47th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Confere AIAA 2006-2025

1 - 4 May 2006, Newport, Rhode Island

UNIFIED RATIONAL FUNCTION APPROXIMATION

FORMULATION FOR AEROELASTIC AND FLIGHT
DYNAMICS ANALYSES

Dario H. Baldelli1 and P.C. Chen2

Zona Technology, Inc., Scottsdale, Arizona, 85258

Jose Panza3 and Joshua Adams4

General Atomics – Aeronautical Systems, San Diego, California, 92127

This paper presents a unified rational function approximation (RFA) formulation to take
into account the influence of aeroelastic effects on the flight dynamic behavior of the whole
aircraft. Standard production flutter engineering models that incorporate translational and
rotational rigid-body degrees of freedom are usually not able to recover the traditional flight
dynamic equations of motion due to several inherent conflicting modeling objectives. This
paper identifies key elements within the RFA to update the unsteady aerodynamic model. By
allowing the inclusion of gravity-related terms, vertical acceleration related stability
derivatives, as well as lift and drag forces due to forward velocity perturbations into the
RFA matrices, it is shown by closed-form solutions that the usual flight dynamic equations of
motion are fully recovered. In this way, open and closed-loop simulations of the
aeroelastic/aeroservoelastic response of the whole aircraft would be available to the flight
control engineer to design and validate dependable flight control systems. Its
implementation will naturally extend the RFA setup into a flight dynamic oriented class. To
illustrate the modeling framework, the proposed technique is demonstrated using two
examples: i) Navion, and ii) Predator B “Hunter Killer” aircrafts.

Nomenclature
[Ai] = aerodynamic force matrix
[C] = structural damping matrix
g = gravity acceleration
[K] = structural stiffness matrix
k = reduced frequency, ωb/V∞
B = aircraft’s wing span
[M] = mass structural matrix
{P} = vector of gravity related components
p = non-dimensional Laplace variable, sL/V∞
[Q] = generalized aerodynamic matrix force
[R] = diagonal aerodynamic root matrix
s = Laplace variable
[TA] = body to stability axes transformation matrix
V∞ = aircraft airspeed
{δc} = vector of control surface deflections
{ξ} = vector of generalized structural displacements
ω = frequency of oscillation

1
Control Engineering Specialist. Member AIAA.
2
Vice President. Member AIAA.
3
Structural Analysis Group Manager, Member AIAA.
4
Structural Engineer.

1
Copyright © 2006 by Dario H. Baldelli. Published by the American Institute of Aeronautics and Astronautics, Inc., with permission.
Subscripts
A = related to aerodynamic lag
ae = Related to aeroelastic plant
C = related to control commands
i = 0,1,2 related to quasisteady model approximation
long = related to longitudinal dynamics plane
lat = related to lateral dynamics plane
r = related to rigid-body states
s = related to structural deflections

I. Introduction

T HIS paper presents a unified rational function approximation (RFA) formulation to incorporate aeroelastic
effects on the flight dynamic behavior of the whole aircraft. It is well known that linear flutter engineering tools
that incorporate rigid-body degrees of freedom are unable to recover the usual flight dynamic equations of motion
due to dissimilar modeling objectives.

Conventional flight dynamics are used to develop flight control laws, improve handling qualities and assess the
maneuver/flight loads, with given distributed aerodynamic forces and mass models, whereby the airframe is
considered rigid or deformed in quasi-static manner by an equilibrium force distribution.1 On the contrary,
aeroelasticity considers the mutual interaction of unsteady aerodynamics, inertial and elastic structural forces
induced by static/dynamic deformation of the structure and external disturbance forces, whereby flutter, gust loads,
aeroservoelasticity are of major concern.2 Clearly, the need of an unified formulation that encompasses flight
dynamic and aeroelasticity in a single body is brought about by emerging design requirements for air vehicles with
predominant aeroelastic characteristics, i.e. Sensorcraft, Switchblade Flying Wing, Morphing UAVs, etc. Due to
these reasons, the flight control engineers will have available a single modeling tool in a consistent format, which
allows for design and validation of robust and dependable flight control laws.

In the past, several techniques were developed to systematically incorporate aeroelastic effects into the aircraft
dynamic equations of motion. Among these, Rodden and Love3 addressed the conservation of the angular
momentum issue; in addition, Whinter4,5 and Dykman6 formulated the p-transform approach to generate reduced-
order models that avoid additional state variables to represent the unsteady aerodynamics. More recently, this topic
was simultaneously revisited by Looye,7,8 as well as the present authors9 in the 2005 IFASD conference. The former
proposes the combination of the aerodynamic models, whereby only dynamic increments are added to the
unchanged flight dynamic equations of motion, while the latter presents a unified RFA formulation to take into
account the influence of aeroelastic effects on the translational and rotational rigid-body modes of the aircraft.

Specifically, the RFA setup originally devised to model the unsteady aerodynamic forces computed for purely
oscillatory motion, is sought here as the aeroelastician’s natural tool to recover the quasisteady longitudinal and
lateral equations of motion ordinarily used by the flight control engineer. To this end, this unified modeling
framework will be composed of the following main tasks: i) coordinate transformations between the principle axes
and the flight dynamic’s stability axes;3 ii) inclusion of measured aerodynamic and control stability derivatives from
wind-tunnel or flight tests10,11 into the RFA matrices; and iii) addition of the related gravity components as well as
the lift and drag derivatives/forces due to forward velocity perturbations within the aeroelastic formulation.9 The last
step constitutes the main contribution of this work leading to an accurate recovery of the complete set of natural
modes that appear in the flight dynamic description of the aircraft. Clearly, its implementation will naturally extend
the RFA modeling setup into a flight dynamic oriented class.

To illustrate the modeling framework, the proposed technique is demonstrated using two examples: i) Navion,
and ii) Predator B aircraft. The Navion aircraft example provides an easy-to-handle lower dynamics system for
which a complete aerodynamic stability set is already available in open literature.12 The General Atomics
Aeronautical Systems Inc (GA-ASI) Predator B represents an actual state-of-the-art UAV, analyzed using standard
production flutter engineering tools.2 A detailed set of flight test estimated static and damping related non-
dimensional stability derivatives are available to the flight control designer. This database is used in this work to
update the quasi-steady aerodynamic matrices.

2
 II. Time-Domain Aeroelastic Formulation
The linearized flight dynamics equations of motion are described by six degrees of freedom models and include
forces and moments due to gravity, aerodynamics and propulsion. Accepted assumptions during the development
account for13,14,

• Vehicle is considered as a rigid-body.

• Earth is considered as an inertial frame of reference.
• Mass distribution of the whole vehicle is constant during the analysis.
• Vehicle has a plane of symmetry.
• The motion of the vehicle is composed of small deviations about a steady flight condition.
• Decoupling of the longitudinal and lateral/directional dynamics.
• Forces and moments are dependent only on the velocities of the vehicle, i.e., quasisteady flow.
• Negligible influence of the altitude gradients for the small altitude perturbations.

