Professional Documents
Culture Documents
reference adaptive control, such as infinite zero structure and signs of high-frequency gain matrices. A
comprehensive study of multivariable model reference adaptive control systems in the presence of damage is
performed to obtain critical design specifications for adaptive flight control, such as system and controller
parameterizations and adaptive parameter update laws. Both analytical and simulation results are given to illustrate
the design and performance of adaptive control systems for aircraft flight control.
transport model, which shows the desired performance of the distribution due to damage, Ixy and Iyz become nonzero, resulting in
adaptive control system in the presence of damage. the following moment equations for the asymmetric aircraft:
II. Modeling of Aircraft with Asymmetric Damage Ix p_ Ixy q_ Ixz r_ Ixy pr Iz Iy qr Iyz q2 r2 Ixz pq
In this section, a detailed study of the modeling of aircraft myw_ qu pv mzv_ ru pw L
dynamics under asymmetric damage is presented. A nonlinear mg cos cos y mg cos sin z (7)
aircraft dynamic model in the presence of asymmetric damage is first
introduced. A linearized aircraft model is then derived about a chosen
equilibrium, and the dynamic coupling features of the damaged
aircraft model are discussed. Several MIMO aircraft systems based Ixy p_ Iy q_ Iyz r_ Ix Iz pr Ixy qr Ixz r2 p2 Iyz pq
on a generic aircraft model are considered, and the invariance of the mxqu w_ pv mzu_ rv qw M
infinite zero structure of the damaged aircraft is illustrated.
mg sin z mg cos cos x zTL TR (8)
A. Nonlinear Model of Aircraft with Asymmetric Damage
1. Force Equations
Several aircraft models under damage conditions have been
proposed recently [1–3]. This subsection introduces the nonlinear Ixz p_ Iyz q_ Iz r_ Iy Ix pq Ixy p2 q2 Ixz qr Iyz pr
aircraft model under asymmetric damage conditions presented in [2]. mxv_ pw ru myrv u_ qw N
Downloaded by MCGILL UNIVERSITY on January 27, 2015 | http://arc.aiaa.org | DOI: 10.2514/1.47996
w_ pv uq pr qx
_ p_ qry p2 q2 z
Nt N0 t Nt (12)
Z=m g cos cos (3)
where L0 , M0 , and N0 represent the moments under nominal
where m is the mass of the aircraft; X, Y, and Z are body-axis conditions, and L, M, and N are the uncertain moment
aerodynamic forces; and are Euler pitch and roll angle of the variations due to damage. The force and moment equations (1–3) and
aircraft body axes with respect to the reference axes; u, v and w are (7–9) are much more complicated than their counterparts without
the body-axis velocity components of the origin of the body-axis damage.
frame; p, q, and r are the body-axis components of the angular
velocity; x; y; zT (zero for undamaged aircraft) are the
3. Controls
coordinates of the center of gravity in the body frame, and they are
assumed to be constant between damage events. The engine thrusts In the above aircraft dynamic equations, only engine thrusts are
on two sides are denoted as TL and TR , and they are assumed to be explicitly included. Other control surfaces (including elevator,
parallel to the body x axis of the aircraft. The mass of the aircraft m is aileron, and rudder) will alter the aerodynamic forces and moments
subject to uncertain change due to the damage. The aerodynamic X, Y, Z, L, M, and N. If the aircraft shape is symmetric about the x-z
forces X, Y, and Z can be expressed as plane, then lateral aerodynamic force and moments Y, L, and N
depend on aileron and rudder deflections. Longitudinal forces and
Xt X0 t Xt (4) moment X, Z, and M depend on the elevator deflection. However,
when the aircraft shape is not symmetric due to damage, the
dependency of aerodynamic forces and moments on the controls is
no longer decoupled. Any aerodynamic force or moment may
Yt Y0 t Yt (5) depend on both longitudinal control surface (elevator deflection) and
lateral control surfaces (aileron and rudder deflections). A generic
case would be that X, Y, Z, L, M, and N are functions of u, v, w, p, q,
r, e , a , and r . The engine thrust T is not only a function of engine
Zt Z0 t Zt (6) throttle t , but also depends on other factors, including the airspeed,
air temperature, and pressure [7,8]. The explicit expression of the
where X0 , Y0 , and Z0 represent the aerodynamic forces under engine thrust with respect to these factors can be found in [8], and
nominal conditions, and X, Y, and Z are uncertain variations partial derivatives of the engine thrust with respect to aircraft velocity
due to the wingspan changes caused by damage. A short time interval and engine throttle can be found in [7]. For the study of aircraft under
is considered for aircraft modeling so that the fuel consumption is steady-state conditions, it is assumed that the air temperature and
negligible. pressure do not change, and the thrust only depends on engine
throttle setting t and aircraft velocity u, v, and w.
2. Moment Equations
In standard aircraft modeling, the mass distribution of the aircraft 4. Kinematic Equations
is assumed to be symmetric about x-z plane, so that the cross products The relation between the rates of Euler angles and the angular
of inertia Ixy and Iyz are zero [7]. However, with asymmetric mass velocities p, q, and r is given by
1502 LIU, TAO, AND JOSHI
_ p tan q sin r cos (13) engine thrust (note that they are the control inputs and not their
perturbations). These control inputs act on the aircraft through the
forces X, Y, and Z and moments L, M, and N (whose expressions are
_ q cos r sin (14) usually complicated). Let x, y. and z denote the change in the
center-of-mass location due to damage.
The elements in fx; U are expressed as
_ q sin r cos (15) f1 qw rv X=m g sin TL TR =m q2 r2 x
cos
pqy prz
The force equations (1–3), moment equations (7–9), and the
kinematic equations (13–15) will be used in the model linearization f2 uq pv Z=m g cos cos prx qry
study in the next subsection. p2 q2 z
B. Linearized Aircraft Model Under Asymmetric f3 M mg sin z mg cos cos x zTL TR
Damage Conditions mxqu pv mzqw rv Ix Iz pr Ixy qr
In this subsection, the linearization of the aircraft model with
damage is conducted. The dynamic coupling in the linearized aircraft Ixz r2 p2 Iyz pq
model in the presence of damage is examined, and a generic aircraft f4 q cos r sin
model structure under damage conditions is obtained.
Downloaded by MCGILL UNIVERSITY on January 27, 2015 | http://arc.aiaa.org | DOI: 10.2514/1.47996
Here, u, v, and w represent the velocity perturbations along body
2 3 axes; p, q, and r are the angular velocity perturbations; , , and
0 0 0 0
6 my 0 Iyz 07 are the pitch, roll, and yaw angle perturbations; e , a , and
6 7 r are the deflection perturbations of the elevator, aileron, and
M 21 6 0 my Ixy 07
6 7 rudder; and tl and tr are the left and right throttle pertur-
4 0 0 0 05
bations. For simplicity of notation, in the rest of this paper, the
0 0 0 0 symbol is dropped, and x and U are used to represent their
2 3
1 x z 0 0 deviations from the equilibrium states xo and Uo ; that is,
6 mx Iz Ixz 0 07
6 7 x u w q v r p T (24)
22 6 mz Ixz
M Ix 0 07 (21)
6 7
4 0 0 0 1 05
0 0 0 0 1
U e tl tr a r T (25)
In the control input vector U, e , a , and r are the deflections of the are used to represent the perturbations (deviations from an
elevator, aileron and rudder, and tl and tr are the left and right equilibrium point) of the system state and control vectors.
