Beruflich Dokumente
Kultur Dokumente
September 1, 2003
1 Abstract
The solutions of the Euler equations are the approximate solutions of the
Naviers-Stokes equations in the limit of vanishing viscosity ( viscosity µ → 0
but µ = 0). These solution are used to predict the lift experienced by airfoils
and wings within the framework of inviscid flow, at a certain angle of attack.
The classical Kutta-Joukowski hypothesis enables us to determine these solu-
tions by imposing the Kutta-Joukowski condition at the sharp trailing edge of
the airfoil.
In this work, we study the question of how the circulation required for lift is
produced when time marching Euler calculations are performed for an airfoil.
We discuss the vorticity production, within the framework of inviscid calcula-
tion, and its role in the generation of the lift within the framework of Euler
codes used in CFD.
2 Introduction
One of the fascinating aspects of aerodynamics is the classical theory of lift.
According to this theory, it is possible to predict the lift of airfoils, wings and
wing-bodies within the framework of inviscid theory, that is without solving
the Naviers-Stokes equations of fluid flow. Sharp edges play a very crucial
role in this theory of lift. For airfoils with smooth trailing edges, it is not
possible to determine the lift without having recourse to the solution of the
Naviers-Stokes equations. The role of the reduced Naviers-Stokes equations
(with various terms neglected) in determining the approximate flow field and
computing various aerodynamic coefficients cannot be ignored in design and
analysis aerodynamics. It is neither always possible nor necessary to solve the
∗
Names listed in alphabetical order
1
full Naviers-Stokes equations at every stage of the design cycle and hence the
importance of reduced Naviers-Stokes equations. One such approximation is
the solution of the Euler equations which further reduces to the solution of
Laplace’s equations in the incompressible limit under the assumption that the
flow is irrotational. There is one category of solutions of the Euler equations
which constitutes a class of approximate solutions of the Naviers-Stokes equa-
tions in the limit of viscosity µ → 0 but µ = 0. These solutions are termed
by Lagerstrom as the relevant Euler solutions as mentioned in Salas[1]. One of
the main disadvantages with this type of solution is that while the prediction of
lift is accurate enough, the prediction of the drag is in general subject to error.
In fact, viscosity causes skin friction which always results in the production of
a non negligible finite drag.
The classical Kutta-Joukowski hypothesis enables us to determine the relevant
Euler solution by imposing the famous Kutta-Joukowski condition, namely, the
flow should leave the sharp trailing edge smoothly. This condition has been
found to pick up the relevant Euler solution to a very good accuracy and has
been widely used to compute the pressure distribution around airfoils, wings
and wing-bodies. In the present work, for the sake of completeness, we give
a brief description of the classical theory of lift and the potential flow for a
circular cylinder. We consider lifting flows for an airfoil, in the framework of
an inviscid compressible flow and discuss the mechanism of generating the lift
for an airfoil in an inviscid fluid. We then review briefly the Euler codes, rang-
ing from those based on the Finite Volume methods to those of node based
methods, applied to lifting flows. In this review, we study how the Euler codes
deal with the Kutta-Joukowski condition. We have tried to classify the dif-
ferent methods of implementing the Kutta-Joukowski into the Weak imple-
mentation of Kutta-Joukowski condition and Strong implementation of the
Kutta-Joukowski condition [4]. We describe the new theory on generation of
circulation [3, 4] , when time marching Euler calculations are performed with
the Kutta-Joukowski condition imposed at the trailing edge. We then discuss
the vorticity production and distribution in the computational domain when an
airfoil with a sharp trailing edge is immersed in an inviscid compressible flow.
We finally present the results of our numerical experiments.
2
in this formulation. Even then, these codes have been used and are being
widely used to compute the flow field and obtain the aerodynamic coefficients
quite accurately. A fluid dynamically interesting question arise at this point
regarding how the Kutta-Joukowski condition is satisfied in this formulation:
Is it unnecessary because it is automatically satisfied?
Let us look at some of the issues involved in more details. We consider the case
of a 2-D airfoil with a sharp trailing edge and further consider finite volumes
near the trailing edge and elsewhere on the body (Fig.(8)). Fig.(8) shows five
cells with centroids a, b, c, d and e. While updating the state variables at the
centroid a, it is necessary to compute the fluxes on all the edges or faces of the
cell. In particular, it is required to determine the flux on the cell face lying
on the body (shown hatched in Fig.(8)). On this face, flow tangency boundary
condition has to be imposed and this can be done in many ways. One way is
to use the Kinetic Characteristic Boundary Condition based on the reflection
principle [7]. This cell face does not pose any special problem in implementing
the solid wall boundary condition. Now consider the cell with centroid b. One
of its edges is (1 , 2 ) and the node 2 is the Kutta point. While imposing the
solid wall boundary condition on that edge, a problem arises because all along
(1 , 2 ), the flow tangency boundary condition has to be imposed. But at the
Kutta point (node 2 ), we have two flow tangency boundary conditions - one
from above corresponding to the normal vector on the upper surface(nu ) and
one from below corresponding to the normal vector on the lower surface(nl ).
