1

Shubhendu Bhasin
6343-4339
EAS 6939: Spring 2008 Test#3
I. PROBLEM 1
Problem: Consider the following mechanical System
My = Mg ky c
1
_ y c
2
_ y [ _ j[
where ', q, /, c
1
, c
2
are all positive constants. Show that the system has a GAS equilibrium point.
Solution: The problem can be written in state-space form as:
_ r
1
= r
2
, (1)
_ r
2
=
1
'
('q /r
1
c
1
r
2
c
2
r
2
[r
2
[).
Note that this is an autonomous system. The above system has an equilibrium point at
r
1e
=
'q
/
, r
2e
= 0. (2)
To facilitate the stability analysis, the states can be dened as
.
1
= r
1

'q
/
, .
2
= r
2:
(3)
So the state space form in (1) can be written using (3) as
_ .
1
= .
2
, (4)
_ .
2
=
1
'
(/.
1
c
1
.
2
c
2
.
2
[.
2
[).
The Lyapunov function candidate can be chosen to be non-negative and radially unbounded as
\ (.) =
1
2
/.
2
1
+
1
2
'.
2
2
(5)
The time derivative can be taken as
_
\ (.) = /.
1
_ .
1
+ .
2
' _ .
2
(6)
From (1) and (3), the expression in (6) can be written as
_
\ (.) = /.
1
.
2
+ .
2
(/.
1
c
1
.
2
c
2
.
2
[.
2
[)
The above equation can be rewritten as
_
\ (.) = c
1
.
2
2
c
2
.
2
2
[.
2
[ (7)
Since \ (0) = 0, \ (.) 0,
_
\ (.) < 0 and
_
\ (.) = 0 when .
2
= 0. If .
2
= 0 then _ .
2
= 0. Substituting in the dynamics (4), we nd
that when .
2
= 0 and _ .
2
= 0, then .
1
= 0. So we conclude that
_
\ (.) = 0 only when .
1
= 0 and .
2
= 0. Now we can use the
corollary to Lasalle's theorem to prove that the equilibrium point is GAS. And so it can be shown using (2) that
r
1

'q
/
, r
2
0 as t .
Hence, the equilibrium point of the system is Globally Asymptotically Stable (GAS).
2
II. PROBLEM 2
Problem: Given that '() is dened as
'() =
_
j
1
+ 2j
3
cos(
2
) j
2
+ j
3
cos(
2
)
j
2
+ j
3
cos(
2
) j
2
_
(8)
where j
1;
j
2;
j
3
are positive constants, use the Mean Value theorem to prove that
|'() '(
d
)| _ c |c| .
Solution: The error c in the problem can be dened as
c =
d
From the mean value theorem we can say that
'() '(
d
)

d
=
0'
0

q=q
b
, (9)
where
b
(,
d
). The notation
@M
@q

q=q
b
can be simplied as '
0
(
b
). Rearranging (9), we can get
'() '(
d
) = '
0
(
b
)c. (10)
Taking the norm of the equation (10), we get
|'() '(
d
)| =
_
_
_'
0
(
b
)c
_
_
_.
From the Cauchy-Schwarz inequality, the above expression can be bounded as
|'() '(
d
)| _
_
_
_'
0
(
b
)
_
_
_|c| . (11)
Now, from (8), the following expression can be written
'
0
(
b
) =
_
0 0 2j
3
sin(
2b
) j
3
sin(
2b
)
0 0 j
3
sin(
2b
) 0
_
. (12)
It can be seen from (12) that all the elements of the matrix '
0
(
b
) can be bounded by a constant. Hence the norm of the matrix
can be bounded as _
_
_'
0
(
b
)
_
_
_ _ c, (13)
where c R is a positive constant. From (11) and (13), the following expression can be obtained
|'() '(
d
)| _ c |c| .
QED.
III. PROBLEM 3
Problem: Design a continuous controller for the following dynamics to yield asymptotic tracking
:_ r = ar + /r
3
+ c + n,
where a, /, c are unknown positive constants with known constant upper bounds. Indicate if you can achieve global, local or
semi-global results. The desired trajectory r
d
(t) is continuously differentiable with bounded derivatives.
Solution: The error can be dened as
c = r r
d:
Without loss of generality the state r is assumed to be a scalar. The open loop error system can be written as
:_ c = :_ r :_ r
d
. (14)
Substituting the dynamics in (14), we get
:_ c = ar + /r
3
+ c + n :_ r
d
.
3
The above expression can be written as
:_ c = 1 0 + n, (15)
where 1 0 R can be written as
1 0 = ar + /r
3
+ c :_ r
d
= [ r r
3
1 _ r
d
]
_

