ELEN90055 Control Systems

The University of Melbourne

Problem Set 4 with Reference Solutions
Prepared by M. Cantoni, April 2012

Flow
hi
yi−1
hi+1
pooli
pi yi

Overshot
Gate
pi+1
typical slope 1:10000
Overshot
DATUM Gate

Figure 1: Side view of an irrigation channel (left); Rubicon Water Pty. Ltd. flume gates at top of
East Golburn Main, Shepparton Irrigation District (right)

The stretch of an irrigation channel between two flume gates is called a pool. This is illustrated
in Figure 1, which shows so-called over-shot gates; under-shot gates are also common. Water flows
along the channel from one pool to the next and out of the channel at supply points in each pool.
This is achieved under the power of gravity alone; i.e. there is no pumping. The gates locally set
flows along the channel so as to meet downstream demand. Flume gates are positioned such that each
off-channel supply point to a farm or secondary channel lies towards the downstream end of a pool.
Correspondingly, regulation of the downstream water-level in each pool is an important operational
objective, since this represents the potential energy available to supply flow both downstream and at
the supply points. A reasonable model for the downstream water-level in pool i is given by

αi ẏi (t) = ui (t − τi ) − (vi (t) + di (t)), (1)

where yi is the water-level (m), ui is the flow over the upstream flume gate (m3 /min), vi is the flow
over the downstream flume gate (i.e. downstream load) and di is the total supply-point flow (i.e. local
load). The model parameter τi is a transport delay (min) and αi depends on the channel geometry
(m2 ), so as to capture mass balance.
The upstream flow ui depends non-linearly on the head hi , which is set by the position pi of the
corresponding gate relative to the upstream water-level yi−1 . Similarly, the downstream flow vi is
determined by the head hi+1 , which depends on the water-level yi and the gate position pi+1 . In
short, adjusting the position of a gate sets the downstream flow. A fast (compared to the dynamics
of the pools) position servo for the gates is used to achieve commanded flow. Over-shot gates used
north of Melbourne are shown on the right in Figure 1.
An approach to regulating the downstream water-level in a pool is to use a so-called distant-
downstream feedback controller. The input of such a controller is a measurement of the downstream
water level yi (t), relative to a reference ri (t), and the output is the flow command ui (t) for the

1
upstream gate. The distant-downstream architecture has merit in terms of distribution efficiency for
completely uncertain flow load, in that it leads to the demand-driven release of water from upstream
and ultimately the supply-channel/reservoir/river.
In the following, the pool model parameters are fixed to be αi = 1 and τi = π/10. Note that these
do not yield a model with the the dynamics of a large-scale pool.

1. Consider a distant-downstream controller with rational transfer function C(s). Construct a

block-diagram for the closed-loop system. Use the corresponding upper-case letter to denote
the Laplace transform of a time-domain signal.

2. For C(s) = 1, generate a Nyquist plot of the corresponding open-loop transfer function.
Hint: To make a MATLAB system object for the required delay term, use the command
exp(-0.1*pi*tf([1 0],1)).

3. With C(s) = K, for what range of the constant gain K is the closed-loop stable? Use your
answer to Part 2.

4. For a constant-gain controller C(s) = K, determine the sensitivity transfer function from the
total load signal Li (s) = Vi (s) + Di (s) to the water-level error signal Ei (s) := Ri (s) − Yi (s).

5. With a constant-gain controller C(s) = K, is it possible to achieve a steady-state water-level

error of less than 0.01 (m) for a step change of 0.1 (m3 /min) in the total flow load?

6. For a constant-gain C(s) = K, find the largest integer n such that the peak in the frequency-
response of the closed-loop sensitivity transfer function from Ri (s) to Ei (s), is less than 5
(≈ 14 dB) for K ∈ [n, n + 1]? Use a collection of Nyquist plots to answer the question, rather
than trying to do it in a completely analytic fashion. Confirm with corresponding Bode plots.

7. In light of the preceding analysis, comment on the effect of increasing K on the response of the
water-level to a step change in the reference.

8. Consider a nominal model Yi (s) = (1/s)(Ui (s) − Li (s)) of the pool dynamics corresponding to
τi = 0 (i.e. ignoring the transport delay associated with the upstream in-flow) and a controller
with transfer function C(s) = K(0.5s + 1)/s, where K > 0.

