2EL10 : Control

Theory
Prof. M. M. Solanki
Electronics Engineering Dept.
B.V.M. Engineering College
 Basics of control design -
Proportional, Integral and Derivative
actions.

2
 Proportional, Integral and Derivative actions

𝑅 𝐸 �� 𝑈𝑝(𝑠 𝑌
𝑠 𝑠 𝐾� + 𝐾
� 𝑠 � � )
𝐺(𝑠 𝑠
�
+� )
+- �
�

𝐾 �
𝑡

𝑈𝑝 𝑠 = (𝐾𝑝+𝐾𝑑𝑠𝐼 + )𝐸 𝑠� ⟷ 𝑢𝑝 𝑡 + + 𝐾𝑖 ∫ 𝑒 𝑑
𝑠 𝐾𝑑 𝑒 𝑡 𝑟 𝑟
Proportional control = 𝐾𝑝𝑒 𝑡 Proportional + Derivative Control 0
Proportional + Integral Control

 Simplest to implement  Stability and improved  Perfect tracking of

 May not be sufficient to time response constant reference inputs
meet the design  Provides more control over  Rejection of constant
requirements pole locations disturbance inputs
 Stability
 Control over pole locations
Proportional + Derivative + Integral Control  Perfect tracking of constant reference inputs
For second order systems  Constant disturbance rejection
3
 Qualitative understanding of Integral and Derivative actions.

 Issues in the implementation of PID controllers.

 Introduction to Lead – Lag Compensation.

 Performance specification in the time and frequency domains.

4
 Steady-state error and Integral
control
Consider a plant with no integrator terms with a proportional controller.
1
𝑅 𝑠 = 𝑌
𝑠 𝐸 𝑠 1
𝑈𝑝(𝑠) 𝐾𝑝 𝐺 𝑠
=
=0.5 𝑠 𝑠
+-

+𝑎

The open-loop system is stable for 𝑎 > 0 and the steady state is at 𝑦 = 0. Through a feedback
controller we are trying to shift the steady state value to 𝑦 = 1.
Let us understand why a proportional controller is incapable of tracking a step input.

Suppose that the error 𝑒 𝑡 has the behaviour-

 𝑒 𝑡 → 0 as 𝑡 → ∞. The steady-state error is

⇒
zero.

 𝑦 𝑡 tracks 𝑟 𝑡 as 𝑡 → ∞.

see if they are feasible with the assumed 𝑒 𝑡 .

Let us now look at the other signals in the loop and

5
 𝑅 𝑠=1 𝐸 𝑌
� 𝑈𝑝
𝑟 𝑡 = � 𝑠 𝑠 1 𝑠
1𝑢(𝑡)
𝑒(𝑡 𝐾𝑝 = 𝐺 𝑠= 𝑦(𝑡
𝑢𝑝 (𝑡 𝑠+
) 0.5 )
) 1
+-

𝑡
𝑦 = 𝑒 −𝑡 𝑦 + 0.5 𝑑
𝑡 0 ∫ 𝑟
𝑒 −𝑐 𝑒𝖴
𝖴 𝑡−𝑟
𝑦(𝑡) → 𝑟 𝑡 𝑦(𝑡) → 00 as 𝑡
as 𝑡 → ∞ →∞
Contradiction

Moreover, if the assumed 𝑒(𝑡) was feasible, it would mean that 𝑢𝑝 (𝑡) → 0 as 𝑡 → ∞ and the controller
applies no input to the system to maintain the system at the new steady state.

The controller must apply some nonzero input to the plant to maintain the output at some nonzero
equilibrium. Therefore the steady state error cannot be zero in this case with a proportional controller.

