Professional Documents
Culture Documents
1.
1.1
INTRODUCTION
Aeroelasticity
Static aeroelasticity
Dynamic aeroelasticity
Static aeroelasticity
Static aeroelasticity studies the interaction between aerodynamics and elastic forces on an elastic
structure. Mass properties are not significant in the calculations of this type of phenomena, since
inertial forces are completely excluded from such analysis.
Dynamic aeroelasticity
Dynamic aeroelasticity studies the interactions among unsteady aerodynamic, elastic, and inertial
forces.
1.2 Aeroelastic flutter
An example of dynamic aeroelastic phenomena is flutter, in which the flexibility and inertia of
the structure play an essential part in the dynamic stability of the total fluid-structure system. It
occurs when a structural system, under flow conditions beyond some threshold (critical) value of
the flow parameter (viz. critical dynamic pressure), is driven into unstable, and self-excited
oscillations due to unsteady aerodynamic forces from the flow. Flutter is basically a phenomenon
of unstable oscillations in a flexible structure. Beyond the critical flow conditions, the onset of
flutter instability is recognised by the exponential increase in the vibration amplitudes of the
structural system with time (Figure 1.1). Aircraft structures that function as lifting surfaces are
prone to flutter instability due to their interaction with the aerodynamic flow. The critical flow
condition that leads to the onset of flutter is called the Flutter Boundary of the structure. The
flutter boundary of an aerospace vehicle is a characteristic design parameter that is very
important for practical design of its lifting surfaces.
The mechanism of flutter can be explained from the physics of energy flow in the total fluidstructure system. Under sub-critical flow conditions, the structural oscillations in the
Displacement
V V f
Converging oscillation
Time
(i)
Displacement
V V f
Time
(ii)
Displacement
V V f
Time
(iii)
Figure 1.1. Nature of dynamic response (displacement) of a structural system subjected to aerodynamic
flow. For free stream flow velocities below a critical value, V V f , the oscillations are stable, as shown
in (i). At the critical flow velocity, the oscillations are un-damped, as shown in (ii). For velocities above
this critical value, V V f , the oscillations are unstable, as shown in (iii).
aerodynamic flow are stable, and thus damped out since the net aerodynamic power flow over
any oscillation cycle is less than what the structure actually dissipates out. At the flutter
boundary (of critical flow velocity), the aerodynamic power input equals the dissipated power,
and steady oscillations, of constant amplitudes, occur. Beyond this critical flow condition the
aerodynamic power input exceeds the dissipated power in each cycle, leading to increase of the
vibration amplitudes in time.
1.3 Prediction and cure
Besides aircraft structures, various other structural systems, like long span bridges, chimneys, tall
buildings etc. are prone to flutter instability. To ensure safety of these structures against
aerodynamic loads, it is necessary that they are designed to withstand severe wind conditions. It
is thus essential that the flutter boundaries of these structures are estimated, and it is ensured that
these are well above the worst aerodynamic loads that these structures are likely to encounter.
Prediction of the flutter boundary of a structure subjected to aerodynamic loads is essential to
ensure its safety against flutter. This involves making a mathematical model of the structure (say
an aircraft) with appropriate inertial and stiffness distributions. This idealized structure is then
analyzed with appropriate aerodynamic load simulations using various aerodynamic theories.
If a structural system is prone to flutter instability, or the safety margin is quite low, appropriate
cure for the problem can be prescribed through some ingenious redistribution of the inertial and
stiffness properties so that an increase in the flutter velocity can be achieved. This kind of
practice requires a reliable knowledge of the effects of changes of the various system properties
upon the flutter boundary.
1.4 Timoshenko Beam:
Timoshenko beam theory which is a higher order beam theory, is known to be superior in
predicting the transient response of the beam over the Euler Bernoulli beam model. In neglecting
the contribution of shearing deformation the Euler-Bernoulli beam theory (EBT) requires that
plane sections remain plane and perpendicular to the neutral axis after deformation.
Consequently, this theory is best suited for thin or slender beams as shear strains have a
considerable influence on the deformation of thick beams. A more accurate representation of
beam flexure which allows for the inclusion of shear strains present in isotropic beams and more
suited for thick beam analysis is the Timoshenko beam theory.
The Euler-Bernoulli beam theory (EBT) frequently used for the analysis of isotropic beams,
which have extensive use in engineering structures, describes beam kinematics completely in
terms of flexural deformation. In neglecting the contribution of shearing deformation the EBT
requires that plane sections remain plane and perpendicular to the neutral axis after deformation.
Consequently, this theory is best suited for thin or slender beams as shear strains have a
considerable influence on the deformation of thick beams. A more accurate representation of
beam flexure which allows for the inclusion of shear strains present in isotropic beams and more
suited for thick beam analysis is the Timoshenko beam theory. This theory, a first order shear
deformation theory (FSDT), relaxes the normality assumption of plane sections evident in the
EBT. By allowing for the inclusion of a constant through thickness shear strain, it violates the
no-shear boundary condition at the top and bottom horizontal beam surfaces, requiring a
problem dependent shear correction factor. Since Timoshenko Beam theory is known to be
superior in predicting the transient response of the beam, it can yield better results and
sometimes entirely new flutter modes and flutter velocities. Since flutter is a dynamic aeroelastic
phenomena, Timoshenko beam theory constitutes an improvement over Euler Bernoulli theory in
predicting the onset of flutter.
1.5
Sensitivity Study:
LITERATURE REVIEW
The earliest study of flutter seems to have been made by Lanchester [1], Bairstow and Fage [2]
in 1916. In 1918, Blasius made some calculations after the failure of the lower wing of Albatross
D3 biplane. But the real development of the flutter analysis had to wait for the development of
Non-stationary airfoil theory by Kutta and Joukowsky.
Glauret [4,5] published data on the force and moment acting on a cylindrical body due to an
arbitrary motion. In 1934, Theodorsen.s [6] exact solution of a harmonically oscillating wing
with a flap was published. The torsion flutter was first found by Glauret in 1929. It is discussed
in detail by Smilg [7] Several types of single degree of freedom flutter involving control surfaces
at both subsonic and supersonic speeds have been found [8,9], all requiring the fulfillment of
certain special conditions on the rotational axis locations, the reduced frequency and the mass
moment of inertia.
Pure bending flutter is possible for a cantilever swept wing if it is heavy enough relative to the
surrounding air and has a sufficiently large sweep angle [10]. The stability of more complicated
motions can be determined by calculating the energy input from the airstream. The bending
torsion case in an incompressible fluid has been calculated by J.H.Greidanus and the energy
coefficient in Bending-Torsion oscillations has been given [11]
The use of .Quasi-steady. Aerodynamic theory for the flutter analysis of the wings and excellent
treatises in the field of aeroelasticity are given by Y.C.Fung [12], E.H.Dowell [13,14],
L.Mirovitch [15] and others.
In the typical wing whose elastic axis (locus of shear centers) and mass axis (locus of center of
gravity) do not coincide, the nature of oscillations is always coupled flexure-torsion. A vast
literature exists on the flexure-torsion problem of engineering structures. Evins [16] has given
comprehensive details about vibration fixture transducers and instrumentation. Bisplinghoff and
H.Ashley [17] has described the elastic characteristics shape and inertial idealization.
A new method for determining mass and stiffness matrices from modal test data is described by
Alvin and Paterson [18]. This method determines minimum order mass and stiffness matrices,
which is used to determine the optimum sensor location. Dugundji [19] examined panel flutter
and the rate of damping. The problem of two and threedimensional plate undergoing cyclic
oscillations and aeroelastic instability is investigated by Dowell [13,14].
Abott [20] has suggested a technique for representing the shape of the aerofoil through analytical
relations. The coupled flexure-torsion vibration response of beam under deterministic and
random load is investigated thoroughly by Eslimy and Sobby [21] by use of normal mode
method. The exact determination of coupled flexure-torsion vibration characteristics of uniform
beam having single cross section symmetry is studied by Dokumaci [22].
At present, subsonic flight is a daily event and supersonic and hypersonic flights are a reality.
Now aeroelastic analysis has become an organic part of the design.
Rayleigh had proposed a formulation for dynamic analysis of beams where Euler beam model is
maintained for stiffness considerations, but rotator inertia has been taken into account. This
formulation yielded results that have lower errors than those of the Euler beam model without
rotational inertia considerations. However at high frequencies, the Rayleigh beam still showed
severe deviations .
Timoshenko realised the source of error, and took a ingenious step to create a beam model, now
popularly known as Timoshenko beam[23]. In his formulation, he maintained the rotatory inertia
terms as had been proposed by Rayleigh, but discarded the Euler beam model
The development of structural and finite element models of the Timoshenko beam theory has
been the subject of numerous papers in the literature[24-29]. The exact, 4 x 4 stiffness matrix of
the Timoshenko beam is derived either using the methods of structural analysis or finite element
formulations. The shear locking is due to the inconsistency of the interpolation used for w and ,
or equivalently, not satisfying the requirement that the shear strain xz = (dw/dx) + is elementwise constant for element-wise constant values of EI. Often, the Timoshenko finite element
models are based on equal interpolation of w and and use reduced-order integration to evaluate
the stiffness coefficients associated with the transverse shear strain and full integration for all
other coefficients.
