Вы находитесь на странице: 1из 184

NAMP for North American Mobility In Higher Education

Program

NAMP

PIECE

Module 5

Controllability
Analysis

Introducing Process integration for Environmental Control in Engineering Curricula

PIECE

NAMP integration for Environmental Control in Engineering Curricula


Process

Paprican

PIECE

PIECE

cole
Polytechnique
de Montral

Universidad
Aut
Autnoma de
San Luis Potos
Potos

University of
Ottawa

Universidad de
Guanajuato

North Carolina
State
University

Instituto
Mexicano del
Petr
Petrleo
Program for North American Mobility in Higher Education

Texas A&M
University

NAMP

NAMP

PIECE

Module 5
This module was
by:
Stacey created
Woodruf

Universidad de
Guanajuato
University of
Ottawa

From

Host
University

Universidad de
Guanajuato

Carlos Carren

University of
Ottawa

NAMP

PIECE

Project Summary
Objectives
Create web-based modules to assist universities to
address the introduction to Process Integration into
Engineering curricula
Make these modules widely available in each of the
participating countries

Participating institutions
Six universities in three countries (Canada, Mexico
and the USA)
Two research institutes in different industry sectors:
petroleum (Mexico) and pulp and paper (Canada)
Each of the six universities has sponsored 7 exchange
students during the period of the grant subsidised in
part by each of the three countries governments

NAMP

PIECE

Structure of Module 5

What is the structure of this module?


All modules are divided into 3 tiers, each with a
specific goal:
Tier I: Background Information
Tier II: Case Study Applications
Tier III: Open-Ended Design Problem
These tiers are intended to be completed in that
particular order. In the first tier, students are quizzed at
various points to measure their degree of
understanding, before proceeding to the next two tiers.

NAMP

PIECE

Purpose of Module 5

What is the purpose of this module?


It is the objective of this module to cover the
basic aspects of Controllability Analysis. It is
targeted to be an integral part of a
fundamental/and or advanced Control course.
This module is intended for students with some
basic understanding of the fundamental
concepts of control.

NAMP

PIECE

Tier I
Background Information

NAMP

PIECE

Statement of Intent
Define Stability
Demonstrate simple methods for stability
analysis, mostly for Single-Input Single-Output
(SISO) systems
Understand interaction between control loops in
Multiple-Input Multiple-Output (MIMO) systems
Demonstrate the Relative Gain Array
Investigate controllability analysis for continuous
and discrete systems
Comprehend singular value decomposition (SVD)

NAMP

PIECE

Stability
A dynamic system is stable if the system
output response is bounded for all bounded
inputs. A stable system will tend to return to
its equilibrium point following a disturbance.
Conversely, an unstable system will have
the tendency to move away from its
equilibrium point following a disturbance.

NAMP

PIECE

Why is the stability of a system important??


When a system becomes unstable it can be

A DISASTER!!!!!

NAMP

PIECE

Example
The concept of stability is illustrated in the
following figure. The sphere in (a) is stable as it
will return to its original equilibrium after a small
disturbance whereas the sphere in (b) is unstable
as it moves away from its equilibrium point and
never comes back. The sphere in (c) is said to be
marginally stable.

(a)

(b)

(c)

NAMP

PIECE

Quiz #1
Why is it important that a system
is stable?
List two examples of systems that
have become unstable.

NAMP

PIECE

There are many ways of determining if a


system is stable such as :

Roots of Characteristic Equation


Bode Diagrams
Nyquist Plots
Simulation

NAMP

PIECE

Roots of Characteristic Equation


One can determine if a system is stable based on the
nature of the roots of its characteristic equations.
Consider the following system:
D(s)

Y*(s) + (s)
-

GC (s)

U(s)

Ym(s)

G1(s)

G4 (s)

M(s)

G3 (s)

G2 (s)

+
+

Y(s)

NAMP

PIECE

From the previous diagram, we can see that the


output Y is influenced in the following manner.

Gc G1G2
G3
Y(s) =
Y*(s) +
D(s)
1 + GOL
1 + GOL
Where
GOL = Gc G1G2G4
GOL is the open loop transfer function.

NAMP

PIECE

For the moment, lets consider that there is only a change in set
point, therefore, the previous equation reduces to the closed loop
transfer function,

Y(s) =

Gc G1G2
1
Gc G1G2
Gc G1G2 (s)
Y*(s) =
=
1 + Gc G1G2G4
s 1 + Gc G1G2G4 s(s - r1 )(s - r2 )(s - r3 )...(s - rn )

The roots r1, r2, r3 rn are those of the characteristic equation


1+GcG1G2G4 =0
and (s) is a function that arises from the rearrangement. The roots
of the characteristic equation (denominator) are the poles of the
transfer function whereas the roots of the numerator are the zeros.

