Aeroacoustics Simulation of an Automotive a-Pillar Rain Gutter

Aeroacoustics Simulation of an Automotive A-Pillar Rain Gutter

Herv Dechipre and Dr. Michael Hartmann Group Research, Automotive Techniques, CAE Volkswagen AG, Wolfsburg, Germany

SYNOPSIS The recent successes achieved to reduce tire and engine noise have resulted in a higher contribution of the wind noise in the overall perception of noise by car passengers. Up to now, wind noise has been largely assessed by wind tunnel testing; however there is an increasing need for numerical methods in order to evaluate the design as early as possible. If direct noise computation (DNC) is not reachable for industrial applications, hybrid methods based on acoustical analogy or stochastic modelling have already demonstrated good aptitudes. Nonetheless, as a first step, an accurate CFD simulation must be performed. In this study, the flow-induced noise of an automotive rain gutter has been investigated for Reynolds numbers based on the rain gutter height between 20 000 and 130 000. As both rain gutters height and oncoming flows direction vary along the A-pillar, two configurations have been designed to investigate these effects separately. In addition, two different profile sections of rain gutter were tested. This paper mainly focuses on the computation of the structures of the flow and surface pressure level; the propagation of sound in far field will be addressed in a further publication. The topology of the flow was assessed using steady RANS computation. Unsteady SAS-SST and DES models have been used to compute surface pressure fluctuations. As the flow was stable for the conventional SAS-SST model, forcing terms were used to switch to unsteady mode. Experimental data will be presented and used to validate the results of the numerical simulations.

1. Introduction

Various aspects determine the comfort inside a vehicle. One of them is the interior noise experienced by the passengers. Due to successes in reducing motor and road/tire noises, airflow induced noise (also called wind noise) becomes more and more important. Over 100 km/h wind noise is generally the dominant noise source and can make it difficult to converse or listen to the radio, but it can also add fatigue on a long highway trip. A potential buyer might even consider high wind noise levels as poor design or build quality and may lead to dissatisfied customers. Therefore, car manufacturers have to pay close attention to wind noise and try to minimise it. If one could expect that low drag vehicles would also have low wind noise level; this is not found to be true in practice. One explanation for this lack of correlation comes from the fact that aerodynamic drag depends largely of the airflow over the rear of the vehicle and its wake

EASC 20094th European Automotive Simulation Conference

Munich, Germany6-7 July 2009

1

while interior wind noise depends largely on details of the exterior airflow around the A-pillar and windscreen. Small openings or gaps around doors, windows, or many other components on the outer body such as roof-racks or side-view mirrors (or under floor) may largely contribute to wind noise; conversely they will have little or no effect on aerodynamic drag. A prominent feature contributing to the overall wind noise level in the cabin is the A-pillar rain gutter (fig. 1). Like a long narrow channel along the A-pillar, its purpose is to collect and drain the water that would otherwise flow from the windshield and past the A-pillar along the side windows; reducing then the visibility.

Fig. 1: Streamlines over a A-pillar

In the last decades, car manufacturers have massively invested to build aero-acoustic wind tunnel to investigate wind noise, however to cope with the continual reduction of cost and time development, there is an increasing need for effective numerical methods to simulate noise induced by external and internal flows and which could be systematically used in the early design phase of a vehicle. Recently, many works have been done to tackle the simulation of the flow induced noise over side view mirrors [1-2] and the pressure fluctuations on the cars lateral windows [3-4]. Few works have already addressed the problem of simulating the noise generated by a simplified rain gutter using a facing elbow mounted on a flat plate [5-6] or more simply: a forward facing step [7-9]. However in most of these works, only one geometry and only constant height of step leading to quasi 2D configurations were generally considered. Firstly, the different configurations tested will be presented as well as some experimental results. Methods and set-up used for the simulations carried out with the commercial Softwares ANSYS CFX and Fluent will then be presented. Finally, an overview of the topology of the flow will be addressed and followed by some comments on the results.