Thus, it turns out that the quasi-static assumption may not be sufficient for aircrafts with predominant aeroelastic
characteristics where the interaction between dynamic structural deformation due to unsteady flow and rigid-body
motion can play an important role. On the other hand, aeroelasticity can be viewed as a feedback interaction
mechanism between the structural dynamics and the unsteady aerodynamic forces. The former being induced by the
static or dynamic deformation of the structure, with the flight parameters, i.e., Mach number and altitude. This
feedback mechanism modifies the rigid-body and vibration modes that characterize the dynamic behavior of the
aircraft.15

In what follows, the time-domain aeroelastic formulation is briefly described following Karpel,16 and it will be used
to identify the key enabling elements involved in the recovering process of the traditional flight dynamic equations
of motion11 from the more general aeroelastic framework. Let’s consider the aeroelastic equation of motion in
generalized coordinates, excited by control surface motion and inertial loads at a given dynamic pressure q∞ ,

[ M s ]{ξ } + [Cs ]{ξ } + [ K s ]{ξ } + q∞ ( { }

Qs ( p ) {ξ } = − [ M c ] δ c + q∞ Qs ( p ) δ c { }) + {P} (1)

where [ M s ] ∈ ns ×ns , [Cs ] ∈ ns ×ns , and [ K s ] ∈ ns ×ns are the generalized mass, damping, and stiffness matrices,
[Mc ]∈ nc ×nc
is the coupling mass matrix between the control and the structural modes, {ξ }∈ ns is the vector of
generalized structural deflections including the rigid body modes; and {δ c } ∈ nc is the vector of control surface
commanded deflections. In addition, [Qs ] ∈ ns ×ns and [Qc ] ∈ nc ×nc are the generalized unsteady aerodynamic
force coefficient matrices associated with the structural and control modes and {P}∈ ns is related to the gravity
components through the Euler angles θ and φ ,11
I3
− g sin θ
I3
{ P} = [ Φ ] [ M ]
T
g cos θ sin φ (2)
I3
g sin θ cos φ

where, [ Φ ] ∈ ns ×m , [ M ] ∈ m×m , and I 3 ∈ 3×3 are the modal matrix, the mass matrix in physical coordinates,
T

and the identity matrix, where m is the number of degree of freedom of the FEM model and g is acceleration due to
the gravity. The most general RFA to the aerodynamic force coefficient matrix is,

−1
Q ( p ) = [ Ao ] + [ A1 ] p + [ A2 ] p 2 + [ D ] ( p [ I ] − [ R ]) [E] p (3)

3
where p = sb / V∞ is the non-dimensional Laplace operator, s is the Laplace variable, b is the reference length and
V∞ is the true airspeed. The aerodynamic data is computed for pure harmonic oscillations in the imaginary axis, and
by the analytic continuity principle their domain of application is extended to the whole complex plane. The
approximation process involves the replacement of p by ik , where k is the non-dimensional frequency,
k = ωb / V∞ . Note that, Eq. (3) can handle both Roger’s17 and Minimum-State18 formulations to fit the tabulated
Q ( ik ) matrices. In general, the [ Ai ] and [ E ] matrices, are column partitioned as,

[ Ai ] = Asi Aci , i = 0,1, 2

(4)
[ E ] = [ Es Ec ]

In this formulation, the [ Ai ] , i = 0,1, 2, coefficient matrices represent the quasisteady aerodynamic forces, by
playing the role of an equivalent aerodynamic stiffness, aerodynamic damping and aerodynamic inertia, and the
remnant terms are used to model the flow unsteadiness by Padé approximants. These effects can be modeled as a
state-space realization,

ξ V∞
{ xa } = [ E s Ec ] + [ R ] { xa }
δc b (5)
{ z a } = [ D ] { xa }

where xa is the aerodynamic vector state. Substituting Eqs. (3), and (5) into (1), while considering p, will allow the
overall open-loop aeroelastic system with external excitation for a fixed flight condition to be written in state-space
form as,

ξ 0 I 0 ξ 0 0 0 δc 0
ξ = Ks Cs D ξ + Kc Cc Mc δ c + MP
xa 0 Es R xa 0 Ec 0 δc 0
Aae Bae
(6)
or in a more compact form,

{xae } = [ Aae ]{xae } + [ Bae ]{u} + {w} (7)

where,

{ }
T
{ } { }
T T
( MP )
T T
xae = ξT ξT xaT , uT = δ cT δ cT δ cT , wT = 0T 0T

and,

−1
q∞ b2 q∞ b
M= [M s ] + As2 , Ks = −M [ K s ] + q∞ Aso , Cs = − M [Cs ] + As1
V∞2 V∞

q∞ b2 q b
M c = −M [M c ] + Ac2 , K c = − q∞ M Aco , Cc = − ∞ M Ac1
V∞2 V∞
V
D = q∞ M [ D ] , R = ∞ [R]
b

4
The feedback mechanism between the structural dynamics and the unsteady aerodynamic model is described by Eq.
(6). It is observed that, by eliminating all the terms related with the aerodynamic lag states, i.e., the last row and
column in the [ Aae ] and [ Bae ] matrices (shadowed areas), the traditional flight dynamic equations of motion can
presumably be recovered. Otherwise, the inclusion of the dynamic model for the unsteady aerodynamic forces will
result in a full aeroelastic system.

Thus, to fully recover the flight dynamic equations of motion, a two-stage transformation process of the aeroelastic
model is proposed. The first stage involves a compatible coordinate axes, while the second stage, identifies key
enabling terms within the unsteady aerodynamic model. The former is an accepted practice using production flutter
engineering tools,2-5 while the latter is implemented by the proper allocation of the gravity components vector, w ,
as well as the lift and drag force terms related with forward velocity perturbations within the RFA matrices.

III. Transformation Between Principle And Stability Axes

The objective of the transformation process is to change
the generalized coordinates of the six rigid-body modes q3
q6 q1
computed by FEM in the principle axes into airframe
states in the stability axes. q5 q4

In this way, the rigid body sub-matrices in the state-

space equations of the aeroelastic system will have the q2
same definition as those of the flight dynamics field,
enabling the well-known separation principle between (a)
the longitudinal and lateral/directional dynamics. Tz
Rz
Figure 1 schematically shows the procedure for such
Ry
transformation. It can be briefly summarized as
follows: Tx
Ty Rx
1. Principle Axes to Body Axes, (Fig. 1(a) to (b)): The (b)
generalized coordinates of the six rigid-body
modes computed by the structural finite element H
model are defined in the principle axes. The θ, q y, v
φ, p
transformation from the principle axes, { xq } , to
the body axes, {ξ R }, is performed through the ψ, r
generic matrices [ RP ] and [ RB ] ,2,3 z, w
(c)
Figure 1. Coordinate Axes Transformations

{x q }
T T
= {q1 , q2 , , q6 } (8)
T
{ξ R }T = {Tx ,T y ,Tz , Rx , R y , Rz } (9)

{xq } = [ RP ]−1 [ RB ] {ξ R } (10)

where [ RP ] is the rigid body modal matrix in the principle axis and [ RB ] is the rigid body modal matrix in the
body axis at the c.g. location, respectively. The vector components in Eq. (9), Tx; Ty and Tz are the fore-aft, the
lateral and the plunge translational rigid-body modes. In addition, Rx; Ry, and Rz are the roll, pitch and yaw
rotational rigid-body modes, respectively. It should be noted that, after the transformation the generalized mass
matrix [ M s ] is no longer necessarily diagonal. In fact, the submatrix associated with the rigid body modes is

5
 identical to the mass matrix in the flight dynamics equation; i.e. the off-diagonal terms contain the products of
inertia,

M rb 0
[M s ] = 0 M ee

where the [ M rb ] and [ M ee ] matrices are defined as,

m 0 0 0 0 0
0 m 0 0 0 0
0 0 m 0 0 0
[ M rb ] = 0 0 0 I xx I xy I xz , [ M ee ] = mii

0 0 0 I xy I yy I yz
0 0 0 I xz I yz I zz

where m, mii , I xx , I yy , I zz , I xz , I xy and I yz are the mass of the whole aircraft, the generalized mass associated
with the ii-th elastic mode, the roll, pitch and yaw moments of inertia and the longitudinal and lateral/directional
products of inertia, respectively. In general due to symmetry assumption, I xy = I yz = 0 .

