Vous êtes sur la page 1sur 25

Mathl. Comput. Modelling Vol. 19, No. 3/4, pp.

l-25, 1994
Pergamon Copyright@1994 Elsevier Science Ltd
Printed in Great Britain. All rights reserved
08957177/94~86.00 + 0.00
0895-7177(94)E0020-N

A New Sensitivity Analysis for Structural


Optimization of Composite Rotor Blades*

C. VENKATESAN, P. P. FRIEDMANN AND K.-A. YUAN


Mechanical, Aerospace and Nuclear Engineering Department
University of California, Los Angeles, CA 90024-1597, U.S.A.

Abstract-This paper presents the mathematical details of the formulation for the sensitivity
derivatives for the structural dynamic, aeroelastic stability, and response characteristics of a rotor
blade in hover and forward flight. The formulation is denoted by the term semi-analytical approach,
because certain derivatives are evaluated by a finite difference scheme. Using this formulation, sen-
sitivity derivatives for structural dynamic and aeroelastic stability characteristics were evaluated for
a composite rotor blade. Useful conclusions regarding the relative merits of the semi-analytical ap-
proach, when compared to a pure finite difference approach, are obtained. In addition, influence of
ply orientation angles on the vibratory hub loads in forward flight is also considered.

NOMENCLATURE
a lift curve slope
length of beam element
c blade chord
number of modes
blade profile drag coefficient
cd, mass, damping, and stiff-
Oj j design variable ness matrices, respectively,
EL longitudinal Youngs in modal space
modulus [MIF, PIF, WIF mass, damping, and stiffness
ET transverse Youngs modulus matrices, respectively, in Ii-
nite element domain
F'h@') vector of forcing function
MH~, MH~, MHZ hub moments in the nonro-
{FH), {MH) vector of hub shears and mo
tating hub fixed coordinate
ments, respectively
system
{F% {M:) vector of kth blade root
root moments of the kth
shears and moments,
blade in the hub fixed rotat-
respectively
ing coordinate system
FHX, FH~, FHZ hub shears in the nonro-
N number of blades in the ro-
tating hub fixed coordinate
tor system
system
ND number of design variables
root shears of the kth blade
in the hub fixed rotating co- td vector of finite element
ordinate system nodal degrees of freedom
longitudinal shear modulus R rotor radius
GLT
kA blade cross-section polar ra- PI transformation matrix be
dius of gyration tween rotating and nonro-
tating hub fixed coordinate
km blade cross-section msss
systems
radius of gyration, k& =
kL -I- k%, tub ith modal vector obtained
in the structural dynamic
principal mass radii of gyra-
analysis
tion of the cross-section

*This research wss supported by NASA Langeley Research Center under Grant NAG l-833, with H. Adelman as
Grant Monitor. The authors wish to express their gratitude to L. A. Schmit, Jr. for the numerous discussions
and his insightful comments on the formulation of the sensitivity derivatives. The comments of R. T. Haftka and
A. Sepulveda are also gratefully acknowledged.
C. VENKATESAN et al.

Ah, Av ply orientation angle in


kth eigenvector obtained in horizontal and vertical
stability analysis walls, respectively
kth eigenvector of the ad- LT longitudinal Poissons ratio
joint matrix obtained in sta- 0 rotor solidity
bility analysis
l/rk Azimuthal position of the
generalized coordinate vec- kth blade
tor in modal space
wk modal frequency of kth
nonlinear time invariant mode
equilibrium position of the
wLltwFl,wTl fundamental rotating lag,
blade
flap, and torsional blade
time dependent perturba- frequencies
tional quantity
angular velocity of rotor
state vector
blade percone angle ;LJj.rLj quantities evaluated at the
baseline design variable, Dj
lock number quantity corresponding to
ok
modal damping in the kth kth mode
mode quantities evaluated about
{ HB
kth eigeUVahe (A, = <k f the equilibrium position
iwk) corresponding to baseline
configuration

1. INTRODUCTION
The use of structural optimization for vibration reduction in helicopters, in forward flight, has
emerged as an important design tool during the last decade. This approach is very attractive
because it reduces the vibrations at their source, i.e., the main rotor, before they propagate into
the fuselage. A substantial body of research on helicopter vibration reduction using structural
optimization with aeroelastic and multidisciplinary constraints has been reviewed in [l-3].
An. important ingredient in the application of optimization techniques to design problems
is the formulation of the sensitivity derivatives of the objective function and constraints with
respect to the design variables. When explicit analytical expressions for the objective function
and constraints are not available, the sensitivity derivatives are usually computed by a numerical
finite difference approach. However, there are certain disadvantages associated with the finite
difference approach, such as
(1) the numerical inaccuracy due to step size; and
(2) solution of the complete design problem is required for every change in the design variable,
leading to increased computer times.
Furthermore, the finite difference approach does not provide insight on the significance of various
ingredients which influence the sensitivity derivatives. In [4-71, the sensitivity derivatives were
formulated using chain rule differentiation (i.e., partial derivatives). This scheme was found to
be more efficient and reliable than the finite difference approach, since various derivatives are
evaluated analytically.
The implicit approach to formulating the aerodynamic loads has distinct advantages over the
previously used explicit approach in aeroelastic stability and response calculations [8]. Most com-
prehensive rotorcraft aeroelastic analyses are usually based on such an implicit type formulation.
However, the implicit formulation does not lend itself to generating explicit analytical expressions
for the sensitivity derivatives, which are useful in structural optimization. For such cases, a new
technique based on the combination of the chain rule for differentiation with a finite difference
scheme is required in order to formulate an effective sensitivity analysis.
The main objective of the paper is to present the mathematical formulation of a new compre-
hensive sensitivity analysis capable of capturing all important effects associated with changes in
the design variables. An attractive feature of this formulation is its compatibility with a recently
developed aeroelastic model for composite rotor blades with straight and swept tips, which is
A New Sensitivity Analysis 3

based on an implicit formulation of the aerodynamic loads in hover and forward flight [9,10]. In
addition to the detailed mathematical formulation, the paper also contains results illustrating the
effectiveness of this new sensitivity analysis by comparing the results with those based on a pure
finite difference formulation. Finally, the effect of changing ply orientation on the hub shears and
moments in forward flight is also shown.

