ROBT303 Linear Control Theory with Lab

Root locus
Tohid Alizadeh

Academic year 2016-2017

Tohid Alizadeh (NU)

ROBT303 Linear Control Theory with Lab

Academic year 2016-2017

1 / 37

Part I

What is the root locus?

Tohid Alizadeh (NU)

ROBT303 Linear Control Theory with Lab

Academic year 2016-2017

2 / 37

Motivation: proportional control

Assume that we have a standard feedback loop, where we neglect the presence
of disturbance terms
If

n (t )

and

d (t )

u (t ) = kp e (t ) = Gc (s ) = kp ,
r

i.e. we have a proportional controller

- n

kp

Gp (s)

PROPORTIONAL

ACTUATOR

CONTROLLER

+ PROCESS

+
+

In order to design the proportional controller, we must choose

the closed-loop system is stable

we respect transient specications (e.g., about

We are then interested in studying how the poles of

Mp

kp

and/or

T (s )

such that

Ts ,2% )

vary as a function of

kp

Actually, these poles give us

complete information about closed-loop stability

approximate information on step response parameters, in case we can make

a second-order approximation

Tohid Alizadeh (NU)

ROBT303 Linear Control Theory with Lab

Academic year 2016-2017

3 / 37

Denition of root locus

Where are these poles? First we assume that

Gp (s ) = K 0 Gp0 (s ) = K 0
where

N (s )

and

D (s )

Then, by dening

KGp (s )

is in Evans form, i.e.

(s + z1 )(s + z2 ) . . . (s + zm )
N (s )
= K0
D (s )
(s + p1 )(s + p2 ) . . . (s + pn )

are coprime polynomials (no zero-pole cancellations)

= kp K 0

T (s ) =

L(s )
=
1 + L( s )

Gp0 (s )
1 + Gp0 (s )

N (s )
D (s ) + N (s )

Therefore, the poles of the closed-loop system are the roots of the characteristic
equation

D (s ) + N (s ) = 0

Denition
The root locus of a feedback system is a plot of the poles of

T (s )

as parameter

changes

Tohid Alizadeh (NU)

ROBT303 Linear Control Theory with Lab

Academic year 2016-2017

4 / 37

A simple example

The shown camera automatically

follows a subject who wears infrared
sensors on their front and back
Tracking commands are relayed to
the camera via a radio frequency link
from a unit worn by the subject
A proportional controller has to be
designed, in order for the camera to
follow the subject

Tohid Alizadeh (NU)

ROBT303 Linear Control Theory with Lab

Academic year 2016-2017

5 / 37

A simple example
In this case, we nd the roots of

D (s ) + KN (s ) = s 2 + 10s + K = 0
Since it is a second-order polynomial,
it is easy to see how the poles change
as a function of

=K 0

in this case

In general, we want a method to

draw the root locus for an arbitrary
number of roots
We call positive locus (PL) the values

0, and negative
locus (NL) the values of the roots for
of the roots for

<0

Tohid Alizadeh (NU)

ROBT303 Linear Control Theory with Lab

Academic year 2016-2017

6 / 37

Phase and magnitude conditions

The characteristic equation can be rewritten in dierent ways

D (s ) + N (s ) = 0,

1+

N (s )
= 1 + Gp0 (s ) = 0,
D (s )

Gp0 (s ) =

N (s )
1
=
D (s )

All points of the complex plane which satisfy any of these conditions are points
of the root locus
Looking at the last condition, and remembering that, for a given complex
number

s , Gp0 (s )

is another complex number, we need

Magnitude condition:



 N (s ) 
1


 D (s )  = ||

Phase condition:

Tohid Alizadeh (NU)

N (s )
+ 2h,
1
=
=
D (s )

0 + 2h,

ROBT303 Linear Control Theory with Lab

> 0 (PL)
< 0 (NL)

Academic year 2016-2017

7 / 37

Phase condition: example

Once a point

satises the phase condition, we can always nd a value of

such that the magnitude condition is satised

Dening

= the angle associated to the zero in

associated to the pole in

pi ,

zi ,

and

= the angle

we get

m
n
X
N (s ) X
=
i
i
D (s ) i = 1
j =1

Figure 6-2 (a) and (b) Diagrams showing angle measurements from open-loop poles and open-loop zero to test point s.