LIU, TAO, AND JOSHI 1503
This study considers the postdamage equilibrium x0 ; U0 0 X=m g sin o T=m (34)
T
xo uo wo 0 o vo 0 0 o o
T (26)
Uo eo to ao ro 0 Z=m g cos o (35)
where all angular velocities (p, q, and r) are zero. In this study,
independently adjustable engine thrusts are not considered. So
TL TR , and the total engine thrust can be denoted as T TL TR .
Such an equilibrium choice satisfies that M x_ o fxo ; Uo 0: 0 M mg sin o z mg cos o x zT (36)
i.e.,
0 X=m g sin o TL TR =m (27)
0 Y=m (37)
0 Z=m g cos o cos o (28)
0 L mg cos o y (39)
0 Y=m g cos o sin o (30) In this case, the existence of a solution implies that a set of control
signals can balance the aircraft to a wings-level steady-state
equilibrium point after damage. The existence of such a solution is
crucial for a control system being able to achieve a safe landing,
0 N mg cos o sin o x mg sin o y TL l y
which is otherwise much more difficult or impossible to achieve if a
TR l y (31) wings-level flight cannot be maintained during landing by some
control inputs.
0 L mg cos o cos o y mg cos o sin o z (32) 4. Linearized Aircraft Model with Damage
The above equations can have some typical solutions depending on The nonlinear equation (16) can be expanded using its Taylor
the location, type, and magnitude of the damage. To investigate the series about the equilibrium point xo ; Uo and keeping only the first-
equilibrium solution of the above equations, it is necessary to study order terms:
the dependence of the aerodynamic forces and moments on system
state and control inputs. As mentioned earlier, when damage occurs, M xo BU
x_ fxo ; Uo Ax Uo (40)
the aerodynamic forces and moments may depend on all linear and
angular velocities as well as the control surfaces. A generic case where
would be that X, Y, Z, L, M, and N are functions of u, v, w, p, q, r, e ,
a , and r . The engine thrust T is a function of u, v, w, and t . The
@f @f
assumed existence of the solution implies that when damage occurs, A ; B (41)
certain control variables can still maintain the system at the equi- @x xo ;Uo @U xo ;Uo
librium state.
Although engine thrusts are assumed to be symmetric in this study, Let x xo x and U Uo U, so that
as asymmetric damage occurs, a yaw moment will still be generated
by engine thrusts due to the lateral center-of-gravity shift. Based on a x_ o x
M BU
_ fxo ; Uo Ax (42)
study in [1], the lateral center-of-gravity shift is comparatively small
and not more than 2.5% of the semiwingspan, even when half of the
x_ o fxo ; Uo , one can
By subtracting the equilibrium solution M
wing is lost. Therefore, the asymmetric yaw moment generated by
symmetric engine thrusts is comparatively small, and it can be have
assumed that the rudder can compensate for the additional yaw
moment. It should be noted that the damage compensating capacity M BU
x_ Ax (43)
would be largely enhanced if engine thrusts can be independently
adjustable. The engine thrusts can then assist the rudder to compen- which is a linear differential equation approximating the motion of
sate for the asymmetric yaw moment. Eq. (16) near the equilibrium point xo ; Uo . Equation (43) can then
be expressed as
3. Wings-Level Flight
A particular case is of interest in which the damage is relatively x_ Ax BU (44)
small. As a special situation, we consider the case when a wings-level
equilibrium with v0 0 0 0 is possible, in addition to zero
where A ≜ M 1 A and B ≜ M
1 B.
angular rates. This special case will capture the key features of the
As mentioned for Eqs. (24) and (25), using x and U to replace x
linearized aircraft model under damage in a mathematically tractable
and U as a simple notation, the linearized aircraft model about the
manner. This case corresponds to the following xo and Uo ,
equilibrium in Eq. (26) can be derived with the following structure:
xo uo wo 0 o 0 0 0 0 0 T 1 1
(33)
A4 4
BU
x_ Ax A 2
4 5 x B4 3 B 2
Uo eo to ao ro T M 4 2 U (45)
3
A 5 4 4
A5 5 B 3
5 3 B 4
5 2
x_ o fxo ; Uo 0, which can be
and they satisfy the equation M
expressed as where
1504 LIU, TAO, AND JOSHI
2 @f @f1 @f1
3 2 3
@u
1
@w @q
g cos o a11 a12 a13 a14 a15 a16 a17 a18 0
6 @f2 @f2 @f2 7 6 a21 a22 a23 a24 a25 a26 a27 a28 07
6 @u g sin o cos o 7 6 7
A 1 6 @f @w @q 7 6 a31 a32 a33 a34 a35 a36 a37 a38 07
4 3 @f3 @f3 @f3 5 6 7
@u @w @q @ 6 0 0 coso 0 0 sino 0 0 07
0 0 cos o 0 6 7
2 3 A 6
6 a51 a52 a53 a54 a55 a56 a57 a58 077
0 vo 0 0 0 6 a61 a62 a63 a64 a65 a66 a67 a68 07
6 7
60 0 vo g cos o sin o 0 7 6 a71 a72 a73 a74 a75 a76 a77 a78 07
A 2 6
40
7 6 7
@f3
@r
@f3
@p
@f3
@
05 4 0 0 tano sino 0 0 tan0 coso 1 0 05
0 sin o 0 0 0 0 0 sino =coso 0 0 coso =cos0 0 0 0
2 3 2 3
0 0 0 g sin o sin o b11 b12 b13 b14
@f6 @f6 @f6 @f6
6 @u @w 7 6 b21 b22 b23 b24 7
6 @q @ 7 6 7
A 3 6 @f7 @f7 @f7 @f7 7 6 b31 b32 b33 b34 7
6 @u w @q @ 7 6 7
4 0 0 tan o sin o 0 5 6 0 0 0 0 7
6 7
0 0 sin o = cos o 0 B 6 7
6 b51 b52 b53 b54 7 (49)
2 @f 3 6 b61 b62 b63 b64 7
5 @f5 @f5
g cos o cos o 0 6 7
@v @r @p 6 b71 b72 b73 b74 7
6 @f6 @f6 @f6 @f6 7 6 7
6 @v 07 4 0 0 0 0 5
6 @r @p @ 7
A 6 @f7
4
@f7 @f7 @f7
07 0 0 0 0
6 @v 7
Downloaded by MCGILL UNIVERSITY on January 27, 2015 | http://arc.aiaa.org | DOI: 10.2514/1.47996
@r @p @
4 0 tan o cos o 1 0 05
0 cos o = cos o 0 0 0 Here, it is assumed that the two engines have identical settings, so
2 @f @f @f 3 that tl and tr are combined: tl tr t . The control vector U can
1 1 1
@ @tl @tr thus be expressed as
6 @fe2 7
6 0 0 7
1
B 6 @e 7; B 2 0 U t e a r T (50)
4 @f3 @f3 @f3 5
@e @tl @tr
0 0 0 For a damage-free system, many terms in the linearized model
2 3 2 @f @f5
3 1 would be block diagonal, and so
would be zero. Specifically, M
5
0 0 0 @ @r would A and B.