2. Body Axes to Stability Axes, (Fig. 1(b) to (c)): In this stage, the transformation from the body axes, {ξ R }, to the
airframe axes, {ξ AS }, is performed through the matrix [TA ] , that is.

T
{ξ R }T = {Tx ,T y ,Tz , Rx , R y , Rz } (11)
ξR
= [TA ] {ξAS } (12)
ξR

{ξAS }T = {x , y , u, T
,h, p,w, r , θ , φ , q,ψ } (13)

where x, y , u, β , h, p, w, r , θ , φ , q, and ψ are the perturbed quantities related with the forward position,
lateral position, forward airspeed, sideslip angle, altitude, roll rate, upward velocity, yaw rate, Euler pitch
angle, Euler roll angle, pitch rate, and Euler azimuth angle, respectively. These are the main perturbed
variables that describe the aircraft behavior in the stability axes system. For a symmetric maneuver, Refs. 3
and 4 described the matrix [TA ]long , while this work further extends the approach delineated in Ref. 4 for

anti-symmetric maneuvers. In this case, the matrix [TA ]lat is defined as follows,

Ty 1 0 0 0 0 0 y
Rx 0 0 0 0 −1 0 β
Rz 0 0 0 0 0 −1 p
= (14)
Ty 0 V∞ 0 0 0 V∞ r
Rx 0 0 −1 0 0 0 φ
Rz 0 0 0 −1 0 0 ψ

6
Finally, for an asymmetric maneuver, the matrix [T A ] ∈ 12×12
in Eq. (12) will be generated by the proper
arrangement of the elements that form the rows and columns of the [TA ]long and [TA ]lat matrices. Therefore, a
seamless axes transformation process can be performed to convert the rigid-body modes in generalized coordinates
from the production finite element codes to the airframe states in the airframe stability axes system.

III. RFA Setup: Updating Quasisteady Aerodynamic Model and Including Gravity Terms
In this section, the RFA concept is further extended for flight dynamic modeling purposes. By allowing the inclusion
of gravity related components, vertical acceleration related aerodynamic stability derivatives, as well as the lift and
drag forces due to forward velocity perturbations, the traditional quasisteady flight dynamic equations of motion will
be fully recovered. Then, a straightforward procedure is outlined to identify the location of the elements that need to
be updated/replaced within the quasisteady aerodynamic matrices, [ A0 ] , [ A1 ] and [ A2 ] , of Eq. (3).

It is known that production flutter engineering tools employ panel methods to compute linear unsteady
aerodynamics. However, the drag related stability derivatives computed by those methods, such as CD , CDq, among
others, are not accurate enough because of the inherent lack of skin friction drag. Hence, it will be particularly useful
to replace the computed translational and rotational rigid-body’s related elements at zero and/or low reduced
frequency in the generalized aerodynamic stiffness and damping matrices, [ A0 ] and [ A1 ] , with wind tunnel and/or
flight-test measured aerodynamic stability derivatives.11

For instance, at zero reduced frequency (p = 0), the RFA Fore-Aft Mode

model is reduce to [Q(0)] = [ A0 ] , which contains Plunge Mode

Pitch Mode
elements associated with the static aerodynamic stability
derivatives, if rigid body modes exist.
3
Specifically, for symmetric structural modes namely the
fore-aft mode, plunge mode and pitch mode the static A0 =
aerodynamics associated with these three rigid body modes
ns-3
are located in the [ A0 ] matrix as shown in Figure 2, where
a11 , a 21 and a 31 are the aerodynamics forces due to the
fore-aft mode and a12 , a 22 and a 32 are due to the plunge ns nc
mode. Figure 2. Rigid-body modes location inside [ A0 ]

At zero reduced frequency, all of these terms are normally zero since the fore-aft and the plunge modes cannot
introduce an angle of attack.

The a13 , a 23 and a 33 are the unsteady aerodynamics coefficients due to the pitch mode. In fact, at zero reduced
frequency, they can be directly correlated with the longitudinal static aerodynamic stability derivatives Dα , Lα , and
M α , i.e. the dimensional induced drag force, lift force and pitch moment at unit angle of attack (in radian) of the
whole configuration, i.e. Lα is shown in Eq. (15),

Lα
a 23 = − 0.5 (15)
q∞
where the negative sign in the above equations is due to the sign convention used for Q ( p ) in ZAERO.2

The terms a1i , a 2i , a 3i and a ii shown in Figure 2 represent the unsteady aerodynamics due to the i-th structural
mode. If this i-th structural mode is a control surface mode, then they can be replaced by the wind tunnel measured
aerodynamic stability derivatives of the control surface, where a1i , a 2i , a 3i and a ii can be correlated with the drag,

7
lift and moment stability derivatives due to control surface deflection, respectively, whereas a ii with its own control
surface hinge moment.

In general, the related symmetric rigid-body modes sub-matrices in As0 and As1 can be expressed as a function
of the following longitudinal static and damping non-dimensional aerodynamic stability derivatives,

(
0 0 − S CDα − CLo − ) mg
q∞
As0
sym
= 0 0 (
− S CLα + CDo ) (16)
0 0 − ScCmα

and

(
2S C Du + C Do ) (
S C Dα − CLo ) (
−0.5Sc CDα + C Dq )
( c 2) ( c 2) ( c 2)
As1 =
(
2 S CLu + C Lo ) (
S CLα + C Do ) (
−0.5Sc C Lα + C Lq ) (17)
sym
( c 2) ( c 2) ( c 2)
(
2 Sc Cmu + Cmo ) ScCmα (
−0.5Sc 2 Cmα + Cmq )
( c 2) ( c 2) ( c 2)
where S and c are the reference area and chord used in the aeroelastic formulation. The coefficients; CDu , CLu and
Cmu are the non-dimensional drag, lift, and pitch moment stability derivatives, with respect to the change of the
forward flight speed; CDo , CLo and Cmo are the non-dimensional drag, lift and pitch moment coefficients, at the trim
condition; CD , CL and Cm are the non-dimensional drag, lift and pitch moment stability derivatives, with respect
to the angle of attack in radian; and CD α + CDq , CL α + CLq , and Cm α + Cmq , are the damping derivatives of the
drag, lift and pitch moment coefficients, where q is the non-dimensional pitch rate defined as q = Qc/2V∞ and Q is
the pitch rate in rad/sec, respectively.