NAMP

PIECE

The nature of the roots of the characteristic equation can dictate if a


system is stable or not due to the fact that if there is one (or more)
root on the right half of the complex plane, the response will contain
a term that grows exponentially, leading to an unstable system.
Imaginar
y Part

Imaginar
y Part

Imaginar
y Part
Real
Part

Real
Part

Negative real
root

time

time

Imaginar
y Part

Stable
Region
Stable
Region

Real
Part

Unstable
Region Rea
l
Part

Positive real root


Imaginar
y Part
Real
Part

time

Complex Roots (Negative real


parts)

time

Complex Roots (Positive real


parts)

NAMP

PIECE

Routh Test
The Routh test (Routh stability criterion) is a very useful tool
in determining whether or not a closed-loop system is stable
provided the characteristic equation is available. The Routh
stability criterion is based on a characteristic equation that is
in the form

ansn + an-1sn-1 + ... + a1s + a0 = 0


A necessary (but not sufficient) condition of stability is that
all of the coefficients (a0, a1, a2, etc.) must be positive.

NAMP

PIECE

Routh Array
When all coefficients are positive, a Routh Array must be
constructed as follows:
Row
The first two rows are
1
an
an -2 an -4
...
filled in using the
2
an -1 an -3 an -5
...
coefficients of the
characteristic
3
b1
b2
b3
...
equation. Subsequent
4
c1
c2
c3
...
rows are calculated as
M
M
M
M
M shown in the next
page.
n+1
The system is stable if ALL the elements in the first
column are positive!

NAMP

PIECE

Routh Array
After the coefficients of the characteristic equation are
input in the array, the coefficients, b1, b2 bn and
subsequently c1cn should be calculated as follows and
input into the array.
Row
an -1an -2 - anan -3
b1 =

an -1

b2 =

an -1an -4 - anan -5
...
an -1

c1 =

b1an -3 - an -1b2
b1

b a - an -1b3
c 2 = 1 n -5
...
b1

1
2
3
4
M
n+1

an
an -1
b1
c1
M

an -2
an -3
b2
c2
M

an -4
an -5
b3
c3
M

...
...
...
...
M

Pivot to calculate all bi

NAMP

PIECE

Routh Test Theorems


Theorem 1- The necessary and sufficient condition for stability (i.e.

All roots with negative real parts) is that all elements of the
first column of the Routh Array must be positive and non
zero.
Routh Test Example 1- Consider the following characteristic equation:

s3 + 4.583s2 + 6.38s + 15.625 = 0

Row
1(s3)

6.38

All of the elements in the first column


of this Routh Array are positive,
2(s2)
4.583
therefore
the system is stable.
15.625
3(s1)

2.97

NAMP

PIECE

Routh Test Example 2- It is possible to determine for which values


of Kc the system remains stable

1+K c
s + 4.583s + 6.38s +
=0
0.384
3

Row
1(s3)

2(s2)

3(s1)

6.38

4.583
(1+Kc)/0.384
29.24 - (1+K c )/0.384
4.583

29.24-(1-Kc)/0.384>0
<10.23
0

1+Kc >0 Kc>-1


positive)

Kc

(Kc is

NAMP

PIECE

Theorem 2- If some of the elements of the first column are


negative, the number of roots on the right hand side of the
imaginary axis is equal to the number of sign changes in the first
column.
Routh Test Example 3 If the characteristic equation of a system is
given by the following equation, is the system stable?
Row

s4 + 6s3 + 11s2 + 36s + 120 = 0

1(s4)

11

120

2(s3)

36

3(s )

4(s1)

-108

5(s0)

120

120
0

There are 2 sign changes.


Therefore, the system has
two roots in the right-hand
plane, and the system is
unstable.

NAMP

PIECE

Theorem 3- If one pair of roots is on the imaginary


axis, equidistant from the origin, and all the other roots
are in the left-hand plane, all the elements of the n th
row will vanish. The location of the pair of imaginary
roots can be found by solving the auxiliary equation:

Cs2+D=
0
where the coefficients C and D are the elements of the
array in the (n-1)th row. These roots are also the roots of
the characteristic equation.

NAMP

PIECE

Routh Test Example 4 Determine the stability of the system


having the following characteristic equation:

s4 + 3s3 + 6s2 + 12s + 8 = 0


Row
1(s4)

2(s3)

12

3(s2)

4(s1)

4(s1)

5(s0)

d
(2 s 2 8) 4 s
ds

The derivative taken


indicates that a 4 should
be placed in the s row (Row
4). The procedure is carried
out.
There are no sign changes
in the first column,
indicating that there are no
roots located on the righthand side of the plane.

NAMP

PIECE

Quiz #2
In what cases can the Routh test be used to
determine stability?
Is the system having the following characteristic
equation stable?

s4 + 7s3 + 6s2 + 1 = 0

If a system has two negative real roots, is the system


stable?
If a system has one negative real root and one
positive real root is the system stable?

NAMP

PIECE

Frequency Response
One very useful method of determining system stability,
even when transportation lags exist, is Frequency
Response.
Frequency response is a method concerning the response
of a process or system to a sustained sinusoidal plot.

Frequency Response Stability Criteria


Two principal criteria:
1. Bode Stability Criterion
2. Nyquist Stability Criterion

NAMP

PIECE

Bode stability criterion


A closed-loop system is unstable if the Frequency Response of the
open-loop Transfer Function, GOL=GCG1G2G4, has an amplitude ratio
greater than one at the critical frequency, c. Otherwise the closedloop system is stable.
Note: c is the value of where the open-loop phase angle is -1800.
Thus,

The Bode Stability criterion provides information on the


closed-loop stability from open-loop frequency response
information.

