Acoustics a and nd T Turbule lenc nce: : Aerodyna ynami mics - - PowerPoint PPT Presentation

Acoustics a and nd T Turbule lenc nce: : Aerodyna ynami mics A Appli lications ns o

• f S

STAR- CCM+ CCM+

Milovan Perić

Use of STAR-CCM+ for aerodynamics applications Which turbulence model for which application? Simulation of acoustics phenomena with STAR-CCM+ “Best-practice” guidelines Examples of application Future developments

Int Introduction n

This presentation is based on reports prepared by CD-adapco experts for Vehicle Aerodynamics (Fred Ross), Defence and Aerospace (Deryl Snyder) and Acoustics (Fred Mendonca).

Vehicle aerodynamics (cars, trucks, sport vehicles) Train aerodynamics Aerodynamics of aircraft and rotorcraft Military applications (airplanes, missiles…) Flow around buildings etc. Main aims of simulation:

– Predict mean forces and moments (optimize geometry) – Predict unsteady loads (reduce vibrations) – Predict turbulence structure (minimize noise)

Us Use o

• f S

STAR-C

• CCM+ f

for A Aerodyna ynami mics

STAR-CCM+ offers many turbulence models (eddy-viscosity type, Reynolds-stress, transition, LES/DES…) CD-adapco collaborates with experts in academia to further develop turbulence models… Optimal model choice depends on flow under consideration and the aim of simulation… Eddy-viscosity type models are usually suitable to predict mean forces and moments… Reynolds-stress model predicts better flows with swirling and turbulence-driven secondary flows… LES/DES type models are capable of predicting all flow details (including acoustics), but are more costly…

Whi hich T h Turbule lenc nce M Model? l?

Coupled and segregated solver in STAR-CCM+ differ in discretization (results not the same)… Coupled solver is recommended for steady-state flows exhibiting strong coupling between variables (compressi- bility, buoyancy…). For transient flows, segregated solver is usually more efficient… It is also more accurate when computing propagation of acoustic waves… Double precision is sometimes important for acoustics computations…

Whi hich S h Solv lver T Typ ype?

Steady-state computations often do not fully converge… The reason is usually inherent local flow unsteadiness… Fine grids resolving details of geometry and 2nd-order discretization capture the flow instability… Averaging intermediate “solutions” over a range of iterations is unreliable (especially if residuals are high). Recommended approach:

– Switch to transient segregated solver; – Select time step to resolve the fluctuations of interest; – Average the result over few periods of oscillation…

Whi hich S h Set-U

• Up?

Overview of acoustics tools in STAR-CCM+

Acoustics i in S n STAR-C

• CCM+, I

, I

Ae Aero roaco coust stics ics Simu Simula latio ion Optio ions s St Steady y st state Tra ransie sient

Bro Broadband Corre rrela latio ions s Syn Synthesize sized Flu luct ctuatio ions s SN SNGR CURLE CURLE

su

surf rface ce PR PROUDMAN MAN

vo

volu lume me GOLDST STEI EIN 2D-a

• axi

xi LEE EE Lille illey Me Mesh sh Fre requency cy Cut-o

• off

LES ES DES ES Tra ransie sient RAN ANS Po Poin int/Su Surf rface ce FFTs s and iF iFFTs Au Auto and Cro ross ss Sp Spect ctra ra – – co cohere rence ce and phase se FW-H

• H

Exp Export rt to pro ropagatio ion co codes s Exp Export rt to Pro Propagatio ion co codes s Dire irect ct Noise ise Pro Propagatio ion 1D (a (and 2D) ) Wave venumb mber r analysis lysis

Essential features for transient analysis in STAR-CCM+:

– Suitable turbulence models (LES, DES) – Non-reflecting boundary conditions (inlet, outlet, far field) – Accurate computation of compressible flow at low Mach no. – Reliable estimate of cut-off frequency on given mesh (a guide for mesh resolution) – Spectral analysis:

• FFT at points and surfaces
• Auto- and cross-spectra
• Frequency and wavenumber Fourier analysis

Acoustics i in S n STAR-C

• CCM+, II

, II

Validation: Generic side view mirror (Daimler; Univ. of Southampton)