Note that, the CDu + CDo , CLu + CLo , Cmu + Cmo coefficients that appear in the first column of the As1 submatrix
sym
are the non-dimensional aerodynamic stability coefficients mainly involved in the phugoid mode, but they cannot be
computed accurately by the panel methods. Accordingly, to achieve a more accurate description of the aircraft
dynamic behavior, the designer must replace these coefficients in the related elements of the As1 matrix.
sym

At this point, a departure in the usual RFA modeling setup is proposed to allow for both flight dynamics and
aeroelastic disciplines to be handled within the unified RFA modeling framework. To this end, gravity related terms
are sought to be included in the [ A0 ] matrix, while the only allowable rigid-body acceleration aerodynamic
derivative terms, i.e. the ones related with the w and v , will be accessed through the generic [ A2 ] matrix. Note
that, these are the only acceleration terms retained in the quasisteady flight dynamic equations of motion to account
for the effect on the tail of the wing/body downwash and sidewash.14 Then, Eq. (16) clearly shows the inclusion of
the gravity related term, − mg q , into the As0 matrix, and it will play a key role in the longitudinal dynamics
∞ sym

8
of the plane to accurately predict the phugoid mode. Additionally, there are others terms in the As2 submatrix
sym
that can provide some important contributions to the phugoid dynamic mode. They are identified in the second
column of the As2 submatrix, as shown in Eq. (18).
sym

0.5ScCDα
a11 a13
c( )
2
0.5ScCLα
As2 = a21 a23 (18)
sym c( )
2
0.5Sc 2Cmα
a31 a33
( 2)
2
c

Clearly, these terms are used to facilitate the inclusion of related vertical acceleration aerodynamic stability
derivative terms, w, into the otherwise quasisteady longitudinal dynamic.

In a similar fashion to the symmetric case, for the three anti-symmetric modes namely the lateral translation, roll,
and yaw modes, the sub-matrices in As0 and As1 can be correlated with the static and damping lateral non-
dimensional aerodynamic stability derivatives, as shown in Eqs. (19) and (20),

mg
0 − SC yβ
q∞
As0 anti = 0 0 + SbClβ (19)
sym
0 0 + SbCnβ

)
+0.5SbC y p

( )(
− SC yβ −0.5Sb
C y − C yr
( 2)
c ( 2)c c
2
β

−0.5Sb 2Cl p

( ) ( )
+ SbClβ −0.5Sb 2
As 1 = −Cl + Clr (20)
anti
sym ( c 2) ( c 2) c
2
β

−0.5Sb 2Cn p

( ) ( )
+ SbCnβ −0.5Sb 2
−Cn + Cnr
( c 2) ( c 2) c
2
β

where b, Cy , Cl , and Cn , are the reference length, the non-dimensional side force, roll moment, and yaw moment
stability derivatives, with respect to side-slip angle in radian; Cyp, Clp, and Cnp , are the non-dimensional side force,
roll moment, and yaw moment stability derivatives, with respect to the non-dimensional roll rate p, where p =Pb/2V
and P is the roll rate in rad/sec; Cyβ + Cyr , Clβ + Clr , and Cnβ + Cnr are the non-dimensional damping derivatives of
the side force, roll moment, and yaw moment, respectively.

For accurate control modeling purposes, the non-dimensional control surface aerodynamic stability derivatives can
be related to the ns+i-th column and the first 3 rows of the [ A0 ] and [ A1 ] matrices, where i-th represents the i-th

9
control surface. For symmetric modes, the corresponding static and damping non-dimensional derivatives in the
[ A0 ] and [ A1 ] matrices are,

−0.5cSCD
δ

− SCDδ
( 2)
c

−0.5cSCL
3 δ 3
Ac0 = − SCLδ ∈ Ac1 = ∈ (21)
sym
−cSCmδ
sym
( 2)
c

−0.5c 2 SCm
δ

( 2)
c

where CDδ , CLδ , and Cmδ are the non-dimensional drag force, lift force, and pitch moment coefficients due to
control surface deflection in radian and CD , CL , and Cm are the non-dimensional drag, lift and pitch moment
δ δ δ
damping derivatives due to the rate of the control surface deflection. Likewise, for anti-symmetric modes, the
corresponding stability derivatives in the [ A0 ] and [ A1 ] matrices are,

−0.5bSC y
δ

( 2)
c
− SC yδ
−0.5b2 SCl
3 δ 3
Ac0 = − SbClδ ∈ Ac1 = ∈ (22)
anti
sym
− SbCnδ
anti
sym ( 2)
c

−0.5b 2 SCn
δ

( 2)
c

Again, if Clδ , Cnδ , Cl , and Cn are defined in the stability axes, their signs must be reversed in accordance with
δ δ
the sign convention used in ZAERO.2

For the whole aircraft structural model containing all six translational and rotational rigid-body modes, the size of
the rigid-body sub-matrices in As0 , As1 and As2 is 6x6. They are built up by combining the related rows

and columns from their corresponding symmetric and anti-symmetric 3x3 matrices, i.e., As0 is generated by
whole
the proper arrangement of the elements in Eqs (16) and (19).

In this manner, the complete aeroelastic system given by Eq. (7) is now reformulated using this updated RFA setup
that already included the gravity components vector, vertical acceleration related aerodynamic stability derivatives,
as well as the lift and drag derivatives due to perturbed forward speed. Hence, a flight dynamic oriented rational
functional approximation matrix, QFD ( p ) , is built-up from the individual matrices described in Eqs. (16) through
(22).

Now, the unified aeroelastic and flight dynamics state-space description is achieved, as defined by Eq. (23).

{xae } = Aae {xae } + Bae {u} (23)

10
where Aae and Bae matrices are obtained through the RFA of QFD ( p ) . Clearly, four possible analysis
scenarios can be built using the unified aeroelastic formulation in accordance with the aerodynamic model used and
whether or not the elastic modes are included, as shown in Table 1.

Table 1. Unified Aeroelastic Formulation – Possible Analysis Scenarios

Elastic Modes

Omitted Included

Quasisteady Flight Dynamics Quasisteady

Aerodynamic

Forces Model Aeroelastic Model

Model

Unsteady Full Frequency Flight Full Aeroelastic

Forces Dynamics Model Model

Among them, the full frequency flight dynamic model needs to be defined. It is the aeroelastic model that results
when only translational and rotational rigid-body modes and the full unsteady aerodynamic model are considered. In
this particular case, the rigid body modes will be subjected to the high frequency unsteady aerodynamic effects
coming from the aerodynamic lag model approximation.

Finally, to avoid numerical discontinuities in the computed unsteady aerodynamic forces using the user’s set of non-
dimensional aerodynamic stability derivatives; a double matching process in ZAERO2 is devised. Technically, the
aerodynamic derivative set is first used to replace the coefficients associated with the translational and rotational
rigid body modes in the generalized aerodynamic force matrix computed in the reduced frequency domain. Once the
RFA matrices are computed in the time-domain, the user’s set of non-dimensional static, damping, and acceleration
related derivatives is included again into the [ A0 ] , [ A1 ] and [ A2 ] matrices. This technique is sought to minimize the
risk of jump discontinuities in any of the elements that compose the matrix QFD ( p ) .