NAMP

PIECE

Bode Stability Criterion- Example 1


G2 (s) =
A process has the following transfer function:

2
(0.5s + 1)3

With a value of G1=0.1 and G4=10. If proportional control is used,


determine closed-loop stability for 3 values of Kc: 1, 4, and 20.
GOL=GCG1G2G4

2K c
2
GOL = Gc G1G2G4 = (K c )(0.1)
(10) =
3
(0.5s+1)
(0.5s+1)3

Solution:

Kc

AROL for Kc

Stable?

0.25

Yes

Marginally

20

No

You will find the Bode


plots on the next
slide

NAMP

PIECE

Bode plots for GOL = 2Kc/(0.5s + 1)3

NAMP

PIECE

Nyquist Stability Criterion


The Nyquist stability criterion is the most powerful stability test
that is available for linear systems described by transfer
function models.
Consider an open-loop transfer function, GOL(s) that is proper and
has no unstable pole-zero cancellations. Let N be the number of
times that the Nyquist plot of GOL(s) encircles the (-1, 0) point in
a clockwise direction. Also, let P denote the number of poles of
GOL(s) that lie to the right of the imaginary axis. Then, Z=N+P,
where Z is the number of roots (or zeros) of the characteristic
equation that lie to the right of the imaginary axis.

The closed-loop system is stable, if and only if


Z=0.

NAMP

PIECE

Example 9.2 Find the amplitude ratio and the phase lag of the
following process for = 0.1 and 0.4.
U(s)

X(s)

1
5s + 1

First system:
1
AR =
=
2 2
+1
Second system:
AR = 1

1
(5)2 ( )2 + 1

; = - = -0.3

Third system:
G(j ) =
AR =

Z(s)

-0.3s

25( )2 + 1

1.2
s3 + 2.3s2 + 1.7s + 0.4

; = tan-1 (- ) = tan -1 (-5 )

180

1.2
1.2
=
(j )3 + 2.3(j )2 + 1.7(j ) + 0.4
0.4 - 2.3 2 + 1.7 - 3 j

1.2

0.4 - 2.3
2

+ 1.7 -

Y(s)

; = tan

-1

- (1.7 - 3 )

2
0.4 - 2.3

NAMP

PIECE

Example 9.2 Find AR and (from known equations)


G(j) = G1(j) G 2 (j) ... Gn (j)

G(j) =

2
3
+ 1.7 -

1.2

1
2
25( ) + 1

0.4 - 2.3 2

G(j) = G1(j) + G2 (j) + ... + Gn (j)


- (1.7 - 3 )
G(j) = tan (-5 ) - 0.3 + tan
2
0.4
2.3

If (0.4 2.3 3) < 0 then or 180o


-1

-1

= 0.1 AR = 2.60 ; = - 0.915 s -1 or - 52.4o


= 0.4 AR = 0.87 ; = - 2.75 s -1 or - 157.3o

NAMP

PIECE

Example 9.2 Find AR and Nyquist plot


90

Im

Re

180

=0.4

=0.1

270

0
3

NAMP

PIECE

Quiz #3
Name two methods of determining stability
using frequency response.
What does an amplitude ratio (AR) of 1 signify?
An amplitude ratio of less than 1?
What does a value of Z=0 signify?

NAMP

PIECE

Multiple Input Multiple Output (MIMO)


Systems
Cooling unit
Reflux Receiver
Naptha

Light gas oil


Heavy gas oil

High boiling Residue

FEED PUMPS
Air FuelGas
CRUDE OIL FEED
STORAGE TANKS

PIPESTILL
FRACTIONATOR
FURNAC E

NAMP

PIECE

When dealing with Multiple Input Multiple


Output systems, we have to ask ourselves
two main questions.
1. How to pair the input and output
variables
2. How to design the individual single-loop
controllers

NAMP

PIECE

Lets consider the following system:


Loop 1

Gc1

m1

G11

y1

G12
G21

+
Gc2

m
2

G22

Loop 2

y1(s) = G11(s)m1(s) + G12(s)m2(s)


y2(s) = G21(s)m1(s) + G22(s)m2(s)

y2

NAMP

PIECE

We will perform 2 small experiments to demonstrate MIMO


system interactions.
Lets consider m1 as a candidate to pair with y1.
Experiment #1
When a unit step change is made to the input variable m 1, with
all loops open, the output y1 will change, and so will y2, but for
now, we are primarily concerned with the effect on y 1. After
steady-state is reached, lets consider the change in y 1 as a
result of the change in m1, y1m ; this will represent the main
effect of m1 on y1.
y1m = K11
Keep in mind that no other input variables have been changed,
and that all loops are open, so no feedback control is required.

NAMP

PIECE

Experiment #2-Unit step change in m1 with Loop 2 closed.