_
a
/
c
:
_

_
.
The control input n can now be designed as
n = 1
^
0 /c, (16)
where / R is a constant positive control gain and
^
0 R
41
is an adaptive estimate of 0, the update lawas of which can be designed
as
:
^
0 = 1
T
c, (17)
where
44
is a constant positive diagonal gain matrix. The closed loop error system can be obtained from (15) and (16) as
:_ c = 1
~
0 /c. (18)
\ (t) is chosen as a non-negative radially bounded function ( a lyapunov candidate) as
\ =
1
2
:c
2
+
1
2
~
0
T

1
~
0 (19)
The time derivative of (19) can be taken as
_
\ = c(1
~
0 /c)
~
0
T
1
T
c
Simplifying the above expression we get
_
\ = /c
2
(20)
Since \ _ 0 and
_
\ _ 0, \ /
1
, and so c(t),
~
0(t) /
1
. Integrating (20), we get
\ (t) \ (0) _ /
t
_
0
c
2
dt (21)
It can be seen from (21) that c(t) is /
2
. Since c(t), r
d
(t) /
1
, it can be shown that r(t) /
1
. Since r(t) /
1
, 1 (r) /
1
.
Since 1 (r),
~
0(t), c /
1
, it can be shown from (18) that _ c(t) /
1
. Since c(t), _ c(t) /
1
, c(t) is uniformly continuous (U.C.).
Since c(t) /
1
/
2
and c(t) is U.C., it can be shown from Barbalat's Lemma that
[c[ 0 as t .
This means Global Asymptotic Stability.
IV. PROBLEM 4
Problem: Consider the following system
: r /
s
r
2
/ _ rr = n,
where :, /
s
, / are unknown positive constants with known upper and lower bounds. Dene the error system as
c , r
d
r,
r , _ c +
1
c +
2
c
f
,
where
1
,
2
are positive constant gains, and c
f
(t) is an auxillary lter variable designed as
_ c
f
=
3
c
f
+
2
c /
1
r,
where
1
,
2
are positive constant gains. Design a continuous controller so that
[c[ 0 as t
4
where only r(t) is available for feedback. Of course, r
d
(t) is a known continuous and bounded desired signal with bounded time
derivatives. Make sure that you prove that your controller is bounded and contains only known and measurable signals.
Solution: The open loop error system can be formed as
:_ r = : c +
1
:_ c +
2
:_ c
f
Substituting the dynamics and using the error system denition, the above expression can be written as
:_ r = : r
d
/
s
r
2
/ _ rr n +
1
:(r
1
c
2
c
f
) +
2
:(
3
c
f
+
2
c /
1
r) (22)
The open loop expression in (22) can be rewritten as
:_ r = : r
d
+ 1 0 n +
2
:/
1
r, (23)
where 1 0 R contains the linear parameterization and can be dened as
1 0 , /
s
r
2
/ _ rr.
In the open loop expression (23), can be dened as
=
1
:(r
1
c
2
c
f
) +
2
:(
3
c
f
+
2
c)
An upper bound for can be written as
[[ _
1
|.| , (24)
where
1
R is a computable positive constant and . is dened as
. = [c c
f
r]
T
.
Adding and subtracting the desired linear parameterization, the following expression is obtained
:_ r = : r
d
+ 1 0 n +
2
:/
1
r + 1
d1
0 1
d1
0, (25)
where 1
d1
0 R can be dened as
1
d1
0 = /
s
r
2
d
/ _ r
d
r
d
.
The expression in (25) can now be rewritten as
:_ r = 1
d
0 +
~
1 0 n +
2
:/
1
r, (26)
where 1
d
0 R is dened as
1
d
0 = : r
d
+ 1
d1
0
and
~
1 R
12
is dened as
~
1 = 1 1
d
The Mean value theorum can be used to get the following upper bound
_
_
_
~
1 0
_
_
_ _
2
|.|
The open loop error system in (26) can be further rewritten as
:_ r = 1
d
0 + n
2
:/
1
r, (27)
where R is dened as
=
~
1 0 +
Using the mean value theorem and using (24), the following upper bound can be developed
|| _ |.| . (28)
The control input can be designed as
n = 1
d
^
0 + c /
1
c
f
. (29)
5
Substituting the (29) in the open loop error system (27), the following expression is obtained
:_ r = 1
d
~
0 +
2
:/
1
r c + /
1
c
f
. (30)
where
~
0 R
21
is dened as
~
0 = 0
^
0.
The adaptive update law for
^
0 can be designed based onm the subsequent stability analysis as
:
^
0 = 1
T
d
r.
In the closed loop expression (30), the gain /
1
can be dened as
/
1
=
1