V. Longitudinal Dynamics: Closed Form Solution

Closed form solutions are computed using MATHEMATICA Software19 for symmetric and anti-symmetric rigid-
body maneuvers with the unified RFA formulation, to emulate as those ordinarily used by the flight dynamic
engineer, i.e. by separate longitudinal and lateral dynamic behavior analysis.14

Specifically, the longitudinal dynamic equations of motion, without considering any additional elastic modes in the
model, are obtained in closed form by incorporating Eqs. (16) to (18) and (21) into Eq. (3) while applying the
similarity transformation [TA ] , that is,
long

xlong = TA−1 AaeTA xlong + TA−1Bae ulong (24)

long long
Along Blong
T
T
xlong = {x u h w θ q} , ulong = δ e

where x, u, h, w, θ , q and δ e are the perturbed forward displacement, forward airspeed, altitude, upward velocity,
Euler pitch angle, pitch rate and the elevator control surface deflection angle, respectively.

After neglecting the first and third rows and columns related with the perturbed forward displacement and altitude
airframe states, x and h , we will proceed to evaluate each non-zero element of the transformed dynamic matrix,
TA−1 AaeTA , along its columns,
long

11
Column #1

TA−1 AaeTA (1,1) =−

(
2q∞ S CD0 + CDu ) Xu
long mV∞

TA−1 AaeTA (2,1) =−

(
2q∞ S CL0 + CLu ) Zu
long mV∞

(
q∞ Sc 2mV∞2 Cm0 + Cmu − q∞ Sc CL0 + CLu Cmα ) ( )
TA−1 AaeTA (4,1) =− ( M u + M w Zu )
long mV∞3 I yy

Column #2

TA−1 AaeTA (1, 2) =−

(
q∞ S CDα − CL0 ) Xw
long mV∞

TA−1 AaeTA (2, 2) =−

(
q∞ S CD0 + CLα ) Zw
long mV∞
q∞ Sc −2mV∞2Cmα + q∞ Sc C D0 + C Lα Cmα ( )
TA−1 AaeTA (4, 2) =− ( M w + M wZw )
long 2mV∞3 I yy

Column #3
TA−1 AaeTA (1,3) = −g
long

Column #4

TA−1 AaeTA (1, 4) =−

(
q∞ Sc C Dq + CDα ) Xw
long 2mV∞

TA−1 AaeTA (2, 4) = V∞ −

(
q∞ Sc C Lq + C Lα ) (V + Zq )
∞
long 2mV∞
TA−1 AaeTA (3, 4) =1
long

TA−1 AaeTA (4, 4) =

{
q∞ Sc 2 2mV∞2Cmq + 2mV∞2 − q∞ Sc C Lq + C Lα ( )C mα } (
M q + M w V∞ + Z q )
long 2mV∞3 I yy

{
as well as the input vector, TA−1Bae
long
, }
Column #1
q∞ S −C Dδ( )
{TA−1Bae (1,1)}long = m
X δe

q∞ S −C Lδ( )
{TA−1Bae (2,1)}long = m
Zδ e

− q∞ Sc −2mV∞2Cmδ + q∞ ScC Lδ Cmα

{TA−1Bae (4,1) }long = 2mV∞2 I yy
( Mδ e
+ M w Zδ e )

12
where the variables X u , X w , etc. are defined in accordance to Table 4.3 of Ref. 14. It is observed that, the resulting
state-space representation accurately recovers the classical form of the longitudinal dynamics referenced to stability
axes as shown in Eq. (25),

Xu Xw −g 0 X δe
u u
w Zu Zw 0 (V∞ + Z q ) w Zδ e
= + δ e (25)
θ 0 0 0 1 θ 0
q ( M u + M wZu ) ( M w + M w Z w ) 0 ( M q + M w (V∞ + Z q )) q ( Mδ e
+ M w Zδ e )
The longitudinal stability quartic is then computed from,

(
det sI − Along = 0 ) (26)

and its solution, in the most general case, will be composed with both free longitudinal oscillatory modes, that is the
short-period and the phugoid modes.

The closed form lateral equations of motion are computed in a similar fashion as in the longitudinal dynamics case.
Thus, Eqs. (19), (20), and (22) are fed into Eq. (3), and by applying the similarity transformation, [TA ]lat , the lateral
equations are,

xlat = [ A]lat xlat + [ B ]lat ulat (27)

T T
T
xlat = {y β p r φ ψ} T
ulat = {δ a δr }

where y , β , p, r , φ , ψ , δ a and δ r are the perturbed lateral displacement, sideslip angle, roll rate, yaw rate, Euler roll
angle, Euler yaw angle, aileron and rudder control surface deflection angle, respectively. The first and last rows and
columns related with the perturbed lateral displacement and yaw angle states, y and ψ , are eliminated.

The resulting state-space representation is able to retrieve the traditional lateral dynamics as shown in Eq. (28),

β Yv Y p* (Yr* − 1) g / V∞ β Yδ*a Yδ*r

p L'
β L'p L'
r 0 p L'
δa L'
δr δa
= + (28)
r r δr
N β' N '
p N r' 0 Nδ'a N δ'r
φ φ
0 1 0 0 0 0

As in the longitudinal case, the solution of the lateral characteristic equation,

det ( sI − Alat ) = 0 (29)

will generate the free lateral dynamic modes. These consist of three modes; they are the roll subsidence, the spiral
and the Dutch-roll modes. Clearly, Eqs. (25) and (28) indicate that both dynamic equations of motion are fully
recovered from the unified aeroelastic formulation, when only the set of rigid-body modes are considered.

VI. Application Examples

For numerical comparison purposes two examples are developed, a simple model related with the Navion aircraft12
and a more realistic model of the full aeroelastic GA-ASI Predator B UAV. The easy-to-handle Navion model is
used to validate the rational function approximation setup using previous published results.

13
The second case is used to evaluate the effects on the classical longitudinal and lateral dynamic modes caused by the
inclusion of several aeroelastic modes. The last example illustrates the application of the proposed methodology
using standard production flutter analysis tools.2

Case Study 1: Navion Aircraft

The selected case is a four to five seat, single engine, retractable tricycle landing gear, metallic monoplane, general
aviation aircraft designed and built in the late forties by North American Aviation, as shown in Figure 3. No
aeroelastic modes are involved in this analysis and the recovered longitudinal and lateral equations of motion that
are obtained using the RFA setup are those ordinarily used by the flight control engineer.

For sea level and Mach 0.158 flight conditions, the complete set of longitudinal and lateral non-dimensional
aerodynamic stability derivatives are shown in Table 2, while the geometric, mass distribution and center of gravity
characteristics are provided in Table 3.