6
 The system given below exhibits a steady
state error of 0.1.

𝑅 𝑠=1
� 𝐸 𝑈𝑝 𝑌
𝑟 𝑡 = � 𝑠 𝑠
𝑠 1
1𝑢(𝑡) 𝑒(𝑡 𝐾𝑝 = 𝑢𝑝 (𝑡 𝐺 𝑦(𝑡
) =
9 ) 𝑠 𝑠+ )
+-

With a steady state error of 0.1, note that all the signals are feasible. In this case, the controller drives the plant

𝑢 𝑝 𝑡 such that lim 𝑢 𝑝 𝑡= 0.9 𝑢 𝑡 .

with
𝑡→∞
7
 We have already seen that the use of an Integral controller leads to
perfect tracking of constant reference inputs for a first order plant.
1
𝑅 𝑠 𝐸 𝑈𝑝 𝑌
𝑠 𝑠
𝑟=𝑡 = 1 𝑠 1
𝑠
𝑢(𝑡) 𝑒(𝑡 𝐺 𝑦(𝑡
� 𝑢𝑝 (𝑡 =
) 𝑠 𝑠+ )
)
+-

1
�

The controller output 𝑢 𝑝 𝑡 is now the integral of the error signal 𝑒 𝑡 . The controller output is
nonzero even though the error has gone to zero.
A similar analysis shows that when there is a constant disturbance the error due to the disturbance
accumulates and is used to reject the effect of disturbance.
8
A few important observations.

 Plants with integrators.

When the plant already has an integrator, integral control may not be essential for zero steady-state error.
However the integrator in the plant does not help in rejecting disturbances. Therefore Integral control is
used for these plants as well.

 Integrator wind up phenomenon.

Note that Integral or Proportional + Integral controllers are not stable. When there is a limit on the
actuation capability, the controller output saturates leading to non-decaying errors. The integrator
blindly accumulates the error and its output can grow to large values.
1
𝑅
= 𝑠 𝑌
𝑠 𝐸 1 𝑈𝑝 1
𝑠 𝑠 𝑠
� 𝑠+
1
+-
� Saturation

We will not discuss how this situation can be dealt with in this course. For those interested,
references will be posted on the course page.
9
 A few important observations.

 Integral control leads to sluggish closed-loop response for a first order system.

Dominant
pole

𝐾 𝐾
𝐺𝑐 𝑠 𝐺 = 𝐺𝑐 𝑠 𝐺 =
𝑠 𝑠+ 𝑠 𝑠(𝑠 +
 3 3)
Integral control of higher order plants may lead to instability. This point was emphasized in the last
lecture using a second order plant.

10
 Derivative and Proportional + Derivative
controllers

𝑅 𝐸 𝑈𝑝(𝑠 𝑌
𝑠 𝑠 𝐾𝑝 + ) 𝐺(𝑠 𝑠
𝐾𝑑 𝑠 )
+-

�
𝑈𝑝 𝑠 = (𝐾𝑝+𝐾𝑑𝑠)𝐸 𝑠 � 𝑢 𝑝 = + 𝐾𝑑 𝑒
⟷ 𝑡 𝐾𝑝𝑒 𝑡 𝑡
 In the last lecture, we saw that a PD controller acts not only on the error but also the rate of
change of error.

 It improves the relative stability as we are effectively adding a zero to the open-loop transfer
function in the left half s-plane.

 We discussed that the derivative term is in some sense anticipatory or predictive and its effect
was explicitly seen in the improved damping in the system.

11
 Note :-

Just the proportional or the derivative term is not sufficient and both the terms
are necessary in most cases.

Consider the following system.

𝑅 𝐸 1 𝑌
𝑠 𝑠 𝐾𝑝 + 𝑠
𝐽𝑠
𝐾𝑑 𝑠
+-
2

The output of the system 𝑌 𝑠

𝐾𝑝 +
𝑌 𝑠 𝑅
𝐽𝑠𝐾𝑑+𝑠𝐾𝑑 𝑠 𝑠
is
=
2

+ 𝐾𝑝
Note that the term 𝐾𝑑𝑠 influences only the 𝑠1 term and 𝐾𝑝 influences just the 𝑠0 term. Both the
terms are necessary to have complete control over the pole locations.

12
A few important observations.

 PD controller is a non-causal system.

