Professional Documents
Culture Documents
Group 06
Group members:
TIMOTHY YEO YU JIN
A0101826N
A0100521H
Submission Date:
Question 1
The objective of experiment 1 is to find the extreme positive parameters for the uncertain process
Gm
Gp
models given the nominal model,
and uncertainty weighting function,
as shown
below.
G m=
l m=
1
s+1
s +0.2
0.4 s+1
The 2 uncertain processes assigned to the group are b) and f) as shown below, where and n are
parameters to be found.
Gb=Gm .
Gf =Gm .
1
s+1
1
0.01 s+1
the worst possible perturbation which allows the extreme positive parameters to be obtained.
Therefore,
As
lm
and
Gm
have been given, the objective of this question is to use trial and error to
find the maximum positive value of the unknown parameter that when substituted into
Gp
l
will result in the value of () being less than or equal to the magnitude of m for the
entire frequency range.
To do this, the magnitude of
figure 1 below.
lm
l m
Gb
Gb
1 exceeded that of the uncertainty weight. The value of
Gm
Gb
Gb
any further will result in Gm 1 exceeding
|l m|
the arrow is pointing. A close-up of this region can be found in figure 3. The code required to
produce this graph can be found in appendix A1.
Gb
1
Figure 2: Plot of
Gm
and
l m
in
Gb
Gf
Gf
, the initial guess for the value of n was 1 with increments of 1 until any point of graph of
Gf
1 exceeded that of the uncertainty weight. The value of n at this point was
Gm
Gf
Gf
1
Figure 4: Plot of
Gm
and
l m
Gf
with varying values of n in
Question 2a)
This question requires a PI controller,
C3
method and to compare the performance of this controller with 2 other controllers,
C2
C1
and
designed based on the ITAE performance index for load and set-point respectively. The
equations for
C1
and
1
6.48 s
1
7.39 s
C1 =0.42 1+
C2 =0.28 1+
C2
is shown below.
Gm
4 3.5s
e
7 s +1
Cohen-Coon tuning makes use of the nominal model parameters to design the controller via the
following equations.
K C=
(0.9+
)
K
12
I =
20
9+
30+
Where,
K Nominal process gain=4
Nominal process time constant =7
Nominal process time delay=3.5
Kc
and
C3
as shown below.
C3 =0.471(1+
1
)
5.803 s
Next, the performance of the controllers is compared to each other for a unit step change in the
set-point using Simulink. A simple feedback loop is designed in Simulink as shown in figure 6.
Figure 6: Simple Feedback loop to simulate unit step set-point changes for
C1 , C2 and C3
The resulting graph is plotted in matlab via the code found in Appendix A3 and is shown in figure 7.
Figure 7: Graph showing the reponses for the 3 different controllers to a unit step set-point change
To determine which controller has the fastest servo response, 2 criteria will be looked at, 1) time
taken for the curves to cross the set point and 2) time taken to reach the first peak. Just by
looking at the graph in figure 7, it is obvious that the Cohen-Coon design gives the fastest servo
response.
Table 1 below shows the specific values for the 2 criteria mentioned above.
Table 1: 1) Time taken for curves to cross set-point and 2) time taken for curves to reach the first
peak for the 3 different controllers.
Controller
Criteria
C1 (Load)
C2
(Set-Point)
C3
(Cohen-Coon)
16.3
19.6
15.85
12.65
16.55
12.05
From the 2 criteria mentioned above, it can be seen from table 1 that the shortest times to achieve
C3
these criteria belongs to
, the controller designed using the Cohen-Coon tuning method.
Therefore
C3
C1
C1
C3
are other factors that determine performance such as peak related criteria. An example of peak
related criteria is the maximum overshoot, the higher the overshoot, the worse the performance
and vice versa. Another example of peak related criteria is the decay ratio which is linked to the
oscillatory nature of the curve. A small decay ratio improves the performance as it reduces the
oscillatory nature of the curve while a large decay ratio worsens performance as it increases the
oscillatory nature of the curve.
