Download as pdf or txt
Download as pdf or txt
You are on page 1of 23

Physical Modeling of Fluids (PMFLU)

February 1, 2016

Prashanth Rajkumar S and Gaurav Bhati

Ecole Centrale de Nantes Computational Mechanics, 1 Rue de la Noe, 44321 Nantes, France

Abstract
The purpose of the project is to study and compare the 2D turbulent ow computation of a classical NACA0012
foil for dierent combinations of Reynolds's number (1e6, 2e6, 3e6, 4e6), grid (Coarse, Medium, Fine, Superne),
angle of incidence (0deg, 3deg, 6deg, 9deg), ow models used (K-w SST, EASM, Spallart-Allmaras), numerical
schemes (UPWIND, GDS, AVL-SMART) with dierent orders and under relaxation parameters. we have also
compared the simulation results with McCroskey 1987 and Gregory 1970 wind tunnel experiments. We compare
the residuals, CPU time, pressure, velocity, viscosity, turbulence, pressure coecient plot for dierent cases.

1
Contents
1 Introduction 3
2 Study of Cd for 0° angle of incidence 3
2.1 Evolution of Cd for dierent Reynolds's number: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 Evolution of velocity near boundary layer: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.3 Comparison of Velocity and Turbulence Kinetic Energy: . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.4 Grid convergence for Wall function approach: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.5 Comparison of computed Cd with McCroskey 1987: . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
3 Study of case with non-zero angle of attack 11
3.1 Comparison for Cd and Cl graph with Gregory 1970: . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3.2 Grid convergence and comparison between dierent ow models: . . . . . . . . . . . . . . . . . . . . 12
3.3 Compare Euler and Turbulent case: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
4 Inuence of numerical parameters 18
4.1 Inuence of order discretization for dierent schemes: . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
4.2 Inuence of under-relaxation parameters: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
4.3 Inuence of rened box at wake: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
5 References 23

List of Figures
1 Evolution of Cd for dierent Renolyd's number . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2 Evolution of velocity near boundary layer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
3 Grid Convergence: Cd vs Diusion depth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
4 Comparison with McCroskey: Cd Vs Re . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
5 Comparison with Gregory: Cd vs Cl . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
6 CPU time for dierent Grids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
7 Cp comparision for Euler and Turbulent K-w model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

List of Tables
1 Evolution of Velocity prole . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2 Evolution of Flow Through Chord . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
3 Evolution of Turbulent Kinetic Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
4 Grid convergence for diferent Mesh . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
5 Evolution of Pressure Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
6 Grid convergence for diferent Mesh . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
7 Kinetic Energy between dierent Flow Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
8 Viscosity between dierent Flow Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
9 Kinetic Energy between dierent Flow Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
10 Residuals of forces for Euler and Turbulent K-w model case . . . . . . . . . . . . . . . . . . . . . . . . 17
11 Turbulent Kinetic Energy for dierent Numerical Schemes . . . . . . . . . . . . . . . . . . . . . . . . . 18
12 Frequency for dierent Numerical Schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
13 Viscosity for dierent Numerical Schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
14 Comparison of order of discretzation schemes between . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
15 Inuence of Under-Relaxation parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
16 Inuence of Rened box . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

2
1 Introduction
The NACA airfoils are airfoil shapes for aircraft wings developed by the National Advisory Committee for Aeronautics
(NACA), USA. The shape of the NACA airfoils is described using a series of digits following the word "NACA":- The
NACA four-digit wing sections dene the prole by:-
1. First digit describing maximum camber as percentage of the chord.
2. Second digit describing the distance of maximum camber from the airfoil leading edge in tens of percent's of
the chord.
3. Last two digits describing maximum thickness of the airfoil as percent of the chord Example- The formula for
the shape of a NACA 00xx- implies that the 00 indicating that it has no camber. The xx indicates that the airfoil has
an xx% thickness to chord length ratio: it is xx% as thick as it is long.
The computations are done on FINE/MARINE software by Numeca International, using the ISIS-CFD Navier-
Stokes solver developed by the METHRIC group of Laboratory of Fluid Mechanics at ECN. this

2 Study of Cd for 0° angle of incidence


2.1 Evolution of Cd for dierent Reynolds's number:
At low Reynolds number, the drag coecient is asymptotically proportional to the inverse of the Reynolds number,
which means that the drag is proportional to the speed. The graph in g.1 shows how the drag coecient varies with
Reynolds number for the case NACA 0012 airfoil.
FD
CD = 2 (1)
Aρv 2
FD is the drag force,ρ is the density of the uid,v is the speed of the object relative to the uid, A is the cross
sectional area and CD is the drag coecient  a dimensionless number. The drag coecient depends on the shape of
the object and on the Reynolds number.
ρvD
Re = (2)
µ
In our case we have assumed the value of ρ = 1 , initial velocity v = 1 and D = 1leading to µ = Re
v
. Finally we get
the relation CD α Re , where CD is inversely proportional to Re. For the aforementioned relation we see that the drag
1

coecient,CD decreases with increase in Reynolds's number, Re.