2. Definition of the Test case and Experimental set-up

2.1. Profiles and configurations

Two different rain gutter section profiles were tested, Fig.2.a. The first profile presents a higher water channel and a sharper edge. Shown on fig 1 and 2, the rain gutters were mounted on a flat plate and extended at their back by another plate. On a real vehicle, the height of the rain gutter is generally not constant but varies along the A-pillar. Thick at it bases, it becomes thinner approaching the roof. The streamlines on a car windscreen (fig.1) illustrates also that the angle of the oncoming flow with the rain gutter



2

varies along the A-pillar. Almost perpendicular (90) at the base of the A-pillar, the oncoming flow can reach an angle with the rain gutter of 30 near the roof. In order to evaluate these two parameters independently, two configurations were designed. Fig.2.b represents a rain gutter with a height varying linearly from 10 to 42 mm (called Linear). The configuration shown in fig.2.c (designated as Curved) was designed to study the impact of different flow angles on the rain gutter. The angle between the flow and the rain gutter varies from 90 to 35. It then leads to 4 cases (2 profiles 2 configurations).

Fig. 2: Geometry of the rain gutter Configurations

2.2. Experimental Set-up

The 4 cases were tested in the Aeroacoustic wind tunnel in Braunschweig (AWB) from the German aerospace centre (DLR). The wind tunnel has an open jet section of 1.20.8 m and the collector has a section of 3 3 m [10].

Fig. 3: Experimental Set-up for the 2 configurations (Linear and Curved)

Shown in red on fig 3, the experimental models were equipped of 3 rows of flush-mounted microphones (1/4 and 1/8 inches) to investigate the effect of the height of the rain gutter and the direction of the oncoming flow. Five flush-mounted microphones per row were used in the linear cases. Additional flush-mounted microphones were placed in the perpendicular direction to the rain gutter for the curved configuration. The local angles investigated were 90, 60 and 40. In addition, 10 free field microphones were used to assess the far field noise. The signals at the microphones were registered during 22s with a sampling frequency of 45 kHz. The signal of the far field microphones was corrected using the shear layer correction of Amiet [11].



3

The experimental spectra obtained for the 5 flush-mounted microphones on the medium section of the linear configuration for both profiles are presented on fig. 4. The 2 furthest upstream microphones (Pos. 1 and 2) present for both profiles almost identical spectra. Only directly in front of rain gutter at position 3, the measured noise level is about 3 dB higher for the first profile. The main differences are found behind the rain gutter. The first profile presents its maximum surface pressure level (SPL) directly after the rain gutter (Pos. 4) while the noise level drops down after the second profile. As additional information, the first profile was measured to be 13 dB noisier than the second one in the far field (at about ~ 1,8 m).

Fig 4: Experimental narrow band spectra for both profiles at flow stream velocity of 45 m/s.

3. Numerical Methodology

Steady and unsteady simulations using different models were used to assess the characteristics of the flow over the different profiles and configurations. After briefly presenting the models and methods, a description of the meshes and set-up will be proposed.

3.1. Turbulence Modelling

The main topology of the flow was assessed by performing steady Reynolds Averaged Navier-Stokes (RANS) model using the SST k- turbulence model with the Software ANSYS CFX [12]. In order to get surface pressure fluctuations, transient computations need to be performed. The most widely accepted unsteady approach is certainly the Large Eddy Simulation (LES) where the Navier Stokes equations are filtered using a spatial operator with a filter-width proportional to the local space grid spacing. This later aspect makes a direct connection between the level of resolution of turbulence scales and the mesh refinement, and can make the method very expensive for certain complex industrial applications. In attempt to overcome this restriction, Spalart [13] proposed a hybrid method which combines features of classical RANS formulations with elements of the LES methods. This concept called Detached Eddy Simulation (DES), is intended to take advantage of both methods by covering the boundary layer by a RANS model and switching into a LES mode in detached regions. This allows the calculation to capture the instability of the shear layer, and the development of the coherent structures in the wake, with more accurate prediction of the unsteady forces than can be obtained by steady or unsteady RANS methods. A major drawback of the method is the explicit grid dependency of the method which can lead to a premature flow separation. A solution was proposed by Spalart [14] so called Delayed DES to overcome this problem. The switch for the DES is achieved by comparing the modelled



4

turbulent length scale and the grid spacing and is obtained by correcting the destruction term as follows:

= k k FDES with FDES = max

1,

DES

t

CL

where Lt is the turbulent length scale predicted by the RANS model, is the local grid spacing and and CDES are 2 constants.

