Sunteți pe pagina 1din 14

MSc Computational Fluid

Dynamics,
School of Engineering
2012/2013

Numerical investigation
of the 3-D flow around
Ahmed Body
Automotive flows Assignment

EXAMINER: STUDENT:

Dr. N. Asproulis Gennaro Abbruzzese


(s197885)
Contents
Abstract ................................................................................................................................................................................................. 1

1 Introduction............................................................................................................................................................................... 1

2 Problem setup ......................................................................................................................................................................... 3

2.1 The computational domain......................................................................................................................................... 3

2.2 Models .............................................................................................................................................................................. 4

3 Results ........................................................................................................................................................................................ 5

3.1 Models comparison....................................................................................................................................................... 5

3.2 Grid convergence index calculation .................................................................... Error! Bookmark not defined.

3.3 Simulation with radius 10mm ................................................................................ Error! Bookmark not defined.

4 Conclusion ...............................................................................................................................................................................10

5 References ...............................................................................................................................................................................10

6 Appendix .................................................................................................................................................................................. 11

6.1 Nomenclature ................................................................................................................................................................ 11

6.2 User defined function for the drag law .............................................................. Error! Bookmark not defined.

6.3 Fortran code for analytic solution ........................................................................ Error! Bookmark not defined.
1

Abstract

1 Introduction
Due to marketing needs, vehicle geometries in automotive industry can have substantial change in shape and

really different flow structures. Nevertheless, it is possible to identify some common features for all the vehicle
geometries. Ahmed et al.(1984) in his work introduced his standard vehicle model. The Ahmed body (fig.1) has
been created in order to capture all the key phenomena regarding the automotive flows, starting from the
observation that, against the common sense, the major contribute to drag on a vehicle comes from the wake

region. This particular geometry allows to study the drag and phenomena as vortex structures and flow

separation occurring in the wake region changing the slant angle 𝜑 over the rear part of the body.
Introduction 2

Fig. 1 The Ahmed body. The slant angle can vary but the length of the sloping surface is constant (222mm)

The present work takes as a benchmark the experimental data collected by S. Becker, H. Lienhart and C.
Stoots for the MOVA project and summarized in a qualitative summarization in H.Lienhart and S. Becker
(2003). According to Ahmed et al. (1984) flow separation, with an inlet velocity equal to 60 m/s, should occur
with a slat angle greater than 30°. For this reason Lienhart et al. Investigated two angles, 25° and 35°, in order

to ensure to cover the separation zone. No separation has been observed for a slant angle equal to 25° for

the 40 m/s inlet velocity case.

The aim of this study is to reproduce numerically the experimental results testing classical turbulence models

with the commercial software Fluent. Particular attention has been given on drag coefficient, x and y
component of velocities in wake region and on the capability of the models to predict the formation of the

trailing vortexes shown in figure 2.

(a) (b)
Fig. 2 The two trailing vortexes shown on a intuitive manner (a) and with u-v planes in the wake region (b)

Several studies have been carried out numerically. Krajnović and Davidson (2004) &&& noted that classical

RANS models performs poorly for this case have found two possible explanations for this failure. One could
be the higher order of unsteadiness of the problem. The second is that RANS miss flow separation, even
Problem setup 3

according to the observation coming from Durand et al.[&&&]: “Most RANS turbulent models would

underestimate or miss the separation entirely and return too optimistic flow conditions”.

2 Problem setup
2.1 The computational domain
According to Lienhart et al. (2003) “the contours of various moments display good symmetry”; for this reason

half geometry has been meshed in order to reduce the time of the computation maintaining a good

refinement. Since stilts doesn’t affect flow on the top of the Ahmed body and its contribution to flow in the
wake region is negligible but they affect the quality of the mesh, no stilts have been included, The entire
computational domain is shown in figure 3. The length of the wake is near to 5 times the length of the Ahmed

body, a good practice for computational domain construction.

Fig. 3 geometry and domain

The figure 4 shows the finest mesh, composed of 908151 cells, unstructured the free stream zones and

structured near walls (Ahmed body and ground) with y-plus=100.


Problem setup 4

(a) (b)
Fig. 4 Finest mesh. General view (a) and particular view of structured elements on ahmed and ground (b)

Furthermore, two coarser grids have been tested to check the grid convergence, with the same properties but
with 288838 and 67567. The number of the cells has been chosen in order to obtain a minimum optimal

refinement of 1.3 [&&&].

