You are on page 1of 16

$66(660(172)675$7(*,(6)25&255(&7,1*/,1($5

8167($'<$(52'<1$0,&686,1*&)'257(675(68/76

R. Palacios1, H. Climent1, A. Karlsson2, B. Winzell2


1
Aeroelasticity and Structural Dynamics Department, EADS-CASA
John Lennon s/n, 28906 Getafe, Madrid, Spain
e-mail: (RP) c81185@casa.es (HC) c80411@casa.es
2
SAAB AB

.H\ ZRUGV Aeroelasticity, Unsteady Transonic Aerodynamics, Doublet Lattice, Time


Linearization, Corrected Aerodynamics

$EVWUDFWThe objective of this paper is to analyse different approaches to handle efficiently


the flutter problem where the assumption of linear aerodynamics is not valid. The aim is to
improve the solution of the flutter problem in the industrial environment, which limits
severely the selection of strategies. Two main research lines were followed: On the one hand,
the assessment of correction techniques to introduce Computational Fluid Dynamics (CFD)
results into the linear unsteady aerodynamics. On the other hand, the development of a time
linearization procedure on the non-linear results to either introduce them in the correction
methods or generate directly air force matrices in its classical form.
After showing the industrial standpoint and its requirements of more accurate tools, a
literature survey of correction procedures is included. The more promising of them are later
developed, as well as the time linearization method to allow the interface with CFD. Finally
methods are applied to models with a well known transonic flutter behaviour: the AMP and
the AGARD445.6 wings.

 ,1752'8&7,21

Linear unsteady aerodynamics needs to be corrected in the transonic regime, where the linear
assumption is no longer valid and when the effects of viscosity and thickness are relevant.
The transonic dip, that appears in flutter velocity when Mach number increases, cannot be
reproduced by the classical wing linear theory, which become then non-conservative. The use
of CFD codes coupled with the flexible structure is the likely way to overcome these
difficulties in the future. Currently, two are the reasons that do not allow the use of CFD
codes for industrial flutter calculations:
1. Lack of maturity of the tool to be used in the aircraft certification process.
2. Relative cost: The cost of a Navier-Stokes simulation, that is needed to obtain adequately
the transonic dip, is four orders of magnitude larger than the conventional Doublet Lattice
Method (DLM). It means that with current computers, each run will cost several weeks to
be obtained. The order of magnitude of flutter runs that are needed in the certification
process of a modern aircraft may be larger than ten thousand. Therefore the pure CFD
approach is still unaffordable.
This paper is devoted to show several alternatives to incorporate information coming from
expensive CFD calculations (which can include also experimental tests) into the traditional
and cheaper flutter analysis.

 %5,()/,7(5$785(6859(<2)&255(&7,1*0(7+2'6

Since the first stages of the DLM1, in the early 70’s, efforts were made to enlarge the
applicability region of the theory and several extensions of the standard calculation methods
were proposed. It was finally accepted despite some important proposals to change the core
of the method2  that the only way to go beyond these limits (transonic regime, prediction of
aerodynamics forces in control surfaces, effects of thickness and viscosity) was the
introduction of DSRVWHULRUL corrections on results. Those corrections are carried out by means
of data obtained with more accurate methods. At first, only experimental results were
available, but the growing of computers and refinements of techniques allowed later non-
linear theoretical approximations.
Several correcting methods can be found in the literature. There are not, indeed, key
differences between them, but the time when they were formulated. The older the methods,
the less correcting information they use. Mostly these methods use non-linear pressure
distributions on the model to correct linear finite element theory. Interpolation (even
extrapolation when experimental data is used) should determine the complete element-by-
element distribution.
Prior to the development of the DLM, <DWHV3,4 concentrated his work in a modified-strip
analyses method. Steady state spanwise local lift distribution and spanwise aerodynamic
centre locations (normal to elastic axis) are required as aerodynamic input parameters for
determining the oscillatory lift and moment distribution. %HUJK and =ZDDQ5 presented a
method based on unsteady pressure measurements for a given single vibration mode and
assumed that the information can be transferred to all modes.
5RGGHQ’s first proposal2 for correcting his own theory was based on the accurate knowledge
of just a little amount of data. These are global aerodynamic coefficients measured on a model
and they should be known in the actual conditions where they are used. Correction of linear
results is done by means of SUHPXOWLSO\LQJGLDJRQDOFRUUHFWLRQPDWULFHV applied either to the
pressure vector at the DLM boxes or to the downwash vector. For the first case, it gives the
following expression
∆& S exp = : S ·∆& S WK (1)

But what it is actually known is a set of aerodynamic coefficients (Ce) and they can be related
with the pressure distribution by means of an integration matrix, 6. That yields equation (2),
whose solution using Lagrange multiplier theory allows the computation of the correction
matrix.
&Hexp = 6 ·∆& Sexp = 6 ·: S ·∆& SWK (2)