These things will happen as a result of the unit step change
in m1.
1- y1 changes because of G11, but because of interactions via
the element G21, y2 changes as well.
2- Under feedback control, Loop 2 wards off this interaction
effect on y2 by manipulating m2 until y2 is returned to its
initial state before the disturbance.
3-The changes in m2 will now affect y1 via the G12 transfer
element.
The changes in y1 are from two diferent sources.
(1) the DIRECT INFLUENCE of m1 on y1 (y1m)
(2) the Indirect Influence, from the retaliatory action from
Loop 2 in warding off the interaction effect of m 1 on y2 (y1r)

NAMP

PIECE

After dynamic transients die away and steady-state is


reached, the net change observed in y1 is given by:
y1*= y1m+ y1r
This net change is the sum of the main effect of m1 on y1
and the interactive effect provoked by m1 interacting with
the other loop. y* K 1 K12 K 21 K *
11
11
K11 K 22
A good measure of how well
be controlled
y1ma system
y1can
m
11 y1 is:
() if m1 is used to control

y *

y1m y1r

NAMP

PIECE

Loop Pairing on the Basis of Interaction Analysis


Case 1 : 11=1
This case is only possible if y1r is equal to zero. In physical
terms, this means that the main effect of m1 on y1, when all
the loops are opened, and the total effect, measured when
the other loop is closed, are identical.
This will be the case if:
m1 does not affect y2, and thus, there is no retaliatory control
action from m2, or
m1 does affect y2, but the retaliatory control action from m2
does not cause any change in y1 because m2 does not affect
y1.
Under these circumstances, m1 is the perfect input
variable to control y1 because there will be NO
interaction problems.

NAMP

PIECE

Case 2 : 11=0
This condition indicates that m1 has no efect on y1,
therefore y1m will be zero in response to a change
in m1. Note that under these circumstances, m2 is
the perfect input variable for controlling y2, NOT y1.
Since m1 does not affect y1, y1 can be controlled
with m2 without any interaction with y1.

NAMP

PIECE

Case 3 : 0 < 11< 1


This condition indicates that the direction of the
interaction effect is in the same direction as that
of the main effect. In this case the total effect is
greater than the main effect. For 11>0.5, the
main efect contributes MORE to the total
effect than the interaction effect, and as the
contribution of the main effect increases, the
closer to a value of 1 11 becomes. For 11<0.5,
the contribution from the interaction efect
dominates, as this contribution increases, 11
moves closer to zero. For 11=0.5, the
contributions of the main effect and the interaction
effect are equal.

NAMP

PIECE

Case 4 : 11>1
This is the condition where y1r is the opposite sign of y1m, but it is
smaller in absolute value. In this case y1* (y1r +y1m) is less than
the main effect y1m, and therefore a larger controller action m1 is
needed to achieve a given change in y1 in the closed loop than in
the open loop. For a very large and positive 11 the interaction
effect almost cancels out the main effect and closed-loop control of
y1 using m1 will be very difficult to achieve.

Case 5 : 11< 0
This is the case when y1r is not only opposite in sign, but
also larger in absolute value to y1m. The pairing of m1 with y1
in this case is not very desirable because the direction of the effect
of m1 on y1 in the open loop is opposite to the direction in the
closed loop. The consequences of using such a pairing could be
catastrophic.

NAMP

PIECE

Quiz#4
What is a MIMO system?
What does 11=1 signify? If this is the case, is
m1 a good input variable to control y1?
If 11 is very large and positive, is m1 a good
input variable to control y1?

NAMP

PIECE

Relative Gain Array (RGA)


The quantity 11 is defined as the Relative Gain between
input m1 and output y1.
ij is defined as the relative gain between output yi and
input mj, as the ratio of two steady-state gains:

ij

y i

all loops

m
j

open

y i

all loops closed

m
j

exceptfor
the mj loop

open-loopgain
ij

closed-loopgain

for loop i under


the control of mj

NAMP

PIECE

When the relative gain is calculated for all of the


input/output combinations of a multivariable
system, the results are placed into a matrix as
follows and this array produces

11

21

n1

12
22

n 2

1n
2 n

nn

THE RELATIVE GAIN ARRAY

NAMP

PIECE

PROPERTIES OF THE
RELATIVE GAIN ARRAY

Properties of the Relative Gain Array


1. The elements of the RGA across any row, or
down any column sum up to 1. i.e.:
n

i 1

ij

ij 1
j 1

2. ij is dimensionless; therefore, neither the units,


nor the absolute value actually taken by the
variables mj, or yi affect it.

NAMP

PIECE

PROPERTIES OF THE
RELATIVE GAIN ARRAY

3. The value ij is a measure of the


steady-state interaction expected in the
ith loop of the multivariable system if its
output (yi) is paired with input (mj); in
particular, ij =1 indicates that mj affects
yi without interacting with the other
loops. Conversely, if ij=0 this indicates
that mj has no effect on yi.

NAMP

PIECE

PROPERTIES OF THE
RELATIVE GAIN ARRAY

4. Let Kij* represent the loop i steady-state gain


when all loops (other than loop i) are closed,
whereas, Kij represents the normal open loop
gain.
1

K ij *

ij

K ij

This equation has the very important implication:


that 1/ij tells us by what factor the open loop
gain between output yi and input mj will be
changed when the loop are closed.