Others have used so-called consistent interpolation based on the recovery of correct constraints
in the thick beam limit (Prathap & Babu [30]; Shi & Voyiadjis [31]; Rakowski [32]; Reddy
[33]). Although such elements do not experience locking, they do not lead to the two-node superconvergent element.
Friedman & Kosmatka [34] and Reddy [35] and Reddy et al [36] have independently developed
the two-node super-convergent element using the exact solution of the homogeneous form of the
Timoshenko beam equations. Hermite cubic interpolation of w and interdependent quadratic
interpolation of 4, was used in developing the element that has the super-convergence character
for static problems.
Reddy[37] has provided a excellent treatise on the dynamic behaviour of all the Timoshenko
beam finite elements.
A tool to find the gradient of a dynamic type constraint variable as a function of design
parameters has wide applications in complex engineering problems. Rogers [38] deduced an
expression for the derivative of eigenvalues and eigenvectors with respect to an arbitrary
parameter of a dynamic system, which can be represented mathematically by a linear, constant
coefficient differential equation. By using the expressions, a set of increments in the design
variables may be selected to yield the desired improvements in the system characteristics of
interest.
Adelman and Haftka [39] surveyed the methods applicable to the calculation of structural
sensitivity derivatives for finite element modelled structures and discussed literature published
on four main topics: derivatives of static response (displacement and stresses), eigenvalues and
eigenvectors, transient response and derivatives of optimum structural designs with respect to
problem parameters. The survey also includes a number of methods developed in non-structural
fields such as control and physical chemistry, which are directly applicable to structural
formulations.
Ringertz [40] has studied the optimal design problem of a wing in incompressible flow with
aeroelastic constraints. The weight of a cantilever wing is minimized using the thickness of the
composite face sheets as design variables subject to constraints on flutter and divergence speed.
A doublet-lattice panel method is used for computation of unsteady aerodynamic loads. Ringertz
discusses several difficulties with optimization of eigenvalues of un-symmetric and complex
matrices.
A methodology for carrying out analytical sensitivity analysis of the flutter phenomenon in long
span bridges has been discussed by Jurado and Hernandez [41]. A nonlinear eigenvalue problem
for the calculation of flutter instability has been modelled and is further used for the sensitivity
analysis of flutter instability with respect to key chosen design variables, moments of inertia of
the bridge deck. Testing of these derivatives has been performed through centred differences
method. They have done detailed studies on Great Belt, Vasco da Gama and Old Tacoma Bridge
based on the presented method.
3.
Reddy[37] has presented superconvergent finite element model for static problemsusing two
alternative approaches: (1) assumed-strain finite element model of the conventional Timoshenko
beam theory, and (2) assumed-displacement finite element model of a modified Timoshenko
beam theory.
The displacement field of the Timoshenko beam theory for the pure bending case is
(
( ),
( )
(3.1)
Where w is the transverse deflection and x, the rotation of a transverse normal line about the y
axis. The strains and stresses of the Timoshenko beam theory are
(3.2)
(3.3)
The equilibrium equations of the beam are
.
/1
....(3.4)
( )
(3.5)
where q(x) is the distributed transverse load, E Young's modulus, G the shear modulus, A the
area of cross section, I the moment of inertia, and Ks the shear correction factor.
3.1 Displacement Finite Element Models:
3.1.1 The General Model:
The displacement finite element model of the Timoshenko beam theory is constructed using the
principle of minimum total potential energy, or equivalently, using the weak form
0
.
(
(
(
,
)
)
(
)
(
.
,
,
)
/.
( )
( )
)
....(3.5)
/-
.
0
/1
/1
(3.6)
q(x)
x
a
Mea
Meb
2
Va
Vb
he
( )
( )
( )
(3.7)
where(W j,j ) are the nodal values of (w, x) and j()(x) ( = 1, 2) are the associated
interpolation functions. Substitution of (7) for w and x, and w = i(1) and x = i(2) into (3.5)
yields the finite element model
[
,
,
,
,
- * +
]2 3
- * +
2*
}
3
+
(3.8)
where
( )
`Kij12 =
Kij22 = (
( )
( )
( )
( )
( )
( )
( )
( )
Kij11 =
( )
( )
( )
( )
(3.9)
10
)[
]{
( )
(3.10)
Where
( )
( )
(3.11a)
(3.11b)
In the thin beam limit, i.e 0, the first and third equations of (3.10) imply the following
relation among (W1,W2,1,2) :
.
/ .
/
Which is equivalent to the Kirchoff constraint x +dw/dx =0 (or shear strain xz=0).
(3.12)
The second and fourth equations of (10), in view of (12), yield the constraint
(
(3.13)
This is equivalent to dx/dx =0, which is an incorrect condition to satisfy as it forces the
curvature and hence bending energy to zero. Thus, (3.10) in an effort to satisfy the constraints
(3.12) and (3.13), will yield the trivial solution W1=W2=1=2=0 (i.e., the element locks).
The Kirchoff condition (3.12) suggests that w and x be interpolated such that dw/dx is a
polynomial of the same order as x. If w is approximated using a linear polynomial ( a minimum
requirement), then x should be a constant. Since the minimum continuity requirement on x is
also linear, it follows that w be approximated using a quadratic polynomial. This is a consistent
interpolation. Unless the weak form of Timoshenko beam theory is modified, we have no
alternative but to use a quadratic approximation for w and linear for x and use full integration to
evaluate the coefficient matrices to obtain an element that does not experience locking. However,
if one approximates both w and x with linear polynomials but treats x as a constant in the
evaluation of the shear strain,
it will also yield the stiffness matrix. This procedure is known in the literature as reduced
integration of the shear stiffness. It amounts to evaluating the second term of Kij22 in (9) using
one-point integration as opposed to two-point integration required to exactly evaluate the
integral. The element equations of the reduced integration element are
( )
)[
]{
( )
(3.14)
(
)
(
)
This element is designated as the reduced integration element (RIE) by Reddy (1993). Alternate
11
derivation of the element without using the reduced integration concepts will be presented in the
sequel. In the thin beam limit, the element equations reduce to only one constraint, namely the
Kirchhoff condition in (3.12). While the element does not lock, it does not yield exact
displacements at the nodes for the static problems, and often a sufficient number of elements sis
needed to obtain accurate deflections.
3.1.3 Consistent interpolation element ( CIE)
As suggested earlier, if we use a quadratic approximation of w and linear approximation of ,
(8) reduces to a 5 x 5 system of equations. By eliminating the mid-side degree of freedom
associated with w, we can reduce the 5 x 5 system to the following 4 x 4 system of equations
(Reddy 1993; Reddy 1999):
( )
)[
]{
}
{
( )
( )
( )
( )
)
( )
( )
( )
}
(3.15)
Where
( )
(
)
(3.16)
(2)
and i
are the quadratic interpolation functions. Here the subscript c is used for the centre
node of the element. Note that the element has the same stiffness matrix as the reduced
integration element but a different load vector. The load vector is equivalent to that of the EulerBernoulli beam element. In fact, for constant q, the load vector in (3.15) is identical to that of the
Euler-Bernoulli beam element.
( )
( )
( )
( )
(3.17)
(3.18)
( )
. /,
12
( )
. / ,(
( )
. / ,(
( )
. / ,(
( )
. /(
( )
. /(
( )
) -
) -
(3.19)
. /(
( )
. /(
(3.20)
* +
* +
(3.22)
Where
( )
( )
( )
( )
( )
( )
( )
/(
)]
( )
(3.23)
(3.24)
And Q1= Va , Q2 = Ma , Q3=Vb , and Q4 = Mb. Equation (3.22) has the explicit form,
/[
]{
{ }
(3.25)
This element leads to the exact nodal deflections in static analyses for any distribution of the
transverse load q(x) and element-wise constant bending stiffness E1 and shear stiffness GAKs .
Therefore, the element is said to be superconvergent. In the thin beam limit, (3.25) reduces to the
Euler-Bemoulli beam equations, and no additional constraints are implied by the system.
3.2
13
Here we develop the finite element model based on a variational form in which the
displacements (w, x) and strains (Kxx, Yxz) are treated as independent field variables. The
variational statement associated with this mixed formulation is given by the stationarity of the
following functional (see Oden & Reddy 1982, p. 116, equation (4.115)):
20
( )
)
(
)
(3.26)
Where
,
,
,
The first variation of R yields the weak forms
.
(3.27)
(
(
(
)
.