As 5RGGHQ showed in his report, the former correction matrices were useful only for
quantitative modification of results, since the correction assumes that the theory represents at
least qualitatively the actual physical phenomenon. However the later recovery of this
correction procedure by 3LWW and *RRGPDQ6 has given fairly good results in the transonic
regime (6LOYD et al7) by pre-multiplication of the downwash by a correction matrix got from
CFD-computed steady pressures for mode shapes or rigid pitch. In addition to this, an
Enhanced Correction Factor Technique has been recently presented by -DGLF8, in which full
correction matrices substitute the diagonal ones of the former formulations. This method deals
with any kind of available CFD or experimental data and gives a level of accuracy in the
correction according with that input.
One of the first attempts to develop a correction procedure based on externally computed
aerodynamics data on DLM boxes, although it uses only quasi-steady transonic pressures, is
due to *DUQHU9 (in Airbus this method is known as Dau-Garner Method). The basic
assumption is that the complex ratio of the unsteady velocity distribution for a given reduced
frequency, N, to the quasi-steady velocity (that is, with N ) remains unchanged if transonic
results are compared with linear ones.
The correction method given later by /XEHU and 6FKPLG10 uses steady transonic pressure
slopes in boxes, ∂Cp∂α , as a reference. They come from measurements (or non-linear
calculations) at different pitch angles of the wing for the desired flight conditions: aircraft
configuration, sweep angle, Mach number, steady attitude of the aircraft. They are further
assumed to be extendable to all modes and reduced frequencies (though limited to low ones).
The authors uses these correcting data in two different ways:
0XOWLSOLFDWLYHFRUUHFWLRQ ∆C p (k ) corr = ∆C p (k ) theo ⋅  C p α Cpα 
 (3)
 exp theo  steady

$GGLWLYHFRUUHFWLRQ (
∆C p (k ) corr = ∆C p (k ) theo + C p α
exp
− Cpα )
theo steady
(α(k ))mode xx (4)

In flutter analysis only very small deflections are considered and that would allow
linearization of the shock wave motion. Nevertheless, steady correcting data gives only the
original position of the shock wave and it should not be expected a good prediction of its
unsteady evolution based on a linear potential theory. That would limit correcting methods
based on steady data to the range of low reduced frequencies.
<RQHPRWR11 offers a new viewpoint to compute transonic pressure phase shifts by using
experimental analysis of shock wave effects. He introduces a semi-empirical equation
correcting chordwise phase results of unsteady linear results. The author refers to
7LGMHPDQ’s12 experimental study to conclude that the amplitude of unsteady transonic
pressure distributions is similar to that observed in the quasi-steady transonic load. Hence, he
proposes Luber’s multiplicative correction for pressure amplitudes in addition to the semi-
empirical equation in phase.
Because of the strong viscous effects, linear unsteady pressures are especially inaccurate at
wing trailing edge, even in subsonic flight. That spoils the prediction of aerodynamic forces in
control surfaces. To reduce this error some constant coefficients are usually employed, for
instance, :LQ]HOO13 proposes to change the local mean velocity (taking FÂ8∞, with F) at the
trailing edge, what means factorising pressures coefficients on the control surfaces. This
coefficient is usually computed to match a given aerodynamic coefficient (i.e. hinge moment).
After unsteady CFD results were widely available, several correcting methods have been
developed to improve linear results using these data, bearing in mind that a full non-linear
flutter analysis is still unaffordable. %DNHU14 corrects linear aerodynamics by defining the
concept of ORFDO HTXLYDOHQFH: A UHSUHVHQWDWLYH deflection shape –similar to the mode shape
involve in flutter- is chosen (for instance, a rigid body pitch) and its non-linear unsteady
pressure fields are computed for different frequencies. To relate these pressures with the ones
derived from the actual wing deflection, the wing is divided into regions (strips spanwise) and
local equivalence of some aerodynamic quantity (local lift) is imposed. That gives the ORFDO
HTXLYDOHQFHIDFWRU that at each region and for each modal shape corrects the linear pressures.
Similarly, 6FKXO]Hand 7LFK\15correct unsteady linear aerodynamics in the transonic regime
by using factors derived from CFD calculations on assumed mode shapes. They make use of a
factorisation coefficient and an additive term, which reads:
∆C p = ∆C p ⋅ F ⋅ e iϕ + C p
corr DL x
(5)
Both correction factors are evaluated locally at DLM boxes from both a reference z-mode and
a reference x-mode. Since DLM airloads do not include in-plane movements, the authors
suggest adding it explicitly, by computing pressure distributions due to a reference x-mode
and adding it stripwise. It reads as follows:
&)'
Lϕ &S] &)' φ[ (\)
) ⋅H = &S[ = &S[ (6)
'/
&S] UHI − [ φ UHI − [ ( \ )
UHI − ] [