Example:

Gp0 (s ) = 1 1 2 3 4

Next, we dene some properties, useful to draw the root locus of a given system
Modern Control Engineering, Fifth Edition

Tohid Alizadeh (NU)Katsuhiko Ogata

Copyright 2010, 2002, 1997 by Pearson Education, Inc.

All rights reserved.
Academic
year

ROBT303 Linear Control Theory with Lab

2016-2017

8 / 37

Root locus with Matlab

The root locus can be plotted by using Matlab command

rlocus()

However, we want to have some intuition on how the root locus is sketched,
especially for synthesizing controllers

Tohid Alizadeh (NU)

ROBT303 Linear Control Theory with Lab

Academic year 2016-2017

9 / 37

Root locus with Matlab

The root locus can be plotted by using Matlab command

rlocus()

However, we want to have some intuition on how the root locus is sketched,
especially for synthesizing controllers

So, let's move to the sketching rules. . .

Tohid Alizadeh (NU)

ROBT303 Linear Control Theory with Lab

Academic year 2016-2017

9 / 37

Part II

Root locus properties

Tohid Alizadeh (NU)

ROBT303 Linear Control Theory with Lab

Academic year 2016-2017

10 / 37

Number of branches and symmetry

Rule 0
The root locus is symmetrical with respect to the real axis

Since

N (s )

and

D (s )

have real coecients, if a complex root is present in the

characteristic equation, then its complex conjugate is also present.

Rule 1
The root locus has 2n branches,

in the positive locus and

in the negative locus

This can obtained from the fact that the polynomial equation

D (s ) + N (s ) = 0
has

complex roots, which continuously depend on the parameter

Tohid Alizadeh (NU)

ROBT303 Linear Control Theory with Lab

Academic year 2016-2017

11 / 37

Origin and end of the branches

Rule 2
For both the PL and the NL,

Gp (s ).

Amongst these,

branches of the root locus originate at the poles of

terminate on the zeros of

Gp (s ),

and

r =nm

at innity

The characteristic equation can be written as

1+

from which

(s + p1 )(s + p1 ) . . . (s + pn )

=0

(s + p1 )(s + p1 ) . . . (s + pn ) + (s + z1 )(s + z1 ) . . . (s + zm ) = 0

We see that for

For

(s + z1 )(s + z2 ) . . . (s + zm )

= 0,

the solutions coincide with the poles of

Gp ( s )

, the equation degenerates into (s + z1 )(s + z1 ) . . . (s + zm ) = 0,

m roots, coinciding with the zeros of Gp (s )

which has

The other

r =nm

Tohid Alizadeh (NU)

branches will end at innity

ROBT303 Linear Control Theory with Lab

Academic year 2016-2017

12 / 37

Asymptotic behavior
Rule 3
The branches that go to innity have asymptotes which meet on the real axis, at the
point of abscissa

xa =
and form with the real axis angles

r a =

m
X
i =1

zi

n
X
i =1

pi

such that

(2h + 1),
2h,

h = 0, 1, . . . , r 1
h = 0, 1, . . . , r 1

(PL)
(NL)

The result is obtained by noticing that

lim

Gp0 (s ) =

lim

sm
= lim r
s s
sn

and then nding the solutions for PL and NL

The computation of

Tohid Alizadeh (NU)

xa

has a meaning only if

r 2

ROBT303 Linear Control Theory with Lab

Academic year 2016-2017

13 / 37

Root locus on the real axis

Rule 4
All the points on the real axis belong to the root locus.
If the total number of real poles + real zeros to the right of the test point is

odd, the test point belongs to the PL

If the total number of real poles + real zeros to the right of the test point is

even, the test point belongs to the NL

Figure 6-8 Determination of the root locus on the real axis.