(3.28)
(
(3.29)
(3.30)
(3.31)
( )
( )
( )
( )
( )
(3.32)
where (Wj, j, j, j) are the nodal values of (w,x ,xx,xz) and j()(x) ( =1, 2, 3, 4) are the
associated interpolation functions whose choice is yet to be made. Substituting (3.32) into
(3.28)-(31), we obtain the following finite element model:
,
,
,
[,
,
,
,
,
, , , , -
, *
, *
{
, *
, -] *
+
+
}
+
+
*
*
{
*
*
+
+
}
+
+
*
*
{
*
*
+
+
}
+
+
Where
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
(3.33)
14
(3.34)
Couple of observations are in order concerning the finite element model in (3.33). We note that
[A] is a vector {A} when xz is approximated as a constant, 0. In addition, the first equation of
(33) has the form
(3.35)
{ }
{ } { }
when w is interpolated using quadratic or higher-order polynomials. The nonzero entries
correspond to the deflection degrees of freedom at node 1 and node m. For linear interpolation of
w, we have m = 2 and (3.35) is alright. However, when m > 2, (3.35) implies that Fi = 0 for i = 2,
---, m - 1, which, in general, is not true. Thus, either the distributed load is zero or it is converted
to generalized point forces at the end nodes through Hermite cubic polynomials. In the latter
case, the force components can be added to Va and Vb and the moment components to Ma and
Mb at nodes 1 and m respectively.
3.2.2 ASD-LLCC element
For linear (L) interpolation of (w, x) and constant (C) representation of (xx, xz), and for
constant values of E1 and GAKs, the element equations become (m = n = 2 and p=q=l)
2
2
*
*
( )
( )
+{
+{
{ }
3
+{
(3.36)
(3.37)
(3.38)
(3.39)
Solving (3.38) and (3.39) for 0 and 0 and substituting into (3.36) and (3.37) (i.e condensing out
0 and 0), we arrive at the following 3x4 system of equations,
( )
/[
]{
( )
(3.40)
(
)
(
)
where = EI/GAKsh2. These are exactly the same equations obtained in the displacement
formulation with the linear interpolation of w and x and using one-point Gauss quadrature to
evaluate the shear stiffnesses, i.e., the reduced integration element (RIE). Thus, the assumed
strain-displacement formulation eliminates the need for reduced integration concepts.
3.2.3 ASD-HQLC element
Suppose that the distributed load is represented using
15
( )
( ) ( )
(3.41)
{ }
(3.42)
. /[
]{
. /0
+{
1{
/{ }
. /*
/0
+{
1{
(3.43)
2 3
(3.44)
(3.45)
where the end nodes of the element are designated as '1' and '2', and the middle node as 'c', and
the interior nodal degrees of freedom associated with w are omitted as they do not contribute to
the equations. Solving (3.44) for {} and (3.45) for 0, substituting the result into (3.42) and
(3.43), and eliminating c, we obtain
.
/0
/0
1{
1{
.
}
/0
1{
/0
1{
{ }
}
}
(3.46)
(3.47)
. /[
]{ }
(3.48)
{ }
The stiffness matrix is the same as that of the superconvergent element derived by Reddy (1997);
however, the load vector is different. It is the same when either the applied load q is elementwise uniform or the load vector is computed using (3.24) with i given by (3.19).
It should be noted that the degree of the polynomial interpolation used for w does not enter the
equations presented in all the models discussed in this section. However, the load representation
implies that w be interpolated with Hermite cubic polynomials or i(1) of (3.19). It can be shown
that the use of the interdependent interpolations of (3.19) and (3.20) for w and x also results in
(3.48).
16
3.3
( )
( )
(3.49)
where wb and ws denote the bending and shear components, respectively, of the total transverse
deflection w (see Reddy 1999), and ~x denotes the shear rotation, in addition to the bending
rotation, of a transverse normal about the y axis. The strains and the stressstrain relations are
given by
.
(3.50)
(3.51)
/
.
(3.55)
/
(3.56)
17
The finite element model of the modified Timoshenko beam theory can be developed using the
standard steps. The first step is to write the weak forms of the three equations over a typical
element. We have
,
)
/.
(
/
(
.
(
/.
/
(
)
(3.57)
Where
(
/1
.
(3.58)
/1
/-
/(3.58)
From the weak form (57) , it is clear that x and ws can be interpolated using the Lagrange
interpolation and wb using Hermite interpolation. The lowest admissible functions are linear for
x and ws and cubic for wb. However, the condition that the shear force be element wise
constant for element wise constant values of EI in turn requires that ws be quadratic.
Let (x,wb,ws) be interpolated as
( )
( )
( )
( )
( )
( )
( )
( )
(3.59)
where i, Wis and Wib denote the nodal values of x, ws and wb, respectively, i(1) and i(2) are
linear and quadratic interpolation functions, respectively, and i are the Hermite cubic
interpolation functions (m = 2, n = 3, p = 4). Substituting the interpolations (59) into the weak
form (57), we obtain the following finite element model:
, [, , -
, , , -
, * +
, - ] {*
+}
, - * +
* +
{ * +}
* +
* +
{* +}
* +
(3.60)
or simply
,
-* +
(3.61)
18
( )
( )
( )
( )
( )
( )
( )
( )
( )
(3.62)
The element equations (60) are not suitable for practical use. The reason is that we only know the
total displacement w = wb + ws and not its bending and shear parts separately. This is also true
about the total rotation (x = -wbx + x). Hence, it is necessary to recast the element equations
(60) in terms of the physical nodal variables.
/0
1{
/0
/0
/0
18
18
9
9
.
.
15 {
{
. /0
18
/0
/0
(3.63)
18
18
. /0
( )
( )
1{
9
}
(3.64)
{
( )
( )
9 (3.65)
19
( )
/[
]{
/[
]{
{ }
( )}
(3.66)
( )
where
( )
( )
(3.67)
Wcs denotes the value of ws and is the specified transverse load at the centre node of the
element. Note that the finite element equations associated with the second equation in (3.60) is
split into a pair of equations for convenience.
As noted earlier, it is necessary to combine the two components of the transverse deflection as
well as the rotation into total deflection and rotation. This amounts to rewriting the algebraic
equations (60) to obtain a model solely in terms of the total deflection w = wb + ws and rotation
x = --wb, x + x at the element nodes. First we condense out Wcs using the second equation of
(66). We have
/
.
where
(3.68)
( )
/0
1{
/0
1{
/0
1{
8{
39
/0
1{
(3.69)
/0
1{
{{ }
( )
( )
( )
( )
( )
( )
}}
(3.70)
(
(
4
.
(3.71)
20
/0
1{
){
/0
(
{
1{
( )
( )
/0
1{
{
.
( )
( )
( )
( )
( )
( )
/0
1{
(3.72)
(
){
( )
( )
( )
)
( )
( )
(3.73)
. /
( )
/[
]{
( )
( )
( )
. /
( )
( )
. /
( )
. /
( )
(3.74)
( )
Where
(3.75)
.
(
/.
,
.
(
(
( )
/.
/.
/
/
.
)(
(
(
)
.
/.
)-
)(3.76)
(3.77)
(
/
(
(
) (3.78)
)
21
(3.79)
respectively, where
(3.80)
, -
, ,
]8
*
*
+
+
,
,
,
,
- * +
]{
}
- * +
*
*
+
}
+
(3.81)
, -* +
* +
* +
(3.82)
, , [ , *
*
{
*
*
+
+
}
+
+
, ,
,
,
,
,
, , -
*
*
{
*
*
,
,
,
,
- {
- *
- *
-] { *
}
+
+
+ }
,
,
,
,
[
,
,
,
,
, , , , -
, *
, *
{
, *
, -] *
+
+
}
+
+
+
+
}
+
+
(3.83)
,
,
,
, ,
,
{ }
-] {{ }}
- { }
, - , {[, - , , - , -
* +
{* +}
* +
, * +
, - ]} {*
+}
* +
, -
(3.84)
22
Where
(
( )
( )
( )
( )
( )
( )
(3.85)
( )
( )
( )
( )
)
( )
( )
( )
( )
(3.86)
. /[
{ }
)[
( )
]{
( )
(3.87)
(
)
(
)
For quadratic interpolation of both w and x, the element matrices are of order 6x6 for
pure bending case.
Interdependent interpolation element (IIE):
For this case, the stiffness matrix and load vector are given in (25) . The mass matrix[M] of (85)
consists of several parts as given below.
23
, .
/[
/[
/[
. /[
)
.
/[
/[
(3.88)
Finite element model with superconvergent stiffness matrix (SCE): Although the
superconvergent form of the stiffness matrix can be derived using various approaches,
only the interdependent interpolation element formulation is readily extendable to the
dynamic case. The other formulations do not permit the algebraic manipulations with
the mass terms in place. Hence, one may choose a mass matrix to go with (48) and(74).
There are several choices (i) use the same mass matrix as in (88) , (ii) use the mass
matrix of the euler-bernoulli beam element, or (iii) use the mass matrix of the IIE
element with =0(hence, =1). The first choice reduces the formulation to IIE, the
second and third choices are the same because of the relationship between i(1), i(2) and
i. Thus for the dynamic case, the finite element model in(74) takes the form
/[
/[
){
(
.