The selection of the reference modes could be in principle arbitrary, but the more similar to
the DFWXDOmode shape is chosen (i.e. modes involved in flutter) the more accurate results may
be expected.
Some other different ways to improved the accuracy of DLM have proposed recently: An
original suggestion due to %DNHU16 again, which is based on 5XEEHUW’s works17, assumes
removing a number of boxes in strips at the wing tip. He achieves then an improvement in the
convergence of DLM results. 5RGGHQ18presented a quartic approximation to the kernel in the
DLM, what allows higher aspect ratio in surface boxes (up to 1:5) and extend frequency limits
for linear unsteady aerodynamics. /XEHU and %HFNHU19has developed a correction procedure
that works at high angles of attack and has therefore no interest in flutter problems, but in
buffet ones. Finally9DQ=\O20has presented recently a suitable way of studying delta wings
by DLM.

 '(9(/230(172)6(/(&7('&255(&7,217(&+1,48(6

The selection of methods to correct the linear theory has been done in the present work taking
into account the variety of available data that can be found. Therefore, /XEHU¶V proposal is
taken due to its simplicity for preliminary corrections. The method needs only a little amount
of non-linear data and may give the first approximate results or at least information about
trends. The method given by 'DX*DUQHU should be also considered, since it introduces
steady data (which many times can be the only available) computed for different mode shapes
in a thoughtful way. Finally, two state-of-art methods (%DNHU and 6FKXO]H7LFK\) have been
taken in order to allow the linear theory to use the enormous power of CFD’s. In particular,
Baker’s development shows a rather more interesting attractive, since the aerodynamic model
is not taken as a whole and corrections follow the local behaviour of the different modal
shapes. But by using Schulze-Tichy formulation it is possible to include data coming from
different assumed mode shapes.
All these methods were included in EADS-CASA software for unsteady linear aerodynamics,
which is based in DLM calculations. As a result program 7+(2'256(1 (THEOretical
DOublet-lattice with tRanSonic ENhancements) has been created.
Each method needs a different set of transonic data, but all of them must be extracted from an
external source. Therefore, a geometric interpolation of the external results to the selected
points in the DLM boxes is needed, as well as a second interpolation to obtain the conditions
(Mach and reduced frequency) selected in the DLM. The first one is achieved by means of the
surface interpolation given by +DUGHU and 'HVPDUDLV21, while the second one uses the classic
$NLPD’s method22.
A critical review of the selected methods is now included with examples of application on the
AGARD445.6 wing23 and the AMP Model. Both have been widely used as reference
configurations for transonic flutter analysis, since experimental results can be found in the
literature. DLM models for both wings are included in ILJXUHVand.

)LJXUH. DLM model of )LJXUH. DLM model of AMP wing


AGARD445.6

)LJXUH  shows the V-g plot of flutter computations on AGARD445.6 using uncorrected
DLM linear aerodynamics.

)LJXUH Flutter results (V-g plot) for AGARD 445.6 using linear unsteady aerodynamics

/XEHU6FKPLG0HWKRG

The method10 requires a steady ∆Cp distribution (or the corresponding Cpα values) over the
model surface for two angles of attack of a rigid pitch mode, computed at the desired flight
conditions. These pressures are later compared with linear ones, as stated in equations (3) and
(4). A comparison between those pressures is shown in ILJXUH  for AGARD445.6 wing,
where it can be seen how the TDS includes the effect of the transonic shock wave that cannot
be predicted by linear aerodynamics.
)LJXUH. TSD vs. DLM ∆Cp on AGARD445.6 for 2 deg pitch

Results are not hopeful at all for the PXOWLSOLFDWLYHPHWKRG, since the corrected pressures by
equation (3) show singularities at the trailing edge and distort the aerodynamics force
matrices.
The DGGLWLYH PHWKRG gives rather more reasonable results in the two examples considered
(AGARD445.6 and AMP). The flutter speed is reduced in the correct sense to matched the
experimental data, although in the AGARD case the result is too conservative and in the AMP
wing only half of the distance between the linear result and the experimental one is achieved.
7DEOH shows the results for both wings at their test conditions (0.78 Mach for AMP and 0.96
Mach for AGARD445.6), showing experimental, linear and corrected values. The latter come
from TSD computations for rigid pitch deflection at 2 and 4 degrees.

$*$5' $03

&RUUHFWHGPRGHV )OXWWHU6SHHG )OXWWHU)UHTXHQF\ 7RWDOSUHVVXUHDW)OXWWHU EDU


.($6 +]
st nd
1 and 2 100.55 10.25 1.120
st th
1 to 6 99.87 10.27 RQO\WZRPRGHVDYDLODEOH

Experimental 136 14.50 1.03


Linear DLM + K-method 146 14.63 1.216

7DEOH Results on AGARD445.6 and AMP for Luber-Schmid additive formulation

Flutter frequency values in AGARD445.6 shows that the coupling mechanism between the
structural modes is strongly modified by this formulation.