The transfer functions 𝑇1 𝑠 = 𝐾𝑑 𝑠 and 𝑇2(𝑠) = (𝐾𝑝+𝐾𝑑 𝑠) are improper transfer functions.
We have seen that systems with such transfer functions are non-causal and cannot be realized in practice
in their pure form.

 Amplification of Noise.

As the derivative term acts on the rate of change of error, it’s output is highly susceptible to noise.
The noise can enter through the output measurement sensors or through the reference signal itself.

When a low noise sensor is available for output measurement the derivative controller can be used
in the feedback path.

𝑅 𝐸 1 𝑌
𝑠 +- 𝑠 𝐽𝑠 𝑠 The noisy reference signal gets
2 filtered in the forward path before
encountering the differentiator.
𝐾𝑝 +
𝐾𝑑 𝑠
13
 A causal approximation to a
derivative controller

To address both the issues the following approximation for the derivative term can be
used
𝑇𝑑 𝑠 = 𝐾𝑑 𝑠 = �
𝑠 𝑠
(𝐾 𝑠) 𝑝1+ 1 +
𝑝
1
�

• 𝑇𝑑 𝑠 is causal and as 𝑝 → ∞,𝑇𝑑 𝑠 → 𝐾𝑑𝑠 and we recover the derivative controller.

the signal before passing it to the differentiator. As 𝑝 → ∞, we have an all-pass filter and
• The multiplicative term acts as a low pass filter and eliminates high frequency content in

we get back a pure differentiator.

14
Lead and Lag compensation.

15
 Lag compensation
The Proportional + Integral controller (PI) provides perfect tracking of constant reference inputs and
helps in rejecting constant external disturbances.

We have seen that the PI controller is not a stable system leading to issues such as integrator wind up
and also the possibility of destabilizing the closed-loop system.

Moreover, if small steady-state errors are acceptable, perfect tracking may not be necessary.

These observations lead us to the following approximation to a PI controller-

𝐾𝐼 𝐾𝐼 + 𝐾𝑝 𝑠 𝐾𝐼
𝐺� (𝑠) � + 𝐾𝑝𝑠
=�𝐾 + = ≈
𝑠 𝑠 𝑠+
𝑝
�

with 𝑝 being very close to zero.

𝐾𝐼 + 𝐾 𝑝 𝑠 𝑠
𝐺�
+𝑧
Further,
𝑠�
where 𝐾 = 𝐾and 𝑧= ≈ = 𝐾
𝐾 𝑠+𝑝 𝑠+
𝐾𝐼
.
𝑝
�
�
𝑝

Instead of a pole at the origin and a zero in the LHP, we now have a pole and zero in the LHP and the
controller is stable. This approximation is called a Lag compensator.
16
 Consider the standard first order plant
with PI controller.

𝑅 𝑠 𝐸 𝑠 𝐾 1 𝑌
𝐾𝑝 𝑇𝑠 + 𝑠
𝑈𝑝(𝑠) + - 𝐼
1
+
𝑠

Static position error constant for this system is

𝐾𝑝 = lim 𝐺 𝑠 𝐺 𝑠 ⇒ steady state error 𝑒 =
1
𝑠 1+𝐾
𝑐 = ∞ = 0
𝑠→
𝑠
0 𝑝
Consider the same plant with a lag

𝑠 𝑈𝑝(𝑠 1 𝑌
compensator
𝑅 𝑠
+𝑧 ) 𝑇𝑠 + 𝑠
𝐸 𝑠 +-
𝐾
𝑠 1
+𝑝
Static position error constant for this system is
𝐾𝑝 = lim 𝐺 𝑠 𝐺 𝑠
𝑠→ 𝑧 ⇒ steady state error 𝑒𝑠 = 1+𝐾
1𝑧
≈0
𝑐 = 𝐾 𝑠 𝑝