This is easily obtainable, since

all pairs of complex poles give an
angle contribution equal to zero
the same for all the real
poles/zeros to the left of the test
point
the real poles/zeros to the right
of the test point give a
contribution of

each, from

which the rule follows

Tohid Alizadeh (NU)

Modern Control
Engineering,
FifthLab
Edition
ROBT303 Linear Control
Theory
with

2010, 2002, 199714

by Pearson
Academic yearCopyright
2016-2017
/ 37Educ

Breakaway and break-in points

Rule 5
A necessary condition for a point of the root locus to be a breakaway or break-in
point is

N (s )D 0 (s ) N 0 (s )D (s ) = 0
for both PL and NL.

N 0 (s )

and

D 0 (s )

represent the derivatives w.r.t.

of

N (s )

and

D (s ).
Root Locus
1

0.982

0.962

0.925

0.86

0.72

0.45

0.8
0.6 0.992

BREAKAWAY POINT

0.4
Imaginary Axis

0.998
0.2
4

0.2
0.998
0.4
0.6 0.992

BREAKIN POINT
0.8
1
5

Tohid Alizadeh (NU)

0.962

0.982
4

0.925
3

0.86

2
Real Axis

0.72

0.45

ROBT303 Linear Control Theory with Lab

Academic year 2016-2017

15 / 37

Angles of departure
Rule 6
The angles of departure

is

q dep =

of a branch of the root locus from a pole of multiplicity

X
m
n
X

i + (2k + 1),
i

i =1
m
X

where the angles

dep

i =1

and

i =1
n
X
i =1

i + 2k ,

k = 0, 1, . . . , q 1 (PL)
k = 0, 1, . . . , q 1 (NL)

are computed as usual, by considering the current pole as

a test point, and neglecting the associated angle (i

= 0).

This rule is obtained by checking the

phase condition in the neighborhood
of the considered pole

Tohid Alizadeh (NU)

ROBT303 Linear Control Theory with Lab

Academic year 2016-2017

16 / 37

Angles of arrival
Rule 7
The angles of arrival

q arr =

arr

where the angles

of a branch of the root locus to a zero of multiplicity

X
m
n
X

+
i + (2k + 1),
i

i =1
m
X
i =1

and

i +

i =1
n
X
i =1

i + 2k ,

is

k = 0, 1, . . . , q 1 (PL)
k = 0, 1, . . . , q 1 (NL)

are computed as usual, by considering the current zero as

a test point, and neglecting the associated angle (i

= 0).

This rule can be also obtained by checking the phase condition, this time in the
neighborhood of the considered zero

Tohid Alizadeh (NU)

ROBT303 Linear Control Theory with Lab

Academic year 2016-2017

17 / 37

Some notes
Remember that, once we know that a point
want to nd the associated gain

kp

s = s is

on the root locus, and we

of the proportional controller, we just need

to use the magnitude condition:





 N (

 D (

 s )  = 1 ||
 s) 

=
 D (



s)
||
N (
s) 
We obtain the sign of

simply checking if the considered point is on the PL or

on the NL, and then

= kp K 0 kp =

K0

In case we are asked to nd the range of values of

(or

kp )

for which the

closed-loop system is BIBO stable, we can apply the Routh criterion, and nd
the values of

for which the rst column of the Routh table associated to

D (s ) + N (s )
has all positive values

Tohid Alizadeh (NU)

ROBT303 Linear Control Theory with Lab

Academic year 2016-2017

18 / 37

Part III

Drawing the root locus: examples

Tohid Alizadeh (NU)

ROBT303 Linear Control Theory with Lab

Academic year 2016-2017

19 / 37

Examples
r

- n

Example 1 (Click

kp

Gp (s)

PROPORTIONAL

ACTUATOR

CONTROLLER

+ PROCESS

+
+

for the corresponding YouTube video)

Draw a sketch of the root locus associated to the standard feedback loop with
proportional controller

Gc (s ) = kp

for a process described by

Gp ( s ) =

s +1

and compare the obtained solution with that obtained with Matlab. In addition:
compute (if possible) the value of
closed-loop system,