%DNHU0HWKRG

This method14 requires a non-linear unsteady ∆Cpdistribution on the surface for a reference
mode and at a range of reduced frequencies. They should be computed at all critical flight
conditions. The reference mode is chosen to be close enough to the modes involved in flutter.
The author recommends to use either one of them or a pitch mode.
Taking the reference modal shape, ϕ, the resulting non-dimensional downwash can be
evaluated, wϕ (L: index on the boxes):
G]ϕ ]ϕ
Zϕ L = + MN L (7)
G[ L EUHI
Using the AIC matrix, the corresponding linear pressure field is then computed, as well as
pressures for the current mode shapes involved in the flutter process. Then, to set the ORFDO
HTXLYDOHQFH between them, different zones in the aerodynamic model must be defined and
locally equivalence in lift or the generalised force is imposed.
If ^+` is the integration vector over the boxes to compute the generalised force for a given
modal shape Mfrom their pressures and it is divided into subarrays ^+ ` for each region ]the ]

ORFDOHTXLYDOHQFHIDFWRU, η , is computed as:


M
]

Fj
z
{H z }T ⋅ {Cp j }
η =
z
=
{H z }T ⋅ {Cp ϕ }
j z (8)

Finally, the expression to compute the corrected pressure field reads:
{Cp corrected
}mode j = ηzj ⋅ {Cp CFD }ϕ + [AIC] ⋅ {w j − ηzj ⋅ w ϕ } (9)
This formulation allows inclusion of aerodynamic body effects, which can be taken into
account in the same way as in linear theory as a subtracting term on the modal downwash.
That means substituting Z by Z Z  in (9) and adding new interference panels on the
M M E

aerodynamic model. The new term Z comes in THEODORSEN from the Miles’ Theory of
E

slender bodies.
Some results will be shown now for AGARD445.6 and MDO wings. In both cases, non-linear
unsteady pressures were computed using an Euler code the concept of time linearization that
will be described in paragraph 4.
First, classical linear flutter calculations were carried out on the AGARD445.6 model at 0.954
Mach. That gave non-conservative results in comparison with experimental data23, in which
the expected WUDQVRQLFGLSappears. Baker’s procedure was then applied to correct the linear
pressure distribution for different sets of mode shapes with the CFD results. 7DEOHshows a
selection of the results.
5HIHUHQFH &RUUHFWHG )OXWWHU6SHHG )OXWWHU6SHHG
0HWKRG &$6$ 6$$%
0RGH PRGHV
st
1 (first bending) 140 140
nd
Pitch 2 (first torsion) 123 128
st nd
Lift equiv 1 and 2 119 115
st th
1 to 6 120
st
1 (first bending) 135
First bending st nd
1 and 2 124 118
st
1 (first bending) 147
generaliced
force equiv nd $*$5'DW )OXWWHU6SHHG
Pitch 2 (first torsion) 168
0DFK  .($6

1
st
and 2
nd
164 ([SHULPHQWDO 136

'/0 XQFRUUHFWHG 146

7DEOH Flutter speed in KEAS for AGARD445.6 at M=0.954 using Baker correction
Some conclusions can be drawn from these data: Equivalence in lift using pitch mode makes
the results to be conservative, while equivalence in generalised forces gives even worst results
than the linear DLM. When the correction is applied to the first mode using non-linear data
from the same mode, both methods (namely, local equivalence either in lift or in the
generalised aerodynamics forces) give the same results. That happens because, in fact,
Baker’s Method (in both formulations) performs for these cases only an interpolation of the
non-linear pressures to the given reduced frequencies and it does not use the DLM pressures
for this mode. With the non-linear pressure data for the first mode, the correction is equivalent
to calculate the terms Q11 and Q21 with the non-linear pressures and the other terms in the
matrix with the linear ones.
)LJXUH shows a comparison between aerodynamics matrices. The evolution of the first four
terms of the unsteady aerodynamics matrices (Q11, Q12, Q21, Q22) of the AGARD445.6 case
versus the reduced frequency N is presented in three cases:
[1] Continuous line with stars: Direct DLM results without correction.
[2] Dashed line with squares: DLM is corrected using the Baker method based on non-linear
bending mode data and applied to the bending mode only (this is why this curve appears
only for Q11 and Q12).
[3] Dotted line with circles: DLM is corrected using the Baker method based on non-linear
pitch mode data and applied to the bending and pitch modes.

extrapolated k

extrapolated k

extrapolated k extrapolated k

)LJXUH. Aerodynamic matrices (first two modes) corrected by Baker Method.