The steady error will not be zero with a lag compensator. However 𝑧 and 𝑝 can be placed to achieve very small
0
�
�

errors.
17
 Effect of the Lag compensator on the root-
locus

Figure 1. Effect of a PI controller Figure 2. Effect of the Lag compensator

Lag compensator Proportional + Integral

𝑠 + 0.15 𝑠+
Proportional Controller controller

𝐺𝑐 𝐺𝑐 𝑠=
Plant
𝐺𝑐 𝑠 =2 = 2 0.15 𝑠
1 𝑠 + poles are at
𝑠 closed-loop 2
𝐺� 𝑠
(𝑠 + 0.5)(𝑠 + −1 ± 1.3229i 0.015
−0.9464 ± 1.2857i −0.9406 ± 1.2813i
Dominant closed-loop poles are at
=�
Closed-loop poles are at Dominant

1.5) 𝑒𝑠𝑠 = 0.0380 𝑒𝑠𝑠 = 0

𝑒𝑠𝑠 = 0.273
18
 Lead Compensation
We have already found a causal approximation to the derivative controller.

𝑇𝑑 𝑠 𝐾𝑑 𝑠 =
𝑠
= �𝑠𝐾+𝑑𝑝𝑠 + 𝑝
�1
Under a similar approximation to the Proportional + Derivative controller we

𝐾 (𝑠
have
𝑝 𝐾𝑝
𝐺� = 𝐾𝑝 + + 𝑧)
𝑠� 𝐾 𝑠 ≈ + 𝐾𝑑 𝑠 = (𝑠
𝐾𝑝
𝑑
𝑠+𝑝 + 𝑝)
where, 𝐾 = 𝑝𝐾 �
and 𝑧 = �.
𝐾
𝑑

We now have a causal and stable controller which is less susceptible to high frequency noise.

Both the pole and the zero are in the left half s-plane with the pole farther away from the origin than
the zero. This approximation is called a Lead compensator.

19
 Effect of the Lead compensator on the root-locus

Figure 1. Effect of a PD controller Figure 2. Effect of the Lead compensator

𝐾 𝑠+
Proportional + Derivative Lead compensator
1
Plant
𝐺� 𝑠
𝐺𝑐 𝑠 𝐺𝑑 𝑠 = 𝐾 𝑠 +𝐾 𝑠1 +
controller
𝑐 𝑝
�2 =�𝐾
= =𝐾 𝑠 𝑑 15
+𝑧
• The PD through the choice of 𝐾 and 𝑧 allows the placement of the closed-loop poles anywhere in the s-plane.
�

• For a given 𝑧, both the PD controller and the lead compensator choices for the closed-loop poles that are very close
to each other.
• As the pole of the lead compensator more farther away the match between the two root-loci in Figure 2 becomes
closer.
20
 Lead and Lag compensators

Both the approximations derived so far have the same general structure.

𝐾 (𝑠
𝐺𝑐 𝑠 + 𝑧)
(𝑠
= + 𝑝)
The relative positions of 𝑧 and 𝑝 determine whether the compensator is a lead or a lag
compensator.

• With 𝑧 > 𝑝, the compensator is a lag compensator.

• With 𝑝 > 𝑧, the compensator is a lead compensator.
Consider ∠𝐺� 𝑗𝜔 = tan−1𝑧 −
𝜔

tan−1 𝜔 . � 𝑝
• When 𝑧 > 𝑝, ∠𝐺𝑐 𝑗𝜔 is negative for all 𝜔. The compensator provides a phase lag to
sinusoidal signals of all frequencies. Hence the name Lag compensator.

• When 𝑝 > 𝑧, ∠𝐺𝑐 𝑗𝜔 is positive for all 𝜔. The compensator provides a phase lead to
sinusoidal signals of all frequencies. Hence the name Lead compensator.

21
 PID Controllers

We have provided an approximation to both PD and PI controllers.

What about PID controllers?

Lead and Lag compensators are used in cascade to approximate the PID control
action.

22
Performance specification
in the time and
frequency domains

23
 Performance specification in the time and
frequency domains
 In the course so far, we have found the description and specification of the closed-loop behaviour in terms
of overshoot, rise time, settling time, steady state errors etc. is very natural and easy.