60 @f6 @f6 7 6 @fa6 7
@f6
6 @tl @tr 7 6 @a 7
6 7
@r
B 3 6
60 0 0 7
7; B 4 6 @f7 7
@f7
6. Wings-Level Model
40 6 @a 7 (46)
0 0 5 4 0
@r
0 5 The linearized model for wings-level steady-state flight and its
0 0 0 0 0 generic structure can be obtained following the same procedure.
Corresponding to the wings-level steady-state equilibrium, the
model linearized about this equilibrium has the following structure:
1 1
All of the derivatives in A and B are evaluated at the equilibrium point A 2 B 2
x0 ; U0 . They are subject to uncertain changes due to the unknown M BU
x_ Ax A4 4 3
4 5 x B4 3
4 3
4 2 U
4 (51)
A 5 4 A 5 5 B 5 3 B 5 2
changes in the aircraft mass, moments of inertia, and aerodynamic
forces and moments due to damage. where
2 @f @f1 @f1
3
@u
1
@w @q
g cos o
6 @f2 @f2 @f2 7
6 @u g sin o 7
A 1 6 @f @w @q 7; A 2 0
5. Generic Structure 4 3 @f3 @f3 @f3 5
@u @w @q @
Generically, the linearized aircraft model is 0 0 1 0
2 3
0 0 0 0
6 @f 6 @f6 @f6 @f6
7
x_ M M
1 Ax 1 BU
Ax BU (47) 6 @u @w @q @ 7
3 6 @f
A 6 @u6 @f6 @f7 @f77
@w @q @ 7
4 0 0 0 0 5
0 0 0 0
with A M 1 A and B M The structure of M
1 B. 1 can be 2 @f5 @f5 @f5 3
determined as follows: @v @r @p
g cos o 0
6 @f6 @f6 @f6 @f6
07
6 @v @r @p @ 7
6 7
A 6 @f7
4 @f7 @f7 @f7
07
2 3 6 @v @r @p @ 7
m11 m12 m13 0 m15 m16 m17 0 0 4 0 tan o 1 0 05
6 m21 m22 m23 0 m25 m26 m27 0 07 0 1
0 0 0
6 7 cos o
6 m31 m32 m33 0 m35 m36 m37 0 07 2 @f @f 3
6 7 1 1 @f1
6 0 0 0 1 0 0 0 0 07 @ @tl @tr
6 7 6 @fe2 7
M 6
1
6 m51 m52 m53 0 m55 m56 m57 0 077 (48) B 1
6 @e
6 @f
0 0 7
7; B 2 0
6 m61 m62 m63 0 m65 m66 m67 0 07 4 3 @f3 @f35
6 7 @e @tl @tr
6 m71 m72 m73 0 m75 m76 m77 0 07 0 0 0
6 7
4 0 0 0 0 0 0 0 1 05 2 3 2 @f @f5
3
5
0 0 0 0 0 0 0 0 1 0 0 0 @ @r
60 @f6 @f6 7 6 @fa6 @f6 7
6 @tl @tr 7 6 @a 7
6 @f @r 7
B 3 6
60 0 0 7 7; B 4 6 7 @f7 7 (52)
40 6 @a 7
0 0 5 4 0
@r
0 5
with mij (i, j 1; 2; . . . ; 9) being nonzero terms. Thus, the structure
0 0 0 0 0
of A and B can obtained as
LIU, TAO, AND JOSHI 1505
All of the derivatives in A and B are evaluated at the equilibrium point contributes to the coupling terms, and other components of the
x0 ; U0 , and they are also subject to uncertain changes due to the center-of-gravity shift do not appear in A 3 when the aircraft is in the
unknown changes in the presence of damage. neighborhood of a wings-level flight. The signs of these terms
depend on the location and nature of the damage and nominal flight
7. Damage-Related Linear Dynamics Derivatives conditions. It should be noted that although the center-of-gravity
We next perform a closer review the coupling terms in the shift components x and z do not appear in the coupling terms,
linearized wings-level aircraft model. These coupling terms in A 3 they contribute to uncertain variations in A 1 and A 4 , whose terms
can be expressed as also vary with the changes of the aerodynamic characteristics due to
damage.
@f6 @TL @TR @N The coupling terms in B 3 can be expressed as
l y l y (53)
@u @u x0 ;U0 @u x0 ;U0 u @f6 @TL
l y (65)
@tl @tl x0 ;U0
@f6 @TL @TR @N
l y l y (54)
@w @w x0 ;U0 @w x0 ;U0 w
@f6 @TR
l y (66)
@tr @tr x0 ;U0
@f6 @N These terms are from the additional moments generated by thrusts
mw0 y (55)
Downloaded by MCGILL UNIVERSITY on January 27, 2015 | http://arc.aiaa.org | DOI: 10.2514/1.47996
where m s is a triangular and stably invertible polynomial matrix, 3) If some rows of En1 are nonzero, the elements in m s in those
and Kp is finite and nonsingular. rows have relative degree 1, and the other rows have relative degree 2.
For multivariable MRAC design, it is important to answer the Kp consists of nonzero rows from En1 (in the same order) and the
following questions: Will the infinite zero structure of the aircraft other rows are the nonzero rows from En2 .
system change when system matrices A and B have uncertain
changes due to damage? Will the signs of principal minors of the B. Case I: m s and Kp for Input u e ; r T and Output
high-frequency gain matrix change? To investigate these key plant y ; T
properties, several case studies will be conducted in the next section. For this case study, the inputs are aileron e and rudder r , and the
outputs are pitch angle and yaw angle .
III. Infinite Zero Structure and High-Frequency
Gain of Linearized Models 1. m s and Kp Under Nominal Conditions
In this section, the results of several case studies based on the For undamaged aircraft, the system matrices A0 and B0 as in [7],
linearized aircraft model will be presented. It will be shown that the En1 , and En2 can be obtained as
infinite zero structure of the aircraft systems remains the same in
the presence of damage, whereas the signs of the high-frequency gain En1 CB0 0 (72)
matrix could change for certain cases.