Ts ,5% = 0.3

kp

such that, given the step response of the

nd the steady-state error

nd the range of values of

kp R

to

r (t ) = step(t )

for the same closed-loop system

such that the closed-loop system is BIBO

stable
Tohid Alizadeh (NU)

ROBT303 Linear Control Theory with Lab

Academic year 2016-2017

20 / 37

Examples (cont'ed)
r

- n

Example 2 (Click

kp

Gp (s)

PROPORTIONAL

ACTUATOR

CONTROLLER

+ PROCESS

+
+

for the corresponding YouTube video)

Draw a sketch of the root locus associated to the standard feedback loop with
proportional controller

Gc (s ) = kp

for a process described by

Gp (s ) =

0.05
s (1 + 10s )

and compare the obtained solution with that obtained with Matlab. In addition:
compute (if possible) the value of

kp

such that the step response of the

closed-loop system is critically damped

nd the steady-state error

to

r (t ) = 5ramp(t )

for the same closed-loop

system
nd the range of values of
Tohid Alizadeh (NU)

kp R

such that closed-loop system is BIBO stable

ROBT303 Linear Control Theory with Lab

Academic year 2016-2017

21 / 37

Examples (cont'ed)
r

- n

Example 3 (Click

kp

Gp (s)

PROPORTIONAL

ACTUATOR

CONTROLLER

+ PROCESS

+
+

for the corresponding YouTube video)

Draw a sketch of the root locus associated to the standard feedback loop with
proportional controller

Gc (s ) = kp

for a process described by

Gp (s ) =

15

s (s + 3)(s + 5)

and compare the obtained solution with that obtained with Matlab. In addition:
nd the range of values of

kp R

such that the closed-loop system is BIBO

stable
compute (if possible) the value of

kp

such that the step response of the

closed-loop system, using (if possible) a second-order approximation, is

undamped
Tohid Alizadeh (NU)

ROBT303 Linear Control Theory with Lab

Academic year 2016-2017

22 / 37

Examples (cont'ed)
r

- n

Example 4 (Click

kp

Gp (s)

PROPORTIONAL

ACTUATOR

CONTROLLER

+ PROCESS

+
+

for the corresponding YouTube video)

Draw a sketch of the root locus associated to the standard feedback loop with
proportional controller

Gc (s ) = kp

for a process described by

Gp ( s ) =

s +1
s2

and compare the obtained solution with that obtained with Matlab. In addition:
nd the range of values of

kp R

such that the closed-loop system is BIBO

stable
compute (if possible) the value of

kp

such that the step response of the

closed-loop system, using a second-order approximation, is critically damped

(i.e., no overshoot!)
Tohid Alizadeh (NU)

ROBT303 Linear Control Theory with Lab

Academic year 2016-2017

23 / 37

Examples (cont'ed)
r

- n

kp

Example 5 (Click

Gp (s)

PROPORTIONAL

ACTUATOR

CONTROLLER

+ PROCESS

+
+

for the corresponding YouTube video)

Draw a sketch of the root locus associated to the standard feedback loop with
proportional controller

Gc (s ) = kp

for a process described by

s +2
Gp (s ) = 2
s + 2s + 2
and compare the obtained solution with that obtained with Matlab. In addition:
nd the range of values of

kp R

such that the closed-loop system is BIBO

stable
for which value of

Tohid Alizadeh (NU)

kp

is it possible to approximate

T (s )

ROBT303 Linear Control Theory with Lab

as a rst-order system?

Academic year 2016-2017

24 / 37

Examples (cont'ed)
r

- n

kp

Example 6 (Click

Gp (s)

PROPORTIONAL

ACTUATOR

CONTROLLER

+ PROCESS

+
+

for the corresponding YouTube video)

Draw a sketch of the root locus associated to the standard feedback loop with
proportional controller

Gc (s ) = kp

for a process described by

Gp (s ) =

s (s + 1)(s 2 + 4s + 13)

and compare the obtained solution with that obtained with Matlab. In addition:
nd the range of values of

kp R

such that the closed-loop system is BIBO

stable
nd the value of

Tohid Alizadeh (NU)

kp

for which the step response will be undamped (if possible)