Question 2b)
There are 3 equations available to determine the frequency response based robust stability
criteria. These equations are different forms of each other and they are shown below. The
equation that will be used in this section is equation (3).
Equation 3 states that the amount of tolerable uncertainty must be more than the actual
uncertainty in order for a controller to be robustly stable. From this equation, we can also imply
that the greater the amount of tolerable certainty a controller is able to provide given a fixed
Gm
, the more robustly stable a controller is as it can accept a greater actual uncertainty. Figure
1
than T 2
as it remains
closer to the value of 1 over a larger range of frequencies. This is indicative of good set-point
1
tracking which can be in turn be interpreted as a fast servo response. Therefore, the closer T
remains to the value of 1 for a larger range of frequencies, the faster the servo response is.
1
It should also be noted that for PI controllers, the values of T
for each of the 3 controllers
tends towards 1 when the frequency goes to 0. This means that at the steady state condition, there
is no offset which is to be expected when using a PI or PID controller.
y-
1
Figure 9: T graph for the 3 controllers
From figure 9, it can be seen that the Cohen-Coon controller has the least tolerable uncertainty,
and hence is the least robust stable. However, it remains closest to the value of 1 over the largest
range of frequencies, giving it the fastest servo response as found in question 2a). Likewise, the
ITAE (set-point) designed controller has the largest tolerable uncertainty and therefore is the
most robust stable. However, it deviates from 1 faster than the other 2 controllers, giving it the
slowest servo response as found in question 2a). The ITAE (load) designed controller is in
between these 2 controllers.
The evaluation of 2a) is consistent with the frequency response based robust stability criterion as
explained above. This also concurs with the theory that robust stability and performance both
cannot be achieved at the same time. It is a problem inherent in feedback control and one will be
achieved at the expense of the other. Therefore, it is up to the designer to decide which is of
greater importance in the process to be controlled and from there decide the parameters for the
controller.
Question 2c)
Assuming the process/model mismatch is entirely due to uncertainty in the process gain, the
Gp
process model
can be represented by the equation shown below.
G p=
K 3.5s
e
7 s +1
Where,
K=unknown process gain
To determine the range of K that each controller can accept, it is important to first note that the
controller gain is positive, therefore the process gain must be positive as well. This acts as the
lower bound for the process gain. The upper bound of the process gain will be determined from
1
l m
equation (3) in section 2b. Equation (3) states that T must be more than
for
all frequencies. Hence the approach to solving this question revolves around finding the
(|T 1|)min
lm
Gp
Gm
In this question, the expression for () , after substituting in the
and
defined
above, is shown below.
( )=
K
1
4
l
Finally, equating () to m the final expression to solve for the upper bound of K is
shown below.
K
1< (|T 1|)min
4
(|T 1|)min
(found from Matlab), the calculated upper limits of K and finally, the
range of K for which the controller can tolerate based on the frequency response robust stability
criteria
C1 (Load)
C2 (SetPoint)
C3 (CohenCoon)
|T-1|min
Upper bound of
K
Range of K
0.573
1.00
0.404
6.292
5.616
0<K<6.29
0<K<8.00
0<K<5.62
From the ranges of K shown for each controller, it can be seen that the most aggressive
controller, the Cohen-Coon controller, provides the least robust stability while the ITAE (set-
point) designed controller which is the most conservative controller provides the most robust
stability.
Question 2d)
To find out the upper limit of the K is tolerable for each controller, the simulink set-up shown in
figure 6 will be used. For each controller, the value of the process gain, K ,will be increased,
from the upper bound of K found via the robust stability criterion until the time-based response
shows sustained oscillations as shown in figure 10 which is indicative of marginal stability. The
new upper bound value, K* is then noted down for each controller and the results have been
recorded in table 3.
C1
(Load)
C2 (SetPoint)
C3
(CohenCoon)
Upper bound of K* as
determined by feedback loop
response
7.31
11.37
6.27
Upper bound of K as
determined by frequency
response robust stability
criterion (2c)
6.292
5.616
From the values shown in table 3, it can be seen that the values obtained from the feedback loop
response is higher than the values obtained from the frequency response robust stability criterion.