(a) With K = 0.1, use a Bode plot to determine the phase margin associated with the nominal
open-loop transfer function Λ(s) = (1/s)C(s).
(b) For what range of K is the phase margin associated with Λ(s) = (1/s)C(s) greater than
45◦ = π/4 rad?
(c) Based on the nominal model of the pool dynamics, comment on the closed-loop water-level
response to step changes in the reference with K = 0.1 compared to the response with the
smallest value of K found in Part (b).
(d) Generate a Bode plot of the relative error (i.e. the multiplicative modelling error) associated
with neglecting the non-zero transport delay τi = π/10 (min) in the nominal model.
(e) Using the small gain theorem, argue that the closed-loop, including the transport delay
τi = π/10 (min), is stable for 0 < K < 3. To this end generate appropriate Bode plots.
Confirm with a Nyquist plot.
(f) What is the steady-state water-level error in response to a step change in the flow load for
0 < K < 3?

2
Reference Solutions
1. Figure 2 shows the block-diagram for the distant-downstream control system.
π
2. When C(s) = 1, the open-loop transfer function is Λ(s) = exp(− 10 s)/s. The Nyquist plot is
shown in Figure 3; the dotted lines were added to the MATLAB plot by hand to reflect what
happens as s follows the ǫ-contour around the pole at the origin. Note that there are N = 0
(clockwise) encirclements of the critical point −1 + j0 and so the corresponding closed-loop is
stable.
Li (s) = Di (s) + Vi (s)

R(s) E(s) U (s) _ Yi (s)

1
_
C(s) e−τi s
αi s

Figure 2: Block diagram of a pool under distant-downstream control

3. From the Nyquist plot in Figure 3, it can be seen that increasing the gain of C(s) = 1 by a
factor of 1/0.2 = 5, causes the Nyquist plot to intersect −1 + j0.1 This give rise to closed-loop
poles on the imaginary axis. A further increase in gain leads to N = 2 (clockwise) encirclements
of the critical point and hence, an unstable closed-loop, since the number of open-loop RHP
poles P = 0. As such, K < 5 is required for stability. Decreasing the magnitude of the gain,
does not lead to any encirclements. Figure 4 shows the Nyquist plot for C(s) = −1. Note
that this change in sign leads to N = 1 encirclements of the critical point, as s follows the
ǫ-indentation around the pole at the origin. There is at least 1 clockwise encirclement for any
negative controller gain. That is, K > 0 is also required for closed-loop stability. To summarize,
the distant-downstream feedback control system is stable for 0 < K < 5.
4. For C(s) = K, the sensitivity transfer function from Li (s) to Ei (s) is given by
Ei (s) 1
=
Li (s) αi s + K exp(−τi s)
1
= π .
s + K exp(− 10 s)
To see this, note that with Ri (s) = 0,
!
1 1
Ei (s) = −Yi (s) and Yi (s) = K exp(−τi s)
(−Li (s)).
αi s 1+ αi s

Note that this is not a rational function.

5. The water-level error response to a 0.1 (m3 /min) step in the total flow load is given by
1 0.1
Ei (s) = π .
s + K exp(− 10 s) s
1
Note that
π π
Λ(jω) = − sin( ω)/ω − j cos( ω)/ω.
10 10
π
So the Nyquist plot intersect the real axis when cos( 10 ω) = 0; i.e. is, ω = 5, 15, 25, 35, . . . rad/min. ℜe(Λ(j5)) =
π
− sin( 10 5)/5 = −0.2.

3
 1
For values of K such that π
s+K exp(− 10 s) is stable, the signal ei (t) approaches a final value