For bodies with a finite angle at the Kutta point, these two normals are different
and if qte is the velocity vector at the Kutta point then, we get
3
∂f ∂f
The gradient terms and are determined from the neighbour-
∂x b ∂y b
hood data and again, no attention is given to the Kutta point. The main point
is that the solid boundary condition suddenly changes from qte · nu = 0 (at a
point vanishingly close to the trailing edge) to qte · nl = 0 at a point on the
lower surface but vanishingly close to the Kutta point. Many schemes of im-
plementing the solid boundary condition do not take into account this sudden
change in boundary condition. More or less the same criticism is valid for all
cell centered finite volume codes. It is interesting and important to note that
these Euler codes give satisfactory results in spite of the above criticism. How
then does the fluid leave the trailing edge smoothly in these computations? It
is argued that the centroids c and d communicate with each other via the cell
face (2 , 3 ). The flux on the cell face (2 , 3 ) is the same in the state update for
centroids c and d . So, the argument can be expressed as follows : The flow will
be prevented from suddenly turning around the Kutta point thus leading to the
satisfaction of the Kutta-Joukowski condition. It is evident that the question of
determining the values of pressure and density at the Kutta point is left open in
these formulations. One can also determine these values from the neighbouring
data but it has been found that Euler codes give widely different values of the
coefficient of pressure at the trailing edge (Cp te ).
In the cell vertex schemes based on dual finite volume approach (sometimes
called covolume methods), it is not possible to skirt the issue of applying the
state update at the Kutta point. A typical dual finite volume around the Kutta
point is shown in Fig. (8). Again the problem of determining the fluxes on the
cell faces lying on the solid wall arises.
The necessity of updating the state variables at the Kutta point comes into
sharper focus for the LSKUM [7] and the q-LSKUM [6]in which we obtain the
space derivatives at the Kutta point 2 (Fig. (2) in terms of the neighbouring
data. The basic question is : Given the flow variables ρnkp , ukp n , v n , pn at
kp kp
time level n at the Kutta point, how to determine ρkp , ukp , vkp , pn+1
n+1 n+1 n+1
kp
using the data in the connectivity of the Kutta point. In order to solve this
problem, we need to find the solution of the partial differential equation of the
flow together with the prevailing boundary condition. It is thus essential to
have a clear mathematical way for updating the state at a sharp edge singular-
ity. This formula will obviously be somewhat different from that at an interior
non-singular point.
We now introduce the following definitions :
(i)Weak implementation of the boundary condition at the Kutta point
(ii)Strong implementation of the boundary condition at the Kutta point
The Weak implementation is employed in the Cell Centered Finite Volume Method,
where the question on state update at the Kutta point, has been discussed
above. Mathur [9] and many others have extensively employed the Weak im-
plementation of the Kutta condition in all their calculations. In the strong
implementation, the state update is directly applied to the Kutta point.
A question can now be raised regarding airfoils with a cusp at the trailing edge
4
instead of a finite angle at the trailing edge considered so far. In such a case,
nu = −nl (3)
and qte · nu = 0 implies qte · nl = 0. Hence, we can not conclude that qte = 0
at a cusp, even under the assumption of a single valued fluid velocity. Suppose
the angle at the trailing edge is θ then we have the following mathematical
situation
qte (θ) = 0 f or θ > 0 (4)
qte (θ) · nu = 0 f or θ = 0 (5)
Assuming single valued q, we have qte (θ) = 0 however small θ is and we do
not know whether it is zero for θ = 0. If we assume that qte is a continuous
function of θ, then, we can assert that the condition (4) is still valid and does
not contradict condition (5). The validity of condition (4) for θ = 0 has to
be investigated and till then, no definite conclusion can be drawn. All the
calculations made in this work are for θ = 0, that is, the airfoils have a finite
angle at the Kutta point. In this section, we are going to examine closely the
flow variables at the trailing edge. We shall be using the following notations:
Pote : total pressure at the Kutta point.
pte : static pressure at the Kutta point.