Acoustic S Sources F From DE m DES, I , I

Volume shape used to control grid refinement in the wake of mirror for a DES-study

Validation: Generic side view mirror, grid at bottom plate

Acoustic S Sources F From DE m DES, II , II

Validation: Generic side view mirror, grid in symmetry plane (2 mm resolution in the near-mirror zone)

Acoustic S Sources F From DE m DES, III , III

Validation: Generic side view mirror, flow visualization

Acoustic S Sources F From DE m DES, IV , IV

Wavenu numb mber A Ana nalys lysis

a+ a- u- a+ a- u+

1D wavenumber-frequency diagram:

• Separated wake region (upper)
• Attached wake region (lower)

2D wavenumber analysis – Power Spectral Density (PSD) in wavenumber space:

• Advection ridge (left)
• Acoustic circle (right)

Under-relaxation in segregated solver can be interpreted as marching in a pseudo-time (one iteration per step)… For Implicit Euler time integration, the relation is: A constant under-relaxation factor corresponds to a variable time step and vice versa… Sometimes one can obtain steady-state solution easier by marching in physical time (using transient method and 1-2 iterations per time step) than in steady mode…

Time me S Step a and nd U Und nder-R

• Rela

laxation, I n, I

When solving transient problems with sufficiently small time steps, under-relaxation is not needed… For typical aero-acoustic studies using segregated solver, the recommended under-relaxation settings are:

– For all transport equations (velocities, temperature and other scalar equations): 1.0 – For the pressure-correction equation: 0.5 to 1.0 (smaller values for highly non-orthogonal grids).

The recommended number of iterations per time step is 2 to 4 (depending on time-step size and grid quality).

Time me S Step a and nd U Und nder-R

• Rela

laxation, II n, II

The reduction of residuals is not a suitable measure for convergence of iterations within time step… For small enough time steps, iterations are not necessary (explicit methods)… One can verify by numerical experiments how many iterations are needed…

Numb mber o

• f It

Iterations ns p per T Time me S Step

10 It/dt 2 It/dt Propagation of an acoustic wave (20 cells per wavelength, 20 time steps per period)

Steady-state RANS computations provide results suitable for

• ptimization studies:

– Mean forces and moments – Effects of shape change – Parametric studies (speed, angle etc.)

Best practice developed for different vehicle types (F1, commercial cars, trucks, motocycles):

– Grid design (refinement zones, cell size distribution, prism layer parameters) – Turbulence model – Solver setup

Vehi hicle le A Aerodyna ynami mics: S : Steady R y RANS, I , I

Personal recommendation for fine grids:

– Design the finest grid according to requirements and available resources, using “Base Size” as the parameter. – Increase the base size by a factor of 8 and generate the coarse grid first; start computation on this grid using default set-up parameters (under-relaxation, CFL-number) and a reasonable limit on the number of iterations. – Then reduce the base size by a factor of 2, generate finer grid and continue computation (the solution will be automatically mapped to the new grid), but increase under-relaxation or CFL- number. – Repeat until the base size of the original fine grid is reached.

Vehi hicle le A Aerodyna ynami mics: S : Steady R y RANS, II , II

Computation on a series of grids requires substantially less computing time (2-4 times less) and provides a set of solutions on different grids, allowing error estimate… Instead of a factor of 2, one can use any fixed number between 1.5 and 2. For a second-order method, the error on the finest grid can be estimated as If the base size ratio between coarser and finer grid is not 2, the actual ratio should be used instead of 2.