 These specifications directly translate into the dominant pole locations in the s-plane, which is very helpful in
the design process.

 However, characteristics such as relative stability, noise rejection etc. are better understood in frequency domain.
Moreover, as we shall see, the design in frequency domain is simpler than the design in time-domain.

 Is there a correlation between the time-domain parameters such as rise time, overshoot etc. and
frequency domain characteristics such as resonant peak, resonant frequency, bandwidth, gain & phase
margin etc. ?

 Exact correlation exists for first and second order systems. For higher order systems, this correlation
holds approximately when the closed-loop system has a pair of dominant poles.

 When the correlation is used for higher-order systems, extensive simulations must be performed before finalizing
the design.

24
 Performance specification in the
time domain
 Consider the closed-loop
system :
𝑅 𝑠 𝑌
𝜔2
𝑛
𝑠 𝑌(𝑠
𝐸 𝑠 𝑠(𝑠 + 𝜔2
)𝑅(𝑠 = 𝑠2 + 2𝜁𝜔
𝑛
2𝜁𝜔𝑛) 𝑛𝑠
+-
𝑛
) + 𝜔2

 The time response of this system to a step input is

characterized by
𝜁 𝜋1−𝜁 2
 Percentage Overshoot 𝑀� = − %
100𝑒
 Settling time 𝑡� = 𝜁𝜔 (2%
4
�

4𝑟 = �
𝑛
tolerance)
 Rise time 𝑡� 𝜋−cos−1
𝜁𝑛
=
𝜔
1−𝜁2
�

 Steady state error in terms of error constants

Typical step Response of a second order 𝐾𝑝 , 𝐾𝑣 and 𝐾𝑎 .
system 25
 Performance specification in the
frequency domain – through the
closed-loop frequency response

 The closed-loop frequency response is characterized by the

following parameters

1
Resonant peak 𝑀𝑟 2𝜁
= 1−𝜁 2

Resonant frequency 𝜔𝑟 = 𝜔𝑛 1 −
2𝜁2
Bandwidth 𝜔𝑏 = 𝜔𝑛 1 − 2𝜁2 + 4𝜁4 − 4𝜁2
+2

• For the system to track inputs quickly, it must have a

Closed-loop frequency
large bandwidth. However, this is not desirable from the
response
noise perspective. In practice, this trade-off is very
important.
26
• 𝑀𝑟 depends only on 𝜁. Therefore, has exact correlation with 𝑀𝑝 and decides the
overshoot in the step response. Just like 𝑀𝑝, 𝑀𝑟 is also indicative of the relative stability.

• The resonant peak exists only for 𝜁

1
2
≤
and the
correlation exists only in the range 0 ≤ 𝜁 2.
1

• This is not of much concern as for 𝜁 > 2

𝑀𝑝 < 0.046 and the overshoot in
1
,
the step
response is very small.

𝑀𝑝 , 𝑀𝑟 v/s 𝜁

• For a given 𝜁, either 𝜔𝑏 or 𝜔𝑟 can be used to specify 𝜔𝑛. Therefore 𝜔𝑏 and 𝜔𝑟 influence the
rise time in the step response. Larger the 𝜔𝑏 or 𝜔𝑟, smaller is the rise time.

27
 Performance specification in the time and
frequency domains
• It must be emphasized again that the exact correlations hold for second order systems.

• For higher order systems, if the system has a pair of dominant poles and the other poles have little
significance, the correlations can be used in the design process.

• As we are interested in the analysis of the response to aperiodic signals rather than
sinusoidal signals, it is natural and easy to specify the system requirements in the time
domain.

• The control system design in the frequency domain is simpler when compared to the design in the time
domain. Therefore the specifications in the time domain and translated to the frequency domain and
the design is carried out.

• The specification in the frequency domain after the design are translated back to time domain to
verify that the requirements are met.

• For higher order system, it is extremely important to verify the design through extensive
simulations, as the correlations are not exact but close.

28