplified landing-gear configurations. The flows around three nose landing-gear configurations of increasing geometrical complexity (Fig. 1) have been studied in dedicated experimental campaigns measuring both aerodynamic and acoustic quantities (Fig. 2) at several operating points. A very significant effort has been done over the last years to investigate the potential of different numerical approaches and perform a step by step validation versus the LAGOON database, as shown by initiatives such as the NASA BANC-III workshop. ∗ Research

Engineer, Department of Acoustics & Environment, [email protected] ´ Engineer, BU Energie Industrie & Services, [email protected] ‡ PhD Student, CFD Team, [email protected] § Research Engineer & project leader, CFD Team, [email protected] † Research

1 of 22 American Institute of Aeronautics and Astronautics Copyright © 2015 by Airbus. Published by the American Institute of Aeronautics and Astronautics, Inc., with permission.

Main assessment criteria for industrial use are simple : 1. High accuracy in simulating noise generating and propagating mechanisms up to the far field. 2. Ability to handle seamlessly very complex geometries of realistic configurations. 3. Turnover time of the whole simulation chain in terms of human and CPU time.


In recent studies, the “classical” approach of solving unsteady Navier-Stokes equations using Detached Eddy Simulations (DES)4, 5 and Large Eddy Simulations6 (LES) on block-structured grids have shown a quite satisfying accuracy, but those methods are unable to tackle highly complex geometries despite chimera techniques. LES with high-order schemes on unstructured grids,7 has achieved an even better accuracy, with a great potential for complex geometries. But these meshing techniques remain very demanding when applied on a real landing-gear configuration. A breakthrough has been done using Lattice-Boltzmann Method8, 9 (LBM), demonstrating better turnover times without compromising the accuracy, regardless the geometrical complexity of the landing-gear. Thus, it has opened promising perspectives for realistic applications.10, 11 The aim of this paper is to go one step further than previous LBM computations8, 9 and focus on a sensitivity analysis regarding various numerical parameters such as the subgrid scale model or the wall law components. In this purpose the newly developed Lattice Boltzmann solver called “LaBS”12 will be used, as it allows the access to the sources of physical models, and gathers all numerical ingredients required for aeroacoustics.

Figure 1.

Geometrical configurations of the LAGOON database

a.

b.

Figure 2. Experiments performed by ONERA at a. open-section wind tunnel; courtesy of ONERA1, 2

F2 closed-section wind tunnel and b.

2 of 22 American Institute of Aeronautics and Astronautics

CEPRA19

II. A.

Numerical Approach

LaBS Solver

The Lattice Boltzmann solver named “LaBS”12 has been applied for all simulation results shown throughout this paper. This new code has been developed within a consortium of industrial companies (Renault, Airbus, CS), academic laboratories (UPMC, Ecole Centrale de Lyon) and strong partnerships with others entities (Onera, Alstom, Université Paris-Sud, Gantha, Matelys) in 2010-2014.


LaBS is built upon the classical Lattice-Boltzmann principles13 and uses a D3Q19 lattice, with a BGK collision model improved with regularization14 for a better robustness / accuracy tradeoff. Turbulence is handled according to LES approach,15 either with high-order explicit filtering (Approximate Deconvolution Model, ADM16 ) or with a dedicated subgrid scale model (Shear Improved Smagorinsky Model, SISM17 ). Near wall turbulence is modelled using wall laws accounting for adverse pressure gradient18 and curvature effects.19 Special care has been paid on two numerical ingredients crucial for aeroacoustic applications: • The dissipation and dispersion errors20 are kept as low as possible to enable proper generation & propagation of acoustic waves. • Among all available boundary conditions, absorbing layers can be enforced21 to avoid spurious acoustic waves reflections. More details about the LaBS solver as well as elementary validation cases can be found in recent publications.19 Solver performance data has been communicated in the frame of the BANC-III workshop.22 Resulting turnover time is almost an order of magnitude faster than other classical LES approaches. Ongoing code optimizations indicate that a supplementary factor 2 in code performance is reachable in the near future.

B.

Far-Field Noise Propagation

Despite the very low dissipation and dispersion properties of LaBS, it is well known that integral methods are more suitable to propagate acoustic waves in homogeneous media for long distances. Therefore, LaBS is being coupled with a Ffowcs-Williams and Hawkings23 (FW-H) solver. The code KIM from ONERA,24 already used extensively for airframe noise computations,7 has been employed in this purpose. The solver KIM uses an advanced-time formulation of the FW-H analogy developed by di Francescantonio25 which allows to perform the integration of the pressure field over solid and porous surfaces. For a proper comparison with CFD, experimental results have been corrected by ONERA from wind tunnel effects such as background noise, acoustic refraction and atmospheric absorption. More details can be found in the works of Manoha et al.2 and Sanders et al.5 Furthermore, these corrections have been applied in order to transpose the experimental propagating medium to an uniform one with the nozzle inlet flow conditions. Hence, the solid formulation of the FW-H analogy has to take into account the convective effects. Unfortunately, coupling works are still ongoing and thus far-field noise comparisons cannot be presented in the current paper which should be considered as a first stage of a broader assessment (see Section VII).


