ENGR 6925: Automatic Control Engineering

Class Supplementary Notes 4

Dr. George Mann,
Faculty of Engineering and Applied Science
Memorial University of Newfoundland
Fall 2007

Feedback Control Systems

A control system consists interconnection of components forming a system that will provide
a desired system response. As the desired response is known, a signal proportional to the
feedback error, which is computed while comparing the actual response against the desired
value, is generated. The use of this error signal to control the process results in a closed-loop
sequence of operations, that is called a feedback system.
A system without a feedback is often known as open-loop system. Thus, an open-loop (di-
rect) system operates without feedback and directly drives a system with a known
input signal. By contrast, a closed-loop with negative feedback uses a measurement
of the output signal and a comparison with the desired output to generate an error
signal that is used by the controller to adjust the actuator.

Controller Process

Measurement
Comparison

Figure 1: A closed-loop system

The block diagram of a closed-loop system is shown below. In this case the feed forward block
G(s) represents actuator, process and controller. The feedback transfer function is denoted as
H(s).

1
ENGR6925: Automatic Control Engineering, Fall 2006 2

R(s) E(s) X(s) R(s) X(s)

G(s) G(s)
+
-
1+G(s)H(s)

B(s)
closed-loop
H(s) transfer function

Figure 2: Block diagram for a closed-loop system

Open-Loop transfer function In the open-loop transfer function we consider no feedback,

in that case,
B(s)
Open − Loop Transfer Function = = G(s)H(s)
E(s)
Feedforward transfer function The feedforward transfer function consists signal amplifier,
controller, actuator, process etc.
X(s)
Feedforward Transfer Function = = G(s)
E(s)

Closed-Loop transfer function From the diagram we can see that

X(s) = G(s)E(s)

E(s) = R(s) − B(s)

E(s) = R(s) − H(s)X(s)
Eliminating E(s) we get

X(s) = G(s) [R(s) − H(s)X(s)]

X(s) G(s)
=
R(s) 1 + G(s)H(s)
Thus the overall closed-loop transfer function can be described as;
G(s)
X(s) = R(s)
1 + G(s)H(s)

Closed-loop system subjected to external disturbance

In practice the control systems are subjected to external disturbances. As an example, when
we drive a load, the load system can be changed suddenly causing an external disturbance.
Also there can be external noise in the system that influence the actuator signal. Thus we
include this disturbance in a block diagram as shown in figure below. The block diagram has
two inputs, reference signal R(s) and disturbance D(s). We can solve in two steps, Consider
the disturbance is zero.
For D(s) = 0 the output for the reference input is YR (s), therefore

YR (s) G1 (s)G2 (s)

=
R(s) 1 + G1 (s)G2 (s)H(s)
ENGR6925: Automatic Control Engineering, Fall 2006 3

Disturbance
D(s)

R(s) E(s) Y(s)

+
+ G1(s) + G2(s)
-

H(s)

Figure 3: Closed-loop system with disturbance

For R(s) = 0 the output for the disturbance input is YD (s) and is given by;

YD (s) G2 (s)
=
D(s) 1 + G1 (s)G2 (s)H(s)

The total output is given by;

Y (s) = YR (s) + YD (s)
Therefore;
G1 (s)G2 (s) G2 (s)
Y (s) = R(s) + D(s)
1 + G1 (s)G2 (s)H(s) 1 + G1 (s)G2 (s)H(s)
G2 (s)
Y (s) = [G1 (s)H(s) + D(s)]
1 + G1 (s)G2 (s)H(s)
The error also can be shown as;

1 G2 (s)H(s)
E(s) = R(s) − D(s)
1 + G1 (s)G2 (s)H(s) 1 + G1 (s)G2 (s)H(s)

The open-loop gain L(s) = G1 (s)G2 (s)H(s)

1 G2 (s)H(s)
E(s) = R(s) − D(s)
1 + L(s) 1 + L(s))

In general we prefer to have a larger loop gain L(s) so that the effect of disturbance is minimized.
Generally this works out at lower frequencies. However, increasing loop gain can cause the
measurement noise to be magnified and therefore can cause some bad control performance.
Generally measurement noise is high at higher frequencies. Thus a controller is desired to have
a lower gain at higher frequencies and higher gain at lower frequencies.