/[
]{
{ }
(3.89)
Where
( )
( )
{ }
( )
({
( )
( )
( )
( )
( )
( )
( )
)
( )
( )
(3.90)
( )
})
Note that when is set to zero in mass as well as stiffness matrices, the equations of IIE and
SCE are reduced to those of the Euler-Bernoulli beam element.
24
Inertia Axis
c
Xcm
XO
+ ve h
+ ve L
+ ve
+ ve M
Figure 4.1 (b) A typical airfoil section showing heave and pitch degree of freedom
Before proceeding to flutter analysis, it is required to define various matrices involved in the
equation of motion viz., inertia matrix, M , structural and aerodynamic stiffness matrices, K
and A and aerodynamic damping matrix, D A . Considering the strip theory of aerodynamics,
equation of motion of a Timoshenko beam element is presented with respect to the assumed
affirmative directions of generalized coordinates (heave h and pitch ) and external generalized
forces (aerodynamic lift L and aerodynamic moment M ).. Another way of defining matrices are
25
through finite element formulation of a beam element with both bending and torsion degree of
freedom.
4.2
The equation of motion of a uniform cantilever beam is derived using two approaches; 1)
through analytical Timoshenko beam formulation and 2) through finite element beam
formulation with bending and torsion degree of freedom. Figure 4.2 shows a uniform cantilever
beam.
Elastic axis and Inertia axis
V
Y
l
26
(4.1)
(4.2)
(4.4)
Moreover the aerodynamic analysis is subject to the quasi-steady assumption, which implies that
only the instantaneous deformation is important and the history of motion may be neglected.
Aerodynamic damping is incorporated through heave and pitch velocities as suggested by Fung
[12] based on the aerodynamic strip theory. The lift gradient for the infinite thin airfoil in a two
dimensional incompressible flow is
. A corrected set of expressions for the lift and
moment coefficients (about support point) with compressibility effects considered is given by,
(4.5)
where
is the free stream Mach number. Here asound represents the isentropic
index (specific heat ratio) of the gas, R is the corresponding gas constant, and T is the gas
temperature. For air, = 1.4, and R=287 J kg-1 K-1. Thus for the free stream flow of air, of
assumed ambient temperature T=T=288.16 K, the velocity of sound is asound = 340.26 m/s.
The effective angle of attack,
, for the computation of these steady aerodynamic forces is
given by the following expression,
(4.6)
27
(4.7)
Where is the density of the beam material, A is the cross section area, , is the Timoshenko
shear coefficient, depends on the geometry and J=I, which is the rotatory inertia.. Normally, =
5 / 6 for a rectangular section . Equation (4.2) remains the same in Timoshenko beam
formulation.
Using equations (4.3), (4.4) and (4.6), equations (4.7) and (4.2), can be rewritten as
0
and
(4.8))
(
The displacements
(4.9)
(4.10)
When
and
equal to zero, equations (4.8) and (4.9) reduce to two independent equations,
one for and one for . The terms involving
and
indicate inertia and aerodynamic
couplings.
Equations (4.8) and (4.9) are linear equations with constant coefficients. For such a (coupled)
system, the solution can be written in the form
(
( )
, and (
( )
(4.11)
where is generally complex. Introducing equation (2.10) into equations (2.7) and (2.8) and dividing
throughout by
, we obtain the ordinary differential equations,
(
/
,
0
(
) -
(4.12a)
/ 1
28
.
20
,
where ( )
) -
( ) and ( )
/ 1(
(4.12b)
( ).
The boundary conditions retain the same form except that and are replaced by
and partial derivatives of and with respect to y by total derivatives.
and , respectively
No closed form solution of equation (4.12) (for values of ) is possible; hence an approximate solution is
used. It will prove instructive to examine the effect of airflow speed
on the parameter that gives the
stability condition of the system. Before the stability analysis of the system, the free vibration
characteristics are investigated.
4.4
Formulation of equations of motion for a wide rectangular cantilever beam of uniform cross
section using finite element beam model
A combined bending-torsion Timoshenko beam model is considered for the analysis. Hence the
two node beam element has three degrees of freedom per node (
and ). Required element
consistent stiffness and inertia matrix for the combined bending-torsion beam element can be
obtained by coupling the individual consistent stiffness and inertia matrices for bending and
torsion elements. Figure 6.1 shows combined bending-torsion beam element with the
corresponding nodal degrees of freedom. The consistent element stiffness and inertia matrices
are as given below,
Intermediate Interpolation Element:
( )
( )
( )
( )
( )
( )
( )
, -
( )
( )
[
( )
( )
( )
( )
( )
(4.13)
( )
( )
( )
29
( )
( )
( )
( )
( )
, -
( )
( )
( )
( )
( )
( )
.
( )
( )
( )
( )
( )
( )
/
[
( )
( )
( )
( )
( )
( )
/
( )
( )
( )
( )
.
( )
/
[
]
( )
( )
.
( )
( )
[
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
(4.14)
[
30
, ( )
( )
( ) (
( )
( )
( )
( ) (
( )
(
( )
( )
( )
( )
( ) (
( )
( ) (
( )
)
( )
]
(4.15)
, -
. /
(4.16)
[
(4.17)
(4.18)
31
The element stiffness and inertial matrices can be assembled to form the global stiffness and
inertial matrices , - and [M] respectively.
4.4.1
Here, while formulating the aerodynamic forces, damping from viscous effects and unsteady aerodynamic
flows are taken into account. The aerodynamic damping is incorporated through heave and pitch
velocities as suggested by Fung [12] based on the aerodynamic strip theory. The aerodynamic lift and
moment at the flexure point, which act on a symmetrical airfoil, according to strip theory, is given by
equations (4.3) and (4.4) respectively. (Section 4.3.1)
(4.3)
(
where,
(4.4)
, represents a corrected set of expressions for the lift and moment coefficients (about support
point) with compressibility effects considered (through Prandtle-Glauert correction factor) and
denotes the effective angle of attack, for the computation of the steady aerodynamic forces.
and
These lift and moment values correspond to unit span case. In our investigation each element is of length
and hence the corresponding force (lift and moment) at each node (for each element) can be calculated
by multiplying the unit span value with respective
. Let be the aerodynamic moment about X axis
due to lift force. In this report this moment force is assumed to be zero; i.e., effect of aerodynamic
bending moment is ignored. Therefore, for the ith element,
* +
{
32
( )
, -
)( )
( )
, -
( )
(
{
|
{
)( )
)( )
)( ) ]
)(
, -
)
}( )
( )
)( )
|
, (
)( )
(
{
)(
)
}( )
]
* +
* +
{
* +
(4.19)
-* +
[, -
, -]* +
* +
(4.20)
where [M] is the inertia matrix and , - is the structural stiffness matrix. Matrices , - and , - can be
termed as aerodynamic stiffness and aerodynamic damping matrices respectively. If there are n elements,
[M], , -, , - and , - will be of the size 3(n+1) 3(n+1). The nodal displacement vector will be of the
size 3(n+1) 1. After imposing the cantilever boundary condition, the size of [M], , -, , - and , reduces 3n3n and the size of * + will be 3n1.
4.5
State-space method
This method modifies a second order ordinary differential equation into a first order ordinary differential
equation. The equation of motion in modal domain, for the uniform beam can be represented as
, M -* +
-* +
[, -
, -]* +
* +
(4.21)
, -* +
(4.42)
33
, -
, -
where , -
Using ,* +
M [, -
, -* +
, -]
* + or [, -
] and * +
-
M ,
1
, -]* +
* +
* +
* +
(4.43)
, -]
(4.44)
The Jones formula [12] for the frequency dependent Theodorsens complex function C(k) is used here to
introduce the phase difference between the aerodynamic loading and the response. This is achieved by
updating the aerodynamic matrices [A] and [DA] by multiplying these by the function C(k),
( )
(4.45)
where
is the non-dimensional reduced frequency obtained from the imaginary part of the
eigenvalue . Convergence in k values for each modal branch is achieved through an iterative method
for a given flow velocity. The flow chart for the above p-k algorithm is presented in Figure 4.4. Here the
updated aerodynamic matrix , and the updated aerodynamic damping matrix , are
given as
, 4.7
( )
, -
and
( )
(4.46)
Stability conditions
Case 1. At subcritical flow velocities in the presence of damping, all the eigenvalues are complex, = r
ii = (
), with negative real parts, r < 0, indicating that the net effective damping is
positive, (since >0), leading to stable oscillations, characterized by decrease in amplitude with time.
The imaginary parts of the eigenvalues give the circular frequencies (i= in rad/s) of the associated
branches from the two modes, while the real parts give the time dependence of the amplitudes.