III. A.

Investigated Configurations & Numerical Setup

Configurations of Interest

Most of the works are focused on the LAGOON1 geometry (Fig. 1) as the related experimental database has been widely disclosed in the frame of the NASA BANC-III workshop. Some elements will be shown on LAGOON2 & LAGOON3 confidential configurations to demonstrate the abilities of the LBM approach to capture the impact of geometrical modifications (such as the tow bar or the wheel rim caps) on the flow-induced noise. The flow operating point closely matches the experiment conducted in the open-jet anechoic CEPRA19 wind-tunnel (Fig. 2-b). The landing-gear model is placed in a free flow with a velocity Uin = 78.99 m/s corresponding to a Mach number M a = 0.23. The static pressure is Pin = 99447 P a and the temperature is set on Tin = 293 K. The Reynolds number based on the wheel diameter D = 0.3 m is ReD = 1.541 · 106 . Flow operating conditions are the same for LAGOON1, 2 &3 configurations.


B.

Mesh Generation

In order to enable a fair comparison with previous studies (Block structured DES5 & Unstructured LES7 ), the meshing directives given to the LaBS solver have tried to reproduce similar spatial discretization of the flow. The main remaining difference being that the LaBS solver seamlessly generates the octree volume mesh ’on the fly’ in parallel at the beginning of each run in a few minutes, whereas classical LES or DES approaches require a more demanding meshing stage prior the computation. Such parallel octree volume meshing is one of the enablers to allow reaching seamlessly billions of cells in the very near future. Figure 3 compares local cell size (log scale) on cuts through the MEDIUM mesh obtained with LaBS with the ones already used in previous LES7 using the solver AVBP. The boundary layers and wakes are meant to be very similarly discretized. Table 1 summarizes the main parameters of the meshes used. LaBS meshing directives (relative cell sizes) are the same, so COARSE & FINE meshes are respectively 1.25 times coarser and finer than MEDIUM mesh everywhere in the computational domain. Mesh COARSE MEDIUM FINE

Number of cells 20 M 40 M 80 M

Smallest cell size (mm) 0.625 mm 0.5 mm 0.4 mm

Resolution zones depth 10 10 10

Time step (s) 1.057 · 10−6 s 8.413 · 10−7 s 6.731 · 10−7 s

Total time simulated (s) 0.3365 s 0.3365 s 0.3365 s

Signal processed (s) 0.24 s 0.24 s 0.24 s

Table 1. Main parameters of the meshes used in LaBS computations

C.

Computational Matrix

Table 2 gathers the run parameters employed for results on the LAGOON1 configuration shown in sections IV & V. Best practices will then be applied on the LAGOON2 & LAGOON3 configurations (section VI). This computational matrix will allow to assess back to back : • Turbulent subgrid scale modeling, with either explicit filtering (i.e. Approximate Deconvolution Model, ADM) or with a dedicated subgrid scale model (SISM). • The mesh convergence statistically speaking, even if refining will resolve finer turbulent structures. • Wall Law main components, by disabling one by one Curvature correction, then Pressure gradient correction, then disabling the wall law itself to obtain a bounce-back “no-slip” wall conditiona .

a Dirichlet

boundary condition is known to be inappropriate for such coarse wall resolution (Y + >> 10)



a.

b.

c. Figure 3. Meshes comparison between unstructured LES7 and current LBM setup; a. Cut at Y=0 mm; b. Cut at Z=0 mm; c. Close up on cut at Z=0 mm

Case Name

Mesh (Table 1)

Turbulence Model

Wall Law

Grad(P) correction

Curvature correction

ADM ADM SISM SISM SISM SISM SISM SISM

COARSE MEDIUM COARSE MEDIUM FINE MEDIUM MEDIUM MEDIUM

ADM ADM SISM SISM SISM SISM SISM SISM

Yes Yes Yes Yes Yes Yes Yes No

Yes Yes Yes Yes Yes Yes No No

Yes Yes Yes Yes Yes No No No

COARSE MEDIUM COARSE MEDIUM FINE MEDIUM NOCURV MEDIUM NOGRADP MEDIUM NOWL

Table 2. Main parameters of the runs used in LaBS computations on LAGOON1 configuration


IV.

LAGOON1 Baseline Assessment

Detailed validation is a prerequisite before moving to any parametric study on wall laws or subgrid scale modelling. The current section only presents ADM COARSE & ADM MEDIUM results (Table 2) as baseline computations. A.

Aerodynamic Results


Let’s focus first on comparing qualitatively (color maps) and quantitatively (profiles) the mean and RMS velocity fields. For conciseness sake, only axial velocity component is shown. Other components are available upon demand and have already been provided to the cross-plotting synthesis of the BANC III.22 The same layout is kept for Fig. 4 to 7. Flow main direction is from top to bottom so that wheels solid walls are slightly above the displayed image upper border. Figure 4 compares the mean axial velocity in the plane Z = 0 just in the wake of the landing gear wheels. The width and intensity of the wake as predicted by LaBS is in very good agreement with reference PIV. Refining from COARSE (Fig. 4-c) to MEDIUM mesh (Fig. 4-d) improves the thickness of the shear layers at the external edge of the wheels (x = −0.15 m; y = −0.15 m) and more globally yields a pattern closer to PIV, more symmetrical with more similar iso-contours.

a.

b.

c.

d.

Figure 4. Comparison of mean axial velocity fields with PIV reference in the plane Z = 0 mm; a. PIV Reference, b. Unstructured LES,7 c. LaBS COARSE, d. LaBS MEDIUM



Figure 5 exhibits the same level of agreement as observed in plane Z = 0 m. An additional refinement (case ADM MEDIUM) even allows to capture the low velocity region near (x = −0.4 m; y = 0 m). The wake width also matches PIV reference.

a.

b.

c.

d.