Vehi hicle le A Aerodyna ynami mics: S : Steady R y RANS, III , III

Vehi hicle le A Aerodyna ynami mics: S : Steady R y RANS, IV , IV

Example: Flow around a 3D wing attached to a wall

• 4 grid levels, base size ratio 2
• Finest grid 460000 polyhedral cells

Section parallel to wall Section normal to wall Wall

Vehi hicle le A Aerodyna ynami mics: S : Steady R y RANS, V , V

Example: Flow around a 3D wind attached to a wall Segregated solver Coupled solver

Vehi hicle le A Aerodyna ynami mics: S : Steady R y RANS, V , VI I

0.3 0.4 0.5 0.6 0.7 0.8

• 15
• 10
• 5

5 10 15 Exp STAR- CCM+

Effect of yaw angle on drag of a truck Effect of underbody geometry on drag of a car

DES-analysis provides:

– Insight into flow features and unsteady phenomena (separation, vortex shedding, pulsation…) – Noise sources

DES is the most accurate approach, but too costly for parametric studies…

Vehi hicle le A Aerodyna ynami mics: DE : DES, I , I

Vehi hicle le A Aerodyna ynami mics: DE : DES, II , II

DES of flow around a truck: details of flow structure in one vertical and one horizontal section (vorticity)

Comparison with experiment is often difficult… Boundary conditions need to be matched for a fair comparison…

Vehi hicle le A Aerodyna ynami mics: DE : DES, III , III

Wind tunnel effects

University of Washington wind tunnel

test configuration Excellent agreement between simulation and experiment for all flap configurations

F16 V Vali lidation S n Study y

Mach 0.2, transition model, 34 million poly-cells, 25 prism layers…

AIA IAA H HiLiftWS1-C

• Conf

nfiguration, I n, I

Comparison of measured and predicted lift

AIA IAA H HiLiftWS1-C

• Conf

nfiguration, II n, II

0 ¡ 0.5 ¡ 1 ¡ 1.5 ¡ 2 ¡ 2.5 ¡ 3 ¡ 3.5 ¡

• ­‑5 ¡

0 ¡ 5 ¡ 10 ¡ 15 ¡ 20 ¡ 25 ¡ 30 ¡ 35 ¡ 40 ¡

CL ¡ Angle ¡of ¡Attack ¡(Degrees) ¡

Experiment ¡ STAR-­‑CCM+: ¡Medium ¡

Workshop conclusions:

– Modeling laminar-turbulent transition is important - simple RANS models do not produce good enough results… – Local grid refinement at wing tip is important - otherwise tip vortex is not well captured…

AIA IAA H HiLiftWS1-C

• Conf

nfiguration, II n, II

Transition AoA=13° AoA=21°

Hub drag is 30% of the total… Need good resolution of geometry details – CAD to mesh in a day for each of two geometries… Need transient simulation to account for rotation…

Rotorcraft H Hub Dr Drag, I , I

Sikorsky UH-60A Hub Sikorsky S-92A Hub

Surface-wrapper provides high geometric fidelity…

Rotorcraft H Hub Dr Drag, II , II

Trimmed grid with prism layers and a sliding interface, ca. 15 million cells

Rotorcraft H Hub Dr Drag, III , III

DES, time step 5° (too large for acoustics, but enough for forces).

Rotorcraft H Hub Dr Drag, IV , IV

Pressure Velocity Magnitude UH-60A S-92A UH-60A S-92A

Studied were variations in drag with adding complexity… Results good for optimization purposes…

Rotorcraft H Hub Dr Drag, V , V

S-92A UH-60A Fro rom: m: M. Dombroski & T.A. Egolf, 68th Annual Forum, American Helicopter, Fort Worth, TX May 1-3, 2012.

Simulation of store separation using overset grids – a validation study

Store S Separation, I n, I

Good agreement between simulation and experiment…

Store S Separation, II n, II

t ¡= ¡0.00 ¡ t ¡= ¡0.16 ¡ t ¡= ¡0.37 ¡

Real application…

Store S Separation, III n, III

Acoustics A Appli lication, V n, Vehi hicle les

Surface FFT (dB) at 500Hz (top) and 1000Hz (bottom)

Acoustics A Appli lication, A n, Airpla lane nes

Noise generation during landing by:

• Wings
• Landing gear

Pressure fluctuation around airfoil Velocity variation around landing gear

Numerics:

– Higher-order discretization – Automatic adaptive mesh refinement

Turbulence:

– Improvements to RANS-models (curvature correction, law of the wall) – Improvements to DES-model (transition from RANS to LES)

Vibro-acoustics:

– Wavenumber analysis – Coupling of flow and structure – Possibly solving special set of equations for noise propagation

Future De Develo lopme ment nts