ROBT303 Linear Control Theory with Lab

Academic year 2016-2017

25 / 37

Examples (cont'ed)
r

- n

Example 7 (Click

kp

Gp (s)

PROPORTIONAL

ACTUATOR

CONTROLLER

+ PROCESS

+
+

for the corresponding YouTube video)

Draw a sketch of the root locus associated to the standard feedback loop with
proportional controller

Gc (s ) = kp

for a process described by

s +b
Gp (s ) = 2
s (s + a )
and compare the obtained solution with that obtained with Matlab, for the following
conditions:

a = 10,

and

b=1

a = 9,

and

b=1

a = 8,

and

b=1

Tohid Alizadeh (NU)

ROBT303 Linear Control Theory with Lab

Academic year 2016-2017

26 / 37

Examples (cont'ed)
r

- n

kp

Gp (s)

PROPORTIONAL

ACTUATOR

CONTROLLER

+ PROCESS

+
+

Example 8
Draw a sketch of the root locus associated to the standard feedback loop with
proportional controller

Gc (s ) = kp

for a process described by

Gp (s ) =

(s + 1)2
(s + 2)2

and compare the obtained solution with that obtained with Matlab. In addition:
nd the range of values of
stable, by using

Tohid Alizadeh (NU)

both

kp R

such that the closed-loop system is BIBO

the magnitude condition and the Routh criterion

ROBT303 Linear Control Theory with Lab

Academic year 2016-2017

27 / 37

Examples (cont'ed)
r

- n

kp

Gp (s)

PROPORTIONAL

ACTUATOR

CONTROLLER

+ PROCESS

+
+

Example 9
Draw a sketch of the root locus associated to the standard feedback loop with
proportional controller

Gc (s ) = kp

for a process described by

Gp (s ) =

(s + 1)2
(s + 2)4

and compare the obtained solution with that obtained with Matlab. In addition:
nd the range of values of

kp R

such that the closed-loop system is BIBO

stable, by directly using the magnitude condition instead of the Routh criterion

Tohid Alizadeh (NU)

ROBT303 Linear Control Theory with Lab

Academic year 2016-2017

28 / 37

Part IV

Design of proportional controllers via root

locus

Tohid Alizadeh (NU)

ROBT303 Linear Control Theory with Lab

Academic year 2016-2017

29 / 37

Design procedure
The design procedure for a proportional controller using the root locus can be
summarized as follows:
1
2

Sketch the root locus for the given system

Translate the transient requirements (typically,

Mp , Tr ,

and

Ts

for a step

response) into a feasible area of the root locus, by assuming that

T (s )

can be

approximated as a second-order system with no zeros

3

Translate the steady-state requirements (typically,

responses) into conditions on the gain

kp :

for step or ramp

these conditions are exact (no

approximation needs to be checked a posteriori)

4

Find a gain

kp

which leads to the satisfaction of both transient and steady-state

requirements (if possible!)

5

Have a look at the additional poles and/or zeros, and plot the step response, to
verify if the transient requirements are actually met (i.e., if the approximation
was good)

If they are not met, further modify

kp ,

and try to meet all the requirements

For the sake of simplicity, we will only consider the case

Tohid Alizadeh (NU)

ROBT303 Linear Control Theory with Lab

kp > 0

Academic year 2016-2017

30 / 37

Transient requirements: percent overshoot

To satisfy the requirement

p ,
Mp < M

as

= q

Consider that

= cos( ),

where

we need to nd the corresponding value of

p /100)
ln(M
2

p /100)
2 + ln (M
is the angle between the line connecting the

pole with positive imaginary part and the origin, and the negative real axis
Therefore, the poles will be constrained in the white sector in the gure, and the
limit angle

will become smaller and smaller as

Tohid Alizadeh (NU)

increases

ROBT303 Linear Control Theory with Lab

Academic year 2016-2017

31 / 37

Transient requirements: rise time

We roughly approximate the rise time as

Tr < Tr ,

we nd

We know that

n = 1.8/Tr ,

Tr = 1.8/n ,
n
n >

so, given a condition

and we impose

is the Euclidean distance of any of the two complex poles

from the origin (i.e., the magnitude of the corresponding complex number)
Therefore, the poles will be constrained in the white area in the gure, and the
forbidden circle will increase as a smaller

