_

Adaptive Control
Basics and Research
Gang Tao
1
_

Feedback Control System

`
_
-
Plant
-
Controller
Feedback

6
- -
r(t) e(t) u(t) y(t)
w(t)

Reference System
- -
r(t) y
m
(t)
Goal of feedback control: lim
t
(y(t) y
m
(t)) = 0
2
_

Issues of Automatic Feedback Control

System modeling
Control objectives
stability, transient, tracking, optimality, robustness
Parametric uncertainties
payload variation, component aging, condition change
Structural uncertainties
component failure, unmodeled dynamics
Environmental uncertainties
external disturbances
Nonlinearities
smooth functions and nonsmooth characteristics
3
_

Adaptive Control Methodology

Adapting to parametric uncertainties
Robust to structural and environmental uncertainties
Aimed at both stability (signal boundedness) and tracking
Self-tuning of controller parameters
Systematic design and analysis
Real-time implementable
Effective for failures and nonsmooth nonlinearities
High potential for applications
Attractive open and challenging issues
4
_

Direct Adaptive Control System

'
_
Adaptive law

(t)
Plant
Controller
C(s; (t))
Reference model
-
-
-
6
-
6 6
6

?
?
u(t) r(t) y(t)
y
m
(t)
(t)
5
_

Indirect Adaptive Control System

Design
equation
Parameter estimator

p
(t)
Plant
G(s;

p
)
Controller
C(s;
c
(t))
-
-
6
-
6 6

?
u(t) y
m
(t) y(t)

c
(t)

p
(t)
6
_

Control System Dynamic Models

Nonlinear models
x = f (x, u, w), y = h(x, u, v)
state vector x R
n
, input u, output y, disturbances w, v; or
x = f (x) +g(x)u+d(x)w, y = h(x, u) +v
Linear state-variable model
x = Ax +Bu+B
w
w, y =Cx +Du+v
Linear time-invariant input-output model
y(t) = G(s)[u](t) +d(t)
G(s) = G
0
(s)(1+
m
(s)) +
a
(s), G
0
(s) = k
p
Z(s)
P(s)

a
(s),
m
(s): additive, multiplicative unmodeled dynamics.
7
_

Aircraft Flight Control System Models

State variables
x, y, z = position coordinates = roll angle
u, v, w = velocity coordinates = pitch angle
p = roll rate = yaw angle
q = pitch rate = side-slip angle
r = yaw rate = angle of attack
8
_

Nonlinear equations of motion (in body axis)

Force equations:
m( u+qwrv) = X mgsin+T cos
m( v +ru pw) = Y +mgcossin
m( w+ pv qu) = Z +mgcoscosT sin
T: engine thrust; : thrust angle; X,Y, Z: aerodynamic forces
Moment equations:
I
x
p+I
xz
r +(I
z
I
y
)qr +I
xz
qp = L
I
y
q+(I
x
I
z
)pr +I
xz
(r
2
p
2
) = M
I
z
r +I
xz
p+(I
y
I
x
)qpI
xz
qr = N
L, M, N: aerodynamic torques
9
_

Linearized longitudinal equations

_

_
u
w
q

_
=
_

_
X
u
X
w
W
0
g
0
cos
0
Z
u
Z
w
U
0
g
0
sin
0
M
u
M
w
M
q
0
0 0 1 0
_

_
_

_
u
w
q

_
+
_

_
X
e
Z
e
M
e
0
_

_
e
output = : pitch angle perturbation
Linearized lateral equations
_

r
p

_
=
_

_
Y
v
U
0
V
0
g
0
cos
0
N
v
N
r
N
p
0
L
v
L
r
L
p
0
0 tan
0
1 0
_

_
_

r
p

_
+
_

_
Y
r
Y
a
N
r
N
a
L
r
L
a
0 0
_

_
_
r
a
_
output = r: yaw rate perturbation
10
_

Adaptive Control versus Fixed Control

System
y(t) = (a
p
+)y(t) +u(t)
Reference model
y
r
(t) = a
r
y
r
(t) +r(t), a
r
> 0
Ideal controller for = 0
u(t) = k

y(t) +r(t), k

= a
p
a
r
Ideal performance for = 0
y(t) = a
r
y(t) +r(t), lim
t
(y(t) y
r
(t)) = 0
Fixed controller for [
1
,
2
]
u(t) = ky(t) +r(t), k < a
p