The boundary conditions set in Fluent are listed below:

 Wall with “no slip” condition for the Ahmed body and ground
 Wall with specified shear stress equal to 0 for side and top surface of the domain

 Outflow condition in outlet surface


 Velocity inlet condition (velocity magnitude=40 m/s, turbulent intensity=20%, turbulent length scale=

0.14m)

 Symmetry condition on the symmetry plane

2.2 Models
Three models have been employed for this simulation:

 Standard k-epsilon
This model employs two equations (for a transport equation for the turbulent dissipation 𝜀 and the turbulent
kinetic energy k). It is the most commonly used. Generally it is not suitable for separation modelling and it is

not enough sensible to sudden geometry changes [Mesiopulos &&&&]


 Realizable k-epsilon
This is a modification of the first model intended to adapt the model for cases of large strain
 K-omega SST

This model is a K-omega including the transport of the turbulent shear stress in order to avoid characteristic
over-prediction of eddy viscosity, which lower the performances of the classical K-omega in case of adverse
pressure gradients.
Results 5

These classical models belong to the RANS (Reynolds Averaged Navier-Stokes) family. The flow properties are

divided in a fixed component and a fluctuating one that is averaged in time, an operation called Reynolds
decomposition.
All of them rely on the Boussinesq assumption, the Reynolds stress tensor is proportional to the mean strain
rate tensor.

Each model includes wall function and a range between 30 and 300 of y-plus is optimal for these models.

3 Results
In this section we discuss results of the simulations. The investigation has been carried on in three phases. The
first has been the identification of the best turbulence mode on the basis of the drag coefficient breakdown. In

the second part the convergence of the grid has been evaluated with the best model in order to check the
reliability of the results. Finally main features of the model have been analyzed and the results have been

compared with experimental data employing the finest mesh.

All the simulations have been run with second order of accuracy of the momentum equation and a

convergence criteria of 1e-04.

3.1 Models comparison


The evaluation of the three models has been carried out comparing the drag coefficient and its breakdown.
The first model, the standard k-epsilon, has shown a straight convergence to 1e-04. The other two has show a

convergence but to a not satisfying level. Nevertheless thanks to this convergence, it has been possible to find
a pattern in the drag coefficient. The figures (&&&) show that the drag coefficient has a periodic oscillation

function of the number of the iterations.

(a) (b)
Fig. 5 Convergence history of the drag coefficient for k-omega SST (a) and realizable k-epsilon (b)
Results 6

Is then licit to assume the drag coefficient for the two models equal to the average. The simulations have

been stopped when the drag coefficient reached this value.

The table 1 shows the drag coefficient breakdown of all cases and experimental investigation and following
there is the summarization of the errors (table 2).

k-w SST R k-e Std k-e Exp (&&&)


slant 7,916E-02 1,034E-01 1,061E-01 1,450E-01
front 1,311E-01 1,471E-01 1,817E-01 1,900E-02
rear 3,801E-02 -1,551E-04 1,852E-02 7,700E-02
viscous 3,577E-02 3,722E-02 4,387E-02 5,700E-02
Net 2,840E-01 2,875E-01 3,501E-01 2,980E-01
Tab. 1 Drag coefficient breakdown for all the models

k-w SST R k-e Std k-e


err -6,584E-02 -4,164E-02 -3,894E-02
slant
rel err -45,41% -28,72% -26,85%
err 1,121E-01 1,281E-01 1,627E-01
front
rel err 589,80% 674,08% 856,17%
err -3,899E-02 -7,716E-02 -5,848E-02
rear
rel err -50,64% -100,20% -75,94%
err -2,123E-02 -1,978E-02 -1,313E-02
viscous
rel err -37,25% -34,70% -23,04%
err -1,400E-02 -1,050E-02 5,212E-02
Net
rel err -4,70% -3,52% 17,49%
Tab. 2 Errors for all models and for all the components of the drag coefficients

As ii is possible to see from the tables the error on the net value of the drag coefficient is rather low for k -

omega SST and the realizable k-epsilon, while it increase for the standard k-epsilon. The breakdown of the
drag coefficient should be important to understand where each model fails or succeed. Actually the errors are
enormously high, and even negative in the rear for the simulation with realizable k-epsilon, for each part even
if the net error seems to be low. Actually this values couldn’t be considered reliable, as the error on the net

value is quite low. The high level of error should be rather ascribed to a wrong attribution of the components

done by the software.