0.1
ei,SS = lim ei (t) = lim sEi (s) = .
t→∞ s→0 K
To achieve ei,SS < 0.01 (m) requires K > 10. But from Part 3, this would not be stable and
the final value theorem does NOT apply. As such, it is NOT possible to achieve a steady-state
error of less than 0.01 (m) for a step change in the total load flow of 0.1 (m3 /min).
1
6. The transfer function from Ri (s) to Ei (s) is given by S(s) = 1+Λ(s) , where Λ(s) = Ks exp(− 10
π
s).
Note that lim|s|→∞ Λ(s) = 0. The Nyquist plot is a (polar) plot in the complex plane of Λ(s)
evaluated at s = jω as ω is varied from −∞ to ∞. The distance from the critical point −1 + j0
to the point Λ(jω) on the Nyquist plot is 1/|S(jω)|. From the Nyquist plot in Figure 3, where
K = 1, it can be seen increasing K moves a point that intersects the real axis closer the critical
point. For K = 4, this point will lie at s = 0.8 + j0, a distance of 0.2 from the critical point,
whereby minω (1/|S(jω)|) ≤ 0.2 and thus, maxω |S(jω)| ≥ 1/0.2 = 5(≈ 14dB). As such, n = 2
is the largest possible value, since the interval [3, 4] contains values of K that yield a sensitivity
peak of more than 5. To confirm, see Figure 5 and Figure 6.
7. From Part 6, increasing K leads to the following:
• a widening of the low-frequency range where |S(jω)| < 1; and
• an increase in the peak value |S(jω)|.
Since the closed-loop sensitivity T (s) from Ri (s) to Yi (s) is related to S(s) by T (s) + S(s) = 1,
increasing K also leads to the following:
• an widening of the low-frequency range where |T (jω)| ≈ 1; and
• an increase in the peak value of |T (jω)|.
See Figure 7. As such, one would expect the water-level response to rise faster and exhibit
increasingly oscillatory behaviour as K is increased. The steady-state error also decreases, but
the settling time increases. This is confirmed in Figure 8.
8. For the delay-free nominal model of the pool dynamics and the specified controller, the open-
loop transfer function is given by ΛK (s) = K(0.5s+1)
s
1
s.

(a) The Bode plot of the open-loop transfer function with K = 0.1 is given in Figure 9. It can
be seen that ω = 0.3 rad/min is (approximately) the loop-gain 0 dB cross-over frequency
(i.e. the loop-gain bandwidth). Note that ∠Λ0.1 (j3) ∈ [−172.5◦ , −170◦ ]. As such, the
phase margin is more than 7.5◦ and less than 10◦ .
(b) With K = 0.1, it can be seen from the Bode plot in Figure 9 that ∠Λ0.1 (j2) ≈ 135◦ . At
this frequency, the loop-gain |Λ0.1 (j2)| ≈ −30 dB. Increasing
√ K does not change the phase
of ΛK (j2). It only changes the gain. Setting K = 10 = 0.1(1030/20 ) in the controller
transfer function corresponds to adding 30 dB to the magnitude plot shown in Figure 9,
without changing the phase plot; in particular, |Λ√10 (j2)| ≈ 1 and ∠Λ√10 (j2) ≈ 135◦ .
This is consistent with a phase margin of 45◦ . √ Since ∠ΛK (jω) increases monotonically
with increasing ω, further increase in K beyond 10 increases the phase margin.
(c) Since with the nominal delay-free pool model the phase margin increases with increasing K,
one would expect the water-level
√ response to step reference changes to be more oscillatory
for K = 0.1 than for K = 10. The corresponding √ increase in the loop-gain bandwidth
(i.e. the 0 dB cross-over frequency) with K = 10 compared to K = 0.1, should also lead
to a faster response. This is confirmed in the step responses shown in Figure 10.

4
(d) The multiplicative modelling (i.e. relative) error is

G − G0
G∆ =
G0
π
= exp(− s) − 1.
10
The Bode plot of this is shown in Figure 11.
(e) The Bode plots for the nominal complementary sensitivity function for controller gains
K = 0.1, K = 1, K = 2 and K = 3 are also shown in Figure 11. It can be seen that the
product of the gain of the complementary sensitivity function and the gain of the relative
error |G∆ (jω)| is strictly less than 1, for all of the controller gains considered. It follows by
the small-gain theorem that the corresponding closed-loop is internally stable. Figure 12
shows the open-loop transfer function (including the delay) Nyquist plot for K = 1. It
can be seen that there are no encirclements of the critical point −1 + j0 (not even as
the ǫ-indentation around the origin is traversed), confirming stability for K = 1. This
continues to be the case as the gain ranges up to K = 3 and down to K = 0.1.
(f) The final-value theorem applies since the closed-loop is stable for all of the specified control
gains. The transfer function from the total load flow Li (s) to the water-level error Ei (s)
is given by