2. MATHEMATICAL DERIVATION
The aeroelastic stability and response behavior of a rotor blade is dependent on the trim (or
equilibrium) state of the helicopter. For a given flight condition, the trim parameters required for
overall vehicle equilibrium are related to the steady state response (or position) of the rotor blades;
and the stability characteristics of the blade are influenced by its equilibrium state. Therefore,
changes in the design variables will modify
(a) the trim parameters,
(b) the blade equilibrium position, and
(c) the stability of the blade.
All these effects can be properly included in the formulation of sensitivity derivatives. However,
such a complete treatment would require additional computer time which may be comparable
to that encountered in a pure finite difference formulation. Therefore, by making certain careful
approximations in the formulation of sensitivity derivatives, the efficiency of the approach can be
preserved. In the present formulation of sensitivity derivatives, it is assumed that the changes in
the trim parameters, induced by small increments in design variables, are negligible. Appropriate
choice of move limits in the design variables during the optimization process will ensure the
validity of this approximation.
A concise mathematical formulation of the sensitivity derivatives with respect to the design
variables is presented in the following section. This approach is denoted as a semi-analytical
approach, beause certain derivatives are computed using a finite difference scheme.

2.1. Sensitivity Derivatives of Natural Frequencies

The procedure used for calculating the sensitivity of natural frequencies of the rotor blade is
based on the scheme described in [ll]. The linear structural dynamic problem of a rotating blade,
based upon the finite element formulation which is described in [9,10], can be written as

[Ml,@) + [Cl .W + [~l,W = 0. (1)

The size of the matrices [M]F, [C]F, and [K]F are, in general, (n x n). The modes required for
aeroelastic analysis are undamped modes, hence, the [C]F matrix in equation (1) is neglected,
and the free vibration problem reduces to

- w2[M] &} + [K] ,{u) = 0, (2)


and any eigenvalue, eigenvector pair satisfies equation (3),

- wf[M],{u}i + [IqF{~}( = 0. (3)

Premultiplying equation (3) with {u}: and differentiating with respect to any design vari-
able Dj (j = 1,2.. . , ND), yields

- 2Wi~{U};[M]F{U}i - w:{u}$!${U), + {&&{U}~

-44&4i + Fl.Wi + -w{u};pqF + {U}TpqF$g = O(4)


I [ I
4 C. VENKATESAN et al.

Recognizing that the mass and stiffness matrices are symmetric matrices, the terms inside the
square brackets are identically equal to zero. Rearranging the terms yields sensitivity of the
natural frequency wi, with respect to the design variable Dj:

where the range of the indices used is: i = 1,. . . , m and j = 1,. . . , ND. Here, m denotes the
total number of modes considered; and ND is the total number of design variables selected for
the optimization problem.
In the study, which is aimed at the structural optimization of composite rotor blades, explicit
expressions for the elements of mass and stiffness matrices are not available. Therefore, the
variation of these matrices with respect to design variables Dj (j = 1,. . . , ND) has to be obtained
from a finite difference approach.

2.2. Sensitivity of the Aeroelastic Stability Boundary in Hover

The aeroelastic stability of the rotor blade in hover is obtained from a set of perturbational
equations linearized about a nonlinear equilibrium position. Therefore, the sensitivity analysis
for the blade stability boundary has to take into account a number of important ingredients:
(a) blade stability analysis is carried out in modal space;
(b) the nonlinear static equilibrium position of the blade is influenced by changes in the design
variables.
It is important to emphasize that variations in the mass, stiffness, and damping matrices due
to changes in the design variables are denoted in this study by the term direct influence, while
modifications in the static equilibrium position due to changes in the design variables are denoted
as an indirect influence.
The equations of motion of the rotor blade in modal space can be written as [9,10]

PflW + [Cl{3)+ PwY~ + IF) = (0). (6)


Equation (6) represents m coupled nonlinear ordinary differential equations corresponding to the
m modes considered in the problem. The vector {F} contains all the structural and inertial
nonlinear terms, in addition to the aerodynamic terms. The solution of equation (6) can be
assumed to be of the form
{Y] = {YO] + {AY(~)], (7)
where {yc} represents the nonlinear time invariant equilibrium position of the blade, and {Ay(t)}
represents the time dependent perturbational motion of the blade.
Since [Ml, [Cl, WI, {F}, and {ys} are all dependent on the design variables Dj, these matrices
are influenced by a small incremental change in these variables from their baseline values. Using
a first order Taylor series approximation, the modified matrices, due to incremental changes
in AD,, can be written as
A New Sensitivity Analysis

{YO) N {YO}D, + $$ADjv


3
and

{Y} 2 {l/o}D, + !&fADj + {&At))-

Using equation (7)) one has

{ti) = {Ati),
LO)= (Ati),
a(Y) a{Yol
-8Dj = aDj (9)

Since the quantity {Ay(t)} d oes not depend on the design variables, its variation with respect
to Dj is zero.
To obtain the nonlinear algebraic equation (representing the equilibrium conditions) and the
perturbational (stability) equation, the vector {F} has to be expanded as a Taylor series about the
nonlinear equilibrium position corresponding to the baseline design configuration. The truncated
series can be written as

where the notation { }IB implies that the function or its derivatives are evaluated at y = ys; p = 0;
5 = 0; Dj representing the equilibrium position corresponding to the baseline configuration.
Substituting equations (8) and (10) into equation (6), and neglecting higher order terms in ADj,
yields three equations, which are obtained after collecting terms in an appropriate manner and
setting them equal to zero. These three equations can be identified, respectively, as
(a) an equilibrium equation corresponding to baseline design;
(b) an algebraic equation which represents the variation in equilibrium position due to changes
in the design variable; and
(c) a perturbational equation representing the stability of the system.
Detailed expressions for these equations are given below.