Figure 5. Comparison of mean axial velocity fields with PIV reference in the plane Z = −104 mm; a. PIV Reference, b. Unstructured LES,7 c. LaBS COARSE, d. LaBS MEDIUM

Figure 6 displays RMS axial velocity fluctuations. Unfortunately, PIV is a bit polluted by spurious optical reflexions. At least, velocity fluctuations patterns have similar aspects. LDV redundant system will allow to overcome this issue (Fig. 9) and to determine the outcome. Figure 7 leads to the same conclusions. PIV is unusable, and only observations can be drawn from comparison with reference unstructured LES.7



a.

b.

c.

d.

Figure 6. Comparison of RMS axial velocity fluctuations with PIV reference in the plane Z = 0 mm; a. PIV Reference, b. Unstructured LES,7 c. LaBS COARSE, d. LaBS MEDIUM

a.

b.

c.

d.

Figure 7. Comparison of axial velocity fluctuation (RMS) with PIV reference in the plane Z = −104 mm; a. PIV Reference, b. Unstructured LES,7 c. LaBS COARSE, d. LaBS MEDIUM 8 of 22 American Institute of Aeronautics and Astronautics

Now, assessing quantitatively the velocity profiles (Fig. 8) in the wake of the wheels with LDV reference confirms previous observation on color maps. The level of agreement is very good on ADM COARSE and becomes excellent on ADM MEDIUM thanks to a better prediction of the stiffness of the profiles (around y = 0.13 m in Fig. 8-a &b) and of the low velocity zone (around y=0 m in Fig. 8-d).

0.2

LDV X=160 Z=0 LaBS COARSE LaBS MEDIUM LES AVBP

0.0

0.0

-0.1

-0.2 20

40

60

80

0

Streamwise velocity (m/s)

a.

0.0

20

40

60

80

-0.2

100

0


b.

0.0

-0.1

-0.2

100

LDV X=128 Z=-104 LaBS COARSE LaBS MEDIUM LES AVBP

0.1

-0.1

-0.2 0

0.2


0.1

y (mm)

y (mm)

0.1

-0.1

20

40

60

80

100

0


c.

20

40

60

80

100


d.

Figure 8. Comparison of mean axial velocity profiles with LDV reference; a. x = 160 mm & z = 0 mm, b. x = 180 mm & z = 0 mm, c. x = 220 mm & z = 0 mm, d. x = 128 mm & z = −104 mm. LES results from7

Unlike PIV, LDV yields very clean and trustful RMS fluctuations velocity profiles (Fig. 9). The level of agreement of LaBS simulations, especially with MEDIUM mesh refinement is quite good. It is to be noted that the subgrid scale energy is not included in these plots, which could explain the slight underprediction in the regions where axial velocity fluctuations are intense (Fig. 9-a,b,c; −0.12 m < y < 0.12 m).

-0.1

0.0

-0.1

-0.2 5

10

15

20

25

30

Rms Streamwise velocity (m/s)

0.0

5

10

15

20

25

30


-0.2 0

c.

0.0

-0.1

-0.2 0

b.

0.1

-0.1

-0.2 0

LDV X=128 Z=-104 LaBS COARSE LaBS MEDIUM LES AVBP

0.2

0.1

y (mm)

0.0


0.2

0.1

y (mm)

0.1

a.


0.2

y (mm)


0.2

y (mm)


y (mm)

0.1

0.2


y (mm)

0.2

5

10

15

20

25

30


0

d.

5

10

15

20

25

30


Figure 9. Comparison of axial velocity fluctuation (RMS) profiles with LDV reference; a. x = 160 mm & z = 0 mm, b. x = 180 mm & z = 0 mm, c. x = 220 mm & z = 0 mm, d. x = 128 mm & z = −104 mm. LES results from7

The overall good agreement in the representation of the flow can be also observed on wall pressure coefficient. The comparison with pressure taps (Fig. 10) is excellent and even confirms the faulty sensor at angle 95o . 9 of 22 American Institute of Aeronautics and Astronautics

Wall Pressure Coefficient

1.0

Pressure Taps (Exp.) LES AVBP LaBS COARSE LaBS MEDIUM

0.5

0.0

-0.5

-1.0

a.

b.

-100

-50

0 Angle (Degree)

50

100


Figure 10. a. Pressure taps location; b. Comparison of wall pressure coefficient along wheel centerline with reference pressure taps measurements

B.

Near-Field Acoustic Results

Near-field acoustics are assessed under the light of Power Spectral Densities (PSD) at the wall. The same signal processing procedure has been applied for all numerical computations, including for reference LES.7 The pressure signal has been sampled at a frequency of 60 kHz for a duration of 0.24 s out of the 0.33 s simulated, and the spectral estimate26 built out of 5 averaged chunks with 50% of overlapping. Hann windowing has been considered as a standard. Figure 11 displays the location of Kulites sensors which will be considered as the reference for the frequency range 300 Hz − 6 kHz. Indeed, a low-cut filter was applied in the experiment (below 300 Hz), and obviously numerical cutoff will happen in the high frequency range due to limited grid resolution (likely above 6 to 10 kHz depending on kulite position). The comparison & analysis of Power Spectral Density (PSD) obtained will be gathered in 3 groups : 1. Figure 12 : The wheel centerline (Kulites K1 to K8), where the progressive growth of the turbulent boundary layer could be observed. 2. Figure 13 : The wheel flank (Kulites K13, K14, K15 & K20), with strong adverse pressure gradient and curvature where the tyre meets the rim of the wheel. 3. Figure 14 : The main leg (Kulites K24 to K27).

b.

a.