PoL : total pressure at a point L on the lower surface and vanishingly close to
the trailing edge.
PoU : total pressure at a point U on the upper surface and vanishingly close
to the trailing edge.
pte,L : static pressure at the L.
pte,U : static pressure at the U .
Mte,L : Mach number at L.
Mte,U : Mach number at U .
qte,U : The fluid velocity at U .
qte,L : The fluid velocity at L.
By the principle of continuity of static pressure near the trailing edge, we have
From this common value of static pressure at the trailing edge, we can obtain
PoL and PoU by using the isentropic relations
γ
(γ − 1) 2
PoL = pte,L 1 + Mte,L γ − 1 (7)
2
and
γ
(γ − 1) 2
PoU = pte,L 1 + Mte,L γ − 1 . (8)
2
Evidently, we need to know MteL and MteU in order to get P0L and P0U .
We now consider the subsonic case and transonic case separately since for the
former, the fluid variables are single-valued while for the latter, some of them
are not.
5
3.1 Subsonic case
Since streamlines cannot suddenly turn at the Kutta point in a subsonic flow,
and further, assuming single-valued property of the flow variables, it is obvious
that the Kutta point is a stagnation point. Thus, qte = 0 and there is no loss
in total pressure as we move along a streamline. Hence,
Since the streamlines can not suddenly turn at the Kutta point, we can assume
as before that MteU = MteL = 0 . This contradicts Eq.(11). We therefore give
up the assumption that MteU = MteL = 0 .
We consider another possibility, that is, the stagnation point is located at L
and the flow is smooth at U (see Fig.(1)). We have PoU < PoL = Po∞ , that
is,
PoU
<1 (12)
PoL
However, Eq.(12) contradicts the fact that the static pressure is always less
than the stagnation pressure at any point within the computational domain,
particularly at L. The only possibility left is the stagnation point located at
U and we shall show that it is indeed the case that the fluid stagnates when
the shock is found on the upper surface of the airfoil. Since U is a stagnation
point, it follows that
pte,U = Po,U . (13)
Using Eq.(6) and Eq.(14), PoU = pte and
γ
(γ − 1) 2
Mte,L γ − 1 > 1
PoL PoL
= = 1+ (14)
Pote PoU 2
As a result, qte,L = 0 and this is consistent with the last assumption. We can
apply the same reasoning in a transonic case with a shock on the lower surface
and show that L is a stagnation point.
We can therefore summarize the flow near the trailing edge as follows :
6
In subsonic flow, the Kutta point is a stagnation point. For all the flow situa-
tions considered, the flow leaves the trailing edge smoothly thus satisfying the
Kutta-Joukowski condition. For the transonic case, except from the pressure,
all the other flow variables are discontinuous across the thick horizontal line
shown in Fig.(8). This suggests that continuity of pressure is one of the crite-
rion that can be enforced in order to satisfy the Kutta-Joukowski condition.
We shall next analyze different ways of enforcing the Kutta-Joukowski condi-
tion when using a node based method while taking into account the state of
the flow at the trailing edge as just described.
7
computed flow.
8
The left hand side of Eq.(16) has the time rate of change of ω and the term
giving vorticity advection. The right hand side of Eq.(16) vanishes when p and
ρ are isentropically related. If ∇p is not parallel to ∇ρ, then this term produces
vorticity. Such a production of vorticity takes place through what is termed as
the baroclinic mechanism. For inviscid flows with the Kutta-Joukowski condi-
tion prevailing at the trailing edge, there is a sudden discontinuity in the wall
boundary condition. At the sharp trailing edge, the Kutta-Joukowski condition
requires u1 = 0 and u2 = 0 , that is, the velocity vector is zero at the trailing
edge while flow tangency boundary condition at the other nodes on the body
requires only the normal component of velocity to be non-zero (qn = q · n
= 0). As a result, sharp velocity gradients are developed. A non-isentropic
change thereby takes place in the flow field because the fluid elements in the
neighbourhood of the Kutta-point are brought to rest abruptly at the trailing
edge. Thus, strong gradients in pressure and density are set up and they in
turn produce vorticity.
Another example where vorticity is produced via the baroclinic mechanism is
in the case of a supersonic flow(M∞ > 1). As shown in Fig.(8), there is a Bow
Shock ahead of the leading edge and the flow is no more isentropic in region
R1 . Therefore, ∇p ∧ ∇(ρ) = 0 and hence, vorticity is produced in that
region.