(1) Baseline Equilibrium Equation (constant term)

[~l~jbLlj + WI. =a (11)


(2) Sensitivity Equation for the Equilibrium state (coeficients of AD, only)

Rearranging the terms of equation (12) and taking inverse, the sensitivity derivative of the
equilibrium position can be written as

aiYo)
-=-
8D.j

MCM 19-3/4-B
C. VENKATESAN
et al.

(3) Perturbation Equation (terms associated with the perturbation quantity (Ay})

[m + MaADj](Aji} + [C + C,AD,]{Ay} + [I?: + K,ADJ{Ay} = 0, (14)

where

m=pqDj+~,
c= [c], +g,
k= [Klo,+$g,
M a=
1~+&[$!g]+~&[~]~}=~7

In equation (14), the matrices l&Z, C, K represent the linearized mass, damping, and stiffness
matrices corresponding to the baseline configuration. The matrices M,, C,, K, indicate the
variation in the mass, damping, and the stiffness due to small changes in the design variable Dj.
In equation (15), all derivatives are evaluated about the equilibrium position, (y = yc; y = 0;
ji = 0), corresponding to the baseline design. The terms inside the summation represent the
effect of the modifications in the equilibrium position of the blade due to the change in design
variable, (indirect effect).
Rewriting equation (14) in first order form yields

where

and

[A211= - [ii?f + M,ADj] -[iiT + K,ADj]

[A221= - [ti + M,ADj] -[c + C,AQ].

When the increment AD, is very small, one can introduce the following approximation:

[.M + M.ADj]- N [iG]- - M-lM,&l-lADj. (17)

Substituting equation (17) in the [A] matrix, and deleting higher order terms in AD,, yields the
modified system matrix [A] due to an incremental change in the design variable ADj,

[A] N [A]Dj + gADj>


3
(18)
A New SensitivityAnalysis 7

dA
where [AID, corresponds to the baseline design configuration and # represents the variation
of the system matrix [A] due to a change in the design variable Dj. These matrices are given by

0
44 _ &-~,u-lk
0
_ fi-K, M-M&f-C _ a-c, (20)
dDj .
I
The eigenvalue problem corresponding to the baseline design can be written as

(2) = [A]Dj{Zl* (21)


The eigenvalue & and the corresponding eigenvector { u}~ in the kth mode satisfy the equation

(22)
The superscript s indicates that the eigenvector is associated with the aeroelastic stability analy-
sis, and thus, it represents an aeroelastic mode.
Since the system matrix [A] D. is a real nonsymmetric matrix, the adjoint eigenvector {rY}E
corresponding to the eigenvalue x k satisfies the equation

.&{v}; = {us};
[A],j.

Premultiplying equation (22) by {v}: and differentiating with respect to the jth design vari-
able Dj yields

where # represents variation of [A]oj with respect to Dj, and it is given in equation (20).
Recognizing from equations (22) and (23) that the terms inside the square brackets are identically
equal to zero, the sensitivity of eigenvalue XI, with respect to the design variable Dj can be written
as

(24)

In the aeroelastic stability analysis, the eigenvalue xk is a complex quantity, and the corre-
sponding modes { uB} k and {v} k are complex modes. Therefore, equation (24) can be separated
into real and imaginary parts with the real part representing the sensitivity of damping and the
imaginary part representing the sensitivity of frequency, thus

d& _Sk*+_.bk (25)


dDj - dDj dDj

2.3. Sensitivity of Response

The sensitivity of the blade response with respect to the design variable is required for the
evaluation of the sensivity of the hub loads. The equations of motion of the rotor blade in
forward flight are nonlinear and have periodic coefficients which can be written as [8]

WI tji) + [Cl {ti) + WI(~1+ @I = 0. (26)


8 C. VENKATESAN et al.

In this equation, the mass [Ml, the damping [Cl, and the stiffness [K] matrices are linear and
periodic. The forcing function {p} contains all the nonlinear terms and the aerodynamic loading.
The vector {y} represents the generalized coordinates associated with the various blade modes,
employed in the aeroelastic stability and response calculation. Assuming a small incremental
change AD, in the baseline value of the design variable Dj, the various terms in equation (26)
can be expanded in a Taylor series about the baseline. Neglecting higher order terms in ADj
and collecting terms which contain AD, as well as those which do not contain ADj produces
two equations. These equations can be interpreted, respectively, as:
(a) equation of motion for blade response corresponding to the baseline configuration; and
(b) the equation providing the sensitivity of the response about the baseline configuration.
The final equations are presented below.

(1) Response Equation for Baseline Configuration (terms without ADj)

In equation (27), the mass, damping, and stiffness matrices are linear and periodic, while the
forcing function {F} Dj is a nonlinear function of the blade response { Y}~, , { G}Dj, { ji},j,
and ?,!J. The periodic solution of equation (27) can be obtained using a quasi-linearization
technique [12,13], or alternatively, it can be solved by harmonic balancing.

(2) Equation for Response Sensitivity (terms with AD,)

I -ifI ,
The derivatives of the forcing function {F} with respect to the generalized coordinates, such as
@ ,;; and @ are all available from the final iteration of the quasi-linearization

proced&e. Equftion (28) is aBlinear differential equation with periodic coefficients. The steady
state solution w, representing the sensitivity of the blade response with respect to the design
variable Dj, can be obtained using the transition matrix or by numerical integration or by
harmonic balancing.