Figure 1: Evolution of Cd for dierent Renolyd's number

3
2.2 Evolution of velocity near boundary layer:
In near wall function, a large number of computational cells
are generally used to resolve the boundary layers with rst
cell located at y + = 1(Viscous sub-layer). In order to avoid
the excessive grid requirement we use wall function which
approximates the ow prole in viscous sub-layers and has
its rst cell at y + = 20 (buer layer), we do this since the
turbulence is negligible in the viscous sub-layer. The gain in
computational eciency is not only by reducing the number
of calculated grid but also to decrease the cell aspect ratio
near the wall.

(a) Y+ Near body wall (b) Wall-function

Figure 2: Evolution of velocity near boundary layer

As expected we see that the Near wall function had captured more accurate velocity near the boundary on compared
to the wall function approach. Though we donot have better accuracy of velocity prole near the boundary we still
get better results beyond it for wall function approach.

2.3 Comparison of Velocity and Turbulence Kinetic Energy:


The simulation results of velocity is observed on leading
edge and trailing edge followed by kinetic energy which was
observed on the turbulent region for the foil. The experi-
ments were conducted for xed values of Renolyd's number
2e6 with 0deg angle of incidence and compared for two dif-
ferent mesh (course and ne) for both near wall and wall
function approach.

4
Evolution of Velocity prole
Leading-Edge
Near-all function Wall function

Trailing-Edge
Near-all function Wall function

Table 1: Evolution of Velocity prole

5
As discussed before on close observation we see that the velocity prole is well captured for Near wall approach
than Wall function approach. It is also evident that velocity eld is well captured for ne mesh than coarse mesh
which relatively has less number of cells. On using near wall approach we observe that the sub viscous layer had
captured the velocity prole with greater accuracy at the trailing edge, the arrows manifest the speed of the velocity
which is almost equal to zero very close to the wall of the airfoil whereas on using wall function we do not observe the
the boundary layer prole as shown by the velocity vectors.

Evolution of Flow Through Chord


Near-all function Wall function

Table 2: Evolution of Flow Through Chord

We observed the turbulent kinetic energy eld at wake for wall and near wall function on two dierent grids. The
computation was done for the same values of Reynolds and incidence angle.

6
Evolution of Turbulent Kinetic Energy
Trailing-Edge
Near-wall function Wall function

Table 3: Evolution of Turbulent Kinetic Energy

On comparing the coarse and ne grid we observe that ne mesh has denser trail than coarse mesh because ne
mesh generally has lesser numerical diusion leading to better capturing of turbulent kinetic energy. On the other
hand we do not see any major dierence between the wall function and near wall in both the computed mesh.

2.4 Grid convergence for Wall function approach:


In this sub section we computed the simulation for four dierent grids coarse, medium, ne and super ne. We have
compared velocity, turbulent viscosity and turbulent kinetic energy elds for the aforementioned grids in the table
below for the same parameters as discussed before.

7
Comparison between dierent Grids
°
Re: 2e6, 0 angle of incidence, Wall-function approach
Trailing-Edge Kinetic Energy Velocity through Chord Trailing-Edge Viscosity

Table 4: Grid convergence for diferent Mesh

After comparing for all four grids we conclude that better mesh quality always leads to better capturing of the
solution elds. The superne mesh with maximum number of cells has the best visualization compared to other
meshes. The intensity of kinematic eddy viscosity eld is more narrow as the number of cells increases, which proves
again that the numerical diusion is lesser as we improve the grid quality.

8
Figure 3: Grid Convergence: Cd vs Diusion depth

The principle concept of grid convergence states that as the mesh resolution is increased in all areas of the domain,
the discretization error in the solution should vanish and the discretized solution should gradually tend towards the
continuous solution. For a second-order accurate scheme GDS, where the discretization error varies as O(h2 ), and
where h represents some measure of the mesh spacing, solution values such as lift and drag plotted versus h2 should
produce a straight-line plot which extrapolates to the continuous solution values in the limit as h  0. The dierence
between the drag coecient for two consecutive grids decreases gradually as we increase the number of cells. So we
see that grid convergence is very sensitive to the grid quality, hence we observe grid convergence as the grid gets ner.