Ei (s) s
= 2 π
Li (s) s + K(0.5s + 1) exp(− 10 s)

Applying the final-value theorem with Li (s) = c/s and c a real constant,
s c
ei,SS = lim s · π · = 0.
s→0 s2 + K(0.5s + 1) exp(− 10 s) s

5
 Nyquist Diagram

2.5

1.5

0.5
Imaginary Axis

−0.5

−1

−1.5

−2

−2.5
−0.4 −0.35 −0.3 −0.25 −0.2 −0.15 −0.1 −0.05 0 0.05 0.1
Real Axis

Figure 3: Nyquist diagram for C(s) = 1 in Q2.

Nyquist Diagram

1.5

0.5
Imaginary Axis

−0.5

−1

−1.5

−2
−0.2 −0.1 0 0.1 0.2 0.3 0.4
Real Axis

Figure 4: Nyquist diagram for C(s) = −1 in Q3.

6
 Nyquist Diagram

0.75

0.65

0.55

0.45

0.35

0.25

0.15
Imaginary Axis

0.05
0.2

−0.05

−0.15

−0.25

−0.35

−0.45

−0.55

−0.65

−0.75
−1 −0.9 −0.8 −0.7 −0.6 −0.5 −0.4 −0.3 −0.2 −0.1 0 0.1 0.2 0.3 0.4 0.5
Real Axis

Figure 5: Nyquist diagrams for Q6: K=1 (blue); K=2 (green); K=3 (red); K=4 (cyan)

Bode Diagram

20
15
10
5
0
Magnitude (dB)

−5
−10
−15
−20
−25
−30
−35
−40
135

90
Phase (deg)

45

−45
−1 0 1 2
10 10 10 10
Frequency (rad/min)

Figure 6: Sensitivity Bode diagrams for Q6: K=1 (blue); K=2 (green); K=3 (red); K=4 (cyan)

7
 Bode Diagram

20

10

Magnitude (dB)
−10

−20

−30

−40

−50
0

−180

−360
Phase (deg)

−540

−720

−900

−1080 −1 0 1 2
10 10 10 10
Frequency (rad/min)

Figure 7: Complementary sensitivity Bode diagrams for Q7: K=1 (blue); K=2 (green); K=3 (red);
K=4 (cyan)

Step Response

1.8

1.6

1.4

1.2
Amplitude

0.8

0.6

0.4

0.2

0
0 1 2 3 4 5 6 7 8 9 10
Time (min)

Figure 8: Water-level response to step reference for Q7: K=1 (blue); K=2 (green); K=3 (red); K=4
(cyan)

8
 Bode Diagram

25
20
15
10
5
0
−5
−10
Magnitude (dB)
−15
−20
−25
−30
−35
−40
−45
−50
−55
−60
−65
−70
−90
−95
−100
−105
−110
−115
−120
−125
Phase (deg)

−130
−135
−140
−145
−150
−155
−160
−165
−170
−175
−180 −1 0 1 2
10 10 10 10
Frequency (rad/min)

Figure 9: Open-loop transfer function Bode for Q8a

Step Response

1.8

1.6

1.4

1.2
Amplitude

0.8

0.6

0.4

0.2

0
−10 0 10 20 30 40 50 60 70 80 90 100
Time (min)

Figure 10: Water-level response to step reference for Q8c: K=0.1 (blue); K=1 (green)

9
 Bode Diagram

20

10

Magnitude (dB) 0

−10

−20

−30

−40
0

−45

−90
Phase (deg)

−135

−180

−225

−270
−1 0 1 2
10 10 10 10
Frequency (rad/min)

Figure 11: Relative modelling error (blue) for Q8(d) and complementary sensitivity functions with
K=0.1 (magenta), K=1 (green), K=2 (red), K=3 (cyan) for Q8(e)

Nyquist Diagram

0.2

0.15

0.1

0.05
Imaginary Axis

−0.05

−0.1

−0.15

−0.2
−1 −0.9 −0.8 −0.7 −0.6 −0.5 −0.4 −0.3 −0.2 −0.1 0
Real Axis

Figure 12: Nyquist plot with K=1 for Q8(e)

10