As an alternative approach, Menter and al. [15] have investigated the development of an improved URANS method which can provide a LES-like behaviour in detached flow regions. The scale adaptive simulation (SAS) concept is based on the introduction of the von Karman length scale into the turbulent equation and defined as follow in 1D:

22'' //

yUyU

USLvK

==

where S is the absolute value of the strain-rate and = 0.41.

The information provided by the von Karman length scale allows the SAS models to dynamically adjust to resolved structures in a URANS simulation, which results in a LES-like behaviour in unsteady regions of the flow field. At the same time the model provides standard RANS capabilities in stable regions. It allows the development of a turbulent cascade up to the grid resolution into the detached regions without or small grid dependency.

3.2. Meshing and set-up

The geometry for the different configurations were realised with Catia V5 before to be imported into ANSYS IcemCFD to be meshed. Block-structured meshes composed of hexahedral cells were realised. To get information on the flow topology over both profiles, steady RANS and unsteady DES computations were firstly performed on a small part of the rain gutter with constant height. Steady RANS computations were then realised on the 4 cases. Finally, unsteady simulations using different turbulence models (SAS-SST, DES and LES) were carried out on profile 1 with linear configuration.

3.2.1. Comparison between profiles 3D geometries with constant height

Fig 5: Computational domain and meshes

To get a first understanding of the flow over the two profiles and the main differences between them, fluid simulations were performed on an extrusion over 40 mm in the span-

0,55 m

0,15 m

0,04 m



5

wise direction for both profiles. The entire meshes consist in 3 400 000 nodes. The resolutions in the span-wise direction are for the first and second profile 0.5 mm and 0.33 mm respectively to ensure the coherence of the structure. At the inlet, a free stream velocity of 45 m/s (162 km/h) corresponding to a Reynolds number based on the height of the rain gutter and a Mach number of 60 000 and 0.13 respectively and a medium turbulence intensity were imposed. Pressure outlet was applied at the outlet, while symmetry was applied at the top and periodic conditions were applied on the sides. The boundary layers was fully resolved with a y+ = 1, with a first cell of 12 m at the wall. The height of the rain gutter is about 20 mm. The time-step used for the unsteady DES simulations was 5 s.

3.2.2. Steady 3D cases

To get information on the flow topology, steady RANS computations were performed on the 4 cases. The tunnel open jet section and the space around the model were reproduced for the computation. The mesh was then composed of 2 domains linked by a GGI interface. The core domain was meshed with a low Reynolds mesh, while a coarse mesh was used for the outer domain. Fig 6 and Table 1 gives information on the computational domain and the size of the meshes. An averaged velocity profile was given at the inlet to get a boundary layer thickness in front of the rain gutter in the same order as in the experiment. Three inlet velocities were tested (30, 45 and 60 m/s); table 2 gathered the Reynolds number corresponding.

Fig. 6: Computational domain dimensions and Mesh Overview

Table 1: Number of nodes for each configuration Profile 1 Profile 2

Conf. Linear Inner: 4 520 000 nodes (6 070 000 in total) Inner: 4 900 000 nodes

(6 500 000 in total)

Conf. Curved Inner: 4 590 000 nodes (5 960 000 in total) Inner: 4 520 000 nodes

(5 910 000 in total)

Table 2: Reynolds number based on the rain gutters height for the 3 studied sections Inlet Velocity Rain gutter 1 Rain gutter 2

0,66 H1 H1 1,6 H1 0,61 H2 H2 1,67 H2 30 m/s 26 500 40 000 64 000 22 500 36 000 60 500 45 m/s 40 000 60 500 97 000 33 500 54 000 90 500 60 m/s 53 000 80 500 129 000 44 500 72 500 121 000