Table 2. Navion Aircraft Non-Dimensional Stability Derivatives

Longitudinal
CL CD CLα CDα Cmα CLα Cmα CLq Cmq CL CD CmM CLδ e Cmδ e
M M
M = 0.158
Sea Level 0.41 0.05 4.44 0.33 -0.683 0.0 -4.36 3.8 -9.96 0.0 0.0 0.0 0.355 -0.923
Lateral C yβ Clβ Cnβ Cl p Cn p Clr Cnr Clδ Cnδ C yδ Clδ Cnδ
M = 0.158 a a r r r
Sea Level -0.564 -0.074 0.071 -0.41 -0.0575 0.107 -0.125 -0.134 -0.0035 0.157 .107 -0.072

Note: All derivatives are per radian.

Substituting the numerical values of Tables 1 and 2 Table 3. Navion Aircraft: Mass, C.G. and Geometry
into the proposed unified RFA formulation, the
Weight, W [lb] 2750
longitudinal and lateral characteristic polynomials CG position [% MAC] 29.5
given by Eqs. (26) and (29) are,
Ixx, [slug-ft2] 1048
Iyy, [slug-ft2] 3000
s 4 + 5.0374 s 3 + 13.0398s 2 + 0.667 s + 0.596 = 0 (30) Izz, [slug-ft2] 3530
Ixz, [slug-ft2] 0
Wing Area, S [ft2] 184
s 4 + 9.4339 s 3 + 14.0907 s 2 + 48.8566 s + 0.3985 = 0 Wing Semispan, b [ft] 33.4
(31) MAC, [ft] 5.7
Weight, W [lb] 2750
The solution of Eq. (30) provides the usual short- CG position [% MAC] 29.5
period and phugoid modes. Similarly, by solving Eq. Ixx, [slug-ft2] 1048
(31), the roll subsidence, Dutch-roll and the spiral Iyy, [slug-ft2] 3000
mode, are obtained. These results accurately match Izz, [slug-ft2] 3530
those presented in Ref. 20 as shown in Tables 4 and
5.

Table 4. Navion Aircraft: Longitudinal Dynamic Roots

Mode Name Linearized Flight Dynamics Unified RFA
Short-Period -2.5018 ± j2.5622 -2.5018 ± j2.5622
Phugoid -0.0169 ± j0.2419 -0.0169 ± j0.2419

Table 5. Navion Aircraft: Lateral/Directional Dynamic Roots

Mode Name Linearized Flight Dynamics Unified RFA
Roll Subsidence -8.4499 -8.4499
Dutch-Roll -0.48789 ± j2.3517 -0.48789 ± j2.3517
Spiral -0.008175 -0.008175

14
Case Study 2: Predator B UAV
The Predator B “Hunter Killer” aircraft is a
pusher-type design (rear mounted Allied-
Signal-powered Turbo-Engine Propeller),
with a composite airframe, and a
retractable tricycle landing gear.

The overall vehicle length is 36 feet (10.97

m) with an overall wingspan of 64-feet
(19.5 m). Each wing contains two large
inboard flaps for varying wing lift and two
large outboard ailerons to control vehicle
roll. Figure 4. GA-ASI Predator B UAV
The vehicle performance specifications include an operational altitude of over 50,000 feet (15,240 m), endurance of
over 30+ hour flights, a minimum take-off speed at sea level of 65 knots (74.8 mph), a sea level cruise speed range
of 198 knots (228 mph), and a sea level dive speed of 277 knots (319 mph). Figure 4 depicts the Predator B UAV’s
aerodynamic model used in this work.

This case analysis describes the results achieved using a simulation tool developed in the ZAERO/ASE2 and
Matlab/Simulink software environments to evaluate the presumably detrimental influence that aerodynamic
unsteadiness and structural flexibility effects can cause on the performance of a generic flight control system
developed using linearized flight dynamic concepts alone.

Aeroelastic Model Analysis

The aeroelastic model for the whole aircraft was created using MSC/NASTRAN and ZAERO2 software systems.
The aerodynamic forces in the frequency-domain were computed through ZONA6, ZAERO’s subsonic unsteady
aerodynamic method, while its time domain realization was performed using the minimum-state approach16
implemented in the ZAERO/ASE module. The Finite Element Model was validated using data from a full Ground
Vibration Test and includes 30 primary elastic structural modes in addition to the complete set of rigid-body modes.

A detailed set of static, damping and translational acceleration related non-dimensional stability derivatives are
available to the flight control designer, and this database is used to update the quasisteady aerodynamic matrices
[ A0 ] , [ A1 ] and [ A2 ] in [QFD ( p)] .
Tables 6 and 7 summarize the computed eigenvalues, using the unified RFA formulation, when an increasing
number of elastic modes are considered in the Hunter Killer UAV aeroelastic model. These values are compared
with the ones achieved using the linearized flight dynamic equations of motion, included in the second column in
both tables.

Note that, the results depicted in Table 6 are achieved using the quasi-steady aerodynamic model, i.e. that is
Es = Ec = R = D = M c 0 in Eq. (6), while Table 7 show the related results when a complete unsteady
aerodynamic model is used. It can be noted that, the first column under the RFA portion only considers the set of
translational and rotational rigid-body modes, whereas the third, fourth and fifth columns have added five, ten and
twenty structural elastic modes, respectively.

From Table 6, it is clearly observed that the addition of elastic modes will only affect the longitudinal dynamic
modes. Specifically, the phugoid mode’s magnitude kept changing, but its qualitative character did not, i.e. it is
unstable without any consideration to the number of modes included in the quasisteady aeroelastic model.

On the contrary, Table 7 shows a great deal of dynamic interactions between the unsteady aerodynamic forces and
the number of elastic modes considered, as expected in a full aeroelastic model. The first eight aeroelastic modes,
presumably related with the classical longitudinal and lateral/directional flight dynamic modes, kept changing their
quantitative and qualitative characterization with the increasing number of elastic modes.

15
 Table 6. Predator B UAV: Quasisteady Aerodynamics and Elastic Modes
RFA : Quasisteady Aerodynamics
Mode Linearized FD Rigid-Body Modes Rigid-Body + 5 EM Rigid-Body + 10 EM Rigid-Body + 20 EM
Long. Dynamics
Phugoid 0.0562 + j0.3355 0.0562 -j0.3357 0.0005 -j0.3935 0.0655 -j0.3516 0.0560 -j0.3618
Phugoid 0.0562 -j0.3355 0.0562 +j0.3357 0.0005 +j0.3935 0.0655 +j0.3516 0.0560 +j0.3618
Short-Period -1.3810 +j0.3252 -1.3787 -j0.3232 -1.5227 -1.3795 -j0.3553 -1.3657 -j0.2841
Short-Period -1.3810 -j0.3252 -1.3787 +j0.3232 -1.6138 -1.3795 +j0.3553 -1.3657 +j0.2841
Lat/Directional Dynamics
Dutch-Roll -0.1662 +j1.1770 -0.1641 -j1.1819 -0.1617 -j1.1738 -0.1610 -j1.1710 -0.1611 -j1.1726
Dutch-Roll -0.1662 -j1.1770 -0.1641 +j1.1819 -0.1617 +j1.1738 -0.1610 +j1.1710 -0.1611 +j1.1726
Roll -7.8432 -8.3335 -8.3318 -8.3294 -8.3322
Spiral -0.0159 -0.0162 -0.0162 -0.0161 -0.0162
FD = Flight Dynamics, EM = Elastic Modes