1
Our numerical experiments show that the term − 2 ∇p ∧ ∇(ρ) is very
ρ
large near the trailing edge. As mentioned earlier in this section, it is believed
that the sharp trailing edge triggers this term and vorticity production takes
place at the Kutta point. In these experiments [4], we note that the vorticity
is a maximum at the trailing edge. As stated earlier, the lift experienced by
an airfoil in a subsonic flow at a given angle of attack is a consequence of the
circulation around the airfoil. Consequently, the existence of the circulation is
given by
Γ = − q.ds = − ω . dS (18)
c s
where ω = ∇ ∧ q, q being the velocity vector at the corresponding node,
implies that the vorticity ω must be non-zero in the domain.
In the numerical experiments performed using q-LSKUM based Euler code and
PLSI at the Kutta point, the vorticity at each node of the domain is computed
when the steady state is reached. It is observed that the vorticity is a maximum
at the trailing edge when compared to the rest of the domain. The numerical
value of the vorticity at the trailing edge increases sharply as the grid is re-
fined.Even though the vorticity is everywhere zero (except at one point) we
still get a finite circulation around the airfoil.
For low Mach number computations also, that is, in the incompressible limit,
our numerical investigations show that the baroclinic mechanism is present
(Fig.(??) and Fig(5)). In other words, the flow is always compressible in the
neighbourhood of the trailing edge. The baroclinic mechanism requires com-
pressibility and it is interesting to note its presence near the sharp edge even
near the incompressible limit.
9
7 The baroclinic term
Fig.(3) shows the distribution of the baroclinic term for 120x60 (coarse) and
160x90 (medium) grids respectively in steady state. It is clear that the baro-
clinic term , is numerically highest at the leading edge; and it is quite large
at the trailing edge. The latter observation confirms that after the vorticity is
being produced (due the baroclinic term (Eq. (17), it is convected along the air-
foil and it accumulates at the trailing edge where it is found to be numerically
largest. Our computations begin with impulsively started initial conditions,
that is, velocity is set to free stream velocity everywhere in the domain at t = 0
and then at t = 0+ its normal component on the surface of the airfoil suddenly
becomes zero. Very large gradients of velocity, density and pressure are thereby
generated close to the surface of the airfoil, causing large values of baroclinic
term and these in turn produce vorticity on the surface of the airfoil. Fig.(3)
shows that the baroclinic mechanism is active on the surface of the airfoil even
in steady state. The same trend is confirmed at low Mach number, the baro-
clinic term is large in the suburb of the solid boundary (Fig.( 5)). Also, In
particular, it is large near the leading edge and the trailing edge. Also, Fig.6
show that the divergence is very large at the trailing edge even in the incom-
pressible limit. It is in fact a maximum at the trailing edge, confirming that
sharp velocity gradients are developed in the region close to the Kutta point
even at low Mach number.
8 Conclusions
We have made an attempt to understand how the circulation necessary for
lift is generated within the framework of inviscid compressible flow. It is
suggested that the baroclinic mechanism, activated by flow tangency and the
Kutta-Joukowski condition, is responsible for generating vorticity and therefore
10
circulation. Our grid refinement study shows that the vorticity distribution at
steady state increases at the trailing edge. That is, vorticity is either produced
by the baroclinic mechanism or near the solid boundary by viscosity. We con-
clude by proposing the following tentative hypotheses [6]:
Hypothesis (1) :
For inviscid compressible flows past lifting airfoils, vorticity is produced by the
baroclinic mechanism.
Hypothesis (2) :
For inviscid compressible flows past lifting airfoils, the vorticity is very large at
the trailing edge.
Acknowledgements
The author is thankful to the University of Mauritius, Rep. of Mauritius and the
CFD Centre, IISc, Bangalore,India for this work.
References
[1] Salas M.D., Foundations for the Numerical solutions of the Euler equa-
tions, Recent advances in Numerical Methods in Fluids, W.G. Habashi,
Editor, Pinewood Press Ltd., Swansea, U.K., pp. 875-912, 1985.
[2] Balasubramaniam R., Numerical experiments with Higher Order Least
Squares Kinetic Upwind Method, M.E. Thesis, Indian Institute of Science,
Bangalore, India, 1997.
[3] Balasubramaniam R., Ramesh V. and Deshpande S.M., On Kutta-Joukowski
Condition, The Seventh Asian Congress of Fluid Mechanics, Chennai, India,
December 1997.
[4] Dauhoo M.Z., Raghuramarao S.V., Ramesh V. and Deshpande S.M., A Strong
Implementation of the Kutta-Joukowski Condition using Peculiar Ve-
locity Based Upwind Method, 16th ICNMFD,Lecture Notes in Physics,
Springer, 1998.