NAMP

PIECE

PROPERTIES OF THE
RELATIVE GAIN ARRAY

5. When ij is negative, it indicates a


situation in which loop i, with all loops
open, will produce a change in yi in
response to a change in mj in totally the
opposite direction to that when all the
other loops are closed. Such
input/output pairings are potentially
unstable and should be avoided.

NAMP

PIECE

COMPUTING THE RELATIVE


GAIN ARRAY

Calculating the Relative Gain Array


There are two ways of calculating the Relative
Gain Array
1. The First Principles Method
2. The Matrix Method

NAMP

PIECE

COMPUTING THE RELATIVE


GAIN ARRAY

First Principles Method


Lets consider a 2x2 system as we encountered before.
First, we must observe that the Relative Gain Array deals
with steady-state systems, and therefore , must only be
concerned with the steady state form of this model which
is:
y1=K11m1+K12m2 (Eq.
y2=K21m1+K22m2

1a
(Eq. )
1b
)

In order to calculate the 11 we defined earlier, we need to


y

evaluate the partial derivatives


as was explained on slide


47.
y

Recall:

all loops
open

ij

all loops closed


exceptfor
the mj loop

NAMP

PIECE

COMPUTING THE RELATIVE


GAIN ARRAY

Due to the fact that the equations found on the previous slide
represent steady-state, open-loop conditions, the
differentiation for the numerator portion of the relative gain is:

y1

m1

K11

allloopsopen
The second partial derivative (the denominator) requires Loop 2
to be closed, so that in response to changes in m 1 , the second
control variable m2 can be used to restore y2 to its initial value
of 0. To obtain the second partial derivative, we first find from
Eq. 1b the value of the m2 must be to maintain y2=0 in the
face of changes in m1, what effect this will have on y1 is
deduced by substituting this value of m 2 into Equation 1a.

NAMP

PIECE

COMPUTING THE RELATIVE


GAIN ARRAY

The computation of the denominator of 11


Set y2=0 and solve m2 in Eq. 1b.
K 21

m2

K 22

m1

Substituting this value of m2 into Eq. 1a. gives:

K 12 K 21
y1 K 11 m1
m1
K 22

Having eliminated m2 from the equation, we now may


differentiate with respect to m1.

y1

m1

loop 2 closed

K 12 K 21

K 11 1
K 11 K 22

NAMP

PIECE

COMPUTING THE RELATIVE


GAIN ARRAY

We then substitute the numerator and denominator into


the definition of 11 which yields:

11

K11

K12K21
K11 1
K11K22

This equation simplifies to the form:

1
11
1

where

K12K21
=
K11K22

NAMP

PIECE

COMPUTING THE RELATIVE


GAIN ARRAY

This exercise should be repeated for all


ijs so that the RGA can be constructed.
For Practice, repeat this exercise and
verify the following.

12 21
1

and

22

1
11
1

NAMP

PIECE

COMPUTING THE RELATIVE


GAIN ARRAY

Thus the RGA for this 2x2 system is given by:


1

1
1

1
1
Note, that if we define

1
11
1

The RGA can be rewritten as follows

NAMP

PIECE

COMPUTING THE RELATIVE


GAIN ARRAY

The Matrix Method for Calculating RGA


Let K be the matrix of steady-state gains of the
transfer function matrix G(s) i.e.:

lim G ( s ) K
s 0

Whose elements are Kij, further, let R be the


transpose of the inverse of this steady state
matrix (K)
1 T

R K

NAMP

PIECE

COMPUTING THE RELATIVE


GAIN ARRAY

With elements rij it is possible to show that the


elements or the RGA can be obtained from the
elements of these two matrices as:

ij K ij rij
It is important to note that the equation above
indicates an element-by-element multiplication
of the corresponding elements of the two
matrices, K and R, DO NOT TAKE THE PRODUCT
OF THESE MATRICES!

NAMP

PIECE

COMPUTING THE RELATIVE


GAIN ARRAY

Example- Matrix Method of Calculating RGA.


Find the RGA for the 2x2 system represented by
Equations 1a and 1b and compare them with the results
obtained using the First Principles Method.
Solution:
For this system, the steady-state gain matrix (K) is the
following.

K 11
K
K 21

K 12

K 22

NAMP

PIECE

COMPUTING THE RELATIVE


GAIN ARRAY

From the definition of the inverse matrix we know that

K 22
K
21

K 12
K 11

Where the determinant of K, |K| is: K K 11 K 22 K 12 K 21


Therefore, by taking the transpose of the K-1 matrix, we
obtain the R matrix

R K

1 T

K 22
K
12

K 21

K 11

NAMP

PIECE

COMPUTING THE RELATIVE


GAIN ARRAY

Since we now have the R and K matrices, we can perform


an element by element multiplication to obtain the
elements (ij) of the RGA ()

K 11K 22
11=
K

K 11K 22
OR 11= K K - K K
11 22
12 21

here is the first element of the matrix. Try on your own to


compute the other 3 elements of the RGA.