2
:
(2 +
2
/
n1
), (31)
where : R is a known positive lower bound on the mass and /
n1
R is a non linear damping gain.
Let \ (t) : T [0, ) R be a continuously differentiable positive denite radially bounded function dened as
\ =
1
2
:r
2
+
1
2
~
0
T

1
~
0 +
1
2
c
2
+
1
2
c
2
f
(32)
The time derivative of the Lyapunov function can be written as
_
\ = r(1
d
~
0 +
2
:/
1
r c + /
1
c
f
)
~
0
T
1
T
d
r + c(r
1
c
2
c
f
) + c
f
(
3
c
f
+
2
c /
1
r)
Simplifying and upper bounding, we get
_
\ _
2
:/
1
r
2

1
c
2

3
c
2
f
+|| [r[
Using (28), the above expression can be further upper bounded as
_
\ _
2
:/
1
r
2

1
c
2

3
c
2
f
+ |.| [r[
Substituting the non linear damping gain /
1
from (31), we get
_
\ _ r
2
, |.|
2
(
2
/
n1
[r[
2
|.| [r[) (33)
Using non-linear damping on the last term of expression (33), we get
_
\ _ r
2
, |.|
2
+
|.|
2
4/
n1
, (34)
where , R is a positive constant dened as
, = min(1,
1
,
3
).
Rewriting (34), we get
_
\ _ r
2

_
,
1
4/
n1
_
|.|
2
(35)
We can say that
_
\ _ 0, provided the following gain condition is satised
/
n1

1
4,
. (36)
From (32), (35) and (36), we can say that \ _ 0 and
_
\ _ 0, \ /
1
, and so r(t), c(t), c
f
(t),
~
0(t) /
1
.Integrating (35), we get
\ (t) \ (0) _
t
_
0
_
r
2
+
_
,
1
4/
n1
_
|.|
2
_
dt (37)
It can be seen from (37) that r(t), .(t) is /
2
. Since r(t), c(t), c
f
(t) /
1
, from the denition of .(t), we can show that .(t) /
1
,
and hence from (28), (t) /
1
. It can also be shown that 1
d
(t) /
1
. Since r(t), c(t), c
f
(t),
~
0(t), 1
d
(t) /
1
. and using the
error system in (30), it can be shown that _ r(t) /
1
. Since r(t), c(t), c
f
(t) /
1
, it can be shown that _ c(t), _ c
f
(t) /
1
. Since
6
_ r(t), _ c(t), _ c
f
(t) /
1
, we can show that _ .(t) /
1
Since r(t), _ r(t), .(t), _ .(t) /
1
, it means that r(t), .(t) are U.C. Since
r(t), .(t) /
1
/
2
and r(t), .(t) are U.C., we can invoke the Barbalat's Lemma to show that
[r(t)[ , |.(t)| 0 as t .
Since |.(t)| 0 as t , it can be proved that
[c(t)[ 0 as t .
Further, since
~
0(t) /
1
, it can be shown that
^
0(t) /
1
. Since 1
d
(t),
^
0(t), c(t), c
f
/
1
and known/computable/measurable,
it can be shown from (29) that the control input n(t) /
1
and contains known and measurable terms. It can be proved that
^
0(t)
contains only position terms
^
0(t) =
^
0(0) + 1
T
d
c(o)

t
0

t
_
0
_
_
1
T
d
c(o) 1
T
d
c(o) 1
T
d
c
f
(o)
_
do.
This is possible because of the DCAL based design, wherein 1
d
(t) contains only desired and hence known terms.