To determine the infinite zero structure for these systems, it is
b032 0
important to verify the relative degrees for the elements in the transfer En2 CA0 B0 n1 CB0 0 1
b (73)
cos 0 064
matrices. In this study, a practical algorithm [9] will be employed for
Downloaded by MCGILL UNIVERSITY on January 27, 2015 | http://arc.aiaa.org | DOI: 10.2514/1.47996
calculating the infinite zero structure and high-frequency gain matrix Thus, it can be determined that all rows of Gs have a relative degree
conveniently. 2. So m s and Kp can be obtained as
The cases for the modeling study include two two-input/two-
output and one three-input/three-output systems. The cases studied s 12 0
are based on the linearized models about the generic and wings-level m s (74)
0 s 12
steady-state equilibria. The inputs and outputs are summarized as
follows:
1) Case I: The inputs are aileron e and rudder r, the outputs are b032 0
pitch angle and yaw angle , and the matrix C is expressed as Kp 0 1
b064 (75)
cos 0
0 0 0 1 0 0 0 0 0 The elements of Kp are as follows: b032 < 0 is the control gain from
C (69)
0 0 0 0 0 0 0 0 1 elevator to pitch acceleration and b064 < 0 is the control gain from
rudder to yaw acceleration.
2) Case II: The inputs are engine throttle t and rudder r, and the
outputs are pitch angle and yaw angle with the same C in Eq. (69).
3) Case III: The inputs are engine throttle t , elevator e, and rudder 2. m s and Kp Under Damage Conditions
r , and the outputs are forward velocity u, pitch angle , and yaw For this model, En1 and En2 can be obtained as
angle , with C in Eq. (69), and with C being
En1 CB 0 (76)
2 3
1 0 0 0 0 0 0 0 0
C 40 0 0 1 0 0 0 0 05 (70) En2 CAB n1 CB
0 0 0 0 0 0 0 0 1
cos 0 b32 sin 0 b62 cos 0 b34 sin 0 b64
sin 0 b cos 0 b sin 0
b cos 0
b (77)
cos 0 32 cos 0 62 cos 0 34 cos 0 64
A. Algorithm for Determining the Infinite Zero Structure and
High-Frequency Gain The rows of Gs have a relative degree 2. So m s and Kp can be
Given the system matrices A, B, and C, the transfer matrix can be obtained as
calculated as
s 12 0
m s 2 (78)
1 0 s 1
Gs E sn1 En2 sn2
E1 s E0 (71)
s n1
cos 0 b32 sin 0 b62 cos 0 b34 sin 0 b64
where Kp sin 0 b cos 0 b sin 0 (79)
cos 0 32 cos 0 62
b cos
cos 0 34
0
b
cos 0 64
n n1
s detsI A s n1 s
1 s 0 ;
En1 CB; where b32 and b34 are the gains from elevator and rudder to the pitch
En2 CAB n1 CB; acceleration, and b62 and b64 are the gains from elevator and rudder to
.. the yaw acceleration.
.
E1 CAn2 B n1 CAn3 B
2 CB; 3. m s and Kp for Wings-Level Flight
E0 CAn1 B n1 CAn2 B
1 CB For this flight condition, 0 0, and En1 and En2 can be
obtained as
For aircraft models without actuator dynamics, the relative degree of
a transfer function is typically 1 (from a control input to a linear or En1 CB 0 (80)
angular velocity) or 2 (from a control input to an angle). Thus, En1
and En2 in Eq. (71) can be evaluated for determining m s and Kp in
this study. The details of this algorithm are given below. b32 b34
En2 CAB n1 CB 1
b 1
b (81)
1) If all of the rows of En1 are nonzero, the elements in m s all cos 0 62 cos 0 64
have relative degree 1, and Kp En1 .
2) If En1 is zero and all of the rows of En2 are nonzero, then the The rows of Gs have a relative degree 2. So m s and Kp can be
degree of the elements in m s is 2, and Kp En2 . obtained as
LIU, TAO, AND JOSHI 1507
s 12 0 where b31 and b34 are the gains from engine throttle and rudder to
m s (82) pitch acceleration, and b61 and b64 are the gains from engine throttle
0 s 12
and rudder to yaw acceleration.
b32 b34
Kp 1
b 1
b (83)
cos 0 62 cos 0 64 3. m s and Kp for Wings-Level Flight
For this special case, En1 and En2 can be calculated as
s 12 0
T m s (92)
C. Case II: m s and Kp for Input u t ; r and Output 0 s 12
y ; T
For this case, the inputs are engine throttle t and rudder r, and the
outputs are pitch angle and yaw angle .
1. m s and Kp Under Nominal Conditions b31 b34
Kp 1
b 1
b (93)
For this case, En1 and En2 can be calculated as cos 0 61 cos 0 64
En1 CB0 0
where b31 and b34 are the gains from engine throttle and rudder to
b031 0 pitch acceleration, and b61 and b64 are the gains from engine throttle
En2 CA0 B0 n1 CB0 CA0 B0 0 1
b064
cos 0 and rudder to yaw acceleration.
Thus, all transfer functions have a relative degree 2, and m s and Kp
can be obtained as
4. Discussion
s 12 0 For this case, the infinite zero structure does not change with the
m s 2 (84) damage. By examining Kp in Eqs. (85) and (93), it can be seen that
0 s 1
there are two situations when the signs of the principal minors may
change:
b031 0 1) If the gain from the engine throttle to the pitch acceleration
Kp 0 1
b (85)
cos 0 064 changes from positive (no damage) to negative (with damage), the
sign change will occur. The physical meaning is, for example, that the
The elements in Kp are as follows: b031 > 0 is the control gain from engine thrust causes a nose-down moment due to vertically down-
engine throttle to pitch acceleration, and b064 < 0 is the control gain shifted center of gravity under damage, instead of a nose-up moment
from rudder to yaw acceleration. The determinant of Kp under under the nominal condition.
nominal conditions is typically negative. 2) If b31 and b64 do not change signs after damage, and b34 (i.e.,
control gain from rudder to pitch acceleration) assumes a positive
2. m s and Kp Under Damage Conditions value, then det KP > 0 when
For the model based on the generic equilibrium, En1 and En2 can
be obtained as b31 b64
b61 < <0 (94)
En1 CB 0 (86) b34
En2 CAB n1 CB CAB The physical meaning of this situation is that two identical engine
thrusts would generate a moment about the negative z axis with the
b31 cos 0 b61 sin 0 b34 cos 0 b64 sin 0 center of gravity shifting to the left (i.e., right wing damage). The
1 b sin b cos 1
b34 sin 0 b64 cos 0 condition for b61 can be derived similarly with a negative b34 , which
cos 0 31 0 61 0 cos 0
corresponds to left wing damage.
(87)
Kp in Eq. (89) is more complicated than that in Eq. (93), because of
and m s and Kp can be obtained as the nonzero 0 . However, it is reasonable to assume that 0 is small.