In case [3] the aerodynamic terms involving the pressures at the first mode (Q11 and Q21)
slightly change to force the flutter results to move in the JRRG direction, whereas in [2] the
corrected force matrices are WRR different.
When all methods were developed in THEODORSEN, a constant extrapolation outside the
known range of reduced frequencies was introduced in the non-linear results. Matrices in
ILJXUH show how this extrapolation is clearly inadmissible. However, it does not affect the
flutter mechanism, which occurs at the lower range of reduced frequencies.

)LJXUH Aerodynamic model for MDO

The MDO wing (a CFD aerodynamic model is shown in ILJXUH was developed within the
European founded project MDO and since then has been widely used as a quasi-industrial test
case for verification of advanced structural and aerodynamic codes. In the present work,
flutter results for this wing are given at Mach=0.88 in terms of DOWLWXGH RI IOXWWHU in the
standard atmosphere. Linear analyses predict that flutter occurs at a very low altitude,
between 1.28 km and 1.75 km, depending on the mesh size and the number of modes included
in the solver (See WDEOH).
Aerodynamics matrices were later computed directly from Euler results by exciting the 18
mode shapes one by one (what we call (XOHUOLQHDUL]DWLRQ), that gave a flutter altitude in the
range from 5.13 to 5.50 km. There is thus a significant gap between linear and non-linear
results for this test case, which has been tried to fill using Baker correction.
7DEOH shows the results of applying the Baker correction method to the MDO wing using
two mesh sizes in the DLM (a coarse panel grid with 11x19 and a dense panel grid with
15x19 elements. It is remarkable that if the coarse panel grid, which is sufficient for the linear
calculations, is used the results get excessively conservative. This is mainly due to lack of
resolution at the leading edge when interpolating between the CFD grid and the panel grid.
The correction does, however, give results fairly close to Euler linearization results and,
thanks to the time linearization procedure, only requires CFD calculation for one mode as
opposed to 18 i.e. a considerable saving in time.

&DVH PRGHV PRGHV PRGHV PRGHV

Linear (coarse) 1.32 1.61 1.69 1.75


Linear (dense) 1.28 1.58 1.66 1.72
Correction (coarse) 8.40 8.50 8.50 8.50
Correction (dense) 6.50 6.55 6.60 6.64
Euler linearisation 5.13 5.38 5.43 5.50
7DEOH Flutter results (flutter height in km) for MDO wing at M=0.88
6FKXO]H7LFK\0HWKRG

This method15 requires non-linear unsteady ∆Cpdistributions on the lifting surface for several
mode shapes and several reduced frequencies. The CFD computations should be performed at
all critical flight conditions and the mode shapes should be the ones involved in the flutter
mechanism, although in fact it depends on the availability of CFD results.
To achieve more accurate results, pressures due to in-plane displacements can be introduced.
As a first approximation to the in-plane component of each assumed modal shape, these
pressures could come from a standard x-mode defined as follows: chordwise in-plane
displacements on the wing changing linearly from zero at the root to a given value at the tip.
For each actual mode in DLM computations these non-linear pressures would be factorised by
the coefficient that better relates the in-plane displacement of the mode with the reference
mode.
After computing these data for the desired range of reduced frequencies, corrected pressures
are directly evaluated at the boxes of the lifting surface by using equations (5) and (6).
Two variations of the method have been analysed: 1) The original method, with correction
factors given by (6) and a modified formulation, in which a dimensioning factor has been
added to both reference pressure distribution. The factor was selected as the global lift
coefficient, and the correction could be presented as follows:
&)'
Lϕ &S] & /'/
) ⋅H = '/ (10)
&S] & /&)'
UHI − ]

It should be noticed that only one reference mode is being used herein to correct either one or
several modes. However, better results could be obtained if information about the complete
set of modes involved in the flutter phenomenon were used, in order to correct each mode for
its own non-linear pressure distribution. Since Euler calculations from just one modal shape
were available for AGARD445.6, there were no options to do that in the present work.
This method can also be useful to introduce directly the steady ratio of linear and non-linear
pressure coefficients for each mode, correcting with this the whole range of unsteady
pressures, if no more CFD information were available.
Flutter calculations for AGARD445.6 using this correcting method in both formulations have
given the results showed in WDEOH (Linear flutter at 146 KEAS; Experimental at 136 KEAS).