K 11K 22

-K 21K 12

-K 12K 21
K

K 22K 11

NAMP

PIECE

Example of RGA for the Wood and Berry


Distillation, using the Matrix Method
Find the RGA for Wood and Berry Distillation column
whose transfer function matrix 12
is.8e s 18.9e 3s

G(s)

16.7 s 1
6.6e 7 s
10.9 s 1

21.0s 1
19.4e 3 s
14.4s 1

Solution: For this system, the steady-state gain matrix is


easily extracted from the transfer function matrix by
setting s=0.
12.8 18.9

K G ( 0)

6
.
6

19
.
4

NAMP

PIECE

The next step is to determine the inverse of the matrix K:

0.157 0.153

0
.
053

0
.
104

Once the inverse is calculated, the transpose of this matrix must


be calculated to yield the matrix R.

0.053
0.157
R (K )

0
.
153

0
.
104

1 T

After these two matrices are computed, it is time to calculate the


RGA by multiplying the matrices element by element.

2 1

1 2

Note that all of the


rows and columns
sum to one.

NAMP

PIECE

LOOP PAIRING USING THE


RELATIVE GAIN ARRAY

Loop Pairing using the RGA


Now that we know how to compute the RGA, we will now
consider how it can be used to guide the pairing of
input and output variables in order to obtain the control
configuration with minimal loop interaction.
On the following slides, we will investigate how to
interpret the elements of the RGA (ij). We will use the
five scenarios presented early to interpret the
implications of the values of ij

NAMP

PIECE

LOOP PAIRING USING THE


RELATIVE GAIN ARRAY

Case 1: ij=1, the open loop gain is the


equal to the closed loop gain.
Loop interactions implications : This situation

indicates that loop i will not be subject to retaliatory


effects from other loops when they are closed, therefore
mj can control yi without interference from other control
loops. If any of the other elements in the transfer
function matrix are nonzero, the ith loop will experience
some disturbances from other control loops, but these
are NOT provoked from actions in the ith loop.

Recommendation for pairing: In this case, the pairing


if mj with yi would be ideal.

NAMP

PIECE

LOOP PAIRING USING THE


RELATIVE GAIN ARRAY

Case 2: ij=0, the open loop gain between


mj and yi is zero.
Loop interactions implications : mj has no direct
influence on yi (keep in mind that mj may still have an
effect on other control loops)

Recommendation for pairing: Do NOT pair yi with mj,


it would be more advantageous to pair mj with another
output variable, since we are led to believe that yi will
not be influenced by the loop containing mj.

NAMP

PIECE

LOOP PAIRING USING THE


RELATIVE GAIN ARRAY

Case 3: 0<ij<1, the open loop gain


between yi and mj is smaller than the
closed loop gain.
Loop interactions implications : The closed loop
gain is the sum of the open loop gain and the
retaliatory effect, from the other loops,
a) The loops are interacting, but
b) They interact in such a way that the retaliatory effect
from the other loops is in the same direction as the
main effect of mj on yi.

NAMP

PIECE

Loop interactions implications :


The loop interactions assist m j on controlling yi, The extent
of this assistance is dependent on how close ij is to 0.5
When:
ij =0.5: the main effect of mj on yi is exactly the same as the
retaliatory effect.
0.5<ij <1, the retaliatory effects are less than the main
effect
0<ij< 0.5, the retaliatory effect is larger than the main
effect.
Recommendation for pairing: If possible, avoid pairing yi
with mj if ij<0.5

NAMP

PIECE

LOOP PAIRING USING THE


RELATIVE GAIN ARRAY

Case 4: ij>1, the open loop gain between yi and mj is


larger than the closed loop gain.
Loop interactions implications : The loops interact, and
the retaliatory effect from the other loops acts in opposition
to the main effect of mj on yi, (which means that the loop
gain will be reduced when the other loops are closed), but
the main effect is still dominant, otherwise ij would be
negative. For large values of ij, the controller gain for loop i
will have to be chosen much larger than when all loops are
open. This would cause loop i to be stable when the other
loops are open.
Recommendation for pairing: The higher the value of ij ,
the greater the opposition m j experiences from the other
loops in trying to control y i. Therefore try not to pair yi with
mj with if the value of ij is large.

NAMP

PIECE

LOOP PAIRING USING THE


RELATIVE GAIN ARRAY

Case 5: ij<0, the open loop and closed loop


gains between yj and mi have opposite signs.
Loop interactions implications : The loops interact,

and the retaliatory effect from the other loops is not only in
opposition, but it is greater in absolute value to the main
effect of mj on yi. This is potentially dangerous because if the
other loops are opened, loop i could become very unstable.

Recommendation for pairing: Avoid pairing mj with yi


because of the retaliatory effect that m j provokes from the
other loops acts in opposition to, and dominates the main
effect on yi.

NAMP

PIECE

Quiz#5
What advantages does the Matrix Method have
over the First Principles Method?
What does with a value of 1 signify, and
should mj and yi be paired together?
What does with a value less than zero of
signify, and should mj and yi be paired together?