Table 7. Predator B UAV: Unsteady Aerodynamics and Elastic Modes

RFA : Unsteady Aerodynamics
Mode Linearized FD Rigid-Body Rigid-Body+ 5 EM Rigid-Body+ 10 EM Rigid-Body+ 20 EM
Long. Dynamics
Aeroelastic #1 (Phugoid) 0.0562 + j0.3355 0.3451 -j0.0372 0.0704 -j1.3256 0.0279 -j0.5622 -0.1549 -j0.2773
Aeroelastic #2 (Phugoid) 0.0562 -j0.3355 0.3451 +j0.0372 0.0704 +j1.3256 0.0279 +j0.5622 -0.1549 +j0.2773
Aeroelastic #3 (Short-Period) -1.3810 +j0.3252 0.9340 -2.2871 -3.1009 -j1.5133 -2.3849 -j0.8295
Aeroelastic #4 (Short-Period) -1.3810 -j0.3252 -0.9563 -2.7127 -3.1009 +j1.5133 -2.3849 +j0.8295
Lat/Directional Dynamics
Aeroelastic #5 (Dutch-Roll) -0.1662 +j1.1770 -2.3673 -j1.1992 -0.1752 -j1.3716 -0.1741 -j1.3688 -0.1471 -j1.2134
Aeroelastic #6 (Dutch-Roll) -0.1662 -j1.1770 -2.3673 +j1.1992 -0.1752 +j1.3716 -0.1741 +j1.3688 -0.1471 +j1.2134
Aeroelastic #7 (Roll) -7.8432 -1.1659 -0.1134 0.2626 -0.1048
Aeroelastic #8 (Spiral) -0.0159 0.0444 -0.0113 -0.0125 -0.0183

The terms from the first 6x6 sub-matrix in [QFD ( p )] are being modified by user input by way of updating the
[ A0 ] , [ A1 ] , and [ A2 ] matrices in the RFA. Because these terms are directly related to rigid-body modes that are
being updated, the RFA will not necessarily match the original AIC data. In contrast, the RFA values for the terms
related to elastic modes in [QFD ( p )] , should be in close approximation of the original AIC data, unless flight or
wind-tunnel test data is also available to update these terms.

Figures 5 through 8 show the changes that the unified RFA approach performs on individual components of the
generalized aerodynamic force (GAF) coefficient matrix, [QFD ( p )] . Figure 5 shows the real and imaginary parts of
the generalized force QFD2,4 ( p ) as a function of the reduced frequency, i.e. the generalized lateral force due to roll
motion. Once the RFA matrices are updated with the user supplied aerodynamic database, it is observed that both a
change in the ordinate at the origin in the real part and a change in the slope of the imaginary part are observed. This
fact can be explained in terms of the aerodynamic static and damping matrices. Specifically, QFD2,4 ( p ), is partially
generated using A0 ( 2, 4 ) = mg q∞ from the static aerodynamic matrix [ A0 ]whole , as well as
A1 ( 2, 4 ) = 0.5SbC yp (c / 2) from the equivalent aerodynamic damping matrix [ A1 ]whole . Hence, the inclusion of
the gravity related term modifies the real part of QFD2,4 ( p ), while the change of Cyp magnitude affects the slope of
the imaginary part of QFD2,4 ( p ) .

Similarly, Figure 6 depicts the real and imaginary parts of the generalized force QFD6,2 ( p ) as a function of the
reduced frequency, i.e. the generalized directional moment due to sideslip motion. In this case, only
A1 ( 6, 2 ) = −0.5SbCnβ ( c / 2) is modified, hence a change in the slope of the imaginary part of QFD6,2 ( p ) is
generated.

16
 Figure 5. Generalized Aerodynamic Force QFD2,4(p)

Figure 6. Generalized Aerodynamic Force QFD6,2(p)

On the contrary, Figures 7 and 8 depict the real and imaginary parts of the GAF for the pitching moment,
QFD5,11 ( p ) , and vertical force, QFD3,16 ( p ) , related with the elastic modes #11 and #16. These modes are the first
symmetric bending VT mode and the second symmetric bending wing mode, as shown in Figures 9 (a) and (b). In
this case, these terms are practically not affected during the quasisteady aerodynamic updating process and as
expected only minor differences are observed between the original and updated RFA values.

Figure 7. Generalized Aerodynamic Force QFD5,11(p)

17
 Figure 8. Generalized Aerodynamic Force QFD3,16(p)

(a) First Symmetric VT Bending Mode (b) Second Wing Bending Mode
Figure 9. PredatorUAV: Elastic modes related with QFD5,11(p) and QFD3,16(p).

These results confirm the statement made in Section IV that no discontinuities are present in any of the elements that
compose the matrix QFD ( p ) .

Closed-Loop Aeroservoelastic Simulation

The aeroelastic models included in Table 7 are used in the Hunter Killer UAV aeroservoelastic simulation. This set
of models, generated throughout an increasing number of elastic modes and unsteadiness effects, can be considered
as a family of uncertain aeroelastic plants for the six-degree of freedom FCS. This set will help to demonstrate the
robustness of the synthesized FCS simultaneously against both aerodynamic unsteadiness and vehicle flexibility
effects.

The simulation tool developed along this work allows the user to build the aeroelastic model from a set of pre-
computed structural and aerodynamic models. The final model will have the rigid-body sub-matrices in the state-
space equations of the aeroelastic system with the same definition as those of the flight dynamics. The flight and
fuel conditions are defined to allocate the proper autopilot’s gains. In addition, doublet command inputs can be
generated, i.e., pitch, roll angles or yaw rate. For this case analysis, the menus appears in the following order:

1. Structural Model: 3. Flight Condition Selection:

a. 6 DOF a. Mach Number
b. 6 DOF + 5 Elastic Modes b. Flight Altitude
c. 6 DOF + 10 Elastic Modes c. Fuel Weight Ratio
d. 6 DOF + 20 Elastic Modes 4. Doublet Command:
2. Aerodynamic Model: a. Pitch Channel
a. Quasisteady Model b. Roll Channel
b. Unsteady Model c. Yaw Rate Channel

18
Each aeroelastic model includes the airframe states, xas = [φ , θ ,ψ , p, q, r ] , considered in the MLDSTAT and ASE
T

cards. The state-space realization of the unified RFA for the aeroservoelastic system presents the following form:

xas A Are xas B

= rr + rr {u}
xe Aer Aee xe Bre
xas
{ y} = [Crr Cee ]
xe
The output vector, y, is composed of three rate-gyro measurements, i.e. roll, pitch and yaw rate devices, while the
control effector vector, u, is built from the SURFSET bulk data card, i.e. the Hunter Killer case has a set consisting
of nine control surfaces. The generated aeroelastic models have a frequency content of 7, 10 and 15 Hz when 5, 10
and 20 structural modes are considered. Figure 10 presents the output and control vector components response to a
doublet in the pitch channel when considering unsteady aerodynamic forces as well as an increasing number of
elastic modes in the aeroelastic model.

