Simulation calculation of aerodynamic performance of Mira model

based on virtual wind tunnel

Fu Yu*a, Zhao zikaib, Ma Xiaolea
a
CATARC New Energy Vehicle Test Center (TianJin) Co.,Ltd., TianJin 300300, China;
b
CATARC Automotive Test Center (TianJin) Co.,Ltd., TianJin 300300, China;
*
Corresponding author: [email protected]

ABSTRACT

To improve the simulation accuracy of vehicle aerodynamic resistance, better benchmarking of simulation and test. In
this paper, a virtual numerical wind tunnel is established according to a domestic wind tunnel's structural size. The
MIRA model is used as the research object to simulate the MIRA model's flow field in the rectangular calculation
domain and the virtual wind tunnel calculation domain. The results show that the estimated value error of the wind
resistance coefficient in the mentioned domains is two counts and three counts. However, the difference between the
simulation results of the flow field at the rear of the two-car bodies is noticeable. The simulation results of the flow field
at the rear of the car body in the numerical wind tunnel calculation domain are more consistent with the experimental
results. Therefore, the numerical wind tunnel can better predict and optimize the aerodynamic performance of the vehicle.
Keywords: aerodynamics, Aearodynamic drag Coefficient, Virtual Wind Tunnel, flow field

1. INTRODUCTION
When the vehicle speed reaches 80km/h, the aerodynamic resistance accounts for 50% of the total vehicle resistance, and
increases exponentially with the increase of vehicle speed. In the vehicle development stage, optimizing the aerodynamic
performance of the whole vehicle can greatly reduce the air resistance under high-speed working conditions, so as to
reduce fuel consumption and pollutant emission. Compared with the road test, the wind tunnel test has higher stability,
can objectively and clearly evaluate the aerodynamic performance of the vehicle, and can compare the actual drag
reduction effects of different schemes[1]. Therefore, wind tunnel test is widely used in the industry as the main way to
verify the aerodynamic performance of vehicles. However, the cost of wind tunnel test is high, and there is no test
sample vehicle in the early stage of vehicle development, so it is impossible to carry out wind tunnel test. Its application
in the process of vehicle aerodynamic performance optimization has certain limitations. Numerical simulation
technology can predict and optimize the index parameters of wind tunnel and the aerodynamic performance of vehicle
model[2].How to improve the accuracy of simulation prediction has become an important problem that CFD engineers
need to solve.At present, the vehicle aerodynamic performance simulation calculation is based on the rectangular
calculation domain, and its structural dimensions are quite different from the actual wind tunnel. This brings obstacles to
the simulation results and test benchmarking.The establishment of virtual wind tunnel based on the actual wind tunnel
structure size can solve this problem.Fischer and kuthada compared the wind tunnel test results of different vehicle
models with the virtual wind tunnel results. The results showed that the simulation results based on the virtual wind
tunnel were closer to the test results[3].At the same time, accurate virtual wind tunnel benchmarking is the basis and key
to realize high-precision open road simulation. Based on a 3/4 open wind tunnel in China, a virtual wind tunnel model is
established in this paper. Mira model is selected for simulation verification, and compared with the measured results.

2. INTRODUCTION OF VIRTUAL WIND TUNNEL AND MIRA MODEL

2.1 Introduction to virtual wind tunnel mode
Based on the actual size of a 3/4 open wind tunnel in China, this paper establishes a virtual wind tunnel model, which
includes contraction section, nozzle, stationary chamber, collection port, diffusion section, pressure balance port and
buffer port, as shown in Figure 1.

2nd International Conference on Applied Mathematics, Modelling, and Intelligent Computing

(CAMMIC 2022), edited by Hari Mohan Srivastava, Chi-Hua Chen, Proc. of SPIE Vol. 12259,
122593F · © 2022 SPIE · 0277-786X doi: 10.1117/12.2638771

Proc. of SPIE Vol. 12259 122593F-1

 Figure.1 Diagrammatic sketch of numerical wind tunnel structure model
2.2 Mira model introduction
The international standard Mira step back model is selected as the simulation model in this paper. The basic angles of the
model are shown in Figure 2.