NAMP

Basic Loop Pairing Rules

PIECE

From what we have learned about loop pairing, it is natural


that the ideal RGA would take the form
1 0 0 0
0 1 0 0

0 0 1 0

0 0 0 1
This is known as the identity matrix, in which each row and
column only contains one non-zero element whose value is
unity (1). This ideal RGA is produced when the transfer
matrix G(s) has one of two forms, only a diagonal element,
or is in lower triangular from. The first situation indicates
that there is no interaction between the loops. The second
case indicates that there is a one-way interaction (which is
explained on the next slide).

NAMP

PIECE

If the G(s) indicates that there is a one-way interaction( the


transfer function matrix is in triangular form), it will yield an RGA
of the identity matrix, but it can not be treated as if there are no
interactions or influences. Please consider the following example.

G ( s)

1
s 1
3
3s 1

0
4

4 s 1

yields an RGA

1 0

0
1

Note that since the element g 12(s) is zero, the input m2 does not
have an effect on the output y 1, however, the input m1 does
influence the output y2 as can be seen due to the fact that the g 21
element is nonzero. Upsets in Loop 1 requiring action by m 1 would
have to also be handled by the controller of Loop 2. So, even
though the RGA is ideal, Loop 2 would be at a disadvantage. Thus,
in deciding on loop pairing, one should distinguish between ideal
RGAs produced from diagonal or triangular transfer function
matrices.

NAMP

PIECE

RULE #1
Pair input and output variables that have positive
RGA elements closest to 1.0.
Consider the following examples to demonstrate
this rule.
For a 2x2 system with output variables y 1 and y2, to be
paired with m1 and m2
0.8 0.2

If the RGA is

0.2 0.8

Then it is recommended to pair m1 with y1 and m2 with y2,


which is quite often referred to a the 1-1/2-2 pairing.

NAMP

PIECE

Now, consider the 2x2 system whose transfer matrix is:

1.5 0.5

0.5 1.5

In this case, a 1-1/2-2 pairing is preferred as to avoid


pairing on a negative RGA element. Usually, we will try to
avoid pairing on RGA elements greater than 1, but pairing
on negative RGA elements is worse.
Recall the Wood and Berry distillation column example we
saw on Slide 65, its RGA
In this case, it
2 is: 1
is desirable for

a 1-1/2-2
1 2
pairing

NAMP

PIECE

On the other hand, for the 2x2 systems whose RGA is

0.3 0.7

0.7 0.3
y1 should be paired with m2 and y2 should be paired with
m1, this is referred to as 1-2/2-1 pairing. (as the
elements 1-2,2-1 are closer to a value of 1 and all
elements in the RGA are positive.)

NAMP

PIECE

Lets consider the following 3x3 matrix:

1.95
0.66
0.29

0.65
1.88
0.23

0.3
0.22
1.52

The same general guidelines, we applied to the 2x2


systems can also be applied here. It can be seen that
although the diagonal elements are all greater than 1,
the other elements are all negative, suggesting that a
1-1/2-2/3-3 pairing would be preferable.

NAMP

PIECE

NIEDERLINSKI INDEX

Niederlinski Index
Pairing Rule #1 is usually sufficient in
most cases, it is often necessary to use
this rule in conjunction with the theorem
found on the next slide developed by
Niederlinski and later modified by
Grosdidier et al. This theorem is
especially useful if the system is 3x3 or
larger.

NAMP

PIECE

NIEDERLINSKI INDEX

Consider the n x n multivariable system whose inputoutput variables have been paired y1-u1, y2-u2..yn-un,
resulting in a transfer function model of the form:
.
y(s)=G(s) u(s)
Let each element of G(s), gij(s) be,
1.Rational, and
2.Open-loop stable

NAMP

PIECE

Let n individual feedback controllers (which have integral


action) be designed for each loop so that each one of
the resulting n feedback loops is stable when all of the
other n-1 loops are open.
Under closed-loop conditions in all n loops, the
multivariable will be unstable for all possible values of
controller parameters if the Niederlinski Index N defined
below is negative.
On the following
slides there are
important points to
help us use this
result properly.

G ( 0)
n

g ii (0)
i 1

(Eq. N)

NAMP

PIECE

NIEDERLINSKI INDEX

Important Points for us to consider:


1.The result is both necessary and sufficient for 2x2
systems; for higher dimensional systems, it only
provides sufficient conditions (if Equation N holds it is
definitely unstable, but if Eq. N does not hold, the
system may or may not be unstable: the stability will be
dictated by the values taken by the controller
parameters).
2.For 2x2 systems the Niederlinski index becomes
K

K 21
N 1 where defined as follows as
K 11K 22
seen on Slide 57

12

NAMP

PIECE

NIEDERLINSKI INDEX

2. For a 2x2 system with a negative relative gain, >1, the


Niederlinski index is always negative; hence 2x2 systems
paired with negative relative gains are ALWAYS
structurally unstable.
3. This theorem is designed for systems with rational transfer
function elements, therefore, this technically excludes
systems containing time-delays. However, since Eq.N
depends on Steady State gains (s=0, therefore, the gains
are independent of time-delays). Due to this fact, the results
of this theorem also provide important information about
time-delay systems as well, but is not very rigorous. USE
CAUTION WHEN APPLYING Eq.N TO SYSTEMS WITH
TIME DELAYS.