Figure 11. Location of Kulites on the a. Wheel, and b. Axle and Leg of LAGOON landing gear; Red : Unsteady Kulites sensors; Green : Pressure taps

Figure 12 demonstrates the ability of LaBS to capture the broadband noise very faithfully along wheel centerline. The numerical cutoff due to the mesh resolution occurs at 6 kHz for upstream angles (K1) and 10 of 22 American Institute of Aeronautics and Astronautics


goes way beyond 10 kHz for downstream angles (K8), which is much better than classical LES despite high-order schemes.27 The flow detachment between K4 & K5 induces a sudden rise of pressure fluctuation levels. Reference LES7 very likely predicts an early detachment whereas LBM remains in excellent agreement beyond K5. Tonal emergence at 1000 Hz and 1500 Hz are also very well predicted. Physical phenomena inducing these tones are very well explained in recent papers7–9 and will not be further investigated. Figure 13 leads to similar conclusions. Moreover, the prediction of the PSD remains very good in the region where the turbulent boundary layer meets strong adverse pressure gradients (K14). Section V will evaluate the main drivers through cases SISM MEDIUM NOCURV, SISM MEDIUM NOGRADP & SISM MEDIUM NOWL. The interpretation of Fig. 14 is harder to perform. Low frequency spurious tone on K24 is very likely caused by a flapping at the stagnation point due to the coarse resolution. The level of agreement on the overall frequency range is fair, but not as good as the one obtained on Figs. 12 & 13. However, the reduced dimension of the main leg may lead to a sensitive flow regime.28 Moreover, the local resolution of the mesh around the leg (100 cells per leg diameter in comparison with 750 cells per wheel diameter) may be slightly insufficient to capture the physics correctly. C.

Far-Field Acoustic Results

Acoustic post-processing are being conducted on each case simulated using the Ffowcs-Williams and Hawkings analogy24 described in section II-B using a solid surface formulation. The results will be compared in a future paper with the ones obtained in the CEPRA19 wind-tunnel in terms of PSD on both flyover & sideline arcs and integrated as OverAll Sound Pressure Level (OASPL).


110

100

100

Pressure PSD (dB/Hz)


110

90

80

70

Kulite 1 LES AVBP LaBS ADM_COARSE LaBS ADM_MEDIUM

60

2

3

4

5

6

7 8 9

2

3

3

4

5

6

7 8 9

10 Frequency (Hz)

10

4

110

110

100

100

90

80

70


3

4

5

6

7 8 9

3

2

3

4

5

6

7 8 9

10 Frequency (Hz)

10

4

80

70


60

2

2

3

4

5

6

7 8 9

2

3

3

4

5

6

10 Frequency (Hz)

7 8 9

10

4

10

d. 130

120

120



2

90

130

110

100

90


80

2

2

3

4

5

6

7 8 9

3

2

3

4

5

6

10 Frequency (Hz)

7 8 9

10

4

110

100

90


80

70

70 10

e.

2

50 10

c.

Kulite 2 LES AVBP LaBS ADM_COARSE LaBS ADM_MEDIUM 10

b.



2

50

2

2

3

4

5

6

7 8 9

2

3

3

4

5

6

10 Frequency (Hz)

7 8 9

10

4

10

f.

130

130

120

120




70

50 10

60

110

100

90


80

2

2

3

4

5

6

7 8 9

3

2

3

4

5

6

10 Frequency (Hz)

7 8 9

10

4

110

100

90


80

70

g.

80

60

50

a.

90

70 10

2

Figure 12. centerline

2

3

4

5

6

7 8 9

3

2

10 Frequency (Hz)

3

4

5

6

7 8 9

10

4

h.

10

2

2

3

4

5

6

7 8 9

3

10 Frequency (Hz)

2

3

4

5

6

7 8 9

10

4

Comparison of Power Spectral Density (PSD) at the wall with kulites measurements along wheel


130

110

120



120

100

90

80


70

100

90


80

60

70 10

a.

110

2

2

3

4

5

6

7 8 9

2

3

3

4

5

6

7 8 9

10 Frequency (Hz)

10

4

10

b.

2

2

3

4

5

6

7 8 9

3

2

3

4

5

6

7 8 9

10 Frequency (Hz)

10

4

130

120

120



100

90


80

100

90


70 10

c.

110

80

70 2

2

3

4

5

6

7 8 9

2

3

3

4

5

6

10 Frequency (Hz)

7 8 9

10

4

10

d.

2

2

3

4

5

6

7 8 9

3

2

3

4

5

6

10 Frequency (Hz)

7 8 9

10

4

130

130

120

120



Figure 13. Comparison of Power Spectral Density (PSD) at the wall with kulites measurements along wheel inner flank

110

100

90


80

2

90

2

3

4

5

6

7 8 9

2

3

3

4

5

6

7 8 9

10 Frequency (Hz)

10

4

Kulite 25 LES AVBP LaBS ADM_COARSE LaBS ADM_MEDIUM 10

b. 130

120

120


130

110

100

90


80

2

2

3

4

5

6

7 8 9

3

2

3

4

5

6

7 8 9

10 Frequency (Hz)

10

4

110

100

90


80

70

c.

100

70 10

a.

110

80

70



110

70 10

2

2

3

4

5

6

7 8 9

3

2

10 Frequency (Hz)

3

4

5

6

7 8 9

10

4

d.

10

2

2

3

4

5

6

7 8 9

3

10 Frequency (Hz)

2

3

4

5

6

7 8 9

10

4

Figure 14. Comparison of Power Spectral Density (PSD) at the wall with kulites measurements around landing gear main leg


V.