Figure.2 Three views of MIRA model

3. SIMULATION CALCULATION OF VIRTUAL WIND TUNNEL MODEL

3.1 Introduction of virtual wind tunnel model volume grid
In order to improve the accuracy of the calculation results of the virtual wind tunnel, the grid of the virtual wind tunnel is
encrypted[4], and the grid size of the refinement area is shown in Table 1.
Table 1 Mesh size of the refinement area

Volume mesh refinement area Size

1st Layer 24
2nd Layer 48
3rd Layer 96
4th Layer 192

Therefinement area is shown in Figure 3.

Figure.3 Numerical wind tunnel model mesh encryption area division

3.2 Physical model and boundary condition setting of virtual wind tunnel
In this paper, Navier Stokes equation is selected as the governing equation for solution, and realizable k-epsilon, two
layer all y + wall treatment is selected as the turbulence model.The inlet boundary type is velocity inlet, the nozzle speed

Proc. of SPIE Vol. 12259 122593F-2

is 120km/h, the outlet boundary type is pressure outlet, the pressure value is 0, the wall boundary type of wind tunnel is
wall, and the ground of contraction section is set as sliding wall[5].
3.3 Analysis of simulation results of virtual wind tunnel
As can be seen from Figure 4, the speed of the air flow increases rapidly after passing through the contraction section,
and the flow velocity at the nozzle reaches 33m/s.After the air flow is ejected from the nozzle, a shear layer is formed
between the air and the surrounding air, that is, there is velocity shear.Along the direction of air flow, the shear layer
gradually thickens.The velocity change in the mainstream area is small, and the flow velocity decreases gradually after
the air enters the collection port. The distance between the center of the turntable and the nozzle of the wind tunnel is
4.7m.During the wind tunnel test, the vehicle will be installed at the center of the turntable.According to the analysis of
6m vehicle length, the vehicle is between 1.7m and 7.7m from the nozzle.The numerical analysis results of flow field
show that the flow field in this area has good stability and meets the test requirements[6].

Figure.4 Numerical wind tunnel velocity distribution at section Y=0

Taking the center of the wind tunnel turntable as the origin, the simulation and test static pressure gradient results at
different locations along the streamline direction at the height of Z=0.5m in the basin are compared.It can be seen from
Figure 5 that the static pressure coefficient of the virtual wind tunnel is slightly different from the measured value in the
range of -4.7m to -2.5m and 8.5m to 10m, which is mainly due to the difference between the boundary conditions of the
wind tunnel pressure balance port and buffer port and the actual situation.However, in the range of -2.5m to 8.5m in the
test section, the error between the static pressure coefficient simulation and the test results is small, Therefore, the
simulation results of the virtual wind tunnel can be used for Mira model simulation[7].

Figure.5 Simulation and experiment results of static pressure gradient in wind tunnel are compared

4. COMPARATIVE STUDY ON FLOW FIELD SIMULATION RESULTS OF MIRA

MODEL BASED ON VIRTUAL WIND TUNNEL AND RECTANGULAR DOMAIN
4.1 Research on flow field simulation grid scheme of Mira model based on virtual wind tunnel
4.1.1 Introduction to Mira model surface mesh
It can be seen from the three views of Mira model in Figure 6 that the basic structure features of Mira model are retained
in the 3D model grid model, which can be used for further simulation calculation.

Figure.6 Three views of MIRA mesh model

Proc. of SPIE Vol. 12259 122593F-3

4.1.2 Research on mesh size independence
The size of the grid directly affects the convergence and accuracy of the calculation. If the grid is too rough, it can not
capture the details of the flow field, resulting in distortion and low accuracy. However, if the grid is too fine, it will cost
a lot of time, cost and hardware resources, and the computing efficiency is low. Therefore, it is necessary to calculate the
grid schemes with different sizes, determine the optimal grid scheme, and reduce the grid scale on the premise of
ensuring the calculation accuracy[4]. In this paper, four mesh schemes, meshsie1 ~ meshsie4, are set up, and different
face mesh and volume mesh sizes are defined. It is proposed to optimize the optimal grid scheme through simulation
calculation[8]. The calculation results of different grid schemes are shown in Table 2.
Table 2 Results of different grid schemes
Mesh size Number of
project Cd Error
（mm） cells
1 10 35043620 0.306 3.38%
2 12 21752823 0.305 3.04%
3 18 9860014 0.310 4.73%
4 24 6183979 0.320 8.11%

From the above simulation results, it can be seen that with the decrease of grid size, the number of grids increases and
the Cd value decreases. When the grid size is reduced to 12mm, the Cd calculation results tend to be stable. When the
surface grid is 24mm, 18mm, 12mm and 10mm, the Cd calculation errors are 8.11%, 4.73%, 3.04% and 3.38%
respectively. Therefore, meshsize2 grid scheme is preferred considering the calculation accuracy and time cost.