[5] Dauhoo M.Z., Ghosh A.K., Ramesh V. and Deshpande S.M., q-LSKUM a new
Higher Order Kinetic Methods For Euler Equations Using The Entropy
Variables , 8th International Symposium on Computational Fluid Dynamics,
Bremen(Germany), September 5-10 1999.
[6] Dauhoo M.Z.The Least Squares Kinetic Upwind Method based on En-
tropy variables and its Application to the Understanding of the Aero-
dynamics of Lift, Ph.D. Thesis, Dept. of Mathematics , Faculty of Science,
University of Mauritius, July 2000.
[7] Ghosh A.K. and Deshpande S.M., A Robust Least Squares Kinetic Upwind
Method for Inviscid Compressible Flows, AIAA Paper No. 95-1735.
[8] John Andersson, Jr., Modern Compressible Flow With Historical Perspec-
tive, Mc Grawhill International Editions, Aerospace Series, 1990.
[9] Mathur J.S. and Deshpande S.M., Reconstruction on unstructured grids
using an upwind kinetic method, Proc. 15th Int. Conf. on Numerical Methods
in Fluid Dynamics, Monterey, California, June 24-28, 1996.
[10] Milne-Thompson L.M., Theoretical Aerodynamics, Third Edition, London,
Macmillan and Co. LTD., New York, St Martin’s Press, 1958.
11
[11] Raghurama Rao S.V., New Upwind Methods Based on Kinetic Theory
for Inviscid Compressible Flows, Ph.D. thesis, Indian Institute of Science,
Bangalore, India, 1994.
3 4
a 1
b c 2 5
1
2 3 7
θ
e d
6
U
U
L L
Figure 1:
Shock
U.
α
L
U
8
Bow Shock
12
KP
(c) The
connec- Boundary Layer
8
the
Kutta (d) Viscous Flow over a Flat Plate
point
Figure 2:
Figure 3: Surface Plots of the Baroclinic term, at steady state, a coarse and a
medium grid respectively ( M∞ = 0.63, α = 2◦ )
13
6 6
0.4
10
5 5
0.2
4 4
5
0
3 3
0
−0.2
2 2
−5 −0.4
1 1
−0.6
0 0
−10
−1 −0.8 −1
−10 −5 0 5 10 −0.2 0 0.2 0.4 0.6 0.8 1 1.2
6 0.02 6
0.04
0.015
5 5
0.02 0.01
4 4
0.005
0
3 0 3
−0.005
−0.02 2 2
−0.01
1 1
−0.04 −0.015
−0.02
0 0
−0.06
−0.025
−1 −1
−0.1 −0.08 −0.06 −0.04 −0.02 0 0.02 0.04 0.06 0.08 0.1 0.98 0.985 0.99 0.995 1 1.005 1.01 1.015 1.02 1.025 1.03
Figure 5: The Baroclinic term for the flow past NACA 0012 at M∞ = 0.16, α
= 2.o on Fine Grid
4 4
0.3
10
3 3
0.2
5
2 0.1 2
0
0 1 1
−0.1
0 0
−5 −0.2
−1 −0.3 −1
−10
−0.4
−2 −2
−10 −5 0 5 10 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
14
4 4
0.06 0.03
3 3
0.04 0.02
0.02 0.01
2 2
0 0
1 1
−0.02 −0.01
−0.04 0 0
−0.02
−0.06 −0.03
−1 −1
−0.08 −0.04
−2 −2
−0.06 −0.04 −0.02 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.85 0.9 0.95 1 1.05 1.1
Figure 6: Contours of Divergence for the flow past NACA 0012 at M∞ = 0.16,
α = 2.o on Fine Grid
18000
4
x 10
16000
9
8 14000
6 12000
5
10000
4
3
8000
2
1 6000
0
1 4000
0.5 1.5
1 2000
0
0.5
−0.5
0
−1 −0.5
Figure 7: A 3-D Surface plot of the Vorticity distribution close to the airfoil for
the coarse grid
4
x 10
8
4
x 10
7
9
8
6
7
6
5
5
4
4
3
2
3
1
0
2
0.4
0.2 1.5
1 1
0
0.5
−0.2
0
−0.4 −0.5
Figure 8: A 3-D Surface plot of the Vorticity distribution (fine grid) close to
the airfoil in the limit of incompressibility for the flow past NACA 0012 at M∞
= 0.16, α = 2.o
15