9
2.5 Comparison of computed Cd with McCroskey 1987:

Figure 4: Comparison with McCroskey: Cd Vs Re

The comparison between the simulation results with the ne grid and the experimental data from McCroskey for transitional
boundary layer showed an excellent agreement, with maximum error of about 3.6%. It was also observed that as the Reynolds
number increased the drag coecient decreased. The reduction of CD with Reynolds number gradually decreases as we increase
the the ow speed. The ne mesh with wall function approach agrees well with the experimental data.

10
Pressure eld around NACA aerofoil
Angle of attack 0 ° Angle of attack 3 °

Angle of attack 6 ° Angle of attack 9 °

Table 5: Evolution of Pressure Field

° ° °
The above table shows the pressure elds around the NACA aerofoil for dierent angle of attack 0 , 3 , 6 and 9 . °
°
In 0 case, pressure eld is symmetric about x axis. However with the increase in the angle of attack we see creation of
low pressure region on the top of aerofoil and high pressure region at the bottom near the leading edge, which results
into lift on the aerofoil. The peak value of pressure is observed at the leading tip of the aerofoil.

3 Study of case with non-zero angle of attack


3.1 Comparison for Cd and Cl graph with Gregory 1970:
The below graph shows the evolution of drag and lift coecient computed using ne mesh, compared with the exper-
imental results of Gregory et al. 1970. The computational results are in quite good agreement with the experimental
results. As we increase the angle of attack of air, we see an increase in lift coecient as well as in drag coecient.
º
Therefore, drag is the price to pay for more lift. The minimum of drag-lift ratio occurs at 0 angle of incidence,
meaning that at this angle we obtain most of the lift for a least drag.

11
Figure 5: Comparison with Gregory: Cd vs Cl

3.2 Grid convergence and comparison between dierent ow models:


On comparison of computational results for dierent grids we nd that turbulent kinetic energy and trailing edge
viscosity are well captured in a ne mesh. Hence we notice a longer trail of turbulence as we increase the mesh quality.
This is because of the fact that as we go for more and more ner mesh we impose less numerical diusion into the
solution, nevertheless its obvious that physical diusion imposed by the model prevails irrespective of the mesh quality.
Therefore, we get more accurate results in a ne mesh as shown in below table.

12
Comparison between dierent Grids
°
Re: 2.88e6, 3 angle of incidence, Wall-function approach, K-w Model
Trailing-Edge Kinetic Energy Trailing-Edge Viscosity

Table 6: Grid convergence for diferent Mesh

On comparing the CPU time for dierent computations, we nd that ne mesh takes more time and memory to
converge because of increased number of cells in the grid. The advantage of using more cells is that we get more
accuracy in the solution. If we compare near wall and wall function computations for same grid wall function takes
more time to converge. Generally near wall needs ner mesh and solves more equation near the viscous layer for this
reason wall function approach was introduced so that we use less number of cells to achieve same accuracy away from
the wall.

13
Figure 6: CPU time for dierent Grids

Here we compare the viscosity and turbulent kinetic energy for both wall and near wall function for three dierent
ow models such as Spallart, EASM and k-w SST model. We observe that the kinetic energy does not vary too much
when we change the ow model whereas the turbulent viscosity shows some dierence where in which k-w model under
predicts the ow and EASM remains near to perfect in this case.

14
Kinetic Energy between dierent Flow Models
°
Re: 2.88e6, 3 angle of incidence, Fine Mesh
K-w SST Flow Model EASM Flow Model

Table 7: Kinetic Energy between dierent Flow Models

Viscosity between dierent Flow Models


°
Re: 2.88e6, 3 angle of incidence, Fine Mesh
K-w SST Flow Model EASM Flow Model Spallart Flow Model

Table 8: Viscosity between dierent Flow Models

15
If we compare CPU time for computation with dierent turbulence models we nd that Spallart model is most
ecient and converge very fast, with EASM taking maximum time to converge. As Spallart is a relatively simple one
equation model

Cp and CPU time comparison for dierent Flow Models


°
Re: 2.88e6, 3 angle of incidence, Fine Mesh
CPU time Near Wall function CPU time Wall function

Cp Near Wall function Cp Wall function

Table 9: Kinetic Energy between dierent Flow Models

The co-ecient of pressure for all three cases are exactly the same, since this is a 2D computation all the ow
models are near to perfect whereas if u compute a complex 3D ow we will observe that k-w SST to be the most
accurate model among all.

3.3 Compare Euler and Turbulent case:


We have plotted the evolution of co-ecient of pressure alone the aerofoil for Eulerian and turbulent models. We
observe that the coecient of pressure is almost the same expect for two peaks at the trailing edge, which are
explained as shock waves.