6

3.2.3. Unsteady 3D case (Profile 1 with height variation)

Due to the low resolution of the previous meshes in the span-wise direction, acceptable unsteady results could not be expected. At the opposite a complete refinement in the span-wise direction of the mesh would not be realistic as it would lead to far too high number of nodes. It was then decided to only compute the unsteady flow in a 100 mm wide box centred on the middle section of the rain gutter. Due to the slope, no periodic conditions or symmetry were acceptable; therefore the chosen alternative was to use interfaces. Three domains with GGI interfaces were then created. The core domain consists in a 7 000 000 hexahedral block structured mesh. The maximum size of the cells is 10.3 mm, allowing theoretically a spatial resolution of wavelength of 21 mm i.e. 16 kHz. The second and the outer domains were respectively composed of 1 120 000 and 273 000 nodes leading to an overall number of nodes of 8 400 000. A hybrid scheme was used for the convective terms, which automatically switches from second order upwind in RANS regions to second order central differencing in scale-resolved flow regions. The time-step used for the simulations was set to a value of 10 s, maintaining a Courant number around 1. Thus, an averaged velocity profile corresponding to 45 m/s, as well as a turbulence viscosity ratio given at the inlet, and an averaged static pressure at the outlet. The flow was simulated on about 50 ms.

Fig. 7: Computational CFD domain and zoom on the mesh in the core domain

4. Results and discussion

4.1. Comparison between the two profiles

Fig 8: representation of the structure for both profiles with a stream flow at 45 m/s.



7

Fig 8 gives a representation of the unsteady (computed with DES) and a steady flow for both profiles. The pressure field is represented on the side. Figures b) and d) represent the mean flow using streamlines and the variable Q at 5106 s-2 was used to represent the turbulent structures at a time step on (a) and (c). The variable Q is defined as follow:

( )ijijijiji

j

j

i SSx

u

x

uQ =

=

21

21

As first comment, the site of the recirculation area is bigger for the first profile. The level of turbulence and pressure fluctuations appear also smaller behind the second profile. This correlates well with the experimental results and tends to confirm that the first profile is noisier than the second one.

4.2. 3D Steady Topology

The friction line patterns of the 3D linear configurations for the part with the highest section for both profiles obtained experimentally (oil visualisation) and from the simulations are represented on fig.9. It can be seen that as expected, the size of recirculation area in front and behind the rain gutter increase almost linearly with its height. Close to the border of the rain gutter (right side), the streamlines are highly deviated to the exterior showing another 3D effect. Compared to the previous quasi 2D cases, the plate behind the rain gutter was a bit higher, reducing the size of the recirculation and leading for the 2nd to almost no recirculation behind the rain gutter. On the 2nd profile, one also see the reattachment line occurred on the middle of the thicker part of the rain gutter which is confirmed by the flow visualisation.

Fig 9: Friction line pattern of the 3D linear configurations of both profiles from experiments and numerical simulation

Fig 10 and Fig 11 present some more representation of the flow for the different configurations. The red lines on fig. 10 represented the investigated section also presented on fig. 11. A recirculation area having a length of about 6h (h being the rain gutters height) could be found for the profile1. This value agrees well with common value of 5 to 8 found in literature for forward facing step. Looking from the top, it can also be seen that low transverse effects occur in the linear cases while for the curved configuration a slight deviation of the flow can already been seen before the rain gutter.



8

Fig. 10: Flow Topology of 3 different configurations: Profile 1 Linear (top, left), Profile 2 Linear (top, right), Profile 1 Curved (bottom, left)

On the curved configuration, the inflow angle tends to reduce the size of the recirculation and one can notice the formation of the vortex on this configuration along the rain gutter which is not without reminding the A-Pillar vortex. One can finally notice that the height of the boundary layer in front of the rain gutter is almost as high as the rain gutter itself.

Fig 11: Representation of the evolution of the flow topology against the height (Profile 1 and 2) and against the flow angle and the inlet velocity for the profile 1

4.3. Unsteady 3D case 4.3.1. 3D flow structures

According to Rung et al. [16], the numerical simulation requires ~12 time steps per oscillation period to accurately represent the harmonic disturbances which gives here a limitation of a maximum frequency of 8 333 Hz. The minimum frequency expected is 20 Hz (T = 50ms). Simulations with SAS-SST, DES and LES models conducted on the same grid. Using the conventional SAS-SST implemented in ANSYS CFX, the simulations was enable to switch to any unsteady mode and always results in a steady solutions (fig. 13 a). In comparison using DES or LES models, the structures behind the rain gutter could be resolved (fig.14). Looking



9

at the DES blending functions, it was also controlled that the core domain was running in LES mode around the rain gutter. Compared to the simulations on the quasi 2D cases (fig. 8) also computed with DES, the structures in front of the rain gutter could not be so well caught although similar span-wise resolution was used. One element of explanation for this difference can be the higher thickness of the boundary layer in this case which tends to damp the instabilities and does not generate resolved turbulence before the rain gutter.