Thus, Kp in Eq. (89) would be close to Eq. (93), and the sign change
s 12 0 cases would also be similar.
m s (88)
0 s 12
2 3 2 P7 P7 P7 3
b11 b12 b14 i1 a1i bi1 n1 b11 i1 a1i bi2 n1 b12 i1 a1i bi4 n1 b14
En1 CB 4 0 0 0 5; En2 CAB n1 CB 4 b31 b32 b34 5
1 1 1
0 0 0 b
cos 0 61
b
cos 0 62
b
cos 0 64
1. m s and Kp Under Nominal Conditions The transfer functions in the first row of Gs have a relative degree 1,
For this case, En1 and En2 can be obtained as and the transfer functions in the second and third rows have a relative
degree 2. So m s and Kp can be obtained as
2 3
b011 b012 0 2 3
En1 CB0 4 0 0 05 s1 0 0
0 0 0 m s 4 0 s 12 0 5 (97)
2
0 0 s 1
En2 CA0 B0 n1 CB0
2 P3 P3 3
i1 a01i b0i1 n1 b011 i1 a01i b0i2 n1 b012 0
Downloaded by MCGILL UNIVERSITY on January 27, 2015 | http://arc.aiaa.org | DOI: 10.2514/1.47996
4 b031 b032 0 5 2 3
1 b11 b12 b14
0 0 b
cos0 064 Kp 4 b31 b32 b34 5 (98)
1 1 1
b
cos 0 61
b
cos 0 62
b
cos 0 64
The transfer functions in the first row of Gs have a relative degree 1,
and the transfer functions in the second and third rows have a relative
degree 2. So m s and Kp can be obtained as
2 3
s1 0 0 4. Discussion
m s 4 0 s 12 0 5 (95) The infinite zero structure m s is also invariant after damage. If
0 0 s 12 b31 < 0 and the following condition is satisfied,
b11 b32
b31 < <0 (99)
2 3 b12
b011 b012 0
Kp 4 b031 b032 0 5 (96)
0 0 1
b064
a sign change would occur. The physical meaning is, for example,
cos 0
that if the center of gravity shifts downward and crosses the path of
the engine thrust, the engine thrust causes a nose-down moment,
The elements in Kp are as follows: b011 > 0 is the control gain from instead of a nose-up moment under the nominal condition.
engine throttle to forward acceleration, b012 > 0 is the control gain The major difference between this case and the previous one is that
from elevator to forward acceleration, b031 > 0 is the control gain the sign change in case II only requires that b31 < 0, whereas in
from engine throttle to pitch acceleration, b032 < 0 is the control case III, b31 is not only required to be negative, but also has a
gain from elevator to pitch acceleration, and b064 < 0 is the control magnitude constraint characterized by Eq. (99). Although the
gain from rudder to yaw acceleration. situation may be physically meaningful, the downward shift of the
The signs of the principal minors can thus be determined as center of gravity would likely be small, thus resulting in no sign
1 > 0, 2 < 0, and 3 > 0. change.
2 3 2 P7 P7 P7 3
b11 b12 b14 i1 a1i bi1 n1 b11 i1 a1i bi2 n1 b12 i1 a1i bi4 n1 b14
En1 CB 4 0 0 0 5; En2 CABn1 CB 4 b31 cos0 b61 sin0 b32 cos0 b62 sin0 b34 cos0 b64 sin0 5
1 1 1
0 0 0 cos0
b31 sin0 b61 cos0 cos0
b 32 sin0 b 62 cos0 cos0
b34 sin0 b64 cos0
2 3 2 3
s1 0 0 b11 b12 b14
m s 4 0 s 12 0 5 Kp 4 b31 cos0 b61 sin 0 b32 cos 0 b62 sin 0 b34 cos 0 b64 sin 0 5
1 1 1
0 0 s 12 cos0
b31 sin 0 b61 cos0 cos0
b32 sin 0 b62 cos0 cos 0
b34 sin 0 b64 cos0
LIU, TAO, AND JOSHI 1509
Proposition 1: For all of the cases studied so far, the asymmetric For the adaptive control design, the following assumptions will be
aircraft damage does not change the infinite zero structure of the considered:
linearized aircraft model. Assumption 1: The parameter matrices A, B, and C are piecewise-
Explanation: A close examination of the case studies shows that constant, with a finite number of unknown and constant jumps Ai , Bi ,
the relative degrees of the transfer functions in any row of Gs do not and Ci (i 1; 2; . . . ; N).
change, and Kp for both undamaged and damaged aircraft systems So there are up to N 1 occurrences of damage in the system. For
remain nonsingular. The condition for the relative degrees to change each value (Ai , Bi , and Ci ), the transfer matrix can be defined as
is that a row in En1 or En2 suddenly changes from a nonzero row Gi s Ci sI Ai 1 Bi .
vector to zero. Considering the structures of A and B under damage Assumption 2: All zeros of Gi s are stable.
conditions (with nonzero terms appearing), based on which En1 and Assumption 3: An upper bound on the observability index of
En2 are calculated, it is not likely to have a row changing from Gi s is known.
nonzero to zero. More specifically, it is not likely that a row in the Assumption 4: Gi s is strictly proper with full rank and has a
matrix CB or CAB would change from nonzero to zero after damage known modified interactor matrix m s such that lims!1 m s
(based on both generic model structure and linearized models of the Gi s Kpi , the high-frequency gain matrix of Gi s, is finite and
NASA generic transport model (GTM). Thus, the key condition on nonsingular.
1
the infinite zero structure for multivariable MRAC would most likely Assumption 5: Wm s m s.
be satisfied under damage conditions. Assumption 6: All leading principal minors of the matrix Kpi are
However, system damage may pose other challenges for multi- nonzero and their signs are known and the same for each i.
variable MRAC designs. First, the case studies show that for some The basic assumptions for multivariable MRAC are as follows:
cases under certain damage conditions, the signs of the principal The plants need to be minimum phase. A common m s is needed for
Downloaded by MCGILL UNIVERSITY on January 27, 2015 | http://arc.aiaa.org | DOI: 10.2514/1.47996
minors of the high-frequency gain matrix may change. Therefore, all Gi s, and signs of the leading principal minors of Kpi do not
MRAC designs that do not depend on the signs of principal minors of change. These assumptions allow a standard multivariable MRAC
Kp are preferable and should be studied. Second, large parameter scheme to handle piecewise-constant parameter variations that can
variations can occur due to damage, and technical issues (including occur due to damage. It should be noted that discontinuous system
stability and output tracking) under these large parameter uncer- parameter changes are allowed, and such jumps will not cause insta-
tainties need to be studied. bility. From Proposition 1 in Sec. III.E, it can be seen that
Assumption 4 can be satisfied based on the case studies, and
Assumption 6 can be satisfied under certain damage conditions.
IV. Multivariable Model Reference Adaptive
Control Design
B. Plant-Model Matching Controllers
From the case studies, it can be seen that in some cases (especially
For model reference adaptive control design, a nominal model
lateral dynamics in the models), sign changes in high-frequency gain
reference controller is needed that achieves the desired control
matrix may not occur when damage is present. In this section, a
objective when the system parameters A, B, and C are known.
multivariable model reference adaptive control design is presented
Parameters of such a controller, which are unknown, are also used in
for aircraft with damage when there are no sign changes in the high-
deriving an error model needed for adaptation of an adaptive
frequency gain matrix.
controller.