Figure.7 Cd results of different grid schemes

4.1.3 Introduction to grid encryption area setting
The interaction between the air flow and the body surface will produce flow separation due to the structural
characteristics of the body surface. In order to more accurately capture the flow separation of the air flow, the grid
around the separation area needs to be densified. The grid encryption scheme is shown in Figure 8.

Figure.8 Diagrammatic sketch of Grid encryption scheme

The grid scheme of vehicle external aerodynamic performance simulation should adapt to the selected turbulence and
wall model. Different turbulence models have different requirements for y +. For the realizable k-epsilon turbulence
model, the y + value should be between 0 ~ 1 or 30 ~ 100. It can be seen from Figure 9 that most of the y + of the grid
model is between 0 and 1, which meets the requirements of the realizable k-epsilon turbulence model[9]. Therefore, the
scheme has good adaptability and meets the basic requirements of simulation.

Proc. of SPIE Vol. 12259 122593F-4

 Figure.9 Y+ Cloud image
4.2 Physical model and boundary condition setting of aerodynamic performance simulation of Mira model based
on virtual wind tunnel
The physical model is Reynolds averaged navies Stokes and realizable K- ε Turbulence, two-layer full y + wall treatment,
nozzle speed of 80km/h, outlet pressure value of 0Pa, body modeling surface and bottom outer surface boundary type of
wall, wheel speed of 80km/h, ground speed of 80km/h[4].
4.3 Description of Mira flow field simulation model based on rectangular domain
In order to clarify the difference between the aerodynamic performance simulation results of Mira model in virtual wind
tunnel calculation domain and rectangular simplified calculation domain, this paper models and solves the Mira model in
rectangular calculation domain, and compares the flow field calculation results of Mira model in virtual wind tunnel
domain and rectangular calculation domain with the wind tunnel test results, so as to evaluate the accuracy and necessity
of virtual wind tunnel domain calculation[10][11].
In order to fully develop the fluid in the calculation domain and reduce the interference of the wall to the flow field
around the vehicle body, the length, width and height of the rectangular calculation domain are 10 times the vehicle
length, 10 times the vehicle width and 5 times the vehicle height respectively, as shown in Figure 10.

Figure.10 Diagrammatic sketch of Rectangular domain

In order to objectively compare the simulation results of virtual wind tunnel and rectangular computing domain, the grid
size, encryption scheme, physical model and boundary condition setting of rectangular computing are consistent with the
grid scheme of virtual wind tunnel.
4.4 Comparative analysis of aerodynamic performance simulation results of Mira model based on virtual wind
tunnel and rectangular domain
4.4.1 Comparison of wind resistance coefficient results
The simulation calculation results in Table 3 show that the CD value of Mira model is 0.298 in the virtual wind tunnel
calculation domain and 0.293 in the rectangular calculation domain. The difference between Mira model and the test
result[5] is 2 counts and 3 counts respectively.
Table 3 Results of MIRA in Numerical wind tunnel and Rectangular domain
Model Cd
virtual wind tunnel 0.298
rectangular domain 0.293
Test result 0.296
4.4.2 Comparison of pressure coefficient results
a. Pressure coefficient comparison of Mira model

Proc. of SPIE Vol. 12259 122593F-5

It can be seen from Figure 11 that the overall positive pressure coefficient of the virtual wind tunnel calculation domain
is slightly larger than that of the rectangular calculation domain, and the difference at the windshield is more obvious.

(a)rectangular domain (b)virtual wind tunnel

Figure.11 Front pressure coefficient of Rectangular domain and virtual wind tunnel domain
It can be seen from Figure 12 that there are differences in the calculation results of the rear pressure coefficient between
the two calculation domains. The rear pressure coefficient of the virtual wind tunnel calculation domain is higher than
that of the rectangular calculation domain.