2.4. Sensitivity of Hub Loads

The hub loads in the nonrotating frame are obtained after summing up the hub loads associated
with the individual blades in the rotating frame. Let FER, FkR, FkR and MiR, MiR, MER
represent the hub shears and the hub moments due to the kth blade, in the rotating frame,
respectively. The hub loads in the nonrotating frame can be written as

(29)

and

(30)
A New Sensitivity Analysis 9

where
tile= $1+ 23 - 11, Ic=l,...,N,

and N is the number of blades in the rotor. The quantities FH=, FH~, FH= and MH~, MH~,
MH= represent the hub shears and hub moments, respectively, in the nonrotating frame. Equa-
tions (29) and (30) can be written symbolically as

tFH) = 5 LT]jF;] (31)


k=l

and

IH) = 5 LT]{F% (32)


k=l

where the transformation matrix [T] is given by

1
cos $k -Sin+k 0

PI = Sin$Jk cos?),k 0 .
[ 0 0 1

The blade loads {Fk}, {ML} are functions of the design variable Dj and the steady state
response of the blade {y}, {g}, {ji}. Taking the derivatives of {FH}, {MH} with respect to the
design variable Dj yields

N
d(FH)
-=
dDj C[
k=l
T

I{Wid + Wk,l a{~) + WiJ


aDj b(y)dDj mao,
a@) + W%l a{U)
mao,
(33)

and

N
d{MH)
-=
dDj C[
k=l
T

I(a@& + W&
8Dj a(y)%
a{~) + aW;l a{ti) + +%) a{ji)
Tao,
(34

The quantities w, w represent the sensitivities of hub shears and hub moments with
respect to the design variable Dj. In equations (33) and (34), the known quantities are the
transformation matrix [!I] and the response sensitivity $&, s, @, from the solution of
equation (28). The unknown quantities are the partial derivatives of the hub loads for a particular
blade in the rotating frame, given by

(35)

and

(36)

Since the hub loads of the individual blade {Fk} and {Mk} are evaluated numerically when
using an implicit formulation, the eight partial derivatives given by equations (35) and (36) have
to be calculated by a finite difference scheme.

3. RESULTS AND DISCUSSION


Using the semi-analytical procedure developed in this study, sensitivity derivatives of the nat-
ural frequencies and the aeroelastic stability characteristics have been evaluated for composite
rotor blades. To validate the mathematical formulation, the sensitivity derivatives evaluated by
the semi-analytical approach are compared with those obtained by a finite difference formulation
10 C. VENKATESANet al.

employing forward differencing. The structural dynamic and aeroelastic calculations required for
the finite difference scheme, as well as those required for generating various derivatives for the
semi-analytical approach, are based on a new aeroelastic model developed for a composite rotor
blade [9,10].
The sensitivity analysis was performed for two rotor configurations having uniform properties
along the span. The cross-section of a composite rotor blade having single-cell box beam con-
figuration is shown in Figure 1. The properties of this rotor configuration are taken from 19,101.
The cross-section of the beam has an outside dimension of 7 width by 2 height with a uniform
thickness of 0.35. The basic parameters of the composite rotor blade and the material properties
are given in Table 1.

Figure 1. Single-cell laminated rectangular box beam.

For sensitivity analysis, two design variables have been chosen for the composite rotor blade.
They are the composite fiber orientation angles A, and Ah. The design variable AV represents the
fiber orientation in the inner half thickness of the vertical wall, while the outer half thickness of
the vertical wall and the horizontal walls have zero ply angles. Similarly, the design variable Ah
represents the fiber orientation in the inner half thickness of the horizontal walls, while the outer
half thickness of the horizontal wall and the vertical walls have zero ply angles. A positive A,
implies that the fibers are oriented toward the top wall of the blade, and for positive Ah, the
fibers are oriented toward the leading edge of the blade.
A New Sensitivity Analysis 11

Table 1. Baseline configuration for single-cell composite rotor blade.

Fundamental, coupled rotating natural frequen-


cies for a straight blade with zero ply angles:
WLr = 1.533
WF~ = 1.187
war = 5.186

7 = 5.0 c/R = 0.08


0 =O.l & = 0.0
k,l/R = 0.0 a = 5.7
k&R = 0.01609 cd,, = 0.01
N=4 cw = 0.005
Q = 360 rpm R = 255.45
Offsets of center of mass, aerodynamic center,
and tension center from elastic axis are zero.
Material Constants:
EL = 30. x lo6 psi
ET = 3. x lo6 psi
GLT = 1.2 x lo6 psi
VLT = 0.3

The results of the sensitivity analysis are presented in three sections:


(i) sensitivity of natural frequencies of the composite rotor blade with respect to ply orien-
tation,
(ii) sensitivity of the aeroelastic stability in hover, and
(iii) sensitivity of vibratory hub loads in forward flight.

3.1. Sensitivity of Natural Frequencies of Composite Blade

The natural frequencies and mode shapes of a rotating composite blade were calculated using
a beam finite element formulation. The rotor blade was idealized with five beam elements. The
nodal degrees of freedom for the beam finite element are shown in Figure 2. The natural frequency
sensitivity is calculated for the first six modes, corresponding to three flap, two lead-lag, and one
torsional mode of the rotor blade. In the case of highly coupled modes, the designation of the
modes as flap, lag, or torsion is for purposes of identification only. The sensitivity derivatives of
the natural frequencies with respect to the design variables A, and Ah are evaluated about two
reference (or baseline) ply orientations of 0 degrees and 20 degrees. A reference ply orientation
of 0 degrees implies that for the baseline configuration, both the vertical and the horizontal walls
have 0 degree ply orientation and the sensitivity derivatives are computed about this configuration
of the blade. A reference ply orientation of 20 degrees implies that the baseline configuration of
the blade has a ply orientation set at +20 degrees in the inner half of either the vertical walls or
the horizontal walls; this second alternative represents a more realistic configuration.
To determine the influence of step size on the sensitivity derivatives of the natural frequencies,
a parametric study was conducted. The step size of the design variables A,, and Ah was varied in
the range l.OE-03 to 2.0 degrees. The sensitivity of natural frequencies was obtained by finite
differences and the results are shown in Figures 3 through 7. The sensitivity derivatives evaluated
about a baseline ply orientation of 0 degrees are shown in Figures 3-5. Figure 3 represents the
sensitivity derivatives of the natural frequencies of the rotating blade with respect to design
variable A,. The results indicate that the sensitivity derivatives are greatly influenced by the
step size, and the derivatives increase monotonically with respect to the step size. At low values of
the step size (below l.OE-02), the sensitivity derivatives have very small magnitude which appear
12 C. VENKATESANet al.

93
3
03
7x93
fxf3

II- I, -
I
Figure 2. Finite element nodal degrees of freedom.