Case 2. Beyond a critical velocity, (
), the real part of at least one of the complex eigenvalues, =
r ii = (
), becomes positive, i.e. r =
> 0. This indicates that beyond this critical
velocity, the net damping is negative, leading to unstable oscillations, characterized by increase in
amplitude with time. At the critical (flutter) velocity (
) i.e. at the flutter boundary, the real part of
the eigenvalues vanishes, (r = 0), indicating purely simple harmonic motion, without any net damping at
all.
Case 3. Divergence is indicated by the condition that the imaginary part of vanishes, i.e. i=
the corresponding real part is positive.
=0, when
34
Find
c
c
pr i
p r ik
2V
2V
C(k)
No
is
c
2V
k=k previous
Yes
Converged eigenvalues for mode at corresponding flow velocity
Go to next modal branch for convergence of next root
Updated flow velocity if all roots have converged
35
Flutter analysis of the wing is also carried out using the same elementary beam model.
The quasi-steady aerodynamic theory is used to obtain the aerodynamic forces interacting with
the structure. First the problem is solved taking one bending mode and one torsion mode as a
first estimate and then the result has been improved taking higher modes.
The flutter speed obtained for the present configuration has shown to be very high and also that
the wing is very stiff. Hence the stiffness of the wing is reduced by reducing the modulus of
elasticity and correspondingly the modulus of rigidity. The results obtained for the wing with
reduced stiffness parameters are typical for subsonic flutter.
The eigenvalue is a continuous function of the air speed U. When U is not zero, but
infinitesimally small, the exponent is no longer pure imaginary but complex, = + i. Of
course, to investigate this case, we must return to the non-self adjoint system. It can be shown
that for sufficiently small U and for (dCL / d) < 2 , the wing is losing energy to the
surrounding air, so that the motion is damped oscillatory, and hence asymptotically stable. The
clear implication is that is negative. As U increases, can become positive, so that at the point
at which changes sign, the motion ceases to be damped oscillatory and becomes unstable. The
air speed corresponding to = 0 is known as critical speed and denoted by Ucr. There are many
critical values of U but, because in actual flight U increases from an initially zero value, the
lowest critical value is the most important. One can distinguish between two critical cases,
depending on the value of . When = 0 and = 0 the wing is said to be in critical divergent
condition. When = 0 and 0 the wing is said to be in critical flutter condition.
The above qualitative discussion can be substantiated by a more quantitative analysis. To this
end, we must derive and solve the complete non-self adjoint eigenvalue problem.we obtain the
eigenvalue problem
[K + U2H + UL + 2M ] a = 0
----------(4.47)
36
5 NUMERICAL RESULTS
Based on the Timoshenko beam finite element formulations given in chapter 4, a MATLAB code
was written for the free vibration analysis and flutter analysis of aircraft wing. The results have
been validated using a standard package NASTRAN and compared with the results predicted by
the Euler Bernoulli formulation. Further the results of some parametric studies have been
presented.
5.1 Free vibration analysis results
5.1.1 Uniform beam
In this section, to ascertain the correctness of the formulation, a bench mark problem of an
uniform cantilever beam is solved.
Numerical data:
The following properties of the cantilever beam are used for the analysis:
Length =0. 5m
Width = 0.1m
Thickness = 0.003m
Youngs Modulus of elasticity = E = 71 * 109 N/m2
Shear Modulus of rigidity = G = 26 * 109 N/m2
Density of the material = s = 2722.77 kg/m3
Density of air = = 1.225 kg/m3
U
0.5m
0.1m
0.003m
37
The natural frequencies of typical uniform beam with the above properties are as shown in the
following Tables 5.1, 5.2, 5.3 and 5.4. Table 5.1 shows the comparison of natural frequencies
between Linear RIE formulation and analytical models. The values of bending frequencies
predicted by the Rayleigh beam model, which considers only rotary inertia but not shear
deformation are also presented. From the comparison of all the analytical formulations it is seen
that rotary inertia mainly contributes to the reduction in flexural frequencies of the beam. Shear
deformation effects are small in case of the first few bending modes. It can be seen that the
frequencies of the RIE formulation converge very slowly and in case of some higher frequencies
the values predicted by the linear RIE formulation are higher than that predicted by the
corresponding Euler Bernoulli beam element (Table 5.3). Linear RIE elements which use linear
shape functions for interpolating the longitudinal displacement w, require large no. of elements
to converge to the frequency predicted by the analytical model. The low value of Shear rigidity
justifies the less pronounced changes in bending frequencies between the analytical models. As
shear deformation effect on the torsion was not considered in this analysis, the torsional mode
frequencies are the same as that of the Euler Bernoulli beam elements. (Table 5.3)
Table 5.1 Natural frequency comparison for wide cantilever beam of uniform cross section
between Reduced Integration Element and analytical models
Type and
Mode no.
Results in Hz
N=10
N=20
N=40
N=80
Analytical
Euler
Bernoulli
beam
Analytical
Rayleigh
beam
Results in
Hz
Analytical
Timoshenko
Beam
Results in
Hz
1 bending
9.910517
9.901581
9.899318
9.8987548
9.89886047
9.898808270
9.898808272
2 bending
63.56639
62.40455
62.11756
62.046026
62.0350996
62.03304941
62.03304985
3 bending
186.4138
176.7054
174.3802
173.80515
173.700075
173.684002
173.684012
4 bending
393.3687
352.427
343.1035
340.82585
340.382821
340.321109
340.321182
92.68698
92.66911
92.664648
92.6631598
1 torsion
92.75845
92.6631598
92.6631598
2 torsion
280.5683
278.633
278.1503
278.02967
277.989479
277.989479
277.989479
3 torsion
475.2967
466.2983
464.0604
463.50188
463.315799
463.3157991
463.3157991
4 torsion
681.5952
656.839
650.6863
649.15280
648.642118
648.6421188
648.6421188
38
Table 5.2 Natural frequency comparison for wide cantilever beam of uniform cross section
between Interdependent Interpolation Element and analytical models
Type
and
IIE Element
Results in Hz
Mode no.
N=10
N=20
N=30
N=40
Analytica
l Euler
Bernoulli
beam
Analytical
Rayleigh beam
Results in Hz
Analytical
Timoshenko
Beam
Results in Hz
1
bending
9.898798
9.898791
9.898791
9.8988604
9.898791 7
9.898808270
9.898808272
2
bending
62.03404
62.03219
62.0321
62.035099
62.03209 6
62.03304941
62.03304985
3
bending
173.7235
173.6826
173.6804
173.70007
173.6801 5
173.684002
173.684012
4
bending
340.6319
340.3306
340.3139
340.38282
340.3111 1
340.321109
340.321182
1 torsion
92.75845
92.68698
92.67375
92.66911
92.663159
8
92.6631598
92.6631598
2 torsion
280.5683
278.633
278.2754
278.1503
277.98947
9
277.989479
277.989479
3 torsion
475.2967
466.2983
464.64
464.0604
463.31579
9
463.3157991
463.3157991
4 torsion
681.5952
656.839
652.2786
650.6863 648.64211
8
648.6421188
648.6421188
Table 5.2 shows the comparison of natural frequencies between Intermediate Interpolation
Element (IIE) formulation and the analytical models. The IIE element frequencies converge
rapidly and good agreement can be seen with the frequency values predicted by the analytical
Timoshenko beam model. The superconvergent two node IIE is superior in predicting the
flexural mode frequencies although it does not represent the pure shear frequencies
accurately(Reddy [37]). The torsional mode frequencies are the same as that of the Euler
Bernoulli beam elements. (Table 5.3). Table 5.3 shows the comparison of natural frequencies
between the Euler Bernoulli beam formulation and the analytical models. Table 5.4 shows the
results obtained from MSC Nastran beam element CBEAM with 10 elements. The NASTRAN
beam elements also yield frequencies which are lower than the Euler Bernoulli beam formulation
39
Table 5.3 Natural frequency comparison for wide cantilever beam of uniform cross section
between Euler Bernoulli Beam and analytical models
Type and
Mode no.
Results in Hz
N=10
N=20
N=30
Analytical
Euler
Bernoulli
beam
N=40
Analytical
Rayleigh
beam Results
in Hz
Analytical
Timoshenko
Beam
Results in Hz
1 bending
9.898869
9.898861
9.898861
9.898861 9.89886047
9.898808270
9.898808272
2 bending
62.03715
62.03523
62.03513
62.03511 62.0350996
62.03304941
62.03304985
3 bending
173.7443
173.7029
173.7006
173.7003 173.700075
173.684002
173.684012
4 bending
340.7072
340.4041
340.3871
340.3842 340.382821
340.321109
340.321182
1 torsion
92.75845
92.68698
92.67375
92.66911 92.6631598
92.6631598
92.6631598
2 torsion
280.5683
278.633
278.2754
278.1503 277.989479
277.989479
277.989479
3 torsion
475.2967
466.2983
464.64
464.0604 463.315799
463.3157991
463.3157991
4 torsion
681.5952
656.839
652.2786
650.6863 648.642118
648.6421188
648.6421188
Table 5.4 Natural frequency for wide cantilever beam of uniform cross section using MSC
Nastran
Type and Mode no
1 bending
9.88733
2 bending
61.78090
3 bending
172.5161
4 bending
342.1795
1 torsion
92.63934
2 torsion
277.3469
40
elements. The beam formulation is suitably adapted to account for the bending-torsion coupling
in the normal modes, due to offset of the shear center from the centroid.