LAGOON1 Sensitivity Analysis

The objective of section V is to investigate the sensitivity of results along with key numerical parameters. Table 2 allows to compare back to back simulations with one single numerical parameter changing at a time. Simulation physical time and post-processing is the same for all runs. A.

Influence of the Subgrid Scale Modeling


The comparison of simulations with different subgrid scale models (ADM MEDIUM & SISM MEDIUM) did not reveal any major difference. Surprisingly, wall pressure spectra (Fig. 15) along landing-gear wheel centerline (kulites K1 to K8) collapse with each other. Mean and RMS flow fields also show extremely similar patterns (not shown here). A possible explanation of such a similar behaviour is that both subgrid scale models are very low dissipative and have minor impact on frequency range of interest. As the SISM subgrid scale model shows a better robustness throughout all computations, it must be considered as the best practice from now on. B.

Influence of the Grid Refinement

Section IV demonstrated the satisfying level of agreement of aerodynamic results for both COARSE and MEDIUM meshes with minor improvement in the landing gear wake for MEDIUM mesh. The comparison of simulations SISM COARSE, SISM MEDIUM & SISM FINE now allows to ensure grid convergence more carefully. Fig. 16 confirms that very minor differences can be observed on wall PSD : • Broadband noise levels for case SISM COARSE slightly exceeds the levels obtained from cases SISM MEDIUM & SISM FINE. • Peak around 1500 Hz emerges more for case SISM COARSE than for cases SISM MEDIUM & SISM FINE. Less turbulent structures of small sizes for case SISM COARSE might lead to more coherent eddies detachments. Thus, the flow could be prone to tonal resonances. Anyway, as the observed differences are subtle, any physical explanation is quite speculative. C.

Influence of the Wall Law

In experimental observations,1, 2 the separation of wheels boundary layers occurs around θ ≈ 110o (see Fig. 11 for angles convention). The influence of wall law components is much more visible under the light of this separation point prediction : • In terms of mean velocity field, Fig. 17 displays velocity magnitude on a cut through wheel central plane (y = 0.1 m). Disabling wall law components one by one (Runs SISM MEDIUM NOCURV, SISM MEDIUM NOGRADP, SISM MEDIUM NOWL) triggers earlier detachment of the flow than what is observed experimentally. Unfortunately, no PIV plane is available for comparison. • In terms of near field pressure fluctuations, Figure 18 shows the consequences of such an early separation point on wall pressure spectra. A fair agreement is still kept from K1 to K4 (where the flow is attached) in all simulated cases, no matter if wall law treatment is partially or even totally disabled (i.e. dirichlet no-slip condition). Then downstream, the detached flow leads to a very different spectral behaviour (i.e. strong overprediction of broadband pressure fluctuation level). Pressure gradient correction term18 has a less striking effect, but still contributes significantly to the good prediction of wall pressure spectra shape, matching with kulites measurements. Curvature correction term (SISM MEDIUM NOCURV) has shown negligible effect on the cases investigated with respect to fully enabled wall law treatment (SISM MEDIUM).


110

100

100



110

90

80

70

Kulite 1 LaBS ADM_MEDIUM LaBS SISM_MEDIUM

60

2

3

4

5

6

7 8 9

2

3

3

4

5

6

7 8 9

10 Frequency (Hz)

10

4

110

110

100

100

90

80

70


3

4

5

6

7 8 9

3

2

3

4

5

6

7 8 9

10 Frequency (Hz)

10

4

80

70


60

2

2

3

4

5

6

7 8 9

2

3

3

4

5

6

10 Frequency (Hz)

7 8 9

10

4

10

d. 130

120

120



2

90

130

110

100

90


80

2

2

3

4

5

6

7 8 9

3

2

3

4

5

6

10 Frequency (Hz)

7 8 9

10

4

110

100

90


80

70

70 10

e.

2

50 10

c.


10

b.



2

50

2

2

3

4

5

6

7 8 9

2

3

3

4

5

6

10 Frequency (Hz)

7 8 9

10

4

10

f.

130

130

120

120




70

50 10

60

110

100

90


80

2

2

3

4

5

6

7 8 9

3

2

3

4

5

6

10 Frequency (Hz)

7 8 9

10

4

110

100

90


80

70

g.

80

60

50

a.

90

70 10

2

2

3

4

5

6

7 8 9

3

2

10 Frequency (Hz)

3

4

5

6

7 8 9

10

4

h.

10

2

2

3

4

5

6

7 8 9

3

10 Frequency (Hz)

2

3

4

5

6

7 8 9

10

4

Figure 15. Comparison of Power Spectral Density (PSD) at the wall with kulites measurements along wheel centerline for two different subgrid scale models ADM & SISM


110

100

100



110

90

80

70

Kulite 1 LaBS SISM_COARSE LaBS SISM_MEDIUM LaBS SISM_FINE

60

2

3

4

5

6

7 8 9

2

3

3

4

5

6

7 8 9

10 Frequency (Hz)

10

4

110

110

100

100

90

80

70


3

4

5

6

7 8 9

3

2

3

4

5

6

7 8 9

10 Frequency (Hz)

10

4

80

70


60

2

2

3

4

5

6

7 8 9

2

3

3

4

5

6

10 Frequency (Hz)

7 8 9

10

4

10

d. 130

120

120



2

90

130

110

100

90


80

2

2

3

4

5

6

7 8 9

3

2

3

4

5

6

10 Frequency (Hz)

7 8 9

10

4

110

100

90


80

70

70 10

e.

