Documente Academic
Documente Profesional
Documente Cultură
WeD2.3
I. INTRODUCTION
'
zr
P( U )
wp
zp
y
K (U)
Fig. 1. Standard LFT control structure.
y Cy ( U ) Dyw ( U )
0 u
Here, w [ wrT wTp ]T and z
279
WeD2.3
^U (t ) C (
1
,
Q S T wr (t )
0 zr (t ) S R zr (t ) dt t 0 t t 0. (3)
Let Q, S, R be scaling matrices such that Q0, R>0. We
assume that is block-diagonal: =diag(1,, r), wherei
denotes a troublemaking component. Note that the bounded
induced L2 norm performance z p 2 d J wp 2 can also be
t wr (t )
) [0, K ) ]
Qi
Scalings S
i
Ri
diag(G1,!, G ni ) G i (t ) I ni
Qi ni uni
Si SiT 0
Ri Qi
' i ( s)
Qi
Si
Ri
1
qI ni
0
Qi
zp
Qp
Sp
Rp
d J wp
T
Q S T Dc DcT S DcT R 0 (6)
Bc P S Cc
RCc
RDc
R
holds for all U(t)SU, then the closed-loop system (5) is robust
stable against the uncertainty block satisfying the IQC (3),
and has the induced L2 norm performance bound J.
Proof. Given the system (5), we assume a parameter
-dependent quadratic function V(xc)=xTc P(U)xc such that
V wT Qw wT S T z z T Sw z T Rz 0.
(7)
The inequality (7) is rewritten as
t
d
(V ( wT Qw wT S T z zT Sw zT Rz )dW ) 0. (8)
0
dt
Note that the second term is always nonnegative. According
to standard arguments from Lyapunov theory, the system is
stable. Here, the function V decreases to zero, but not
necessarily monotonically. Thus, V is not a Lynpunov
function in the conventional sense. Condition (7) is
equivalent to (6) by Schur complement.
C. Control Synthesis
Based on the analysis condition, in this subsection we aim
to present a constructive procedure to design the controller
(4). With the controller system matrices notated as
A Bk
:k k
(9)
,
Ck Dk
Theorem 1 can be reformulated as
Z E T :k F F T :kT E 0
J I n p
0
(1 / J ) I n p
(10)
by the fact that the :cl of the closed-loop system (5) has affine
dependence on :k. Partitioning P and P1 as
Y N 1 X M
P T
(11)
, P
T
N
M
with X, Y, M, Nnun. Projection Lemma[7] leads to the
following solvability conditions for controller :k.
Theorem 2 (Solvability conditions). Consider the LPV
plant of the form (1) with scheduled parameter vector U(t)
defined as (2), and let NX and NY denote orthonormal bases of
the null spaces of >BTu , DTzuR, DTzuS@ and [Cy, Dyw] respectively.
If there exist parameter-dependent symmetric matrices X(U),
Y(U)nun satisfying the following inequalities for all
U(t)SU, then there exists an nth-order LPV controller :k to
robustly stabilize the closed-loop system and has the induced
L2 norm bounded by J.
sy ( Ap X ) X XCzT R Bw XCzT S
T
N X RCz X
R
RDzw
N X 0 (12)
BT S T C X DT R Q sy ( DT S )
z
zw
zw
w
WeD2.3
sy (YAp ) Y
N YT BwT Y S T C z
RCz
CzT R
T
T
Q sy ( Dzw
S ) Dzw
R N Y 0 (13)
RDzw
R
NY
X I
(14)
I Y ! 0, N X N X , NY
I nr n p
3)
YBw CzT S
0
T
Dyw
0
Dyw
Q S Dc
T
DcT S
RDc
0
B T
DcT R k
R
Cy
T
T
Bw Y S C z S Dzu Dk C y
RC z RDzu Dk C y
T
Dzu
R
C
T
Dc R k
BuT
T
T
( Bw Bu Dk Dyw XCz S )
RCz X
(17)
0
T
S Dzu
RD
zu
Q
T
Dzu
S
T
S Dc
R
DcT S
RDc
(18)
4)
L1
T
T
( Bw Bu Dk Dyw XCzT S C kT Dzu
S )T
RCz X RDzu C k
T
T
(YBw B k Dyw CzT S C Ty DkT Dzu
S )T
L2
RCz RDzu Dk C y
T
NA M
YB C B C X Y ( A B D C ) X
k
(19)
(20)
D - T r h RE ,
where V, T, q, -, h, D, P, RE are speed, flight path angle, pitch
rate, pitch angle, altitude, angle of attack, gravity constant
and earth radius, respectively. Control inputs are fuel
equivalence ratio I[0.1 1.2] and control surface deflection
Ge[-20 20] deg. Control actuator dynamics are represented
by first order filters with bandwidth 100 rad/sec for I and 30
rad/sec for Ge. Following [9], a curve-fitted model (CFM) has
been derived for control design. The approximations of the
forces and moments are given as follows
L | qSCL ( Ma, D , G e ), D | qSCD ( Ma, D , G e )
T | qS [CT ,I ( Ma, D )I CT ( Ma, D )]
(21)
(25)
M | zT T qScCM ( Ma, D , G e )
(22)
(23)
281
WeD2.3
0.01
Phugoid Mode
Imaginary Axis
0.005
-0.005
Short-period Mode
-0.01
-0.015
-2
-1.5
-1
-0.5
0.5
1.5
Real Axis
Fig. 2. Open-loop system poles for the extremal values of q and Ma.