Numerical data
The numerical data used for the actual wing and also for the wings with reduced stiffness
parameters are as shown below.
Case (1): Actual wing:
Youngs Modulus of elasticity = E = 72 * 109 N/m2
Poissons ratio = = 0.3
Shear Modulus of rigidity = G = 27.69 * 109 N/m2
Case (2): With reduced stiffness parameters:
E* = 0.1E and G* = 0.1G
Youngs Modulus of elasticity = E* = 7.2 * 109 N/m2
Poissons ratio = = 0.3
Shear Modulus of rigidity = G* = 2.769 * 109 N/m2
41
Table 5.5 Mass distribution and mass densities of the beam element
Ref [42] PD ST 0314
Sl. No.
Ele L in
m (lr)
C/S Area
Density in
in m
-6
Kg/m
x 10 (Ar)
(s) r
0.350
12871.0
9309.30
119.82
0.315
12037.0
7803.98
93.94
0.285
12318.0
9609.27
118.37
0.300
11400.0
9784.31
111.54
0.325
15438.0
7654.20
118.16
0.315
7348.1
7968.55
58.55
0.315
6826.9
49346.43
336.88
0.325
6470.2
41645.03
269.45
0.325
5314.3
44344.80
235.66
10
0.325
5110.9
40029.30
204.58
11
0.325
4947.1
34392.09
170.14
12
0.325
4817.5
29235.08
14.84
13
0.325
4153.8
26966.23
112.01
14
0.325
3939.4
22792.80
89.79
15
0.325
3455.0
17246.97
59.58
16
0.350
3259.6
9963.28
32.47
17
0.350
3132.6
6385.20
20.00
18
0.300
3153.5
6178.57
19.48
19
0.210
2803.0
9275.77
25.99
20
0.350
2557.7
6660.95
17.04
21
0.370
2009.6
7063.66
14.20
22
0.370
2119.7
6485.07
13.75
42
Sl.
Ele L in m
No.
Izz in m4
Iyy in m4
J in m4
x 10-4
x 10-4
x 10-4
Chord
length(m)
0.350
4.0477
35.3980
70.3185
2.402
0.315
3.2949
29.8140
45.8763
2.326
0.285
3.1249
29.5620
25.7400
2.256
0.300
2.8388
23.3110
15.0045
2.191
0.325
3.7622
23.3330
9.0301
2.120
0.315
1.5680
11.6677
7.1700
2.047
0.315
1.4159
10.4640
5.6790
1.975
0.325
1.2581
9.3388
5.2815
1.902
0.325
0.9428
6.8012
4.3770
1.828
10
0.325
0.8215
6.4647
3.8130
1.754
11
0.325
0.7414
6.0293
3.3300
1.680
12
0.325
0.6706
5.4199
2.7525
1.606
13
0.325
0.5127
4.1906
2.3640
1.532
14
0.325
0.4474
3.6573
2.0250
1.458
15
0.325
0.3453
2.8600
1.6335
1.384
16
0.350
0.2958
2.3924
1.3110
1.307
17
0.350
0.2537
2.0252
1.0515
1.227
18
0.300
0.2295
1.8596
0.8730
1.153
19
0.210
0.1901
1.5036
0.7470
1.095
20
0.350
0.1525
1.2544
0.5865
1.031
21
0.370
0.1036
0.7833
0.4125
0.949
22
0.370
0.9409
0.6224
0.2940
0.865
43
Ele L in m
No.
ZG in m
YG in m
x 10-3
x 10-3
0.350
-355.430
36.300
0.315
-432.248
-137.681
0.285
-379.560
-199.990
0.300
-434.200
-278.100
0.325
-161.000
-9.200
0.315
-1.648
39.906
0.315
-9.040
11.800
0.325
-6.333
7.034
0.325
-8.275
-3.881
10
0.325
-19.418
-0.664
11
0.325
-18.830
-3.497
12
0.325
-10.425
13.077
13
0.325
-16.588
-4.718
14
0.325
-14.829
-5.609
15
0.325
-6.369
-12.955
16
0.350
-0.028
-10.495
17
0.350
-2.468
8.142
18
0.300
2.517
-8.635
19
0.210
1.100
-7.300
20
0.350
-1.173
-6.608
21
0.370
-0.851
-5.770
22
0.370
-1.952
-5.314
The above numerical data are used for the analysis of the subsonic wing. The wing is visualized
as a collection of stepped beam elements, each having its respective properties as shown in the
above tables. The natural frequencies obtained for the wing for each case are given below.
44
Euler
Bernoulli
beam
element
RIE
results in
Hz
IIE
results
in Hz
results
in Hz
Function space
approach
3D model
NASTRAN
Ref [32]
results in Hz
Table 4.3
Ref[32]
PD ST-0314
Table 4.3
PD ST-0314
1
bending
7.2169346
7.157122
7.156243
7.124
7.087
2
bending
21.140967
20.77725
20.6735
20.786
20.481
3
bending
50.403696
48.7578
48.06263
48.538
47.781
4
bending
101.41677
95.77767
93.12571
1 torsion
56.832777
56.78816
56.83033
2 torsion
121.05991
120.9533
120.9788
3 torsion
175.83192
175.1956
175.1329
4 torsion
249.07651
241.5394
241.8016
56.385
56.338
The values of frequencies predicted by different methods are given in the table (5.8). The
Function space approach frequencies correspond to the method developed by Mukerjee and
Prathap for explaining the locking phenomena[ref]. It can be seen that both the Reduced
Integration element and Intermediate Interpolation Element yield good agreement with the
values predicted by function space approach and 3-d NASTRAN results. Thus the Timoshenko
beam element finite element formulations constitute an improvement over Euler Bernoulli beam
formulations in predicting the dynamic characteristics of a complex structure such as an aircraft
wing. The element wise values are given in the table 5.9
45
Ele L in m
No.
1
0.350
1.1125
0.315
1.1955
0.285
1.3535
0.300
1.1991
0.325
0.9999
0.315
0.932
0.315
0.9058
0.325
0.7978
0.325
0.7279
10
0.325
0.6595
11
0.325
0.6149
12
0.325
0.5711
13
0.325
0.5064
14
0.325
0.466
15
0.325
0.4101
16
0.350
0.321
17
0.350
0.2865
18
0.300
0.3504
19
0.210
0.6665
20
0.350
0.2109
21
0.370
0.1632
22
0.370
1.4052
46
47
Vf = 147.061m/s
Figure 5.5 Variation of damping (g) and frequency ( ) with free stream velocity for the case of RIE
formulation for a continuous cantilever beam of uniform cross section in combined bending-torsion
vibration.
Vf = 147.058 m/s
Figure 5.6 Variation of damping (g) and frequency ( ) with free stream velocity for the case of IIE
finite element formulation for a cantilever beam of uniform cross section in combined bendingtorsion vibration.
48
1.80E+02
2nd Bending
1.60E+02
1st Torsion
0.4
3rd Bending
0.2
0
-0.2
100
200
Frequency (Hz)
0.6
Damping g
2.00E+02
1st Bending
0.8
1.20E+02
1.00E+02
8.00E+01
6.00E+01
-0.4
4.00E+01
-0.6
2.00E+01
-0.8
0.00E+00
-1
1st Bending
2nd Bending
1st Torsion
3rd Bending
1.40E+02
-2.00E+01 0
100
200
Vf = 150 m/s
Figure 5.7 Variation of damping (g) and frequency ( ) with free stream velocity for a cantilever
beam of uniform cross section in combined bending-torsion vibration using MSC NASTRAN (strip
theory of aerodynamics with p-k method of flutter analysis).
Table 5.10
Uniform beam Flutter results
Flutter speed (m/s)
Euler
Bernoulli
beam
element
RIE
element
IIE element
MSC
NASTRAN
N=5
N=10
N=20
N=5
N=10
N=20
145.828
146.806
147.057
46.231
45.413
45.192
145.978
146.826
147.061
45.941
45.346
45.175
145.829
146.807
149
147.058
46.231
45.412
42.7
45.191
Table 5.10 also shows the convergence characteristics of the various elements. It can be seen that
the RIE elements slowly converge to the value predicted by the IIE element. This shows the
delayed convergence characteristics of the RIE element in predicting the flutter boundary. The
IIE element although showing a marginal increase in the flutter velocity validates with the free
vibration frequency which doesnt show any significant difference between the Euler Bernoulli
and the Timoshenko model.