Proportional, Integral and Derivative (PID) Control

Proportional Control
Proportional control provides a control signal proportional to the error feedback, given by;

up = Kp e(t)
ENGR6925: Automatic Control Engineering, Fall 2006 4

Thus
Up (s) = Kp E(s)
where Kp is the proportional gain of the controller. It is important to note here that when error
converges to zero, the proportional signal also becomes zero. Therefore if a system does not
require a control signal once the target has been reached, the proportional control can provide
the driving force to reach the target.
Integral Control
Integral control action provides a signal proportional to the accumulated feedback error over a
period of time. Therefore
Zt
ui = Ki edt
0

Thus
Ki
Ui (s) =
s
Integral control action will have a finite value once error converges to zero. Therefore at steady
state the integral action will be active with a constant control output.
Derivative Control
The derivative control intends to provide damping to a control system and is proportional to
the error rate.
de(t)
ud = Kd
dt
Thus
Ud (s) = Kd sE(s)
We sometimes write using derivative time constant as;

Ud (s) = Kp (τd sE(s))

Proportional control of a first-order process

R(s) E(s) Y(s)

+ Kp 1
- Ts + 1

Figure 4: Proportional control of a 1st order process

Consider the first-order process with unity feedback system as shown in the figure above.
The overall transfer function is given by;
Y (s) G(s)
=
R(s) 1 + G(s)
ENGR6925: Automatic Control Engineering, Fall 2006 5

where
Kp
G(s) =
Ts + 1
The error signal is given by;
E(s) = R(s) − Y (s)
E(s) Y (s) 1
=1− =1−
R(s) R(s) 1 + G(s)
E(s) 1
=
R(s) 1 + G(s)
E(s) 1
=
R(s) 1 + TKs+1
p

E(s) Ts + 1
=
R(s) T s + Kp + 1
For a unit-step input R(s) = 1/s;
Ts + 1 1
E(s) =
T s + Kp + 1 s
To find the steady state response we use the final-value theorem, given as;
ess = lim e(t) = lim sE(s)
t→∞ s→0

s(T s + 1) 1 Ts + 1
ess = lim =
s→0 T s + Kp + 1 s T s + Kp + 1
1
ess =
Kp + 1
It is clear that the proportional control of a first-order will result in a steady state error value.

y(t)

1
1+KP

Figure 5: steady state error with Kp control of 1st order system

In order to reduce it is possible to increase the proportional gain so that the fraction 1/(1 + Kp )
will become appreciably small. However, such increase requires a considerable input to the
process.
ENGR6925: Automatic Control Engineering, Fall 2006 6

Integral control of a first-order process

R(s) E(s) Y(s)

KI 1
+
- s Ts + 1

Figure 6: Integral control of a 1st order process

Now consider the same process being controlled by an integral controller. For this system;
KI
G(s) =
s(T s + 1)
E(s) 1
=
R(s) 1 + s(TKs+1)
I

E(s) s(T s + 1)
=
R(s) T s2 + s + KI
For a unit-step input;
s(T s + 1) 1
E(s) =
T s2 + s + KI s
(T s + 1)
E(s) =
T s2 + s + KI
Using final value theorem;
s(T s + 1)
ess = lim sE(s) = lim
s→0 s→0 T s2 + s + KI

ess = 0
The integral control output can made to be faster by employing a higher gain for KI .
However, we can see from the transfer function that the second order characteristic equation
will start producing complex roots, which will cause oscillation in the output. The roots of the
Characteristic equation can be found by
 
2 2 s KI 1 2 KI 1
T s + s + KI = T (s + + ) = T (s + ) + −
T T 2T T 4T 2
For non-oscillatory response,
KI 1
− ≤0
T 4T 2
i.e.
1
KI ≤
4T
ENGR6925: Automatic Control Engineering, Fall 2006 7

However, such a response would be quite low. To increase the response rate, we use PI (Pro-
portional plus Integral) control. The proportional control would provide additional force to
drive the system towards the target, and the integral controller would maintain the steady state
control. We can perform the following analysis to determine the overall characteristic equation
and also a disturbance analysis to control a first order system using PI control.