2
11
_

Closed-loop system
y(t) = a
r
y(t) +(a
p
++k +a
r
)y(t) +r(t),
e(t) = y(t) y
r
(t) =
a
p
++k +a
r
s +a
r
1
s a
p
k
[r](t)
Tracking performance (for r(t) = 1)
e
ss
= lim
t
e(t) =
a
p
++k +a
r
a
r
(a
p
++k)
Adaptive controller
u(t) = k(t)y(t) +r(t)

k(t) = e(t)y(t), > 0

with k(0) being arbitrary, leading to lim
t
e(t) = 0.
Observation: an adaptive controller ensures desired stability and
tracking, despite any large parameter uncertainty .
12
_

Our Recent Work

G. Tao and P. V. Kokotovi c, Adaptive Control of Systems with
Actuator and Sensor Nonlinearities, John Wiley & Sons, 1996.
G. Tao and F. L. Lewis, eds., Adaptive Control of Nonsmooth
Dynamic Systems, Springer, London, 2001.
A. Taware and G. Tao, Control of Sandwich Nonlinear Systems,
Springer, Berlin, 2003.
G. Tao, Adaptive Control Design and Analysis, John Wiley & Sons,
Hoboken, New Jersey, 2003.
G. Tao, S. H. Chen, X. D. Tang and S. M. Joshi, Adaptive Control of
Systems with Actuator Failures, Springer, 2004.
13
_

Adaptive Control of Aircraft with Synthetic Jet Actuators

14
_

Synthetic Jets for Aircraft Flight Control

Physics of synthetic jet
piezo-electric sinusoidal voltage acts on diaphragm
diaphragm vibrations cause cavity pressure variations
ejection and suction of air, creating vortices
jet is synthesized by a train of vortices
lift is produced on the airfoilvirtual shaping.
15
_

Tailless aircraft with jets (top view)

16
_

Simulation Results
System state variables
lateral velocity: x
1
(t) roll rate: x
2
(t)
yaw rate: x
3
(t) roll angle: x
4
(t)
System model
A =
_

_
0.0134 48.5474 632.3724 32.0756
0.0199 0.1209 0.1628 0
0.0024 0.0526 0.0252 0
0 1 0.0768 0
_

_
, B =
_

_
0
0.0431
0.0076
0
_

_
D. L. Raney, R. C. Montgomery, L. L. Green and M. A. Park, Flight Control
using Distributed Shape-Change Effector Arrays, AIAA paper No.
2000-1560, April 3-6, 2000
17
_

Control gain K
LQR design with Q = I
4
, R = 10
K =
_
1.0113 77.1793 115.8959 9.1691
_
P =
_

_
0.751 14.980 159.812 8.2617
14.980 27181.878 138979.668 7843.345
159.813 138979.668 723352.800 40670.052
8.262 7843.345 40670.052 2301.187
_

_
Reference signal:
r(t) =
_
_
_
1.5sin(t) 0 t 60
1.5sin(t) +3sin(2t) t 60
Adaptation gains:
1
= 1,
2
= 2
18
_