Tohid Alizadeh (NU)

Tr

is required

ROBT303 Linear Control Theory with Lab

Academic year 2016-2017

32 / 37

Transient requirements: settling time

Ts ,2% < Ts ,2% (same for Ts ,5% < Ts ,5% ), we know that
d = n is (minus) the real part of any of the complex

To satisfy the requirement

Ts ,2% = 4/d ,

where

poles
Therefore, we nd

d = 4/Ts ,2%

and impose

d
d >

The poles will be constrained in the white half plane in the gure, and the
dividing line will move left as

Tohid Alizadeh (NU)

Ts ,2%

increases

ROBT303 Linear Control Theory with Lab

Academic year 2016-2017

33 / 37

Steady-state requirements
The fact that the steady-state error for a given polynomial reference will be zero
does not depend on the chosen gain
the process

kp ,

but only on the number of integrators of

Gp ( s )

Instead, if we had a nite value of

smaller by increasing

for a given reference, we can make it

kp

For instance, if

Gp ( s ) =
then

L(s ) =

s +3
kp
s +3

In this case, if the requirement is

e = 0 for a step reference, then we cannot

e < 10% of the reference step signal, then
e =

1
kp

1+ 3

do it

< 0.1 kp > 27

nite for a ramp or parabolic reference, it is impossible, since

for all values

Tohid Alizadeh (NU)

kp > 0
ROBT303 Linear Control Theory with Lab

Academic year 2016-2017

34 / 37

Steady-state requirements (cont'ed)

Instead, if

Gp (s ) =
then

L( s ) =

kp
s

In this case, if the requirement is

e = 0 for a step reference, it will always

e < 20 for a unit ramp reference, then
e =

kp

be satised

< 0.2 kp > 5

e = 0 for a ramp reference, then we cannot do it

e nite for parabolic reference, it is impossible, since e
kp > 0

for all

Analogous considerations hold for the rejection of process disturbances

d (t )

Important: for some systems, increasing the gain too much leads to instability

Tohid Alizadeh (NU)

ROBT303 Linear Control Theory with Lab

Academic year 2016-2017

35 / 37

Example: DC motor
Considering again the previously-analyzed DC motor, we recall that its transfer
function was

Gp (s ) =

0.01
0.005s 2

+ 0.06s + 0.1001

Let us consider the following requirements:

Ts ,2% < 2

s,

Mp < 25%, e < 10%

for a step reference

By using the

rltool

setting, for instance,

of Matlab, we can verify that the requirements are met by

kp = 87.7

(no need to check the approximation)

Step Response
1.4

1.2

Amplitude

0.8

0.6

0.4

0.2

Tohid Alizadeh (NU)

0.2

0.4

ROBT303 Linear Control Theory with Lab

0.6
Time (sec)

0.8

1.2

Academic year 2016-2017

36 / 37

Example: DC motor
Let us assume now that the same motor is used for a high-precision application,
for which we need
By using the

Ts ,2% < 2

rltool

s,

Mp < 5%, e < 1%

of Matlab, we can verify that the requirements cannot be

met by using a proportional controller, since not enough degrees of freedom are
provided
Step Response: kp=27

Step Response: kp=1025

0.8

1.8

0.7

1.6
1.4

0.6

1.2
Amplitude

Amplitude

0.5
0.4

1
0.8

0.3
0.6
0.2

0.4

0.1
0

0.2

0.2

0.4

0.6
0.8
Time (sec)

1.2

1.4

To have

Mp < 5%,

we can set at most

To have

e < 1%,

we must set at least

Tohid Alizadeh (NU)

kp = 27,

0.2

0.4

0.6
Time (sec)

for which

kp = 1025,

for which

ROBT303 Linear Control Theory with Lab

0.8

1.2

is too large

Mp

is too large

Academic year 2016-2017

37 / 37