Neglecting all the errors of the components the realizable results the most suitable with an under-prediction

of the drag coefficient equal to the 3,52%. The k-omega SST presents an error more than 1% higher, while the
standard k-epsilon supplies a drag coefficient rather unreliable with an over-prediction of 17,5%.

3.2 Grid convergence analysis


Results 7

The grid convergence has been tested on a set of three grid with realizable k-epsilon model. The table 4

summarizes the details of experiment and results. The key parameter for the convergence evaluation is still the
drag coefficient.

mesh ID cells drag coefficient


1 908151 2,875E-01
2 288838 3,096E-01
3 67567 4,471E-01

For unstructured mesh it is rather difficult to calculate the refinement but it can be approximated by the

formula (3.1) (NASA &&&):

1
𝑁1 𝐷 (3.1)
𝑟=( )
𝑁2

With N number of cells and D dimensionality of the domain (3 for three-dimensional cases). Another difficulty

comes from the calculation of the observed order of accuracy p for not constant refinement. For this reason

the system of equations suggested by (&&&) (3.2), (3.3), (3.4) has been solved iteratively:

1
𝑝= |𝑙𝑛|𝜀32 /𝜀21 | + 𝑞(𝑝)| (3.2)
ln⁡(𝑟21 )

𝑝
𝑟21 − 𝑠
𝑞(𝑝) = ( 𝑝 ) (3.3)
𝑟32 − 𝑠

𝑠 = 𝑠𝑖𝑔𝑛(𝜀32 /𝜀21 ) (3.4)

with 𝜀𝑖𝑗 = 𝐶𝐷𝑖 − 𝐶𝐷 and 𝑟𝑖𝑗 refinement between the mesh i and j. The q becomes 0 when the refinement is
𝑗

constant. Then Richardson extrapolation and GCI have been computed, respectively with the formula (3.5) and

(3.6):

𝑖𝑗 𝑝 𝑗 𝑝
𝐶𝐷,𝑒𝑥𝑡𝑟𝑎𝑝𝑜𝑙𝑎𝑡𝑒𝑑 = (𝑟𝑖𝑗 𝐶𝐷𝑖 − 𝐶𝐷 )/(𝑟𝑖𝑗 − 1) (3.5)

𝑖𝑗 𝐹𝑠 𝑒 𝑖𝑗
𝐺𝐶𝐼𝑓𝑖𝑛𝑒 = 𝑝 (3.6)
𝑟𝑖𝑗 − 1
𝑗
𝑖𝑗 𝐶𝐷 − 𝐶𝐷𝑖
𝑒𝑓𝑖𝑛𝑒 = | 𝑗
| (3.7)
𝐶𝐷

With 𝐹𝑠 a safety factor equal to 1.25 when three grids are used. The results are summarized in table 5.

𝒊𝒋 𝒑 𝒊𝒋 𝒊𝒋
i -j 𝒓𝒊𝒋 𝒆𝒇𝒊𝒏𝒆 𝑪𝑫,𝒆𝒙𝒕𝒓𝒂𝒑𝒐𝒍𝒂𝒕𝒆𝒅 𝑮𝑪𝑰𝒇𝒊𝒏𝒆
2-1 1,465 0,07685 3,568 0,2799 3,31%
Results 8

3-2 1,623 0,444 3,568 0,2799 11,99%


Tab. 3 Summarization of the convergence study

The extrapolated value is 0.2799, lower than the value given by the finest mesh. It means that the real value
supplied by the model is farther from the experimental and the real error is actually higher. It is not possible to
consider the mesh refined enough, as the GCI is higher than 1% but it is possible to proceed with the

experiment to understand main features of the model.

3.3 Model analysis


The plot in figure 6b shows that the first cells on the walls have been correctly dimensioned in order to

maintain the 𝑦 + under the threshold ≈ 100 even in the zone of maximum velocity on the front and before the
slant angle (figure 6a). The condition for the application of the wall function is then satisfied.