Since the system parameters A, B, and C may take any of the values
A. Problem Formulation Ai , Bi , and Ci (i 1; 2; . . . ; N), there is a set of corresponding
Consider a linear system of the form nominal controllers, and each controller has the structure
_ Axt BUt; yt Cxt
xt (100) U t T T
1 !1 t 2 !2 t 20 yt 3 trt (104)
where A 2 Rn n , B 2 Rn m , and C 2 Rm n are unknown parameter where !1 t FsU t, !2 Fsyt, Fs AF s=s,
matrices, and xt 2 Rn , Ut 2 Rm , and yt 2 Rm are the state, AF s I; sI; . . . ; s2 IT , and s is a monic stable polynomial of
input, and output vectors. To represent an aircraft model, A and B can degree 1, with the upper bound on the observability indices of
be expressed as Gi s. The nominal parameters 1 11 ; . . . ; 11 T
, 2
T m m
, 20 , 3 , and ij 2 R
21 ; . . . ; 21 (i 1, 2 and
A A0 A (101) j 1; . . . ; 1) are for plant-model matching and are derived next.
The following notation,
B B0 B (102) Gi s Ci sI Ai 1 Bi Zi sP1
i s (105)
where A0 and B0 are the nominal parameter matrices for aircraft is introduced for some m m right coprime polynomial matrices
without damage, and A and B contain the unknown coupling Zi s and Pi s, with Pi s being the column proper (i
terms and parameter changes caused by damage (they are zero when 1; 2; . . . ; N).
there is no damage). With the specification of s, m s, Pi s, and Zi s, there exist
The objective is to design a control vector Ut such that the plant 1 , 2 , 20 , and 3 Kpi
1
such that
output yt tracks a given reference output
T T
1 AF sPi s 2 AF s s20 Zi s sPi s
ym t Wm srt 2 Rm (103)
3 m sZi s
for a stable m m transfer matrix Wm s and a bounded reference
signal rt 2 Rm , despite the unknown damage. The designed control (106)
signal Ut is applied to the linearized aircraft system. The above Since s and Zi s are stable, we can have the plant-model transfer
reference model is written in its operator form, where ym t≜ matrix matching equation:
L1 Wm sRs, with L1
being the inverse Laplace transform
operator and Rs being the frequency-domain counterpart of rt. I T T 1
1 Fs 2 FsGi s 20 Gi s 3 Wm sGi s
The transfer function matrix Wm s in Eq. (103) can be seen as the (107)
operator that maps from rt to ym t.
Such a notation is instrumental in representing an adaptive control from which the plant-model matching parameters 1 , 2 , and 20
system, involves both linear time-invariant dynamics and nonlinear can be determined with 3 Kpi 1
. For each Ai , Bi , and Ci
time-varying dynamics, and will be frequent hereafter. (i 1; 2; . . . ; N), a set of constant parameters j (j 1, 2, 20, 3)
1510 LIU, TAO, AND JOSHI
C. Adaptive Control Scheme With the LDS decomposition in Eq. (108), Eq. (112) can be
To design the adaptive control scheme, a high-frequency gain expressed as
decomposition will be first introduced. As shown in [10,11], such a
decomposition-based design relies less on the a priori knowledge of
Ds SUt T T
1 !1 t 2 !2 t 20 yt 3 rt
the high-frequency gain matrix. Then a model reference adaptive
control design is proposed, and the desired stability and tracking L1
s m sy ym t (113)
properties are established.
From Eq. (113) and the adaptive controller (110), the following
1. LDS Decomposition of Kp equation can be obtained:
Let i (i 1; 2; . . . ; m) denote the leading principal minors of the
high-frequency gain matrix Kp 2 Rm m and assume that i ≠ 0 ~ T t!t
(i 1; 2; . . . ; m). The gain matrix Kp then has a nonunique m sy ym t 0 m sy ym t Ds S (114)
decomposition [10,11]:
where Ds is in Eq. (109); S ST > 0 in Eq. (108); t ~
Downloaded by MCGILL UNIVERSITY on January 27, 2015 | http://arc.aiaa.org | DOI: 10.2514/1.47996
0
Ds diagfs1 ; s2 ; . . . ; sM g
6
21 0 0
07
6 7
2 m 6
31
32 0
07
diag sign1 1 ; sign ; . . . ; sign (109) 6 7
1 2 m1 m 0 6 .. .. .. .. .. 7 (115)
6 . . . . .7
6 7
4
m1m2 0 05
such that i > 0 (i 1; . . . ; m) may be arbitrary. All Kpi can have the m11
same Ds based to Assumption 6. m1
mm2 mm1 0
Ut T T 1
1 !1 t 2 !2 t 20 yt 3 Wm syt (111) Choose fs as a stable monic polynomial whose degree is equal to
the maximum degree of m s. Introduce the filter hs 1=fs;
which, together with the reference model (103) and Assumption 5, define the filtered tracking error,
yields
m shset e 1 t; . . . ; e m tT
et (118)
Fig. 1 Block diagram of the model reference adaptive control system. Thus, we have
LIU, TAO, AND JOSHI 1511
where t ≜ T t hsT !t and t is the estimate of . i Ti > 0 (i 2; 3; . . . ; m), and T > 0.
An estimation signal can then be defined as
5. Stability Analysis
t et ^ et
et 0; 2T 2 t; 3T 3 t; . . . ; m
T
m tT To demonstrate the stability of the closed-loop system, a
piecewise-continuous Lyapunov function can be used. Based on
tt (123)
Assumption 1, there are N 1 finite jumps due to the damage and
totally N choices of Ai , Bi , and Ci . Assuming that the asymmetric
which leads to damage occurs at time instant tj (j 1, 2; . . . ; N 1), the following
Lyapunov-like function can be chosen:
t 0; ~2 t2 t; ~3 t3 t; . . . ; ~m tm tT
T T T
X
m
T
hsT !t T t tt V ~i 1 ~ ~ T 1 ~ ~ ~T
i i tr trS (129)
T T T i2
0; ~2 t2 t; ~3 t3 t; . . . ; ~m tm tT T t
t T t tt for time intervals tj1 ; tj (j 1; . . . ; N), with t0 0 and tN 1.
Because of the changes in the system parameters after the damage
T T T
0; ~2 t2 t; ~3 t3 t; . . . ; ~m tm tT (which are finite) and the finite jumps of the nominal parameters,
there would be a finite jump in V for each jump in the system
~ T t tt
Ds St ~ (124) parameters Ai , Bi , and Ci : i.e.,
1.5
1
Tracking error of pitch angle θ (deg)
0.5
−0.5
−1
−1.5
−2
−2.5
−3
0 20 40 60 80 100 120 140 160
Time (s)
Fig. 2 Tracking error of pitch angle t.
1512 LIU, TAO, AND JOSHI
10
0
Downloaded by MCGILL UNIVERSITY on January 27, 2015 | http://arc.aiaa.org | DOI: 10.2514/1.47996
−5
0 20 40 60 80 100 120 140 160
Time (s)
Fig. 3 Tracking error of roll angle t.