Simulation I: Adaptive inverse performance

0 50 100 150 200
10
0
10
Plant state x
1
(t) (solid), reference state x
m1
(t) (dotted) vs. time (sec)
f
t
/
s
e
c
0 50 100 150 200
0.1
0
0.1
Plant state x
2
(t) (solid), reference state x
m2
(t) (dotted) vs. time (sec)
d
e
g
/
s
e
c
0 50 100 150 200
0.05
0
0.05
d
e
g
/
s
e
c
Plant state x
3
(t) (solid), reference state x
m3
(t) (dotted) vs. time (sec)
0 50 100 150 200
0.5
0
0.5
d
e
g
Plant state x
4
(t) (solid), reference state x
m4
(t) (dotted) vs. time (sec)
Figure 1: Plant and reference states.
19
_

0 50 100 150 200

10
0
10
Tracking error e
1
(t) vs. time (sec)
f
t
/
s
e
c
0 50 100 150 200
0.2
0
0.2
Tracking error e
2
(t) vs. time (sec)
d
e
g
/
s
e
c
0 50 100 150 200
0.05
0
0.05
d
e
g
/
s
e
c
Tracking error e
3
(t) vs. time (sec)
0 50 100 150 200
0.5
0
0.5
d
e
g
Tracking error e
4
(t) vs. time (sec)
Figure 2: State tracking errors.
20
_

Simulation II: Comparison with a xed inverse

0 50 100 150 200
10
5
0
Tracking error e
1
(t) vs. time (sec)
f
t
/
s
e
c
0 50 100 150 200
0.1
0
0.1
Tracking error e
2
(t) vs. time (sec)
d
e
g
/
s
e
c
0 50 100 150 200
0.1
0
0.1
d
e
g
/
s
e
c
Tracking error e
3
(t) vs. time (sec)
0 50 100 150 200
1
0.5
0
d
e
g
Tracking error e
4
(t) vs. time (sec)
Figure 3: State tracking errors with a xed inverse.
21
_

Adaptive Actuator Failure Compensation

Actuator failures
common in control systems
uncertain in failure time, pattern, parameters
undesirable for system performance
Adaptive control
deals with system uncertainties
ensures desired asymptotic performance
is promising for actuator failure compensation
has potential for critical applications
22
_

Effective methods for handling system failures

multiple-model, switching and tuning
indirect adaptive control
fault detection and diagnosis
robust or neural control
Direct adaptive failure compensation approach
use of a single controller structure
direct adaptation of controller parameters
no explicit failure (fault) detection
stability and asymptotic tracking
Potential applications include
aircraft ight control
smart structure vibration control
space robot control
23
_

Systems with Actuator Failures

System Models
x = f (x) +
m

j=1
g
j
(x)u
j
, y = h(x)
x = Ax +
m

j=1
b
j
u
j
, y =Cx
state variable vector: x(t) R
n
output: y(t)
input vector: u = [u
1
, . . . , u
m
]
T
R
m
whose components may
fail during system operation
f (x), g
j
(x), h(x), A, b
j
, C with unknown parameters.
24
_

Actuator Failures
Loss of effectiveness
u
j
(t) = k
j
(t)v
j
(t), k
j
(t) (0, 1), t t
j
Lock-in-place
u
j
(t) = u
j
, t t
j
, j {1, 2, . . . , m}
Lost control
u
j
(t) = u
j
+

d
jk

jk
(t) +
j
(t), t t
j
, j {1, . . . , m}
Failure uncertainties
the failure values k
j
, u
j
and

d
jk
, failure time t
j
, pattern j, and
components
j
(t) are all unknown.
How much, how many, which and when the failures happen??
25
_

Examples
aircraft aileron, stabilizer, rudder or elevator failures
their segments stuck in unknown positions
their unknown broken pieces (including wings)
satellite motion control actuator failures
MEM actuator/sensor failures on fairing surface
heating device failures in material growth
generator failures in power systems
transmission line failures in power system
power distribution network failures
cooperating manipulator failures
bioagent distribution system failures
etc.
26
_

Block Diagram
,
.

,
.

Controller
System
-
-
1
1
.
.
.
.
.
.
-
?