Fig. 13 Visualisation of the turbulent structures using isosurface of Q (Q =5106 s-2) shaded with velocity simulated with the SAS model without (left) and with (right) forcing

Fig. 14 Visualisation of the turbulent structures using isosurface of Q (Q =5106 s-2) shaded with velocity simulated with DES (left) and LES (right)

This reason was also advanced to explain the unability of the SAS-SST model to switch to unsteady mode. To overcome this problem, it was decided to use the SAS-F model with forcing terms recently introduced by Menter [17]. The idea is to introduce forcing terms in the momentum equations in order to transfer modelled turbulence energy into the resolved energy for flows which does not exhibit sufficient strong instability to switch to unsteady mode. A volume stochastic source term and sink term are then respectively introduce in the momentum equations and modelled turbulent k in a confined user-specified flow region. In this test-case, the zone corresponds to the volume in the core domain in front of the rain gutter. The computation was done with ANSYS Fluent 12 and the results presented on fig13b shows that the structures in front and behind the rain gutter could then be resolved.

4.3.2. Surface Sound Pressure Level

The experimental and computed with DES and SAS-F spectra for 3 flush-mounted microphones are represented on fig. 15. Microphones 1 and 5 were outside the core domain. At point 2, both models could not catch the pressure fluctuations so far (83 mm) in front of the rain gutter confirmed by fig13 and 14. At point 3, the SAS-F presents a fair agreement with the experimental results while for the DES simulations, the spectrum is very low confirming the unability of the model to catch the fluctuations in front of the rain gutter. For



10

point 4, the SAS-F presents again a good agreement with the experiment. The results of the DES simulations are only up to 1-3 kHz in good agreement with the experiment. After 3 kHz, on all DES spectra, a brutal variation of the slope can be observed which might indicate a cut-off frequency due to the mesh (~ 1mm). This would also tend to confirm the lower dependency of the SAS model to the size of the mesh.

Fig 15: Spectra for the middle section (Umean =45m/s); Point 2 (left), Point 3 (Middle) and Point 4 (right)

5. Information Computational Aeroacoustics

The high resolution level and long simulation time required for getting the spectrum or the pressure fluctuations in the far field made the direct approaches not usable for industrial applications. Therefore hybrid approaches separating flow and acoustic simulation appears more appropriate for such application. This work will be the focus on a further paper which will be presented at the Euronoise conference 2009, however few insights will be presented here. + The first method already applied to a duct [18, 19] consists of firstly computing a steady

flow solution and then an unsteady aeroacoustic field by injecting a vortex [20] in the linearised Euler equations or to model the acoustic sources by a stochastic modelling [21].

+ The other alternative is to compute the acoustic field by using the results of the unsteady CFD simulations. Derived of the original approach developed by Sir Lighthill in 1952, the software Actran/LA [22] developed by FFT allows us to compute the Lighthill sources from the unsteady variables (p, , U) and then to compute the sound propagation. A coupling between CFX & Fluent with FFT Actran is already available.

The 2D sound field computed for both profiles are represented on fig 16. The first profile is the noisiest which is confirmed by the spectra computed for both geometries.

Fig. 16: Computed Acoustic field and Spectra using respectively injection vortex method and Stochastic modelling for both profiles (2D computations)



11

6. Conclusions In this study, the topologies of simplified 3D configurations of A-pillar rain gutter have been investigated using steady and unsteady flow simulations. The 1st profile presents the highest level of turbulence and the higher noise level. It was clearly shown the difficulty to get the unsteady flow structures particularly in front of the rain gutter. The addition of the forcing terms in the SAS-SST model turns out to be necessary in this case and allows getting good agreements with the experimental results. At the opposite, the DES simulations could only catch the structure after the separation. The authors would also point out that only short simulation times were performed and that longer simulation times would allow to smooth the spectra and to improve their quality to be compared with the experiment. Further work particularly on the acoustical part will conducted and will be presented in a further paper.

7. Acknowledgement The authors would like to thank M. Menter and M. Mller from ANSYS Germany for their advice and contributions in the CFD part as well as Prof. Delfs and M. Pott-Pollenske for their advice on the experimental part.