- 1st flap -w 2nd flap -+I* 3rd Flap

-0.0040-

-0.0060-
.
-0.0060- t
.
-.
.
-0.0100- :
i
:
-0.01 zo- b
:
-0.0140- :
S
:
-0.0160 S
I I I
1E-02 lE-01 lE+OO

STEP SIZE ( DEG. )

- 1St Lag -c 2nd Lag e** 1 rt Torsion

0.2007

0.150-
/
O.lOO- i
.r

-0.150- \
t
-o.zOO-
1
-0.250 I I I
1E-02 (E-01 lE+OO

STEP SIZE I DEG. )


Figure 3. Influence of step size on the sensitivity of natural frequencies of a singltxell composite
rotor blade. Design variable A,, baseline ply orientation 0 degrees.
A New Sensitivity Analysis 13

as zero in these figures. To illustrate the exact magnitude of the sensitivity derivatives, the results
corresponding to the three flap modes, shown in Figure 3, are presented in a logarithmic scale in
Figure 4. In the log-scale, the sensitivity derivatives exhibit a linear variation with respect to the
step size over the complete range. For low values of the step size, the magnitude of the sensitivity
derivatives are of the order of l.OE-05 to l.OE-06, indicating that for all practical purposes
one can assume these values to be zero. A similar observation was also made for the sensitivity
derivatives corresponding to lag and torsion modes, which are not shown here for brevity.

* 1st Flap m 2ndflap .=a3rd Flap

1E-02 lE-01 lE+OO

STEP SIZE ( DEG. )


Figure 4. Influence of step size on the sensitivity of natural frequencies of a single-cell composite
rotor blade. Design variable A,, baseline ply orientation 0 degrees.

Figure 5 shows the sensitivity derivatives of the natural frequencies with respect to the design
variable Ah. In this case too, the sensitivity derivatives exhibit a monotonic increase with respect
to the step size. At low values of the step size (less than l.OE-02), the sensitivity derivatives
have very small magnitude which can be assumed to be equal to zero.
The influence of step size on the sensitivity of natural frequencies, for a configuration having
a ply orientation of 20 degrees, is presented in Figures 6 and 7. Figure 6 shows the variation of
sensitivity derivatives of natural frequencies with respect to the design variable A,. The baseline
ply orientation is set at A, = 20 degrees. In this case, the sensitivity derivatives in all modes have
finite magnitudes which remain almost constant over the complete range of step size variation,
indicating that the frequency sensitivity derivatives are not influenced by the change in step size
in the range l.OE-03 to 1.0 degree. A similar observation can also be made for the case when
the design variable is Ah, as shown in Figure 7.
A comparison of the magnitudes of the sensitivity derivatives shown in Figures 6 and 7 clearly
indicates that the flap and torsion frequencies are more sensitive to variations in the ply orienta-
tion in the horizontal wall (Ah) than the variations in the vertical wall (A,,). The lag frequencies
are equally sensitive to variations in ply orientations in both horizontal and vertical walls. Fur-
thermore, the results depicted in Figures 6 and 7 indicate that higher modes in flap are more
sensitive to variations in ply orientation than the lower modes. Similarly, the second lag mode
frequency is more sensitive to variations in ply orientations than the first lag mode frequency.
Comparisons of the sensitivity derivatives for the natural frequencies obtained by finite differ-
ence scheme and the semi-analytical approach using equation (5) are shown in Figures 8-11. The
study was carried out for those cases where the baseline ply orientation wss set at 20 degrees. The
quantities in parantheses indicate the multiplication factor for the respective sensitivity deriva-
tives. The sensitivity derivatives of natural frequencies with respect to the design variable A,,
14 C. VENKATESANet al.

are evaluated about the configuration with a reference ply orientation of A, = 20 degrees, while
the derivatives with respect to Ah are calculated about the configuration with a ply orientation
of Ah = 20 degrees. The sensitivity derivatives are formulated using two different step sizes,
namely 0.2 degree and 1.0 degree. Figures 8 and 9 show the comparison of sensitivity deriva-
tives obtained by finite difference and the semi-analytical approaches for the design variables A,
and Ah, respectively, employing a step size of 0.2 degree. From Figure 8, it can be seen that
both finite difference and semi-analytical approaches provide the derivatives which are in excel-
lent agreement for all flap, lag, and torsional frequencies. A similar correlation can also be seen
for the sensitivity derivatives with respect to Ah, shown in Figure 9. The maximum percentage
difference in all the cases shown in Figures 8 and 9 is less than 1%.
W- 1st Flap-* 2nd flap -*a 3rd flap

I**

-O.lO- 1 ?
*.
i.
ao -0.20-
i
.-.
ahh .
-0.30- :
II
:
*
:
-0.40- :
:
i

-0.50 I I I

1E-02 lE-01 lE+OO

STEP SIZE [ DEG. )

* 1st Lag * 2nd Lag +V 1st TO&

0.60-I
I
o.so- .i:
i
:
OAO- F
aa .i
0.30-
.i
.12
0.20-
J
J.
O.lO- ...*
. . . . . . . . . . . . *
&....... ...*-
O.OO~I

1E-02 lE:Ol 1EkXl

STEP SIZE ( DEG. )


Figure 5. Influence of step size on the sensitivity of natural frequencies of a single-cell composite
rotor blade. Design variable Ah, baseline ply orientation 0 degrees.

Figures 10 and 11 present the sensitivity derivatives obtained for the case having a reference
ply orientation of 20 degrees and employing a step size of 1.0 degree. The sensitivity derivatives
obtained by finite difference and semi-analytical approaches are in excellent agreement for all the
frequencies.
A New Sensitivity Analysis 15

w 18t Flap * 2nd flap + 3rd flap

O.OO-) m

-0.1 o-
,-----c-L----L--m-c-

aa -0.20-

ab
-0.30-

-0.40-
,I...........................~...........................-............~.......~..~

-0.50 I I I
1E-02 lE-01 lE+OO

STEP SIZE ( DEG. 1

I- 1st Lag b*Znd Log 1.~. 1atTarsianI

----c----_)---- #

-2.00 I I I
1 E-02 lE-01 lE+00

STEP SIZE t DEG. 1


Figure 6. Influence of step size on the sensitivity of natural frequencies of a single-cell composite
rotor blade. Design variable A,, baseline ply orientation 20 degrees.