(a) (b)
Fig. 6 Plots of velocity magnitude in planes cutting by half the geometry (a) and the actual y+ on along the Ahmed body (b)

The key features analyzed for the evaluation of the model (the realizable k-epsilon) have been three:
1. Flow structure
2. Flow separation
3. Velocities

For the flow structure a qualitative investigation has been performed. The figure 7, obtained through a custom
field function in Fluent to obtain the q-criterion (= 0.5 ∗ (𝑣𝑜𝑟𝑡𝑖𝑐𝑖𝑡𝑦 2 − 𝑠𝑡𝑟𝑎𝑖𝑛⁡𝑟𝑎𝑡𝑒 2 )), shows the structure of

the vortexes in the wake. The simulation gives satisfying results in the reproduction of the counter-rotating
trailing vortexes, which extend for almost all the length of the domain. The horseshoe vortexes described by
Ahmed (&&&) are not clearly defined and they are mixed with the previously described vortexes, but this
behaviour can be attributed to the unsatisfying mesh refinement as these are present.
Results 9

Fig. 7 Q-criterion iso-surface. The trailing vortexes are clearly visible. The horseshoe vortex is present but it has not a smooth shape

As said, the influence of the bluff body on the flow protracts for all the length of the domain but the
characteristic recirculation bubble in the wake extends until about 0.4m in the wake (fig. 8a) and its influence

on the kinetic energy until 0.5m (fig. 0.5).

(a) (b)
Fig. 8 recirculation bubble in the wake. (a) velocity magnitude contours. (b) kinetic energy contours

Nevertheless, after 500mm the flow is still heavily influenced by the body. This is shown in the contours of x -
velocity on y-z plane at 500mm from the rear (figure 9), which demonstrate that vertical phenomena are still
present and affect the velocity of the flow.
Conclusion 10

Fig. 9 Contours of velocity on the y-z plane at 500mm from the rear. Perturbations of the stream are still visible.

The table 4 shows the prediction of the velocities on

4 Conclusion

5 References
Celik, I., 1993, “Numerical Uncertainty in Fluid Flow Calculations: Needs for Future Research,” ASME
JOURNAL OF FLUIDS ENGINEERING, 115, pp. 194-195.

[1]. Tsoutsanis, P., (2013), Automotive flows, Computational fluid dynamics MSc, Cranfield University

(unpublished notes).
[2]. FLUENT user manual, 12.0 © ANSYS, Inc. 2009-01-23.
Roache, P. J., 1998, Verification and Validation in Computational Science and Engineering,

Hermosa Publishers, Albuquerque, NM

L. Durand, M.Kuntz, F. Menter, Validation of. CFX-5 for the Ahmed Car Body, 10th joint. ERCOFTAC. (SIG-15)
Workshop on Refined Turbulence Modelling, October 10-11, 2002

S.Kapadia, S. Roy, and K. Wurtzler, (2003),Detached eddy simulation over a reference Ahmed car model” AIAA
paper no.2003-0857,2003

S.R. Ahmed, G. Ramm, G. Faltin, (1984), Some salient features of the time-averaged ground vehicle wake,
SAE Paper 840300
Appendix 11

ERCOFTAC Database, case 9, http://bit.ly/jOILe2

6 Appendix
6.1 Nomenclature

𝛼𝑞 volume fraction of phase p

𝜌𝑞 density of the phase p

𝑣⃗𝑞 velocity of phase p

𝑚̇𝑝𝑞 mass transfer from phase p to q

𝑝𝑞 pressure of phase p

𝑝 pressure shared by all the phases

𝜏̿𝑞 stress tensor of phase p

𝑔⃗ gravity acceleration

𝐹⃗𝑞 external body force

𝐹⃗𝑙𝑖𝑓𝑡,𝑞 lift force

𝐹⃗𝑛𝑚,𝑞 virtual mass force

𝐾𝑝𝑞 interaction force between phases p and q

𝑣⃗𝑝𝑞 ⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡interphase velocity

ℎ𝑝𝑞 interphase enthalpy

⃗q⃗ q heat flux

𝑆𝑞 source term

⃗⃗𝑝𝑞
𝑄 intensity of heat exchange
Appendix 12

𝐶𝐷 drag coefficient

𝑅 radius of paricles

S-ar putea să vă placă și