References [1] R. Siegert et al., Numerical Simulation of aeroacoustical sound generated by generic bodies

placed on a plate: Part II Prediction of radiated sound pressure AIAA Paper 99-1985, 1999 [2] H. Posson, F. Prot, Far Field evaluation of the noise radiated by a side mirror using LES and

acoustic analogy, 12th AIAA/CEAS Aeroacoustics Conference Massachusetts, 2006 [3] B. Arguillat et al., Measurements of the Wave-Number Frequency Spectrum of Wall Pressure

Fluctuations under Turbulent Flow, 11th AIAA/CEAS Aeroacoustics Conf. AIAA Paper, 2005 [4] S. Debert et al., Fluctuations under Turbulent Flows Enhanced Methods for Separation of

Propagative Wavenumbers form Wall Pressure Dataset, JASA, vol. 123-5, 2008 [5] A. Kumarasamy, K. Karbon, Aeroacoustics of an Automobile A-Pillar Rain Gutter: Computational

and Experimental Study, SAE Paper 1999-01-1128 [6] N.N. Automotive Rain Gutter Noise, Application briefs from FLUENT, EX227, 2005 [7] S. Becker et al., Experimental and Numerical Investigation of the flow induced noise from a

forward facing step, AIAA 2005-3006, 2005 [8] I. Ali et al., Aeroacoustic Study of a forward facing step using linearized Euler equations,

Physica D: Nonlinear Phenomena, Volume 237, Issues 14-17, Aug. 2008, pp 2184-2189 [9] Y. Bae et al., Investigation of Flow-Induced Noise from a Forward Facing Step, 19th Symposium

of Japan CFD, Tokyo 2005 [10] M. Pott-Pollenske, J. Delfs, Enhanced Capabilities of the Aeroacoustic Wind Tunnel

Braunschweig, 14th AIAA/CEAS AIAA-2008-2910, Aeroacoustics Conference, Vancouver, 2008 [11] R.K. Amiet, Correction of Open Jet Wind Tunnel Measurements for Shear Layer Diffraction,

AIAA 75-532, 1975 [12] F.R. Menter, Zonal 2-equation k- turbulence model for aerodynamic flows, AIAA 2906, 1993. [13] P.R. Spalart et al., Comments on the feasibility of LES for Wings and on a Hybrid RANS/LES

Approach, Advances in DNS/LES, Proc. 1st AFOSR Int. Conf. on DNS/LES, 1997. [14] P.R. Spalart et al., A new version of detached eddy simulation, resistant to ambiguous grid

densities. Theor. Comput. Fluid Dyn. 20: 181-195, 2006 [15] F.R. Menter et al., A Scale-Adaptive Simulation Model for Turbulent Flow Predictions, AIAA

paper 2003-0767, 41st Aerospace Science Meeting & Exhibit, Reno, Nevada, USA, 2003 [16] Rung et al., Sound radiated of the vortex flow past a generic side mirror, AIAA-2002-2340, 2002 [17] F.R Menter et al., Scale Adaptive Simulation with Artificial Forcing, 3rd Symposium on Hybrid

RANS-LES Methods, Gdansk, Poland, 2009 [18] H. Dechipre et al., Aeroacoustic Simulation Based on Linearised Euler Equations and Stochastic

Sound Source Modelling, Acoustics 08 Paris, pp 1071-1076, 2008 [19] M. Hartmann and H. Dechipre, Numerische Aeroakustik fr Anwendungen im Automobilbereich,

ATZ VDI-FVT-Jahrbuch 2009, pp 90-99, 2009 [20] R. Ewert, Broadband slat noise prediction based on CAA and stochastic sound sources from a

random particle mesh (RPM) method, Computers & Fluids, Vol. 37-4, May 2008, pp 369-387 [21] S. Caro et al. A New CAA Formulation based on Lighthills Analogy applied to an Idealized

Automotive HVAC Blower using AcuSolve and Actran/LA, AIAA05-3015



12

Aeroacoustics Simulation of an Automotive a-Pillar Rain Gutter

Documents

interior wind noise

interior noise

kmh wind noise

engine noise

low wind noise level

flowinduced noise

high wind noise levels

overall perception of