1
-
-
1
m
-
-
?
-
-
-

m
-
r
u
m
u
1
y
u
1
.
.
.
u
m
v
1
.
.
.
v
m
27
_

Research Goals
Theoretical framework for adaptive control of systems with uncertain
actuator (sensor, or component) failures
Guidelines for designing control systems with guaranteed stability
and tracking performance despite parameter and failure uncertainties
Solutions to key issues in adaptive failure compensation: controller
structures, design conditions, adaptive laws, stability, robustness
New adaptive control techniques for critical systems (e.g., aircraft) to
improve reliability and survivability.
28
_

Example: Boeing 737 Landing

System model
x(t) = Ax(t) +Bu(t), y(t) = , B = [b
1
, b
2
]
T
x = [U
b
,W
b
, Q
b
, ]
T
: forward speed U
b
, vertical speed W
b
, pitch angle
, pitch rate Q
b
; u = [dele
1
, dele
2
]
T
: elevator segment angles
Study of an aircraft with two elevator segments
Output feedback output tracking design
One elevator segment fails during landing at t = 30 sec.
Simulation results
response with no compensation (xed feedback)
response with adaptive compensation.
29
_

0 10 20 30 40 50 60 70 80 90 100
0
0.02
0.04
0.06
0.08
time (sec)

y
(
t
)
,

y
m
(
t
)

(
r
a
d
)
y(t)
y
m
(t)
0 10 20 30 40 50 60 70 80 90 100
0
0.005
0.01
0.015
time (sec)

e
(
t
)

(
r
a
d
)
0 10 20 30 40 50 60 70 80 90 100
4
3
2
1
0
time (sec)

v
(
t
)

(
d
e
g
)
30
_

0 10 20 30 40 50 60 70 80 90 100
0
0.02
0.04
0.06
0.08
time (sec)

y
(
t
)
,

y
m
(
t
)

(
r
a
d
)
y(t)
y
m
(t)
0 10 20 30 40 50 60 70 80 90 100
0.01
0
0.01
0.02
time (sec)

e
(
t
)

(
r
a
d
)
0 10 20 30 40 50 60 70 80 90 100
4
3
2
1
0
time (sec)

v
(
t
)

(
d
e
g
)
31
_

Example: Boeing 737 Lateral Motion

MIMO system model
x = Ax +Bu, y =Cx
x = [v
b
, p
b
, r
b
, , ]
T
: lateral velocity v
b
, roll rate p
b
, yaw rate r
b
, roll
angle , yaw angle
y = [, ]
T
: roll angle , yaw angle
u = [d
r
, d
a
]
T
: rudder position d
r
, aileron position d
a
,
segmented into: d
r1
, d
r2
, d
a1
, d
a2
Actuator failures
d
r2
fails at t = 50, d
a2
fails at t = 100 seconds
Simulation results
32
_

0 20 40 60 80 100 120 140 160 180 200

0
1
2
3
4
5
6
7
Roll angle (t): , reference output
m
(t):
d
e
g
0 20 40 60 80 100 120 140 160 180 200
0
2
4
6
8
10
Yaw angle (t): , reference output
m
(t):
d
e
g
33
_

Research Interests
Adaptive control theory
actuator/sensor/component failure compensation
multivariable and nonlinear systems
actuator and sensor nonlinearity compensation
Adaptive control applications
aircraft ight control
fairing structure vibration reduction
space robot cooperative and compensation control
synthetic jet actuator compensation control
satellite motion control
high precision pointing systems
dynamic sensor/actuator networks
34
_

Some On-Going Research Projects

Rudder failure compensation by engine differentials
aircraft model with engine differentials
adaptive failure compensation control
Adaptive compensation control for aircraft damages
dynamic modeling of aircraft damages
direct adaptive damage compensation control
Adaptive compensation control for synthetic jet actuators
Adaptive failure compensation for space robots
Adaptive compensation of sensor failures
Adaptive control of spacecraft with fuel slosh.