49
x 2 and f
2
Hence the v-g and v-f curves can be plotted from the eigen values.The velocity v/s Real part and
velocity v/s Imaginary part are also plotted for the wing with reduced stiffness parameters. Some
of the typical graphs obtained are shown below in the Figs 3.5, 3.6, 3.7 and 3.8
Vf = 637.530 m/s
Figure 5.8 Variation of damping (g) and frequency ( ) with free stream velocity for the case of RIE
formulation for the aircraft wing in combined bending-torsion vibration.
50
Vf = 636.642m/s
Figure 5.9 Variation of damping (g) and frequency ( ) with free stream velocity for the case of IIE
formulation for the aircraft wing in combined bending-torsion vibration.
Te flutter speeds obtained from the graphs for the actual wing and for the wing with reduced
stiffness parameters are shown in the following Table 5.9 and 5.10
Table 5.11 Flutter results of the subsonic wing (actual wing)
Beam
Elements
Euler
Bernoulli
beam
element
IIE
element
RIE
element
635.803
39.832
636.642
39.603
637.530
39.534
51
Vf = 201.864m/s
Figure 5.10 Variation of damping (g) and frequency ( ) with free stream velocity for the case of
RIE formulation for the aircraft wing (with reduced stiffness ) in combined bending-torsion
vibration.
Vf = 201.287m/s
Figure 5.11 Variation of damping (g) and frequency ( ) with free stream velocity for the case of
RIE formulation for the aircraft wing (with reduced stiffness parameters) in combined bendingtorsion vibration.
52
Table 5.12
Flutter results of the subsonic wing (with reduced stiffness parameters)
Beam
Elements
Euler
Bernoulli
beam
element
IIE
element
RIE
element
201.016
12.584
201.287
12.533
201.864
12.333
The flutter velocities and flutter frequencies of the various elements are given in Tables 5.11 and
5.12 . Here the effect of shear deformation on flutter frequency is more pronounced and
Timoshenko beam elements show a increase in the flutter velocity and decrease in flutter
frequency. The flutter modes correspond to the 3rd Bending and 1st Torsion mode frequencies.
From table 5.8 , it can be seen that the 3rd bending mode frequencies significantly come down for
Timoshenko beam elements. Thus there is a increased gap between the bending and torsion
mode frequencies leading to increased flutter velocities and decreased flutter frequencies. The
RIE predicts higher flutter velocities and lower flutter frequencies over IIE. This is because of
the delayed convergence characteristics of the RIE.
53
6 SENSITIVITY ANALYSIS
The following section is dedicated for the development of an analytical expression for the
derivative of flutter velocity with respect to the design parameters of the aircraft wing, which is
modeled as a uniform cantilever beam. The design variables x d considered in this analysis are
Lengthof the beam(l),Bending Rigidity (EI), Torsional Rigidity (GJ), Mass per unit length ( )
and Inertia per unit length (Im).
6.1 Analytical expression for the derivative of flutter velocity with respect to the design
variables:
Using the state space method, the eigenvalue eigenvector problem has been defined as,
S I x 0
(6.1)
, [
, -
M [, 1
M ,
1
, -]
* +
* +
* +
(6.2)
where [S] is a matrix composed of aerodynamic stiffness [A] and aerodynamic damping [ DA ]
matrices and * + is the vector of the natural coordinates. These matrices depend upon the free
stream air velocity, V and reduced frequency parameter, k ( c 2V ); where is the
imaginary part of the complex eigenvalue, = r ii =(- i). At the critical condition in
which the instability of flutter begins, the real part of one of the complex eigenvalue becomes
zero and hence for that particular mode, f = if. Subscript f denotes the values at flutter
velocity.
Hence the reduced frequency parameter at flutter speed can be written as,
kf
which gives,
c f
2V f
f i f i
(6.3)
2V f k f
c
(6.4)
S I x 0
f
(6.5)
I x 0
Sf i
c
(6.6)
54
yT S f f I 0
(6.7)
symmetric matrix, both left and right eigenvectors ( y and x respectively) will be complex in
nature.
Differentiating equation (6.5) with respect to the design variable, xd and pre-multiplying by the
left eigenvector, yT ,
yT
x y S x dk
x
k
dx
Sf
x dV
V
dx
Sf
dV f
T
2 dk f
Vf
k f y I x 0
c dx d
dx d
(6.8)
x;
Sf
mad
hsk yT
Sf
k f
and hsV yT
Sf
(6.9)
V f
dk
dV
yT I x f hsV i f yT I x f 0
hsm hsk i
c
c
dmad
dmad
(6.10)
Now, the complex terms appearing in the above equation can be written as,
g k hsk i
2V f
c
yT I x
and
gV hsV i
2k f
c
yT I x
(6.11)
gk
dk f
dxd
gV
dV f
dxd
hsm
(6.12)
where dk f dmad and dV f dmad are real numbers. Pre-multiplying equation (6.12) by the
complex conjugate of g k , written as g k , gives,
gk
dk f
dxd
g k gV
dV f
dxd
g k hsm
(6.13)
55
Im(g k hsm )
(6.14)
Im(g k g V )
Im(g V hsm )
Im(g V
(6.15)
gk )
6.2 Formulation of equations of motion for a wide rectangular cantilever beam of uniform
cross section using Timoshenko finite element beam model
A combined bending-torsion Timoshenko beam model is considered for the analysis. Hence the
two node beam element has three degrees of freedom per node (
and ). Required element
consistent stiffness and inertia matrix for the combined bending-torsion beam element can be
obtained by coupling the individual consistent stiffness and inertia matrices for bending and
torsion elements. Figure 6.1 shows combined bending-torsion beam element with the
corresponding nodal degrees of freedom. The consistent element stiffness and inertia matrices
are as given below,
( )
( )
( )
( )
( )
( )
( )
, -
( )
( )
[
( )
( )
( )
( )
( )
(6.16)
( )
( )
( )
56
( )
( )
( )
( )
( )
, -
( )
( )
( )
( )
( )
( )
.
( )
( )
( )
( )
( )
( )
/
[
( )
( )
( )
( )
( )
( )
/
( )
( )
( )
( )
.
( )
/
[
]
( )
( )
.
( )
( )
[
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
(6.17)
[
57
The element stiffness and inertial matrices can be assembled to form the global stiffness and
inertial matrices , - and [M] respectively.
6.3.
Expressions for
Sf
x d
k f
V f
m i
xd
, -
, ( )
, -
, ( )
( )
, ( )
[m]i
m i
1
[m]i
xd
, -
(6.18)
(6.19)
(6.20)
[
, ()
(6.21)
]
( )
(6.22)
58
, ( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
.
( )
/
( )
( )
( )
( )
( )
( )
[
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
(6.23)
( )
( )
( )
( )
, ( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
)
[
( )
( )
( )
( )
( )
( )
( )
( )
[
( )
( )
]
( )
( )
( )
( )
59
/
[
]
)
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
(6.24)
[
]
)
[m] i [m] i
[0]6 x 6
V f
k f
(6.25)
Now, the partial derivatives of k i with respect to the design variables are
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
, (
( )
, ( )
, -
(6.27)
( )
[
, ( )
(6.26)
]
, ( )
, -
(6.28)
60
The updated stiffness and damping matrices for the m dof aircraft wing at the flutter speed is
given by,
[ ]
(
(
[
.
(
0
).
/
/
.
(
(6.29)
)]
).
/
/
(6.30)
0.165
0.335
0.0455
0.3
1
i 1
i
kf
kf
(6.31)
Now, the partial derivatives of , - with respect to Vf and kf are be found to be,
[k]i [k ]i
06 x 6
V f
k f
(6.32)
And also
[
()
1
(
1
(
1
()
(6.33)
1
(
(6.34)
Partial derivative of the aerodynamic matrix, [ ] with respect to V f can be derived as follows.
af
V f
1
2
2
V f c
2
V f 2
1 V f a sound
0
0
0
0
0 X O X cp
0
0
0 0 0
0 0 0
C (k f )
0 0
1
0 0
0
0 0 X O X cp
0 0 0
61
(6.35)
(2 M f )
V f
(1 M f ) V f
af
(6.36)
1
2
2
V f c
2
k f 2
1 M f
af
k f
0
0
0
0
0 0 0
0 0 0
C (k f )
0 0
1
0 0
0
0 0 X O X cp
0 0 0
0 X O X cp
0
0
(6.37)
af
k f
where
1
2
V f2 c
2
2
1 M f
0
0
0
0
0 0 0
0 0 0
C (k f )
0 0
1
k f
0 0
0
0 0 X O X cp
(6.38)
0 0 0
0
0
0 X O X cp
0
0
0
C (k f ) is given by,
k f
C (k f )
2
k f
(k f 0.0455 i)
(k f 0.3 i) 2
(6.39)
Partial derivative of the aerodynamic matrix, D A f with respect to V f can be derived as follows.
.
(
0
).
/
/
.
).
/
/
]
(6.40)
62
0
1
(1 M f ) V f
2
(6.41)
(
0
).