The doublet command starts at t= 5 sec, with ±5 degrees of pitch command and has a duration, ∆d, of 6 seconds. In
the figure, three aeroelastic model responses, including 5 (green), 10 (red) and 20 (light blue) elastic modes, are
compared with those achieved using rigid-body dynamics (blue) derived from quasisteady flight-dynamics
approach.

Along the time traces of the state and control variables in Figures 10, longitudinal and lateral/directional coupling
effects are visible as depicted in the roll, φ, and roll rate, p, plots, but the overall closed-loop behavior is quite
acceptable, considering that the autopilot was designed using only rigid-body assumptions. As expected, as the
number of structural modes considered is increased, major dynamic interactions effects will take places between the
longitudinal and lateral/directional states with the vehicle flexibility and flow unsteadiness.

Note that the greater excursions of the vehicle lateral/directional states and longitudinal control surface
displacements noted in Figures 10 (a) and (b), are presumably related with the existence of an extra unstable mode
in the aeroelastic model with 10 elastic modes, that is mode #7 in Table 7. Finally, the aeroelastic model including
20 elastic modes clearly represents the worst-case condition for the FCS in terms of the required lateral/directional
control surface deflections to track the doublet in the pitch command channel.

(a) Closed-Loop Output Vector Variables

19
 (b) Longitudinal Control Surface Deflections

(c) Lateral/Directional Control Surface Deflections

Figure 10. Hunter Killer UAV: Aeroservoelastic Responses to a Pitch Doublet Command
The results presented so far confirm the use of the unified RFA formulation as an enabling tool to take into account
the aeroelastic effects on the translational and rotational modes of the Predator B UAV.

20
 VII. Conclusions
In this paper, a unified formulation using the rational function approximation setup is presented to recover the
longitudinal and lateral equations of motion ordinarily used by the flight control engineer. This approach is able to
identify key elements within the rational function approximation to update the unsteady aerodynamic model. By
allowing the inclusion of gravity components, vertical acceleration related aerodynamic stability derivatives, as well
as the lift and drag derivatives due to forward velocity perturbations into the RFA matrices, the linearized
quasisteady flight dynamic equations of motion are fully recovered using the proposed unified aeroelastic
formulation.

Hopefully, this modeling framework will become the aeroelastician natural enabling tool by providing a common
state-space representation that accurately models the influence of the vehicle flexibility and flow unsteadiness
effects on the dynamic behavior of the whole aircraft. It is shown by closed-form solutions that quasisteady
longitudinal and lateral dynamics are exactly recovered, when no elastic modes are presented in the aeroelastic
model. The Navion and Predator B Hunter Killer aircrafts are used to demonstrate the modeling framework.

Currently, work is being performed to extend this unified RFA tool to also calculate gain and phase margins for each
gain connection in the flight control system, as well as perform robust stability analysis using the singular value
approach. This work will extend the benefits of this unified formulation and give the aeroelastician/flight
dynamicist more tools to design and analyze aeroelastic vehicles.

Acknowledgments
The authors would like to thank Dr. John Hutcheson and Shadi Ghandchi of GA-ASI for their valuable contribution
with the Hunter Killer UAV flight dynamics model.

References
1
Roskman, J., “Airplane Flight Dynamics abd Automatic Flight Controls, Part II,” Design, Analysis and Research Corporation,
Lawrence, Kansas, USA, 1995.
2
ZONA Technology, Inc, “ZAERO User’s Manual,” Ver. 7.3, Scottsdale, AZ, October, 2005.
3
Rodden, W. P., and Love, J. R., “Equations of Motion of a Quasisteady Flight Vehicle Utilizing Restrained Static Aeroelastic
Characteristics,” Journal of Aircraft, Vol. 22, No. 9, 1985, pp. 802-809.
4
Winther, A. A., Hagemeyer, D. A., Britt, R. T., and Rodden, W. P., “Aeroelastic Effects on the B-2 Maneuver Response,”
Journal of Aircraft, Vol. 32, No. 4, 1995, pp. 862-867.
5
Winther, B. A., Goggin, P. J., and Dykman, J.R., “Reduced Order Dynamic Aeroelastic Model Development and Integration
with Nonlinear Simulation,” Journal of Aircraft, Vol. 37, No. 5, September-October, 2000, pp. 833-839.
6
Dykman, J. R., and Rodden, W., “Structural Dynamics and Quasistatic Aeroelastic Equations of Motion,” Journal of Aircraft,
Vol. 37, No. 3, 2000, pp. 538-542.
7
Looye, G., “Integration of Rigid and Aeroelastic Aircraft Models Using the Residualised Model Method,” International Forum
on Aeroelasticity and Structural Dynamics 2005, Paper IF-046, Munich, Germany, 28 June – 01 July, 2005.
8
Reschke, C., and Looye, G., “Comparison of Model Integration Approaches for Flexible Aircraft Flight Dynamics Modeling,”
International Forum on Aeroelasticity and Structural Dynamics 2005, Paper IF-154, Munich, Germany, 28 June – 01 July, 2005.
9
Baldelli, D. H., Chen, P. C., and Panza, J. L., “ From Aeroelastic Formulation to Flight Dynamic Equations and its Application
to Predator UAV,” International Forum on Aeroelasticity and Structural Dynamics 2005, Paper IF-104, Munich, Germany, 28
June – 01 July, 2005.
10
Wieseman, C. D., “Methodology for matching Experimental and Analytical Aerodynamic Data,” Structures, Structural
Dynamics and Materials Conference, 29th,. AIAA1988-2392, pp. 1412-1422, Williamsburg, VA, Apr. 18-20, 1988.
11
Lind, R. and Brenner, M., “Robust Aeroservoelastic Stability Analysis: Flight Test Applications,” Springer-Verlag, London,
1999, Ch. 4.
12
Nelson, R. C., “Flight Stability and Automatic Control,” McGraw Hill Company, 2nd Edition, 1998, Chs 4 & 5.
13
Wright Laboratory Report WL-TR-96-3099. Wright Paterson AFB, OH 45433-7562, Part II, Section IV, May 1996.
14
McRuer, D., Ashkenas, I., and Graham, D., “Aircraft Dynamics and Automatic Control,” Princeton University Press,
Princeton, 1990, Ch. 4 to 6.
15
Bryson, A. E., “Control of Spacecraft and Aircraft,” Princeton University Press, Princeton, New Jersey, 1994, Ch. 15.
16
Karpel, M., “Size-Reduction Techniques for the Determination of Efficient Aeroservoelastic Models,” Control and Dynamic
Systems, Vol. 54, 1992, pp. 263-295.
17
Roger, K. L., “Airplane Math Modeling Methods for Active Control Design,” Proceedings of the 44th AGARD Structures and
Materials Panel, AGARD-CP-228, Lisbon, Portugal, April 1977, pp. 4.1-4.11.

21
18
Karpel, M., “Design of the Active Flutter Suppression and Gust Load Alleviation using State-Space Aeroelastic Modeling,”
Journal of Aircraft, Vol. 19, No. 6, 1982, pp. 221-227.
19
Wolfram, S., “Mathematica 5,” Wolfram Media, Inc., Champaign, IL, USA, 2003.
20
Nelson, R. C., “Flight Stability and Automatic Control,” McGraw Hill Company, 2nd Edition, 1998, Ch. 4 & 5.

22