16
Figure 7: Cp comparision for Euler and Turbulent K-w model

The below table shows the evolution of residual of global forces with number of iterations for Euler and K-w SST
°
model. The ow is modeled using a ne mesh with Reynolds no. 2.88e6 and angle of attack 3 . The residual in x
direction is quite low at last iteration using Euler approach compared to K-w model with near wall and wall function
approach. Euler approach consists of less no. of equations to solve to compute the ow around aerofoil.

Global Forces for Euler and Turbulent case


°
Re: 2.88e6, 3 angle of attack, Fine Mesh
Fx Vs Total span Fy Vs Total span

Table 10: Residuals of forces for Euler and Turbulent K-w model case

17
4 Inuence of numerical parameters
4.1 Inuence of order discretization for dierent schemes:
On changing the order of discretization scheme from default order 2 to 1 for upwind scheme, we see it captures very
poor turbulent kinetic energy eld. This is obvious because higher order of discretization scheme improves the accuracy
of the solution. We nd same results for the AVL-SMART scheme as shown below. Overall, upwind scheme with
order 2 gives a better visualization of turbulence eld around the wake.

Turbulent Kinetic Energy for dierent Numerical Schemes


°
Re: 2.88e6, 3 angle of incidence, Fine Mesh, Near Wall function
UPWIND Scheme AVL-SMART Schme GDS Scheme

Table 11: Turbulent Kinetic Energy for dierent Numerical Schemes

However, there is no eect on computation of turbulent frequency eld on changing the order of discretization
scheme as shown below.

18
Frequency for dierent Numerical Schemes
°
Re: 2.88e6, 3 angle of incidence, Fine Mesh, Near Wall function
UPWIND Scheme AVL-SMART Schme GDS Scheme

Table 12: Frequency for dierent Numerical Schemes

We nd the same results for Kinematic eddy viscosity eld around the wake as in the case of turbulent kinetic
energy, second order discretization gives better results. We see high eddy viscosity eld near the wake region which is
not visible with rst order discretization.

19
Viscosity for dierent Numerical Schemes
°
Re: 2.88e6, 3 angle of incidence, Fine Mesh, Near Wall function
UPWIND Scheme AVL-SMART Schme GDS Scheme

Table 13: Viscosity for dierent Numerical Schemes

If we compare residual of velocity and pressure, Upwind scheme with order 1 has lowest order of residuals whereas
GDS scheme comes out to be most inaccurate.
The CPU time for both upwind order 1 and order 2 is same but GDS scheme converges in very short time.

20
Residual and CPU time comparison for dierent Numerical Schemes
°
Re: 2.88e6, 3 angle of incidence, Fine Mesh
CPU time Near Wall function ResU vs Iterations

ResV vs Iterations ResP vs Iterations

Table 14: Comparison of order of discretzation schemes between

4.2 Inuence of under-relaxation parameters:


1. Increasing relaxation parameter for velocity with constant pressure relaxation

ˆ Robustness of solution is still maintained, order of the residual remain same for complete range.
ˆ For very high velocity relaxation there is a signicant increase in residuals of velocity.
ˆ Few initial iterations have better convergence for high relaxation parameter as dierence between two consecutive
iteration is high but the nal residual remain as mentioned above.

2. Increasing relaxation parameter for pressure with constant velocity relaxation

ˆ Evolution of residual of velocity remain same for whole range except little variation in Y component
ˆ There is a very small increase in residual of pressure

21
Inuence of Under-Relaxation Values
Evolution of covergence and CPU time
ResU for Constant Pressure relaxation ResV for Constant Pressure relaxation

ResV for Constant Velocity relaxation ResV for Constant Velocity relaxation

ResP for Constant Velocity relaxation

CPU time for Constant Pressure relaxation CPU time for Constant Velocity relaxation

Table 15: Inuence of Under-Relaxation parameters

22
4.3 Inuence of rened box at wake:
On increasing the global no. of iterations and lowering cell size limit to half, we get a better capturing of turbulent
kinetic energy around the wake as compared in below table. This is because of the same reasons as discussed before
the diusion imposed by the ow models are well approximated as we rene the wake region whereas without the
rening box the ne mesh has more numerical diusion.

Inuence of Rened box


°
Re: 2.88e6, 3 angle of incidence
Turbulent Kinetic Energy for Fine Mesh Turbulent Kinetic Energy for Fine Mesh with Rened box

Table 16: Inuence of Rened box

5 References
1. Lecture notes by Researcher Michel Visonneau and Prof. Alban leroyer.

2. Mc Croskey 1987 and Gregory et al.1970.

3. www.wikipedia.com

4. www.cfd-online.com

Acknowledgement
We would like to express our deepest gratitude for all the support and patience of our advisors Researcher Michel
visonneau and Professor Alban leroyer who taught us to work on this FINE/MARINE suite. We extend our sincere
thanks to the professors for all their assistance.

23

You might also like