2

50 10

c.

Kulite 2 LaBS SISM_COARSE LaBS SISM_MEDIUM LaBS SISM_FINE 10

b.



2

50

2

2

3

4

5

6

7 8 9

2

3

3

4

5

6

10 Frequency (Hz)

7 8 9

10

4

10

f.

130

130

120

120




70

50 10

60

110

100

90


80

2

2

3

4

5

6

7 8 9

3

2

3

4

5

6

10 Frequency (Hz)

7 8 9

10

4

110

100

90


80

70

g.

80

60

50

a.

90

70 10

2

2

3

4

5

6

7 8 9

3

2

10 Frequency (Hz)

3

4

5

6

7 8 9

10

4

h.

10

2

2

3

4

5

6

7 8 9

3

10 Frequency (Hz)

2

3

4

5

6

7 8 9

10

4

Figure 16. Comparison of Power Spectral Density (PSD) at the wall with kulites measurements along wheel centerline for different mesh refinements



a.

b.

c. Figure 17. Comparison of time-averaged velocity magnitude on wheel central plane (y = 0.1 m) with the various terms of wall law progressively disabled : a. SISM MEDIUM; b. SISM MEDIUM NOGRADP; c. SISM MEDIUM NOWL


110

100

100



110

90

80

70

Kulite 1 LaBS SISM_MEDIUM LaBS SISM_MEDIUM_NOGRADP LaBS SISM_MEDIUM_NOWL

60

2

3

4

5

6

7 8 9

2

3

3

4

5

6

7 8 9

10 Frequency (Hz)

10

4

110

110

100

100

90

80

70


3

4

5

6

7 8 9

3

2

3

4

5

6

7 8 9

10 Frequency (Hz)

10

4

80

70


60

2

2

3

4

5

6

7 8 9

2

3

3

4

5

6

10 Frequency (Hz)

7 8 9

10

4

10

d. 130

120

120



2

90

130

110

100

90


80

2

2

3

4

5

6

7 8 9

3

2

3

4

5

6

10 Frequency (Hz)

7 8 9

10

4

110

100

90


80

70

70 10

e.

2

50 10

c.


10

b.



2

50

2

2

3

4

5

6

7 8 9

2

3

3

4

5

6

10 Frequency (Hz)

7 8 9

10

4

10

f.

130

130

120

120




70

50 10

60

110

100

90


80

2

2

3

4

5

6

7 8 9

3

2

3

4

5

6

10 Frequency (Hz)

7 8 9

10

4

110

100

90


80

70

g.

80

60

50

a.

90

70 10

2

2

3

4

5

6

7 8 9

3

2

10 Frequency (Hz)

3

4

5

6

7 8 9

10

4

h.

10

2

2

3

4

5

6

7 8 9

3

10 Frequency (Hz)

2

3

4

5

6

7 8 9

10

4

Figure 18. Comparison of Power Spectral Density (PSD) at the wall with kulites measurements along wheel centerline with the various terms of wall law progressively disabled


VI.

LAGOON2 &3 Perspectives

Simulations on LAGOON2 & LAGOON3 configurations (see Fig. 1) have been conducted using the best practices established in sections IV & V. This section aims at illustrating the abilities of the current Lattice-Boltzmann Method to capture physical mechanisms such as: • The noise induced by vortex shedding from the tow bar (LAGOON2). • The additional broadband noise emitted from the wheels caps removal (LAGOON3). As these configurations have not been publicly disclosed, only very selective elements will be displayed.

Impact of the Tow Bar

A tow bar located upstream the gear axle, right between the wheels, is one of the main geometrical elements added to configuration LAGOON2 (see Fig. 1) with respect to LAGOON1 configuration. Experiments reveal the emergence, from the broadband noise, of an extra tone around 1100 Hz. This frequency peak can be observed on both the most upstream kulites (K1 to K4) and in the far field. Previous investigations29 allowed us to related it to the tow bar vortex shedding frequency. The vortex shedding mechanism is confirmed by present works by plotting an isosurface of Q criterion restricted to the region located between the wheels. The turbulent structures roll-up is clearly visible in Fig. 19.

10dB

10dB

Isosurface of Q criterion in the vicinity of tow bar (left wheel hidden) for LAGOON2 configuration

Pressure PSD (Arbitrary units)

Figure 19.

Pressure PSD (Arbitrary units)


A.

Kulite 1 LaBS SISM_FINE LaBS SISM_MEDIUM

a.

10

2

2

3

4

5

Kulite 2 LaBS SISM_FINE LaBS SISM_MEDIUM 6

7 8 9

3

2

10 Frequency (Hz)

3

4

5

6

7 8 9

10

4

b.

10

2

2

3

4

5

6

7 8 9

3

10 Frequency (Hz)

2

3

4

5

6

7 8 9

10

4

Figure 20. Comparison of Power Spectral Density (PSD) at the wall with kulites measurements in the vicinity of tow bar (a. Kulite K1 & b. Kulite K2) for LAGOON2 configuration.

Figure 20 exhibits the level of agreement between LBM computations and experimental data for kulites K1 & K2. Even if the simulation using MEDIUM mesh does not capture the shedding peak, FINE mesh brings additional wall grid refinement and the vortex shedding peak appears. However, its shedding frequency


(1200 Hz) slightly differs from the experimental one (1100 Hz). The Reynolds number based on the tow bar diameter (Red ≈ 500 000) indicates that the flow is in critical regime, which may explain the frequency missmatch. Overall broadband noise (wall pressure fluctuations) also shows a good level of agreement with experiment. Capturing the shedding peak thanks to this supplemental refinement make sense as the number of grid points per tow bar diameter is quite limited (of the order of 40 for MEDIUM mesh) B.