The results shown for the single-cell composite rotor blade configuration can be summarized
as follows:
(4 when the baseline ply orientation is 0 degrees, the sensitivity derivatives of natural frequen-
cies are greatly influenced by the variation in step size of the design variable, whereas for
those cases having a baseline ply orientation of 20 degrees, the sensitivity derivatives are
not influenced by the variation in step size (in the range l.OE-03 to 1.0 degree);

(b) flap and torsional frequencies are more sensitive to ply orientations in the horizontal wall
than to variations in the vertical wall, whereas lag frequencies are more or less equally
sensitive to variations in ply orientations in both vertical and horizontal walls; and

(cl the higher modes in flap and lag are more sensitive to variations in ply orientations than
lower modes.
16 C. VENKATESAN et al.

-2.50 I I I
1E-02 lE-01 lE+OO

STEP SIZE ( DEG. )

-1st Lag -2nd Lag +lstTomion

lE-02 lE-01 lE+OO

STEP SIZE t DEG. )


Figure 7. Influence of step size on the sensitivity of natural frequencies of a single-cell composite
rotor blade. Design variable Ah, baseline ply orientation 20 degrees.

3.2. Sensitivity of Aeroelastic Stability in Hover

It should be noted, see [1,2], that maintainingaeroelastic stability margins in hover is one of
the more important constraints encountered in the optimization process. The sensitivity of the
aeroelastic characteristics was analyzed for a single-cell composite rotor blade configuration. The
aeroelastic stability calculations were performed using three flap, two lag, one torsional, and one
axial mode. The rotor is operating with a thrust coefficient of CT = 0.005. The sensitivity results
axe presented in two sections:
(1) sensitivity of the equilibrium position, and
(2) sensitivity of the modal damping and modal frequency representing the aeroelastic stability
characteristics of the rotor blade in hover.
A New Sensitivity Analysis 17

OAO-

aa OJO-

al\, 0.20-

1st (-k-01) 2nd (-iE-00) Zkd (-lE-00)

flap modes

I I I I
l.l(-1GOo) 2nd (-lE-00) 1st (lE-00)

lag modes torsion mode

Figure 8. Sensitivity derivative of natural frequencies with respect to A,, , for a single-cell composite
rotor blade. Baseline ply orientation of A, = 20 degrees and increment of 0.2 degree.

3.2.1. Equilibrium position sensitivity in terms of the generalized coordinates

The equilibrium sensitivity for the generalized coordinates associated with seven modes used
in the aeroelastic analysis were obtained from equation (13) and the results are compared with
those from pure finite difference approach. The results presented in Figure 12 show only the
sensitivity for the flap, lag, and torsional modes. The axial mode is not an important mode for
the aeroelastic stability, and therefore, results for this mode are not shown.
Figure 12 depicts the comparison of the sensitivity of the equilibrium position for the case with
ply orientation A, 8s the design variable. Poor correlation between the derivatives obtained by
the two approaches is evident. The results also indicate that the derivatives are of the order of
l.OE-03, which is very small. A possible reason for the poor correlation can be attributed to the
assumptions made in the formulation of the derivatives using the semi-analytical approach. It is
shown in [15] that when design variable was chosen as the cross-sectional flap bending stiffness,
the sensitivity derivatives of the equilibrium position have a magnitude of the order l.OE-01,
which is two orders of magnitude greater than that shown in Figure 12.
It should be noted that the results showing the sensitivity of equilibrium position are presented
for illustrative purposes only. The equilibrium position of the rotor blade in hover is not used as
a constraint in the structural optimization of the rotor blade. A much more important parameter
is the sensitivity of the aeroelastic stability characteristics of the rotor blade, which is presented
next.
18 C. VENKATESANet al.

at0 l*w-
alZh l.oa-

I ,
lmt (-lE-01) 2nd (-lE-00) 3rd (-(E-W)

flap modes

3.00-

2.50-

2.00-

at0
ahh .w-
l.oo-

OSO-

O.OU- I I ,

1~4(-lE-00) 2nd (-lE-00) lmt (lE-00)

lag mode8 torsion mode


Figure 9. Sensitivity derivative of natural frequencies with respect to Ah, for a single-cell composite
rotor blade. Baseline ply orientation of l\h = 20 degrees and increment of 0.2 degree.

3.2.2. Sensitivity of modal damping and frequency

The computation of the sensitivity of the blade stability in hover, by the semi-analytical ap
preach, requires the formulation of linearized mass, damping, and stiffness matrices together with
their derivatives with respect to the design variable. Equation (15) implies that in the formu-
lation of these matrices, the second derivatives of the forcing function {F} with respect to the
design variable Dj and the generalized coordinate (y} and its time derivatives are also needed.
The total number of function evaluations required for the formulation of the various derivatives
of {F}, given in equation (15), can be expressed by the relation

(m-1)
Nit = m(3m + l)(iVD + 1) + 3m3 - mZI,

where ND is the number of design variables and m is the number of modes considered in aeroelas-
tic analysis. For seven modes and two design variables, the total number of function evaluations
required is 1344. Performing such a large number of iterations, in order to evaluate the matrices
li%, C, k, M,, C,, K,, is inefficient. To reduce the number of computations substantially, all
the second derivative terms in equation (15) were neglected, and thus only the first derivative
terms are retained. With this simplification, the sensitivity derivatives for the aeroelastic stability
of the composite blade were rederived, using equation (24).
The sensitivity of modal damping and frequency with respect to the ply orientation A, of
the composite blade is shown in Figure 13. The sensitivity derivatives of the modal damping
and frequency are presented for the fundamental modes in lag, flap, and torsion. The semi-
analytical procedure, employing adjoint eigenvectors, produces derivatives which are in fairly
A New Sensitivity Analysis 19