/
.
(
0
)(
).
(6.41)
/
/
(
(
)
(
)(
(6.42)
where
C (k f ) is given by,
k f
C (k f )
k f
(k f 0.0455 i) 2
(k f 0.3 i) 2
Now, [
Sf
(6.43)
Sf
x d
k f
V f
S f
0 0
0 0
1
M [ K ] [ A f ]
1 0
0 1
[ D] [ D A f
M
1
The partial derivatives of S f with respect to xd, Vf and kf can be written as,
(6.44)
63
Sf
0 0
0 0
1
M
[K ] [ A f ]
x
d
x d
M
x d
[ D] [ D A f ]
(6.45)
(6.46)
0 0
0 0
x d
where
Sf
V f
where
0 0
0 0
1 A f
V f
0 0
0 0
1 D A f
M
V
f
Af
V f
V f
Sf
k f
where
0 0
0 0
1 A f
M k
f
0 0
0 0
1 D A f
M
k
f
(6.47)
Af
k f
k f
6.5 Numerical evaluation of the derivative of flutter velocity with respect to the design
variable by finite central difference method
Finite central difference method can be used to verify the gradient information, obtained
analytically. The derivative can be found by using the formula as given below.
dV f
dxd
xd ( xd ) k
V f ( x d ) k 0.5 V f ( x d ) k 0.5
(6.48)
where the parameter , determines the level of accuracy for the obtained gradient value. In this
particular case testing is carried out with = 1% and 5% of xd. However, this is a numerical
method and it lacks precision and errors occur through truncation and rounding off.
64
Analytical Method
Central Differnece
Method
N=10
=0.1
Parameter
Lengthof
the
beam(l)
0.5 m
-238.0587
-277
Bending
Rigidity
(EI)
15.975 Nm2
-0.2435
-.25039
Torsional
Rigidity
(GJ)
23.4 N m2
3.3032
3.31623
Mass per
unit
length ( )
0.816831 N/m
23.7109
24.1175
The result indicate that flutter velocity gradients obtained from the analytical method match well
for most parameters. There is a large deviation in length term. This occurs because all the terms
involved in dynamic analysis critically depend on the length and hence higher order terms are
required to accurately predict the flutter sensitivity. Length sensitivity is highly nonlinear.
bending rigidity sensitivity is negative because as bending rigidity increases second bending
mode frequency also increases leading to decrease in gap between the second bending and first
torsional mode. Hence flutter velocity decreases. The bending rigidity sensitivity is observed to
be small. This explains why the flutter velocities of this beam were not extremely sensitive to
Timoshenko formulations as only the bending stiffness is altered. Torsional rigidity sensitivity is
positive due to the same reason that an increase in torsional frequency will lead to increase in
gap between the second bending and first torsional mode leading to increase in flutter velocity.
Thus any effect on torsional frequencies affects flutter velocity in a significant way.
65
7 CONCLUSIONS
7.1 Discussion of Results:
The aeroelastic sensitivity analysis of subsonic flutter using Timoshenko beam finite element
formulations indicates the critical effect of the frequency spectrum on the onset of flutter.
Hence the flutter sensitivity depends on the particular bending and torsion modes in flutter.
Timoshenko dynamic analysis of flutter is even more significant in case of real aircraft
structures which have a higher torsional rigidity and hence flutter at higher bending modes. As
Timoshenko dynamic analysis constitutes an improvement over the Euler Bernoulli model , it
predicts more precise values of flutter velocities and flutter frequencies.
Also the exact prediction of dynamic characteristics is crucial in control techniques aimed at
active vibration control and active flutter control.
From the free vibration and flutter analysis carried out in the previous sections, it is found that
the two noded superconvergent Timoshenko beam element yields better results over the
conventional Reduced Integration element as it converges faster than the RIE. Flutter analysis
results also indicate better convergence characteristics of the superconvergent element (IIE).
The flutter sensitivity analysis by analytical formulations yielded flutter gradients which are
comparable with those obtained by central difference method. The sensitivities obtained further
validate the earlier results of flutter analysis by Timoshenko beam finite element.
7.2 Further scope :
1. The present work is limited to the clean wing analysis that doesnt show flutter in the
subsonic regime. However it is necessary to check if the wing with control surfaces is
prone to subsonic flutter. The present method can be easily extended to determine flutter
boundaries of wing with control surfaces
2. The flutter analysis of the T-tail is critical from the point of design. The quasi-steady
method can easily be extended to the T-Tail assembly consisting of Horizontal tail,
Vertical tail, Rudder and Elevator. Since the aspect ratio of tail assembly surfaces is small
, shear deformation effect play a significant role in determining flutter velocities.
3. The present analysis can be extended to include shear deformation effects on torsional
frequencies as the torsional rigidity sensitivity is found to be high in the present analysis.
4. Sensitivity analysis can be extended to other parameters like shear centre offset, span of
the beam and position of centre of mass and aerodynamic centre of the wing like
structure.
66
REFERENCES
1. Lanchester, F.W.: Torsional vibration of the Tail of an Aeroplane., Aeronaut.Research
Com.R & M.276, part i (July 1916).
2. Bairstow, L. and A. Fage, Oscillations of the Tail Plane and Body of an Aeroplane in
Flight, Aeronaut.Research Com.R & M.276, part ii (July 1916)
3. Blasius,H:,Umber
Schwingungsercheiningen
Unterflugeln.Z.Flugtech.u.Motorluftschif.16,39-42 (1925)
an
Einholmigen
67
19. John Dugundji, Theoretical consideration of Panel flutter at high supersonic Mach No.,
AIAA J. Vol.4,No.7,July 1966
20. Abott.I.H. and Von Doenhoff, Theory of Wing section, McGraw Hill,Newyork,1949.
21. S.H.R.Eslimy-Islahany and A.J.Sobby, Response of bending torsion coupled beam to
deterministic and random loads, J. of Sound and Vibration,195(2),1996.
22. E.Dokumaci, An Exact solution for coupled bending and torsion vibration of uniform beam
having single cross section symmetry, J.of Sound and Vibration,119(3),1987
23. Timoshenko, S. (1921) On the correction for shear of differential equation for transverse
vibrations of prismatic bars. Philos. Mag. 6 , p. 744.
24. Kapur, K. K : Vibrations of a Timoshenko beam, using a finite element approach, J. of
the Accoustical Society of America, Vol 40, pp 1058~1063, 1966
25. Nickel, R and G. Secor : Convergence of consistently derived Timoshenko beam finite
elements, International J. of Numerical methods in Engineering, Vol.5, pp 243~253,1972
26. Davis, R.R.D. Henshell and G. B. Warburton : A Timoshenko beam element,J. of
sound and Vibration, Vol 22, pp 475~487, 1972
27. Thomas D. L. , J. M. Wilson and R.R. Wilson : Timoshenko Beam Finite Elements , J.
of Sound and Vibration, Vol 31, pp 315~330 ,1973
28. Thomas J and B. A. H. Abbas : Finite element model for dynamic analysis of
Timoshenko Beam, J of Sound and Vibration, Vol 41, pp 291-299, 1975
29. Thomas D. L : Comments on Finite element model for dynamic analysis of Timoshenko
Beam, J. of Sound and Vibration, Vol 46, pp 285~290, 1976
30. Prathap G, Babu C R 1986 Field-consistent strain interpolation for the quadratic shear
flexible beam element. Int. J. Numer. Methods Eng. 23:1973-1984
31. Shi G, Voyiadjis G Z 1991 Simple and efficient shear flexible two-node arch/beam and
four-node cylindrical shell/plate finite elements. Int. J. Numer. Methods Eng. 31 : 759-776
32. Rakowski J 1991 A critical analysis of quadratic beam finite elements. Int. J. Numer.
68
36. Reddy J N, Wang C M, Lam K Y 1997 Unified finite elements based on the classical and
shear deformation theories of beams and axisymmetric circular plates. Commun. Numer.
Methods Eng. 13:495-510
37. Reddy J N On the dynamic behaviour of the Timoshenko beam finite elements Sadhana,
Vol. 24, Part 3, June 1999, pp. 175-198
38. Rogers, L.C., Derivatives of Eigenvalues and Eigenvectors, AIAA Journal, Volume 8,
Pages 943-944, May 1970.
39. Adelman, H.M., Haftka, R.T., Sensitivity Analysis of Discrete Structural Systems,
AIAA Journal, Volume 24, No. 5, Pages 823-832, May 1986.
40. Ringertz, U.T., On structural optimization with aeroelasticity constraints, Structural
Optimization, Volume 8, Pages 16-23, 1994
41 Jurado, J.A., Hernandez, H., Sensitivity analysis of bridge flutter with respect to
mechanical parameters of the deck, Structural Multidisciplinary Optimization, Volume 27,
Pages 272-283, 2004.
42.Dr. S.Mukherjee, Manju, NAL Project Document, Dynamic characterization SARAS wing
and empennage. PD ST-0314