Review On Robust Control For SISO Systems
Review On Robust Control For SISO Systems
e(s)
k(s) g
N
(s)
+
+
u(s)
d(s)
y(s)
+
+
n(s)
Figure 1: Nominal system
The closed-loop system is
c
N
(s) =
g
N
(s)k(s)
1 + g
N
(s)k(s)
. (1)
For the system to be stable, the poles of Eq. 1 must be located in the left
half s-plane.
1
2.1 Performance
For the performance analysis we consider:
y(s) =
g
N
(s)k(s)
1 + g
N
(s)k(s)
r(s) +
1
1 + g
N
(s)k(s)
d(s)
g
N
(s)k(s)
1 + g
N
(s)k(s)
n(s) (2)
and
e(s) =
1
1 + g
N
(s)k(s)
r(s)
1
1 + g
N
(s)k(s)
d(s)
1
1 + g
N
(s)k(s)
n(s) (3)
2.1.1 Reference tracking
Considering only the contribution of r(j) to e(j):
e(j) =
1
1 + g
N
(j)k(j)
r(j) (4)
and being
r
the set where r(j) has its energy most signicant part. Generally,
r
is given by
r
= { :
r
}, (5)
where
r
is known. Suppose that a function (usually constant)
r
() 1
(
r
) is used as design specication. Thus,
|e(j)|
|r(j)|
r
(). (6)
So, it is possible to state that
|1 + g
N
(j)k(j)|
1
r
1 (7)
resulting in
y(j) r(j). (8)
So, the gain of the transfer function must be high where the reference signal
has its energy most signicant part.
2.1.2 External disturbance rejection
Considering only the contribution of d(j) to e(j):
e(j) =
1
1 + g
N
(j)k(j)
d(j) (9)
and being
d
the set where d(j) has its energy most signicant part. Generally,
d
is given by
d
= { :
d
}, (10)
where
d
is known. Suppose that a function (usually constant)
d
() 1
(
d
) is used as design specication. Thus,
2
|e(j)|
|d(j)|
d
(). (11)
So, it is possible to state that
|1 + g
N
(j)k(j)|
1
d
1 (12)
So, the gain of the transfer function must be high where the disturbances
have its energy most signicant part.
2.1.3 Plant variations insensitivity
For the nominal system
y(s) =
g
N
(s)k(s)
1 + g
N
(s)k(s)
r(s), (13)
consider a transfer function variation g(s), so that the output is y(j)+y(j):
y(j) + y(j) =
[g
N
(j) + g(j)]k(j)
1 + [g
N
(j) + g(j)]k(j)
r(j) (14)
Using Taylor series for the linearization,
y(j) =
k(j)
[1 + g
N
(j)k(j)]
2
g(j)r(j). (15)
Dividing Eq. 15 by Eq. 13,
y(j)
y(j)
=
1
1 + g
N
(j)k(j)
g(j)
g
N
(j)
(16)
Suppose that g(j)\g
N
(j) is signicant to
, being
= { :
}, (17)
where
() 1
(
d
) is used as design specication. Thus,
|y(j)\y(j)|
|g(j)\g
N
(j)|
(). (18)
So, it is possible to state that
|1 + g
N
(j)k(j)|
1
1 (19)
So, the gain of the transfer function must be high where the variation
g(j)\g
N
(j) is more signicant.
3
2.1.4 Measurement error rejection
Considering only the contribution of n(j) to y(j):
y(j) =
g
N
(j)k(j)
1 + g
N
(j)k(j)
n(j) (20)
and being
n
the set where n(j) has its energy most signicant part. Generally,
n
is given by
n
= { :
n
}, (21)
where
n
is known. Suppose that a function (usually constant)
n
() 1
(
n
) is used as design specication. Thus,
|y(j)|
|n(j)|
n
(). (22)
In order to guarantee the measurement error rejection, assume:
|g
N
(j)k(j)| 1 (23)
So, it is possible to state that
|g
N
(j)k(j)|
n
1 (24)
So, the gain of the transfer function must be small where the measurement
error is high.
2.2 Conclusion
It can be observed that the requirements for reference tracking, external distur-
bance rejection and plant variations insensitivity (which high gains are required)
are contradictory to the measurement error rejection one (where small gains are
required).
Fortunately, in most of practical problems
r
,
d
and
M
(s) =
g
R
(s) g
N
(s)
g
N
(s)
(25)
so
g
R
(s) = [1 +
M
(s)]g
N
(s). (26)
It is assumed that the designer is able to establish a upper bound to |
M
(j)|,
being:
4
dB
0
g
N
(j)k(j)
- Ref. tracking
- Dist. rejection
- Var. insensibility
- Meas. error
rejection
Figure 2: Low and high frequencies boundaries for the nominal system
r(s)
+
e(s)
k(s) g
R
(s)
+
+
u(s)
d(s)
y(s)
+
+
n(s)
Figure 3: Real system
|
M
(j)| e
M
(). (27)
Typically, e
M
() is small in low frequencies and big in high frequencies, as
it is shown in Fig. 4.
e
M
()
1
Figure 4: Function e
M
()
5
3.1 Stability
The system
c
R
(s) =
g
R
(s)k(s)
1 + g
R
(s)k(s)
(28)
must be stable for all
M
(j).
At this point, it is required to assume that the number of unstable poles
of g
R
(s) is equal than g
N
(s). This assumption is ensured i, by deforming the
Nyquist diagram from g
N
(j)k(j) up to g
R
(j)k(j), there is no crossing
through the point 1 + j0.
1
The condition for the no crossing on this point is
expressed by:
[1 +
M
(j)]g
N
(j)k(j) = 1 + j0. (29)
As g
N
(j) is supposed asymptotically stable, 1 + g
N
(j)k(j) = 0, Eq. 29
can be re-written as
1 +
M
(j)
g
N
(j)k(j)
1 + g
N
(j)k(j)
= 0. (30)
Or in other words
1 +
M
(j)c
N
(j) = 0, (31)
being equivalent to
|c
N
(j)| <
1
e
M
()
, (32)
which is denominated as Stability Robustness Condition.
3.2 Performance
According to Section 2.1, the performance of the nominal system is depict by
Eq. 7, 12 and 19. Changing to the real system:
|g
R
(j)k(j)| p() (33)
in which p() can be
r
(),
d
() or
c
Figure 5: Low and high boundaries for the real system
7