X
Vt
j Vtj < 1; j 1; 2; . . . ; N 1 (130) 2 m
T
V_ ~ t
i ti t
m2 t i2 i
From the adaptive laws, the time derivative of V in each tj1 ; tj can
be obtained as tr ~
~ T
tT t trtSD T
s
t t
X
m
V_ 2
T
~i 1 _ ~ T 1 _ ~ _T 2
T t
t
i i 2 tr 2trS 0 (131)
i2 m2 t
15
x−axis velocity u (ft/s)
10
−5
0 20 40 60 80 100 120 140 160
Time (s)
1
z−axis velocity w (ft/s)
−1
−2
0 20 40 60 80 100 120 140 160
Time (s)
1
Pitch rate q (deg/s)
0
−1
−2
−3
−4
0 20 40 60 80 100 120 140 160
Time (s)
Fig. 4 Longitudinal states: u, w, and q.
LIU, TAO, AND JOSHI 1513
(j 1; . . . ; N), it follows that V_ 0, which implies that V is The smallness of its loop gain is ensured by the L2 properties of
bounded. With the fact that V only has finite jumps at finite time the adaptive laws. The asymptotic tracking property follows from the
instants, it can be concluded that V is bounded for 0; 1. So it can be complete parameterization of the estimation error equation (123), the
concluded that i t 2 L1 , t 2 L1 , and t 2 L1 . L2 properties, and the signal boundedness of the closed-loop system.
Integrating both sides of Eq. (131) for the time interval tj1 ; tj The piecewise-continuous Lyapunov function technique provides a
yields powerful tool to the stability analysis of MRAC for aircraft with
Zt multiple damage occurrences. Thus far, a multivariable MRAC
T
j 2
scheme has been developed that is applicable to the compensation of
d Vt
j1 Vtj (132)
tj1 m2 asymmetric damage in aircraft system.
For N intervals 0; t1 for t1 ; t2 ; . . . ; tN1 ; 1, the above equation V. Simulation Results
holds. Summing both sides of Eq. (132) for j 1; . . . ; N yields
In the simulation study, the MRAC scheme developed is applied to
Z1 T aircraft models from the NASA GTM model to verify the control
2
d V0 Vt1 Vt
1 Vt2 Vt2 scheme performance. The NASA GTM testbed is a remotely piloted
0 m2
5.5% dynamically scaled model for the investigation of aircraft
Vtj Vt
j Vtj1 Vtj1
VtN1 control under adverse conditions. A nonlinear high-fidelity simu-
X
N1 lation platform based on Simulink is available for GTM, which uses
Vt Vt experimentally validated aerodynamic models extracted from wind-
N1 V1 V0 j Vtj V1 (133)
j1 tunnel data and can simulate several failure and damage scenarios
Downloaded by MCGILL UNIVERSITY on January 27, 2015 | http://arc.aiaa.org | DOI: 10.2514/1.47996
0.4
0.2
0
−0.2
−0.4
0 20 40 60 80 100 120 140 160
Time (s)
3
Yaw rate r (deg/s)
2
1
0
−1
−2
0 20 40 60 80 100 120 140 160
Time (s)
15
Roll rate p (deg/s)
10
5
0
−5
−10
0 20 40 60 80 100 120 140 160
Time (s)
Fig. 5 Lateral states: v, r, and p.
1514 LIU, TAO, AND JOSHI
The plant outputs are chosen as pitch angle and roll angle . The so that
control inputs are the elevator position e and aileron position a .
The nominal model is trimmed and linearized about a wings-level
steady-state flight with a total airspeed of 75.89 kt, and its system Kp0 lim m sG0 s
0:7486 0:0859
(142)
matrices A0 , B0 , and C are s!1 0:00001 0:7675
2 3 2 3
0:019 0:1364 9:7778 32:0829 0:0018 0:0004 0 0 0:0056 0:0423
6 0:2804 2:7567 120:1968 2:42 0:0001 0 0:0004 0:0061 7 6 0:6119 0:1579 7
6 7 6 7
6 0:0205 0:3106 3:5393 0 0:007 0:0328 0:0014 0 7 6 0:7486 0:0859 7
6 7 6 7
6 0 0 1 0 0 0:0002 0 0:0002 7 6 0 0 7
A0 6
6
7;
7 B0 6
6
7
6 0 0:0027 0 0:0005 0:5765 125:9974 10:4690 32:0829 7 6 0 0:0223 7
7
6 0 0 0:0255 0 0:2245 1:4053 0:2794 0 7 6 0 0:0232 7
6 7 6 7
4 0 0 0:0018 0 0:629 1:9689 5:4759 0 5 4 0 0:7657 5
0 0 0 0:0002 0 0:0754 1 0 0 0
0 0 0 1 0 0 0 0
C (139)
0 0 0 0 0 0 0 1
Downloaded by MCGILL UNIVERSITY on January 27, 2015 | http://arc.aiaa.org | DOI: 10.2514/1.47996
The NASA GTM model contains several damage and failure which is finite and nonsingular. Its principal minors are 01
scenarios for simulation study under adverse conditions. In this 0:7486 and 02 0:5745. It can be verified that the observability
study, the damage case with the loss of outboard left wing tip is index of G0 s is 0 4. So 0 can be chosen as 4.
considered. The damaged aircraft model can be trimmed about a It is also verified that the damaged aircraft model Gd s CsI
wings-level steady-state flight at a speed of 75.89 kt. The linearized Ad 1 Bd has stable zeros, is strictly proper, and has a full rank. The
system matrices are as below: same m s in Eq. (141) yields a finite and nonsingular high-
2 3 2 3
0:0197 0:113 9:6802 29:902 0:0111 0:0005 0 0 0:0057 0:0276
6 0:2646 2:6051 120:6413 11:876 0:0629 0 0:0005 0:0086 7 6 0:6219 0:1737 7
6 7 6 7
6 0:0042 0:2005 3:4108 0 0:0026 0:0285 0:0697 0 7 6 0:7379 0:0437 7
6 7 6 7
6 0 0 1 0 0 0:0003 0 0:0002 7 6 0 0 7
Ad 6
6 0:0017
7; Bd 6 7 (140)
6 0:0262 0 0:0034 0:571 125:9696 10:4707 29:902 77
6
6 0 0:0035 7
7
6 0:001 0:0487 0:0119 0 0:229 1:4374 0:2793 0 7 6 0:0040 0:0229 7
6 7 6 7
4 0:0672 0:8106 0:7199 0 0:6811 2:4110 5:0337 0 5 4 0:0879 0:5194 5
0 0 0:0001 0:0003 0 0:3972 1 0:0001 0 0
0.3
0.25
0.2
Selected parameters of Θ(t)
0.15
0.1
0.05
0
Downloaded by MCGILL UNIVERSITY on January 27, 2015 | http://arc.aiaa.org | DOI: 10.2514/1.47996
−0.05
−0.1
0 20 40 60 80 100 120 140 160
Time (s)
Fig. 6 Adaptation of selected parameters in t.
adaptive law in Eq. (126)]. Based on Theorem 1, this adaptive control The tracking error et yt ym t is shown in Fig. 2 (tracking
scheme is able to ensure closed-loop stability and asymptotic output error of the pitch angle) and Fig. 3 (tracking error of the roll angle).
tracking in the presence of unknown parametric changes due to Asymptotic output tracking is eventually achieved for the
damage. The closed-loop structure of the model reference adaptive postdamage system with the adaptive control scheme.
control system can be seen in Fig. 1. The other states of the system are shown in Figs. 4 and 5, which
show the disturbance and recovery of these signals when the damage
occurs.