R(s) E(s) Y(s)

+ KP + KI 1
- s Ts + 1

Figure 7: PI control of a 1st order process

To find the general response, consider the case when D(s) = 0.

  
KI 1 KP s + KI
G(s) = KP + =
s Ts + 1 s(T s + 1)

KP s+KI
Y (s) s(T s+1)
=
R(s) 1 + Ks(TP s+KI
s+1)

Y (s) sKp + KI
= 2
R(s) T s + (KP + 1)s + KI
We can also find E(s) using;

E(s) Y (s) T s2 + s
=1− =
R(s) R(s) T s2 + (KP + 1)s + KI

Thus for unit-step response, R(s) = 1/s and

T s2 + s 1
ess = lim sE(s) = lim s 2
=0
s→0 s→0 T s + (KP + 1)s + KI s

This is understandable, when Integral controller is included the steady state error always be-
comes zero for a first order process. To find the load disturbance properties, consider the case

When R(s) = 0;
Y (s) 1/(T s + 1)
= =
D(s) 1 + (kP + KI /s)(1/(T s + 1))
Y (s) s s
= = 2
D(s) s(T s + 1) + (Kp s + KI ) T s + (Kp + 1)s + KI
ENGR6925: Automatic Control Engineering, Fall 2006 8

We know E(s) = R(s) − Y (s) = −Y (s). Therefore;

E(s) −s
= 2
D(s) T s + (Kp + 1)s + KI

Let us assume a constant disturbance input, such as d(t) = Td ; then

Td
D(s) =
s
Thus using final value theorem;

−s2 Td
ess = lim 2
=0
s→0 T s + (Kp + 1)s + KI s

Now we can see from the above analysis, even under a load disturbance, the PI controller has
the ability to converge the output to its target state and this proves its robustness in controlling
a first order process.

Proportional position control of a motor

The position control of a motor using a proportional control is shown below. Values J represent
inertia of rotating parts and b represents damping (or frictional resistance) of the system. First

D(s)

R(s) E(s) 1 Y(s)

KP
+
+ +
- s(js+b)

Figure 8: Proportional control of a motor

consider the case when there is no disturbances. In that case;

E(s) 1 s(Js + b)
= KP
= 2
R(s) 1 + s(Js+b) Js + bs + Kp

Assume a unit-step input, R(s) = 1/s. Then

s(Js + b) 1 (Js + b)
E(s) = =
Js2 + bs + Kp s Js2 + bs + Kp

s(Js + b)
ess = lim sE(s) = =0
s→0 Js2 + bs + Kp
ENGR6925: Automatic Control Engineering, Fall 2006 9

This indicates the Proportional control is sufficient for position control of a motor. Also, we see
in the process model the term s/s, which provides self-integration ability for the motor. Now
consider the load disturbance aspects.

For R(s) = 0;
1
Y (s) s(Js+b)
= KP
D(s) 1 + s(Js+b)
Y (s) 1
= 2
D(s) Js + bs + KP
As E(s) = R(s) − Y (s) = −Y (s)
−D(s)
E(s) =
Js2 + bs + KP
For a constant load disturbance of d(t) = Td
Td
D(s) =
s
Then
−Td
E(s) =
s(Js2 + bs + KP )
Using final-value theorem;
−sTd
ess = lim sE(s) = lim
s→0 s→0 s(Js2 + bs + KP )
−Td
ess = lim
s→0 (Js2
+ bs + KP )
−Td
ess =
KP
We can see that the motor tries to provide an opposite torque at the steady state to overcome
the load disturbance. This proves the proportional controller is insufficient to provide load
disturbance properties, although it can work well for a position control. Now we will look at
the PI control system.

Proportional and Integral position control of a motor

For the load disturbances we can show;
−D(s)s
E(s) =
Js3 + bs2 + KP s + KI
For a step disturbance of D(s) = Td /s, this can be shown as;
ess = lim sE(s) = 0
s→0

I will let you to do the simplification and prove this as an exercise.