)OXWWHU6SHHG .($6
5HIHUHQFH &RUUHFWHG
0RGH PRGHV Original Modified
Formulation Formulation
$*$5' )OXWWHU6SHHG
st
1 (first bending) 132 145 DW0DFK  .($6

nd
Pitch 2 (first torsion) 128 138 ([SHULPHQWDO 136
st nd
1 and 2 115 139 '/0 146
XQFRUUHFWHG
st
1 135 145
First bending st nd
1 and 2 126 139
7DEOH. Flutter results for AGARD445.6 at M=0.954 using Schulze-
Tichy correction

Results show that using rigid pitch as the reference mode, the Modified Formulation seems to
give rather accurate results if applied at least to the first torsion mode, however it remains
always non-conservative and its use cannot be recommended. The ineffectiveness of this
modified method occurs because it implies a reduction factor on the transonic pressures, since
transonic flow in the same geometric conditions as linear flow gives higher loads. This load
increase is one of the reasons of the WUDQVRQLFGLS
If data come from the actual mode involved in the flutter process (here first bending) the
Original Formulation gives better results. The reason is that the correction performs in this
case only an interpolation in the non-linear results, not using the DLM pressures. Therefore,
calculations are exactly the same that when the Baker’s method corrects this mode (used in
the same conditions). Of course, both, Baker and Original Schulze-Tichy give in this case the
same results.
When the Original Formulation is applied using rigid pitch as reference, results are not so
predictable. It can be seen that if only one of the modes is corrected, the result matches rather
well the experimental data, but that changes if more than one mode is corrected.

 7,0(/,1($5,=$7,21352&('85(6

&RQFHSWV

In this chapter a time linearization strategy for the flutter onset problem is presented. Different
approaches to time linearization will be discussed and the reasons to keep with one of them
will be presented. Such a procedure is required to achieve an interface between CFD results
and classical linear flutter analysis, which can be done in two different ways: 1) By generating
directly air force matrices from non-linear results and 2) by supplying the necessary
information for correction procedures to linear aerodynamics.
Non-linear equations, like the Euler’s or Navier-Stokes’ equations, are written in conservative
variables 4:
∂4 / ∂W + GLY ( JUDG ( ) ( 4 ))) = 0 (10)
Assuming that a steady state 4 has been obtained, small unsteady perturbations ∆4 at least


formally satisfy a linear system of partial differential equations:


∂∆4 / ∂W + + ( 40 )∆4 = 0 (11)
with a rather complicated matrix + 4 . There are basically two different ways of making use
R

of this expression.
 7LPHGRPDLQDSSURDFK
The linear equation for ∆4 can be used for time stepping. This requires the same kind of
stability handling as for the fully non-linear equations and similar flux evaluation must be
done. It is not clear how much time is it possible to save with this approach and how to
arrange it so as to be computationally considerably more efficient than running the non-linear
system. It was chosen not to use that approach.
 )UHTXHQF\GRPDLQDSSURDFK
The linear system of equations can be solved for harmonically oscillating components of ∆4.
In fact, there will be a linear system of equations to be solved for the complex valued
amplification function of the unsteady flow field component. Since only a steady state has to
be found, this approach can be cost effective if the unsteady forces for just one or a few
frequencies are sought for.
However, for flutter analysis it is customary to compute unsteady forces for a larger number
of reduced frequencies, certainly more than just one or two. Therefore there has been much
interest in a special technique, based on exciting a mode shape by a time dependent amplitude
function of certain properties. This is what will describe next.

6PDOOSHUWXUEDWLRQVDQGWUDQVIHUIXQFWLRQV

Assume that the response J W of a system due to an input δ W can be written as a sum of a
static part and one that is fluctuating with time, and that the same holds for the input:
J( W ) = J 0 + J1 ( W )
(12)
δ (W ) = δ 0 + δ1 ( W )

Now assume that there is an approximate linear correspondence between δ W and J W such  

that J /δ . Then the linear mapping / can be represented in the Fourier domain by /̂ in (13).
 

J$ 1 (ω ) = /$ ⋅ δ$(ω ) (13)

This /̂ is the transfer function for the response due to the exciting input, and it acts as an
amplification factor (magnitude and phase) for a harmonic input with frequency ω.
In the context of unsteady aerodynamics, the unsteady aerodynamic coefficient at a fixed
location is such an amplification factor. An example is pitch oscillation of a wing with
( )
amplitude α (W ) = α  + ε ⋅ VLQωW = α  + ε ⋅ ,P H LωW , which results in the pressure coefficient
at a certain point ; being & S ( ; W ) = & S + ε ⋅ ,P(∆& S ( ; ω )H LωW ) . Here the complex
VWHDG\

number ∆&S ;ω is a transfer function for pressure due to pitch (it is also called XQVWHDG\
SUHVVXUH due to pitch). On the other hand, if pressure is integrated to form generalised forces,
it is possible in the same way to introduce transfer functions for generalised forces due to
excitation of any mode shape, in particular natural mode shapes. The unsteady airforce
matrices used in linear theory are exactly such transfer functions and linear potential theory
programs directly compute them.
Now, again assuming small perturbations, the time dependent part of pressure fluctuations (or
maybe also viscous forces fluctuations) can be analysed using Fourier transform. As a matter
of fact, the transfer function /̂ above can be obtained from (13) if δÖ (ω ) is non-zero. This
suggests a post processing method to find unsteady pressure and unsteady generalised forces
in the frequency domain. This is done by exciting the non-linear aerodynamic computation by
a suitable pulse, recording the instantaneous pressure or force as a time series and then
computing the quotient of their Fourier transforms.