Influence of the Wheel Rim Caps

Figure 21.

Isosurface of Q criterion for LAGOON3 configuration

10dB

A direct influence has been observed on the wall pressure spectra. Kulite 17, which is located at the center of wheel outer flank (see Fig. 11-a), has evidenced a slight shift (few dB) of broadband noise levels over the low to mid frequency range. The current LBM approach captures this effect successfully and matches experiment as shown in Fig. 22.



The behaviour of the flow around LAGOON3 configuration remains very close from the one observed on LAGOON2 configuration (typical bluff-body flow). However, removing the side caps from the outer flank of the wheels leads to earlier detachment and faster development of turbulent structures in the boundary layer. Figure 21 illustrates this phenomenon with an isosurface of Q criterion.

Kulite 17 LaBS SISM_FINE LaBS SISM_MEDIUM

10

2

2

3

4

5

6

7 8 9

3

10 Frequency (Hz)

2

3

4

5

6

7 8 9

10

4

Figure 22. Comparison of Power Spectral Density (PSD) at the wall with kulites measurements along wheel outer flank (K17) for LAGOON3 configuration


VII.

Conclusion & Future Works

Lattice-Boltzmann simulations have been performed using the solver ’LaBS’12 on all LAGOON configurations. The assessment of code accuracy on the baseline LAGOON1 configuration is globally very satisfactory: excellent agreement has been observed on both aerodynamics (Mean & RMS velocity fields) and near field acoustics (Pressure PSD at the wall). On some aspects of the validation process, LBM accuracy even outperforms classical LES approaches. Complementary investigations have been carried out on LAGOON1 to analyse the sensitivity of the simulations to various numerical parameters. The latter have established that :


• Results are almost insensitive to subgrid scale modeling (SISM vs ADM). This is not a general statement, but rather an observation specifically related to the current solver implementation where both approaches are not too dissipative. • Mesh convergence seems to be reached with MEDIUM mesh. Such a resolution is sufficient to capture flow and wall pressure fluctuations for LAGOON1 configuration. However, for LAGOON2 configuration, the vortex shedding from the smaller tow bar (see section VI-A) requires extra refinement to allow the peak emergence. This should be considered as standard meshing criteria to tackle realistic geometries (including springs, dressings and other small components). • Wall treatment is a crucial aspect of LBM modeling (and more generally of wall-modeled LES). On landing-gear wheel, the use of a basic log-law instead of dirichlet ’no-slip’ condition is of the utmost importance but is not sufficient. Additional correction terms are needed to take into account mean pressure gradients and curvature breaks so that separation point is faithfully predicted. Considering the major impact observed on wall PSD, capturing the separation point is a mandatory stepstone before assessing far-field pressure fluctuations. Best practices have been successfully applied on both LAGOON2 &3 configurations. The tow bar and wheels rim caps effects have been evidenced (among other components contributions), and LBM simulation successfully captures their influence on the baseline configuration. However, the one and only determining factor for aeroacoustics simulations is far-field noise predictions. Developments & validations are ongoing to couple the current LBM solver with Ffowcs-Williams & Hawkings integral method23–25 before focusing on far-field PSD and directivity patterns (OASPL) assessment. Then, application on more realistic geometries at full scale will be the next challenge.

References 1 Manoha, E., Bult´ e, J., and Caruelle, B., “LAGOON : An Experimental Database for the Validation of CFD/CAA Methods for Landing Gear Noise Prediction,” AIAA Paper 2008-2816, 14th AIAA/CEAS Aeroacoustics Conference, Vancouver, Canada. May 5-7 , 2008. 2 Manoha, E., Bult´ e, J., Ciobaca, V., and Caruelle, B., “LAGOON: further analysis of aerodynamic experiments and early aeroacoustics results,” AIAA Paper 2009-3277, 15th AIAA/CEAS Aeroacoustics Conference, Miami, USA. May 11-13 , 2009. 3 “Advisory Council for Aeronautics Research in Europe,” http://www.acare4europe.com, 2020. 4 Sanders, L., Manoha, E., Khelil, S. B., and Francois, C., “LAGOON : CFD/CAA Coupling for Landing Gear Noise and Comparison with Experimental Database,” AIAA Paper 2011-2822, 17th AIAA/CEAS Aeroacoustics Conference, American Institute of Aeronautics and Astronautics, 2011. 5 Sanders, L., Manoha, E., Khelil, S. B., and Francois, C., “LAGOON: New Mach Landing Gear Noise Computation and further analysis of the CAA process,” AIAA Paper 2012-2281, 18th AIAA/CEAS Aeroacoustics Conference, 2012. 6 Liu, W., Kim, J. W., Zhang, X., and Caruelle, B., “Simulation of a Generic Two-Wheel Nose Landing Gear Using Highorder Finite Difference Schemes,” AIAA Paper 2012-2278, 18th AIAA/CEAS Aeroacoustics Conference, American Institute of Aeronautics and Astronautics, 2012. 7 Giret, J.-C., Sengissen, A., Moreau, S., and Jouhaud, J.-C., “Prediction of LAGOON landing-gear noise using an unstructured LES Solver,” AIAA paper 2013-2113, 19th AIAA/CEAS Aeroacoustics Conference, Berlin, Germany. May 27-29 , 2013. 8 Ribeiro, A., Casalino, D., Fares, E., and Noelting, S., “CFD/CAA Analysis of the LAGOON Landing Gear Conguration,” AIAA Paper 2013-2256 , 2013. 9 Casalino, D., Ribeiro, A. F., Fares, E., and N¨ olting, S., “Lattice-Boltzmann Aeroacoustic Analysis of the LAGOON Landing-Gear Configuration,” AIAA Journal, 2014, pp. 1–17. 10 Casalino, D., Ribeiro, A., Fares, E., N¨ olting, S., Mann, A., Perot, F., Li, Y., Lew, P., Sun, C., and Gopalakrishnan, P., “Towards Lattice-Boltzmann Prediction of Turbofan Engine Noise,” AIAA Paper 2014-3101, 15th AIAA/CEAS Aeroacoustics Conference, Atlanta, USA. June 16-20 , 2014.