2nd (-lE-CO)

flag modes

1.60-

1.60-
1.40-

1.20-

lst(-iE_aO) Znd(-iE-60) ls! (IE-00)

lag modes torsion mode


Figure 10. Sensitivity derivative of natural frequencies with respect to A,,, for a single-cell com-
posite rotor blade. Baseline ply orientation of AV = 20 degrees and increment of 1.0 degree.

good agreement with those evaluated by the finite difference method. The sensitivity of modal
damping in lag and flap modes indicate that the semi-analytical procedure overestimates the
derivatives by about 30% when compared to the values obtained by the finite difference scheme.
The reason for the difference in the derivatives can be attributed to the assumptions made in the
semi-analytical approach. In addition, it should be noted that when formulating the sensitivity
derivatives by the semi-analytical procedure, all the second derivative terms were neglected in
equation (15). It is worth mentioning that the quality of the correlation shown in Figure 13
is much better than that presented in [15]. In the present case, the semi-analytical procedure
employs the eigenvectors of the adjoint problem, whereas in [15], the semi-analytical procedure
was based only on the eigenvectors of the original problem, and thus required the assumption of
invariance of the eigenvector with respect to the design variables.

3.3. Sensitivity of Hub Loads in Forward Flight

Results illustrating the sensitivity of hub shears and moments in forward flight for a soft-in-
plane composite blade having a two-cell type cross-section are presented next. Figure 14 shows
the two-dimensional finite element model employed for the composite cross-section analysis from
which the cross-sectional properties were obtained. The leading edge has a semi-circular shape
with a radius of 1.2, and the straight portion has a total length of 6. The middle wall is
located 2.8 behind the leading edge semi-circle. All of the walls have a thickness of 0.1. The
baseline configuration for this blade is shown in Table 2, where the material constants correspond
to glass/epoxy type composite material. Hub shears and moments are calculated for a straight
20 C. VENKATESAN et al.

ltt(-iE-01) Znd(-(E-00) Srd(-lE-00)

flap modes

3.00-.

2.50-

0.00 '1
lst(-lE-00) 2nd (-lE-00) 1st (l&00)

lag modes torsion mode

Figure 11. Sensitivity derivative of natural frequencies with respect to I\h, for a singhzxell com-
posite rotor blade. Baseline ply orientation of I$, = 20 degrees and increment of 1.0 degree.

Table 2. Baseline configuration for two-cell composite rotor blade.

Fundamental, coupled rotating natural frequen-


cies for a straight blade with zero ply angles:
wL1 = 0.765
wFr = 1.096
wTr = 3.356

7 = 5.0 c/R = 0.06


c7 =O.l p* = 0.0
k,,lR = 0.004 a = 5.7
kmzlR = 0.02439 c&J = 0.01
N=4 cw = 0.005
R = 360 rpm R = 250

Material Constants:
EL = 6.2 x 10 psi
ET = 1.6 x lo6 psi
GLT = 0.8 x 10 psi
VLT = 0.25
A New Sensitivity Analysis 21

-o.wzw

-0.co250 j

flap modes

0.0060-

0.0050-

O.C040-

0.0030-

aA, o.oo20-

0.0010-

o.oooo-

-0.0010 J 1st 2nd 1 st

lag modes torsion mode


Figure 12. Sensitivity derivatives of equilibrium values of the generalized coordinates with respect
to Au, for a single-cell composite rotor blade. Baseline ply orientation A,, = 20 degrees and
increment = 1.0 degree.

blade with ply angle variation in either the horizontal walls or the vertical walls at an advance
ratio /J = 0.3. The thrust coefficient CT is maintained at a constant value of 0.005.

Two cases were considered. In the first case, the laminate in the inner half of the horizontal
walls is oriented at ply angle Ah, while the remaining walls have zero ply angles; in the second case,
the laminate in the middle vertical wall and the inner half of the rear vertical wall is oriented at
ply angle A,,, while the remaining walls have zero ply angles. The ply angles Ah and A,, are varied
from 0 to 90 in both positive and negative directions. The baseline configuration corresponds
to the case where Ah and A,, are equal to zero.

Figure 15 illustrates the sensitivity of the 4/rev vibratory hub shears and moments, respectively,
as a function of the ply angle Ah. Figure 15 shows that there is a 10% to 15% variation in
the vibratory hub shears compared to the baseline values. The variation in the vibratory hub
moments is up to 30%, consisting mostly of reduction in roll and pitch moments and increase in
yaw moment.

Figure 16 illustrates the sensitivity of the 4 /rev vibratory hub shears and moments, respectively,
as a function of the ply angle A,,. Figure 16 shows that the variation in longitudinal and lateral
shears is about 10% to 15%, from the baseline configuration, while the vertical shear is not
sensitive to the variation in A,. The variation in the roll and pitch moments is within lo%, while
the maximum variation in the yaw moment is about 25%.
MCM
19-3/4-c
22 C. VENKATESANet al.

0.040-l

0.030-

0.020-

0.01 o-
torsion
o.ooo--

-O.OlO-

-0.020-

-0.030-

-0.04ooJ
(a) Modal damping sensitivity.

0.601

0.50

0.40

0.30

a(0 0.20 i

al\, O.lO-
lag
I
torsion

(b) Modal frequency sensitivity.

Figure 13. Aeroelsstic stability sensitivity derivatives with respect to A,, , for a single-cell composite
rotor blade. Baseline ply orientation A, = 20 degrees and increment = 1.0 degree.

3.4. Computational Requirements

The computational times cited are based on the IBM ES/9000 Model 900 system. For the
evaluation of natural frequency sensitivity derivatives with respect to one design variable, the
semi-analytial formulation requires a CPU time of 2.35 set, whereas the finite difference approach
requires a time of 2.82 sec. For the evaluation of stability sensitivity derivatives with respect to
one design variable, the finite difference procedure requires a computational time of 6.42 set,
while the semi-analytical approach requires 3.82 sec.