&KRLFHRISXOVHSHUWXUEDWLRQDQGDLUIRUFHJHQHUDWLRQ

The amplitude function in (14) has the Fourier transform written in (15).
 π W 
δ  (W ) =   − FRV  for 0≤t≤t0 and 0 elsewhere
W 
(14)

L ( 2π / W 0 ) 2
δ$1 (ω ) = ⋅ (1 − H − ω ) ⋅
L W0
(15)
2 ω ⋅ (ω 2 − ( 2π / W 0 ) 2 )
The last function has removable singularities for ω  and ω πW with non-zero limit values


W  and -W  respectively for those frequencies. )LJXUHVand  plots this Fourier transform
 

δ
versus ω π  . It is non-zero for angular frequencies lower than πW . Notice that up to
Ö

W   W 


twice that typical frequency, the Fourier transform of the pulse is non-zero, and therefore it
can be used to compute transfer functions in a well-defined frequency range.
0.5 0.5
Real part
Imaginary part
0.4

0.3
0
0.2

0.1

0 −0.5
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
Frequency in units of 2π/t0 Frequency in units of 2π/t0

)LJXUH Magnitude of the Fourier )LJXUH Real and imaginary part of the Fourier
transform of the pulse amplitude transform of the pulse amplitude

$SSOLFDWLRQRQDWHVWFDVH

Time linearization results has been compared with experimental ones for a low aspect ratio
wing with 55º wing leading edge tested at Mach 0.95 described in 24 and showed in ILJXUH.
Unsteady aerodynamics was induced by the harmonic motion of a control surface.

)LJXUH 55º delta wing used as test case

Euler simulations were carried out at SAAB by means of the FFA’s code EURANUS,
running the unsteady analysis by using a perturbation grid from a well-converged steady state.
Time step was set to ÂW  whereas excitation amplitude of the pulse perturbation function


was fixed around half a degree in pitch.


For each point on the aircraft surface and each mode shape, the time series of pressures are
then computed. Computation of their Fourier integral of such a time series for a selected range
of frequencies I I and division by the Fourier transform of the pulse amplitude function
 1

for each of the frequencies gives the set complex numbers representing the unsteady pressure
distributions for all frequencies in one and the same run. )LJXUH  shows a comparison
between measured data and computations using this method for the 55º delta wing at mild
transonic conditions (0.94 Mach) with a 48-Hz rigid oscillation of the control surface.
Although there was only 33 1/3 steps in the pulse excitation, there can be seen a rather good
agreement between both results.
)LJXUH Unsteady pressure on a 55o delta wing with oscillating flap, computed with pulse
excitation (lines) and harmonic excitation (symbols). Real part (left) and imaginary part (right).

If this procedure is applied to a set of N mode shapes, the corresponding NxN airforce
matrices (as many as desired frequency values) can be computed after only N runs of the CFD
program, since all frequencies are obtained in one single run. In ILJXUH  we present
comparison between airforce matrices, computed with the CFD pulse excitation method and
other analysis of the 55-degree delta wing. For the sake of simplicity we will only regard the
first airforce element, A11, which is formed from the first elastic bending mode of the wind
tunnel model.

)LJXUH Airforce element A11 as a function of reduced frequency in subsonic flow (Mach
0.7 left) and transonic (Mach 0.94 right).

 &21&/86,216

In the present work a complete set of tools has been presented that allows the correction of
linear unsteady aerodynamics with results of non-linear computations. Several procedures has
been analysed for the correction algorithm, in order to provide the higher degree of accuracy
of the corrected results for different types of non-linear calculations (steady or unsteady, for
one or several rigid or elastic assumed deflections). An efficient method of linearization in
CFD results has been also tested and it gave excellent results both in time-reduction in
calculations and accuracy of the approximation.
All correction methods analysed gave conservative results for flutter prediction, although
pitch steady data as input may give too low flutter speeds. Unfortunately, the main test case,
AGARD445.6, has shown a rather unstable numerical behaviour in flutter calculations and it
has given even problems in linear calculations (slight variations in methods have given big
differences in results). We cannot be conclusive based on results with AGARD445.6 and a
more exhaustive relative comparison between correction methods is therefore still pending.
This analysis methodology is the basis for the transonic aeroelastic capabilities at EADS-
CASA and allows high flexibility to face up to the flutter problem of a production aircraft in
the transonic flight.
 $&.12:/('*0(17

This work was performed as a part of the UNSI project (Unsteady Viscous Flow in the
Context of Fluid-Structure Interaction), a collaboration between ALEINA, BAE, EADS-
CASA, EADS-M (Co-ordinator), DASSAULT, DERA, DLR, FFA, IMFT, IRPHE, NUMCA,
ONERA, SAAB, TUB and UMIST/QMV. The project was funded by the European
Commission, DG XII, Brite/Euram, under the IMT initiative (Project Ref: BRPR-CT97-
0583).