(a)rectangular domain (b)virtual wind tunnel domain

Figure.12 Back pressure coefficient of Rectangular domain and virtual wind tunnel domain
b. Comparison of pressure coefficient results of each measuring point at the tail
Through the above analysis, it can be seen that the pressure results of the two calculation domains at the rear of the
vehicle are quite different. In order to further analyze the differences between the simulation results of the two
calculation domains, this paper arranges 30 monitoring points in the model with the same position as the test measuring
points, which are respectively located at the rear of the trunk and the surface of the rear windshield, and counts the
pressure coefficients of each measuring point of the two calculation domains[5], See Figure 13 and Figure 24 for the
specific location of measuring points.

Figure.13 Diagrammatic sketch of the trunk pressure coefficient position

As can be seen from figures 14 to 17, compared with the simulation results of rectangular computing domain, the
simulation results of the first and second rows of virtual wind tunnel are closer to the test results.

Proc. of SPIE Vol. 12259 122593F-6

 Figure.14 The first row of pressure points at the rear of trunk

Figure.15 pressure points analytical error of the first row of trunk

Figure.16 The second row of pressure points at the rear of trunk

Figure.17 pressure points analytical error of the second row of trunk

It can be seen from figures 18 and 19 that the results of the pressure measuring points in the third row at the rear of the
trunk show that the calculated pressure coefficient in the virtual wind tunnel domain is less than the rectangular
calculation domain and the measured value, and the error is large compared with the calculated pressure coefficient in
the rectangular domain.

Figure.18 The third row of pressure points at the rear of trunk

Proc. of SPIE Vol. 12259 122593F-7

 Figure.19 pressure points analytical error of the third row of trunk
From Figure 20 to figure 23, it can be seen that in the virtual wind tunnel area, due to the blocking effect of air flow, the
pressure at the front lower part of the vehicle head is large, and the pressure difference between the front and the bottom
of the vehicle is large. Therefore, the air flow velocity at the bottom of the vehicle body is large, the pressure at the rear
of the trunk is low, and the degree of negative pressure is obvious.

Figure.20pressure coefficient of Rectangular domain

Figure.21 pressure coefficient of virtual wind tunnel domain

Figure.22 velocity of Rectangular domain

Figure.23 velocity of virtual wind tunnel domain

In addition, compared with the actual wind tunnel structure, some areas in the virtual wind tunnel model are simplified.
For example, in the actual wind tunnel, in order to eliminate the influence of the incoming boundary layer, the suction

Proc. of SPIE Vol. 12259 122593F-8

structure is set in front of the vehicle head, but the virtual wind tunnel model does not establish the structure, but the
front of the vehicle head is set as a sliding wall, which is different from the actual situation, so it brings calculation errors.

Figure.24 Diagrammatic sketch of the rear windshield pressure coefficient position

It can be seen from figures 25 to 30 that there are certain differences between the simulation and test results of each
measuring point in the rear windshield area, but the error of the pressure measuring point results of the virtual wind
tunnel is less than the calculation results of the pressure measuring points in the rectangular area.

Figure.25 The first row of pressure points at the rear windshield

Figure.26 pressure points analytical error of the first row of rear windshield

Figure.27 The second row of pressure points at the rear windshield

Figure.28 pressure points analytical error of the second row of rear windshield

Proc. of SPIE Vol. 12259 122593F-9

 Figure.29 The third row of pressure points at the rear windshield

Figure.30 pressure points analytical error of the third row of rear windshield
From the pressure measuring points of the rear trunk and rear windshield, compared with the results of the rectangular
calculation domain, the simulation results of the pressure coefficient in the virtual wind tunnel domain are closer to the
test measured values, and the overall error is less than that in the rectangular calculation domain.
4.4.3 Comparison of flow field calculation results
It can be seen from figure 31 that there are three vortices of different sizes in the rear windshield and the rear of the trunk.
Compared with the PIV test results[4], the z-direction heights of the three vortices in the rectangular calculation domain
and the virtual wind tunnel calculation domain are basically the same. Compared with the rectangular calculation domain,
the two vortices at the rear of the trunk in the virtual wind tunnel calculation domain are farther away from the vehicle
body, which is consistent with the test results.