4. CONCLUDING REMARKS
This paper describes the mathematical formulation of the sensitivity derivatives needed for
the structural optimization of a composite rotor blade with structural dynamic and aeroelastic
constraints. Using the approach presented in this paper, sensitivity derivatives for the structural
dynamic and aeroelastic stability characteristics of a composite rotor blade were formulated. The
mathematical formulation was validated by comparing the sensitivity results with those obtained
from a pure finite difference approach. The most important observations are summarized as
follows.
A New Sensitivity Analysis 23

Figure 14. Two-cell composite box beam configuration.

*.a. LONGllUDlNAl -. MEltAL - ERTIUL

q , , . , * , . , . , . ,

-90 -60 -30 60 90


PLY ORIENTATiN (DEGEES)

. . . . ROLL -. PITCH -YAW

r3-
E

B
22
I
> _
e
\
Wl-
i
B .
B
=0 I I . I , . I * 1
-90 -60 -30 0 30 60 90
PLY ORIENTATION (DEGREES)

Figure 15. Variation of 4/rev vibratory hub shears and moments with respect to ply orientation Ah
for a two-cell composite rotor blade.
24 C. VENKATESANet al.

.... LONCllUMNAL~--LATERALI--MRTIm

........*.
..-* *.........._
& .... . t...*e-
:, . - - .. _ _ ~ J....... _. . ..-............. ,--_:l

-mm-- C--
-w-c -.-*

6-

0 1 8. I * I * I n I
-90 -60 -30 60 90
PLY ORIENT*TPON (O&ES)

.... ROLL - . PITCH -YAW

B,i, , . , . , . , , , . ,
-90 -60 -30 60 90
PLY ORlENT&N (O&ES)

Figure 16. Variation of I/rev vibratory hub shears and moments with respect to ply orientation A,,
for a twecell composite rotor blade.

(1) Natural frequencies in flap and torsional modes of the rotor blade are more sensitive to
ply orientations in the horizontal wall than to variations in the vertical wall, whereas lag
frequencies are more or less equally sensitive to variations in ply orientations in both vertical
and horizontal walls.
(2) The natural frequencies of the higher modes in flap and lag are more sensitive to variations
in composite ply orientations than the natural frequencies corresponding to the lower modes.
(3) In the calculation of stability sensitivity of the rotor blade in hover, the semi-analytical
method requires the formulation of certain second derivatives whose evaluation is compu-
tationally inefficient compared to a pure finite difference approach.
(4) In the formulation of stability sensitivities, the semi-analytical approach employing the
eigenvectors of the adjoint problem provides derivatives which are in good agremeent with
those evaluated by finite difference scheme.
(5) In forward flight, the vibratory hub loads exhibit a 10% N 30% variation, due to changes
in the composite ply orientation angle.
A New Sensitivity Analysis 25

REFERENCES
1. P.P. Friedmann, Helicopter vibration reduction using structural optimization with aeroelastic/multidiscipli-
nary constraints-A survey, AIAA Journal of Aircmft 28 (l), 8-21 (January 1991).
2. P.P. Friedmann, Impact of structural optimization with aeroelastic/multidisciplinary constraints on helicopter
rotor design, Aerospace Design Conference, Irvine, CA, February 3-6, 1992, AIAA Paper No. 92-1001.
3. H.M. Adelman and W.R. Mantay, Integrated multidisciplinary optimization of rotorcraft: A plan for devel-
opment, NASA TM 101617 (May 1989).
4. J.W. Lim and I. Chopra, Dkgn sensitivity analysis for an aeroelsstic optimization of a helicopter blade,
AIAA Paper No. 87-0923, Proceedings 2ph AIAA Dynamics Specialists Conference, Monterey, CA, April
1987, pp. 1093-1102.
5. J.W. Lim and I. Chopra, Stabiltij sensitivity analysis for the aeroelastic optimization of a helicopter rotor,
Proceedings AIAA/ASME/ASCE/AHS 2gth Structures, Structural Dynamics and Materials Conference,
Williamsburg, VA, April 1988, Paper No. 88-2310, pp. 813-824.
6. J.W. Lim and I. Chopra, Aeroelsstic optimization of a helicopter rotor, Proceedings 44th Annual Forum of
the American Helicopter Society, Washington, DC, June 1988, pp. 545-558.
7. J.S. Sobieski, Sensitivity of complex internally coupled systems, AIAA Journal 28 (l), 153-160 (January
1990).
8. R. Celi and P.P. Friedmann, Rotor blade aeroelasticity in forward flight with an implicit aerodynamic for-
mulation, AIAA Journal 26 (12), 1425-1433 (December 1988).
9. K. Yuan, P.P. Priedmann and C. Venkatesan, A new aeroelastic model for composite rotor blades with
straight and swept tips, Proceedings of the 33Pd AIAA/ASME/AHS/ASC Structures, Structuml Dynamics
and Materials Conference, Dallas, TX, April 13-15, 1992, AIAA Paper No. 92-2259, pp. 1371-1390.
10. K. Yuan, P.P. Priedmann and C. Venkatesan, Aeroelastic behavior of composite rotor blades with swept tips,
Proceedings of the 48th Annual Forum of the AHS, Washington, DC, June 3-5, 1992, Vol. 2, pp. 1039-1059.
11. R.T. Haftka and Z. Gurdal, Elements of Structural Optimization, 3 rd Edition, Kluwer Academic Publishers,
(1992).
12. SM. Roberts and J.S. Shipman, Two-Point Boundary Value Problems, American Elsevier, (1972).
13. P.P. Friedmann and S.B.R. Kottapalli, Coupled flap-lag-torsional dynamics of a hingeless rotor blade in
forward flight, Journal of the American Helicopter Society 27 (4), 28-36 (October 1982).
14. H.L. Thomas, A.E. Sepulveda and L.A. Schmit, Improved approximations for control augmented structural
synthesis, AIAA Journal 30 (11, 17-179 (January 1992).
15. C. Venkatesan, P.P. Priedmann and K. Yuan, A new sensitivity analysis for structural optimization of compos-
ite rotor blades, Proceedings of the 34th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics
and Materials Conference, La Jolla, CA, April 19-22, 1993, AIAA Paper 93-1644, pp. 2952-2973.

Vous aimerez peut-être aussi