 5()(5(1&(6
1 Giesing J.P, Kalman T.P, Rodden W.P.³'LUHFWDSSOLFDWLRQRIWKHQRQSODQDU'RXEOHW
/DWWLFHPHWKRG´Technical Report AFFDL-TR-71-5, part I, vol I and II. Nov 1971
2 Giesing J.P, Kalman T.P, Rodden W. P. &RUUHFWLRQIDFWRUWHFKQLTXHVIRULPSURYLQJ
DHURG\QDPLFSUHGLFWLRQPHWKRGVTechnical report NASA CR-144967. May 1976
3 Yates E.C. “Modified-Strip-Analysis Method for Predicting Wing Flutter at Subsonic or
Hypersonic Speeds”. J. Aircraft, Vol.3, Nº1, 1966
4 Yates E.C, Bennett R.M. “Analysis of Supersonic-Hypersonic Flutter of lifting Surfaces
at Angle of Attack”. J. Aircraft, Vol.9, Nº7, 1972
5 Bergh H, Zwaan R.J. “A Method for Estimating Unsteady Pressure Distribution for
Arbitrary Vibration Modes from Theory and from Measured Distributions for one Single
Mode”. NLR-TR F.250, 1966
6 Pitt D.M and Goodman C.E. “Flutter Calculations using Doublet Lattice Aerodynamics
modified by the Full Potential Equations”. Proc. Fo the 28th AIAA/ASME/ASCE/AHS
Structures, Structural Dynamics and Materials Conference. AIAA-87-0882-CP,
Monterrey, CA.1987
7 Silva R, Mello O and Azevedo J. “Transonic Flutter Calculations based on Assumed
Mode Shapes Corrections”. International Forum of Aeroelasticity and Structural
Dynamics, Madrid. June 2001
8 Jadic I, Hartley D, Giri J. “An enhanced correction factor technique for aerodynamic
influence coefficient methods”. MSC’s proceedings for 1999 Aeroespace User’s
Conference
9 Garner H.C. "A practical framework for evaluation of oscillating aerodynamics loadings
on wings in supercritical flow". AGARD CP-226, April 1977
10 Luber W, Schmid H. "Flutter investigations in the transonic flow regime for a fighter type
aircraft". AGARD RN703, September 1982
11 Yonemoto K."A practical method predicting transonic wing flutter phenomena". ICAS-
1984, pp.724-732
12 Tidjeman H. "Investigation of the transonic flow around oscillating airfoils". NLR-RL-
77090-U
13 Winzell B. “Recent Applications of Linear and Nonlinear Unsteady Aerodynamics for
Aeroelastic Analysis”. AGARD-CP-507. March 1992
14 Baker M.L. "CFD based corrections for linear aerodynamic methods". AGARD R822,
October 1997
15 Schulze, Tichy. "An efficient aero-correction method for transonic flutter calculations".
DASA B08/98
16 Baker M.L and Rodden W.P. “Improving the Convergence of the Doublet-Lattice
Method through Tip Corrections”. International Forum on Aeroelasticity and Structural
Dynamics. Williamsburg, June 1999
17 Rubber P.E. “Theoretical Characteristics of Arbitrary Wings by a Non-Planar Vortex
Lattice Method”. D6-9244. Boeing, 1964
18 Rodden W, Taylor P and McIntosh S. “Further Refinement of the Substonic Doublet-
Lattice Method”. Journal of Aircraft, Vol. 35 No. 5. Sept 1998
19 Luber W, Becker J. “High Incidence Unsteady Aerodynamics for Aeroservoelastic
Predictions” AGARD R-822. October 1997
20 Van Zyl L.H. “Application of the Subsonic Doublet Lattice Method to Delta Wings”. J.
Aircraft, Vol.36, Nº3. May 1999
21 Harder R.L, Desmarais R.N. “Interpolation using Surface Splines”. Engineering Notes,
Vol 9, No 2, pp. 189-191. Feb 1972
22 CASA NT-T-ADD-98001. Appendix B: Akima Method
23 Yates, J.C. “AGARD Standard Aeroelastic Configurations for Dynamic Response. I-
Wing 445.6”. AGARD R765, July 1988
24 Karlsson A., Winzell B., Eliasson P., Nordström J., Trongren L. And Tysell L. “Unsteady
Control Surface Pressure Measurements and Computation”, Proc AIAA 14th Applied
Aerodynamic Conference, New Orleans 1996, Paper 96-2417

You might also like