Figure.31 Wake vortex of simulation results and experimental results

As shown in Figure 32, comparing the structure of the tail vortex between the rectangular calculation domain and the
virtual wind tunnel calculation domain, it can be seen that there is no obvious vortex under the trunk tail in the
rectangular calculation domain, while there is an obvious vortex under the trunk tail in the virtual wind tunnel calculation
domain, which is in good agreement with the test results.

(a)Rectangular domain

Proc. of SPIE Vol. 12259 122593F-10

 (b)virtual wind tunnel domain
Figure.32 Comparison of Wake vortex simulation result detail diagram

5. CONCLUSION
The virtual wind tunnel is established according to the actual structural size of the wind tunnel. In the empty wind tunnel
mode, the flow field in the test section is stable, and the axial static pressure gradient curve is in good agreement in the
test section.Meshsize2 grid scheme, combined with the grid encryption setting of the separation area around the vehicle
body, can reduce the calculation time and cost on the premise of ensuring the simulation accuracy.The Cd value of Mira
model is 0.298 in the virtual wind tunnel calculation domain and 0.293 in the rectangular calculation domain. Compared
with the test result of 0.296, the difference is 2 counts and 3 counts respectively. There are great differences in the
calculation results of the flow field at the rear of the vehicle body between the virtual wind tunnel calculation domain
and the rectangular calculation domain. Through the comparative analysis with the test pressure measuring points and
PIV results, the overall error of the calculation results of the measuring points in the virtual wind tunnel calculation
domain is lower than that in the rectangular domain, which can ensure the reliability of the optimization scheme at the
rear of the vehicle body. Therefore, virtual wind tunnel can better predict and optimize the aerodynamic performance of
vehicles.In order to make the virtual wind tunnel better serve the vehicle development, the follow-up will carry out the
simulation and experimental comparative research of the real vehicle full detail model based on the virtual wind tunnel,
so as to improve the accuracy of the model to meet the engineering design requirements.

REFERENCES

[1] Pang Jia Bin,Lin Zhi Xing.Correetion Methods for Automotive Model Tests in TJ-2 Wind
Tunnel[J].Aotomotive Engineering,2002,24(5):371-375.
[2] Yang, Z.,Schenkel.Assessment of Closed Wall Wind Tunnel Blockage Using CFD[C].SAE Technical
Paper,2004-01-0672
[3] Fischer O,Kuthada T,Wiedemann J,et al.CFD validation study for a sedan scale model in an open jet wind
tunnel[C].SAE Paper 2008-01-0325.
[4] Zhou Hua,Yang Zhi Gang,Zhu Hui.Aerodynamic calculation of MIRA model correlated with wind tunnel
test[J].Journal of Jilin University(Engineering and Technology Edition),2019,49(4):1043-1053.
[5] Hu Xingjun,Zhu, Yunyun.Low-wind-drag numerical research on fenders and wheels of car model[J]. Applied
Mechanics and Materials,2013,(397-400):599-602.
[6] Hu Xingjun,Yue, Lei.Influence of regulating seam on aerodynamic characteristics of virtual wind
tunnel.Advanced Materials Research,2013,( 774-776 ):290-294.
[7] Song Xin,li Shu Ya,Yan Jie.Study on the Correlation Between Digital Wind Tunnel Simulation and Open Road
Simulation[J].Aotomotive Engineering,2020,42(6):759-764.
[8] Hu Xingjun, Bo Yang, Yulong Lei.Automotive shape optimization using the radial basis function model based
on a parametric surface grid[J].Proceedings of the Institution of Mechanical Engineers Part D-Journal of
Automobile Engineering,2016,230(13):1808-1821.
[9] Du, Qianqian; Hu, Xingjun; Li, Qifei.Numerical Optimization Research on Rear Characteristic Angles Based on
MIRA Model for Aerodynamic Drag Reduction.Advanced Materials Research,2013,( 774-776 ):428-432.
[10] JOHN M,BUGA S,MONTI I,et al.Experimental and numerical study of the DrivAer model
aerodynamics[C].SAE paper 2018-01-0741.
[11] Hu, Xingjun,Li tengfei,Guo peng.Influence of wheel spoiler on aerodynamic drag of passenger car[J].Journal
of Jilin University(Engineering and Technology Edition),2012(42):32-36.

Proc. of SPIE Vol. 12259 122593F-11