-
Firstly, a robust LPV controller is designed ignoring the
saturation operator . The parametric uncertainties originate
from the derivatives of lift L and pitching moment M with
respect to angle of attack D and control surface deflection Ge,
with uncertainty level of r20%. To avoid excitation of elastic
modes, a multiplicative input uncertainty 'd(s) with the cover
function Wd(s)=2.5[(s+2)/(s2+9.76s+381.4)] taken from [10]
is placed at the control input. With the parametric
uncertainties extracted from the system in LFT way and
rescaled to [-1 1], we end up with the control interconnection
shown in Fig. 3.
Wzr 2
'd
Gl I
Gm I
wr 2
Wd
wp
wr1
LPV
Controller
Act
Ge
z r1
zp
We
Plant
q -
282
zp
zr 2
We
Gl I
zGrm1 I
wr 2
Wd
wp
'd
LPV
Controller
<
Act
LPV Anti-windup
Compensator
wr1
zr1
G e Plant
q -
WeD2.3
Reference
Trajectory
KP
1
Bout
( Aout diag(a12 , a22 ))
KI
1
Bout
diag( a11 , a21 )
(34)
G e Actuator
Inner-loop
Controller
0.9
0.8
3000
Tref
0.7
Velocity (ms-1)
2900
2800
2700
2600
0.6
0.5
0.4
0.3
0.2
2500
2300
0.1
V
Vref
2400
50
100
150
200
250
(a) Time (s)
300
350
400
-0.1
450
4.5
4.5
50
100
150
200
250
(b) Time (s)
300
350
400
450
4
Pitch Angle (deg)
4
3.5
3
2.5
3.5
2.5
I Vref
x PI
xPI 2
(32)
O2 T ref
Bout K I Aout Bout K P
where O2 denotes the 2u2 zero matrix, and I2 denotes the 2u2
identity matrix. Now select KI and KP to achieve the desired
closed-loop dynamics
O2
I2
(33)
Aideal
[q , Ma]T
Vref
T
ref
V
T
50
100
150
200
250
(c) Time (s)
300
350
400
1.5
0
450
17
0.9
16
50
100
150
200
250
(d) Time (s)
300
350
400
450
50
100
150
200
250
(f) Time (s)
300
350
400
450
15
0.8
Equivalence Ratio
-cmd
1.5
14
0.7
Elevator (deg)
D. Outer-loop Control
The outer-loop controller commands fuel equivalence ratio
and pitch angle to track velocity and flight path angle. The
engine and inner-loop dynamics are ignored in design
because the outer-loops bandwidth is much lower than the
engine and inner-loop dynamics. Replacing the control
deflection Ge in (26) with Ge,trim=-(A(3,4)D+B(3,1)I)/B(3,2)
which provides a zero pitching moment, the outer-loop
dynamics model is developed as
V
V
I
(29)
Aout ( U ) Bout ( U )
T
-
T
Here, the outputs to be controlled are given by [V, T]T. To
guarantee precise trajectory tracking, integrators are
considered to charge up the input values required to trim
the vehicle. Hence, the out-loop control is designed as
gain-scheduled proportional-integral controller
(V Vref )dt
I
V
,
(30)
- K P T K I
(
)
dt
T
T
ref
0.6
0.5
13
12
11
0.4
10
0.3
0.2
50
100
150
200
250
(e) Time (s)
300
350
400
450
283
WeD2.3
1.2
3000
T
Tref
2900
Flight Path Angle (deg)
Velocity (ms-1)
0.8
2800
2700
2600
0.6
0.4
0.2
2500
V
Vref
2400
50
100
150
200
250
(a) Time (s)
300
350
400
-0.2
450
2
Pitch Angle (deg)
2300
1
0
-1
-3
-4
-4
100
150
200
250
(c) Time (s)
300
350
400
-5
450
1.2
150
200
250
(b) Time (s)
300
350
400
450
-3
50
100
-1
-2
50
-2
-5
-cmd
50
100
150
200
250
(d) Time (s)
300
350
400
450
50
100
150
200
250
(f) Time (s)
300
350
400
450
312
314
316
25
1.1
20
1
Equivalence Ratio
0.9
15
Elevator (deg)
0.8
0.7
0.6
10
0.5
0.4
0.3
0
50
100
150
200
250
(e) Time (s)
300
350
400
-5
450
25
25
20
20
15
15
Elevator (deg)
Elevator (deg)
0.2
10
-5
98
10
100
102
104
106
108
(g) Time (s)
110
112
114
116
-5
298
300
302
304
306
308
(h) Time (s)
310
V. CONCLUSION
The dynamic response characteristics of longitudinally
maneuvering air-breathing hypersonic vehicles vary
substantially with flight conditions. These changes require
scheduling of the flight control system to be compatible with
changing operating conditions. For dealing with the modeling
uncertainties and saturation nonlinearities associated with the
vehicle model, we extended the standard LPV control
algorithm. Based on LPV modeling of the nonlinear vehicle
model, the extended LPV method is applied to design a
gain-scheduling flight control system. The controller design
combines a pair of robust LPV controller and LPV
anti-windup compensator and a gain-scheduled proportional
-integral controller as inner/outer loop structure, thereby
achieving attitude stability and precise trajectory tracking
performance. The introduced LPV anti-windup compensator
can successfully overcome the actuators momentary
saturation caused by abrupt atmospheric disturbance. The
nonlinear simulation results appear to be sufficiently
promising to warrant further study of the developed LPV
control algorithm and application to flight control system
design.
REFERENCES
[1]
284