ENGR6925: Automatic Control Engineering, Fall 2006 10

D(s)

R(s) E(s) 1 Y(s)

+
+ KP + KI +
- s s(js+b)

Figure 9: PI control of a motor

Proportional and Derivative control of a second order system

Consider the PD control, wehre
de(t)
upd = KP e(t) + KD
dt
or in Laplace form,
UP D (s) = K + pE(s) + KD sE(s) = (KP + KD s)E(s)
It can easily be shown that the closed-loop transfer function as;

E(s)
R(s)
KP + sKD
1 Y(s)
+
- s(js+b)

Figure 10: PD control of a second order system

Y (s) KP + KD s
= 2
R(s) Js + (b + KD )s + Kp
From the Characteristic equation it is clear that the addition of derivative control able to
increase the overall damping effect of the system. The effective damping coefficient can now be
found as;
b + KD
ξ= √
s KP J

Some illustrative examples on Closed-Loop systems

Example 1: Consider the closed loop transfer function shown below. The unit-step response
shows 25.4% peak overshoot and the peak overshoot time is 3 seconds. Determine K and T of
the model. The closed loop transfer function is given by;
ENGR6925: Automatic Control Engineering, Fall 2006 11

R(s) K X(s)
+
- s(Ts + 1)

Figure 11: Example 1

X(s) G(s)
=
R(s) 1 + G(s)

where G(s) = K/s(T s + 1),

X(s) K/s(T s + 1)
=
R(s) 1 + K/s(T s + 1)
X(s) K
=
R(s) T s2 + s + K
Thus the closed-loop transfer function is a second order system. The oscillatory response is due
to under-damping conditions. We can re-write the same as;

X(s) K/T
= 2
R(s) s + s/T + K/T

Therefore; p
ωn = K/T , 2ξωn = 1/T
Thus;
X(s) ωn2
= 2
R(s) s + 2ξωn s + ωn2
The unit-step response is given by when R(s) = 1/s. Therefore;’

ωn2 1
X(s) =
s2 + 2ξωn s + ωn2 s

Using inverse Laplace the response is;

1
x(t) = 1 − p e−ξωn t sin (ωd t + φ)
1−ξ 2

p p 
where ωd = ωn 1 − ξ 2 and φ = tan−1 1 − ξ 2 /ξ
The peak overshoot occurs when dx(t)/dt = 0. While differentiating we can write;

dx(t) 1
−ξωn e−ξωn t sin (ωd t + φ) + ωd e−ξωn t cos (ωd t + φ) = 0
 
= −p
dt 1 − ξ2
ENGR6925: Automatic Control Engineering, Fall 2006 12

e−ξωn t [−ξωn sin (ωd t + φ) + ωd cos (ωd t + φ)] = 0

While expanding this we can show;
sin(ωd t) = 0
Therefore at peak over-shoot
ωd tp = 0, π, 2π, ...
Since the peak overshoot is at the first oscillation,

ωd tp = π
φ
tp = p
ωn 1 − ξ 2
The maximum overshoot is when t = tp and ωd tp = φ. Thus
1
x(tp ) = 1 − p e−ξωn tp sin (π + φ)
1−ξ 2

we can also argue that the peak overshoot occurs at the half-cycle time. Therefore,
1
x(tp ) = 1 + p e−ξωn π/ωd sin (φ)
1 − ξ2

The peak overshoot Mp = x(tp ) − 1. Therefore

1
Mp = p e−ξωn π/ωd sin (φ)
1 − ξ2
√
It can be shown that sin φ = 1 − xi2 . Therefore
√
1−ξ 2
Mp = e−ξωn π/ωd = e−ξπ/

In this case, Mp = 0.254, therefore

−ξπ
p = ln(Mp )
1 − ξ2
We will get;
(ln Mp )2
ξ2 =
π 2 + (ln Mp )2
0.5
(ln 0.254)2

ξ=
π 2 + (ln 0.254)2
Therefore
ξ = 0.4
As the time to overshoot is 3 seconds
π π
tp = = p =3
ωd ωn 1 − ξ 2
ENGR6925: Automatic Control Engineering, Fall 2006 13

Therefore
π
ωn = √ = 1.14 rad/s
3 1 − 0.42
Therefore
1
= 2ξωn = 2 × 0.4 × 1.14
T
Therefore
T = 1.09 s
As
K
= ωn2
T
K = T ωn2 = 1.09 × 1.142 = 1.42