Selected elements of t and the parameters 2 t and t are
C. Simulation Results shown in Figs. 6–8. The parameters autonomously adapt after the
For the numerical study, the reference input is selected as damage occurs and approach a new set of steady-state values for
rt 0:035; 0:07T , which generates the reference pitch and asymptotic output tracking.
roll angles of 0:5 and 1 deg, respectively. The adaptive control The control signals are shown in Fig. 9. The elevator and aileron
scheme is expected to bring back the system outputs to the reference inputs adjust for the compensation of the damage as a result of the
signals after an uncertain damage occurrence. control parameter adaptation.
1.5
1
θ2(t)
0.5
−0.5
−1
0 20 40 60 80 100 120 140 160
Time (s)
Fig. 7 Adaptation of 2 t.
1516 LIU, TAO, AND JOSHI
0.5
−0.5
−1
Ψ(t)
−1.5
−2
Downloaded by MCGILL UNIVERSITY on January 27, 2015 | http://arc.aiaa.org | DOI: 10.2514/1.47996
−2.5
−3
0 20 40 60 80 100 120 140 160
Time (s)
Fig. 8 Adaptation of t.
2
Elevator angle δe (deg)
−1
−2
0 20 40 60 80 100 120 140 160
Time (s)
4
Aileron angle δa (deg)
−1
−2
0 20 40 60 80 100 120 140 160
Time (s)
Fig. 9 Elevator angle e and aileron angle a .
The results of the adaptive control scheme applied to linearized tracking adaptive control approach, a linearized model of the aircraft
nominal and damaged aircraft models for output tracking demonstrate dynamics with damage was derived and its invariance properties,
that the aircraft can maintain its flight in a neighborhood of the such as infinite zero structure and signs of the high-frequency gain
equilibrium condition (e.g., wings-level steady-state flight), and the matrix, were investigated. These properties are important for the
specified outputs (pitch angle and roll angle in this case) closely design of multivariable model reference adaptive control (MRAC)
track the desired trajectories in spite of the uncertain wing damage. schemes. A multivariable MRAC design is illustrated for the
compensation of aircraft damage. An effective analysis method for
MRAC of piecewise linear systems, using a piecewise Lyapunov
VI. Conclusions function, is employed to show that desired stability and tracking
Adaptive control of aircraft in the presence of structural damage properties are ensured despite jumps in the system parameters due to
has been studied in this paper. To investigate our multivariable output damage. A simulation study of the application of the proposed
LIU, TAO, AND JOSHI 1517
adaptive control scheme to the NASA generic transport model has Control,” AIAA Atmospheric Flight Mechanics Conference, AIAA
been conducted. The simulation results show that the proposed Paper 2007-6717, Hilton Head, SC, 2007.
adaptive control scheme ensures closed-loop stability and desired [5] Liu, Y., and Tao, G., “Multivariable MRAC for Aircraft with Abrupt
Damages,” Proceedings of the 2008 American Control Conference,
tracking performance in the presence of uncertain wing damage.
Inst. of Electrical and Electronics Engineers, Piscataway, NJ, June 2008,
Further study is required in several aspects of this research, including pp. 2981–2986.
design and simulation of adaptive control schemes for systems with [6] Liu, Y., and Tao, G., “Multivariable MRAC Using Nussbaum Gains for
sign changes in the high-frequency gain matrix. Aircraft with Abrupt Damages,” Proceedings of the 47th IEEE
Conference on Decision and Control, Inst. of Electrical and Electronics
Acknowledgments Engineers, Piscataway, NJ, 2008, pp. 1397–1342.
[7] Bryson, A. E., Jr, Control of Spacecraft and Aircraft, Princeton Univ.
This research was supported in part by NASA under grant Press, Princeton, NJ, 1994, pp. 144–148.
NNX08AB99A and the National Science Foundation under grant [8] Stengel, R. F., Flight Dynamics, Princeton Univ. Press, Princeton, NJ,
ECS0601475. 2004, p. 105.
[9] Zheng, D., Linear System Theory, Tshinghua Univ. Press, Beijing,
1999, pp. 29–31.
References [10] Tao, G., Adaptive Control Design and Analysis, Wiley, Hoboken, NJ,
[1] Nguyen, N., Krishnakumar, K., Kaneshige, J., and Nespeca, P., “Flight 2003, pp. 445; 405–412.
Dynamics and Hybrid Adaptive Control of Damaged Aircraft,” Journal [11] Imai, A. K., Costa, R. R., Hsu, L., Tao, G., and Kokotović, P. V.,
of Guidance, Control, and Dynamics, Vol. 31, No. 3, 2008, pp. 751– “Multivariable Adaptive Control Using High-Frequency Gain Matrix
764. Factorization,” IEEE Transactions on Automatic Control, Vol. 49,
doi:10.2514/1.28142 No. 7, 2004, pp. 1152–1156.
Downloaded by MCGILL UNIVERSITY on January 27, 2015 | http://arc.aiaa.org | DOI: 10.2514/1.47996
[2] Bacon, B. J., and Gregory, I. M., “General Equations of Motion for a doi:10.1109/TAC.2004.831134
Damaged Asymmetric Aircraft,” Proceedings of 2007 AIAA Guidance, [12] Jordan, T., Langford, W., Belcastro, C., Foster, J., Shah, G., Howland,
Navigation, and Control Conference, AIAA Paper 2007-6306, Hilton G., and Kidd, R., “Development of a Dynamically Scaled Generic
Head, SC, Aug. 2007. Transport Model Testbed for Flight Research Experiments,” AUVSI’s
[3] Liu, Y., and Tao, G., “Model-Based Direct Adaptive Actuator Failure Unmanned Systems North America Symposium and Exhibition,
Compensation Techniques with Applications to Aircraft Flight Control Anaheim, CA, Aug. 2004.
Systems,” 2006 AIAA Guidance, Navigation, and Control Conference, [13] Jordan, T., Langford, W., and Hill, J., “Airborne Subscale Transport
AIAA Paper 2006-6554, Keystone, CO, Aug. 2006. Aircraft Research Testbed-Aircraft Model Development,” AIAA
[4] Lombaerts, T. J. J., Chu, Q. P., Mulder, J. A., and Joosten, D. A., “Real Guidance, Navigation, and Control Conference, AIAA Paper 2005-
Time Damaged Aircraft Model Identification for Reconfiguring Flight 6432, San Francisco, Aug. 2005.