11 Khorrami, M. R., Fares, E., and Casalino, D., “Towards Full Aircraft Airframe Noise Prediction: Lattice Boltzmann Simulations,” AIAA Paper 2014-2184, 15th AIAA/CEAS Aeroacoustics Conference, Atlanta, USA. June 16-20 , 2014. 12 “Lattice-Boltzmann Solver LaBS,” http://www.labs-project.org, 2014. 13 Chen, S. and Doolen, G., “Lattice Boltzmann Method for fluid flows,” Ann. Rev. Fluid Mech. , Vol. 30, 1998, pp. 329–364. 14 Latt, J. and Chopard, B., “Lattice Boltzmann method with regularized pre-collision distribution functions,” Mathematics and Computers in Simulation, Vol. 72, No. 2, 2006, pp. 165–168. 15 Sagaut, P., Large Eddy Simulation for incompressible flows, Scientific computation series, Springer-Verlag, 2000. 16 Malaspinas, O. and Sagaut, P., “Advanced large-eddy simulation for lattice Boltzmann methods: The approximate deconvolution model,” Phys. Fluids , Vol. 23 - 105103, 2011. 17 L´ evˆ eque, E., Toschi, F., Shao, L., and Bertoglio, J.-P., “Shear-improved Smagorinsky model for large-eddy simulation of wall-bounded turbulent flows,” J. Fluid Mech. , Vol. 570, 2007, pp. 491–502. 18 Afzal, N., “Wake layer in a turbulent boundary layer with pressure gradient- A new approach,” IUTAM Symposium on Asymptotic Methods for Turbulent Shear Flows at High Reynolds Numbers, Bochum, Germany, 1996, pp. 95–118. 19 Touil, H., Ricot, D., and Leveque, E., “Direct and large-eddy simulation of turbulent flows on composite multi-resolution grids by the lattice Boltzmann method,” J. Comput. Phys. , 2014, pp. 220–233. 20 Ricot, D., Mari´ e, S., Sagaut, P., and Bailly, C., “Lattice Boltzmann method with selective viscosity filter,” J. Comput. Phys. , Vol. 228, 2009, pp. 44784490. 21 Xu, H. and Sagaut, P., “Analysis of the absorbing layers for the weakly-compressible lattice Boltzmann methods,” J. Comput. Phys. , Vol. 245, 2012, pp. 14–42. 22 Manoha, E. and B, C., “Summary of the LAGOON Solutions from the Benchmark problems for Airframe Noise Computations-III Workshop,” 16th AIAA/CEAS Aeroacoustics Conference, Dallas, USA. June 20-25 , 2015. 23 Ffowcs-Williams, J. and Hawkings, D., “Sound generation by turbulence and surfaces in arbitrary motion,” Philosophical Transactions of the Royal Society of London. Series A, Mathematical and Physical Sciences, Vol. 264, No. 1151, 1969, pp. 321– 342. 24 Rahier, G., Prieur, J., Vuillot, F., Lupoglazoff, N., and Biancherin, A., “Investigation of integral surface formulations for acoustic post-processing of unsteady aerodynamic jet simulations,” Aerospace science and technology, Vol. 8, No. 6, 2004, pp. 453–467. 25 Francescantonio, P. D., “A New Boundary Integral Formulation For The Prediction Of Sound Radiation,” J. Sound Vib. , Vol. 202, No. 4, 1997, pp. 491 – 509. 26 Welch, P. D., “The use of FFT for the estimation of power spectra,” IEEE transactions on audio and electroacoustics, Vol. 15, No. 2, 1967. 27 Colin, O. and Rudgyard, M., “Development of high-order Taylor-Galerkin schemes for unsteady calculations,” J. Comput. Phys. , Vol. 162, No. 2, 2000, pp. 338–371. 28 Blevins, R. D., Flow-induced vibration, Vol. 1, 1977. 29 Giret, J.-C., Simulations aux grandes ´ echelles des ´ ecoulements instationnaires turbulents autour des trains d’atterrissage pour la pr´ ediction du bruit a´ erodynamique, Ph.D. thesis, Toulouse, INPT, 2014.

Acknowledgments The authors want to thank B. Caruelle from Airbus Operations SAS and E. Manoha for the LAGOON database initiative.1, 2 D. Ricot & P. Bobillier from Renault and B. Gaston & R. Cuidard from CS are greatly acknowledged for their support on Lattice Boltzmann Solver “LaBS”.12 The authors want to mention G. Rahier from ONERA for providing the FW-H code KIM.24 All numerical simulations have been conducted on Airbus’ HPC resources. Present work has been supported by French funded projects LABS & CLIMB in the frame of the “Programme d´ Investissement d´ Avenir ’Calcul Intensif et Simulation Numérique’“.