NAMP

PIECE

RULE #2
Any loop pairing is
unacceptable if it leads to a
control system configuration
for which the Niederlinski
Index is negative.

NAMP

PIECE

Summary of using RGA for Loop Pairing


1. Given the transfer matrix G(s), obtain the steady-state
gain matrix K=G(0), and from this obtain the RGA, ,
also calculate the determinant of the K and the product
of the elements on the main diagonal
2. Use Rule #1 to obtain tentative loop pairing suggestions
from the RGA by pairing the positive elements which
are closest to one.
3. Use the Niederlinski condition (Eq. N) to verify the
stability status of the of the control configuration
obtained using Step 2, if the selected pairing is
unacceptable, choose another.

NAMP

PIECE

Applying Loop Pairing Rules


Loop Pairing Example 1: Calculate the RGA for the system
whose steady-state gain matrix is given below and investigate
the loop pairing suggested upon applying Rule #1.

5
3

K = G(0) = 1

1
1
3
1

NAMP

PIECE

First, we need to take the inverse of this matrix, then


take the transpose of this matrix to obtain R, being:

10 4.5 4.5
4.5
1
4.5
4.5 4.5
1
The next step is to determine the RGA by multiplying the
elements of the K and R matrices.

R 4.5
4.5

4 .5 4 .5
3
4 .5

4.5
3

NAMP

PIECE

Rule #1 would suggest a 1-1,2-2,3-3 pairing


To calculate the Niederlinski Index we need to find :
The determinant of the K matrix which is :|K|=-0.148
The product of the main diagonal which is :

5
K ii

3
i 1
n

1

3

5
1

27
3

It is clear that when the determinant is divided by the


product of the elements of the main diagonal it will
yield a negative number which leads to a

NEGATIVE NIEDERLINSKI INDEX which violates


Rule #2.

NAMP

PIECE

This example provides a situation where the pairing


suggested by Rule #1 is disqualified by Rule #2. Due
to this fact, we need to investigate another loop
pairing. Lets try the possible pairing of 1-1,2-3,3-2,
which would give a RGA of:

10 4.5 4.5

4.5 4.5
1
4.5
1
4.5

NAMP

PIECE

The new K is:


5
3

K G ( 0) 1

1
1
1
3

1
1

3
1

It is clear that the element in 2-2 has been


interchanged with the element 2-3 and the
element 3-3 has been interchanged with the old
element 2-2.

NAMP

PIECE

We need to calculate the determinant and product of the


elements of the main diagonal of the new matrix K:
|K|=0.1481 while the product of the elements is equal to
5/3.
Therefore, the Niederlinski Index is

0.148

0
5/3

ii

i 1

Clearly, this Niederlinski Index is positive, so we


come to the conclusion that this system is no
longer structurally unstable.

NAMP

PIECE

Loop Pairing Example 2: Consider the system with the


steady state gain matrix as seen below

1 0.1
1
K G (0) 0.1 2
1
2 3
1

The determinant of this matrix is 0.53.


The RGA is :

1.89 3.59 0.7


0.13 3.02 1.89
3.02 5.61 3.59

NAMP

PIECE

From the RGA seen, there is only one feasible pairing,


because all of the other pairings violate Rule 2. The
only feasible pairing is a 1-1,2-2,3-3 pairing, but you
will notice that this pairing violates Rule 1, as the RGA
element 1-1 is negative, but according to the
Niederlinski Theorem this system would NOT be
structurally unstable.
If the first loop is opened (the m1, y1 elements dropped
from the process model) the new steady-state gain
matrix relating the 2 remaining input variables with the
2 remaining output variables
~
2 is: 1

K
3

NAMP

PIECE

It is easy to see that if the first loop is open, the


Niederlinski Index of the remaining two loops would be
negative, indicating that the system would be
structurally unstable. As a consequence, this system
will only be stable if all loops are CLOSED, such a
system is said to have a low degree of integrity.

There are some examples of higher order systems


where it is possible to pair on negative RGA
values and still have a structurally stable
system (this is NOT possible for 2x2 systems).

NAMP

PIECE

Summary of Loop Pairing using RGA


Always pair on positive RGA elements that are
the closest to 1 in value. Thereafter, use the
Niederlinski Index to check if the resulting
configuration is structurally stable. Whenever
possible, try to avoid pairing on negative RGA
elements; for 2x2 systems such pairings
always lead to unstable configurations, while
for systems of higher dimension, they can lead
to a condition which, at best has a low degree
of integrity.

NAMP

PIECE

Quiz #6
What does a positive Niederlinski Index
indicate?
According to Rule 1, should elements be
paired on positive or negative elements?
In what case should a favourable pairing
from Rule 1 be discarded?

NAMP

PIECE

LOOP PAIRING FOR NONLINEAR SYSTEMS

Loop Pairing for Non-linear systems.

Example 1- RGA and Loop pairing of non-linear systems.


The process shown is a blending process, the objective is
to control both the total product flow rate (F) and the
product composition (x) as calculated in terms of the mole
fraction of A in the blend. Obtain the RGA for the system
and suggest which input variable to pair with each output.

F
FA
FB

Blending

G
C

Cx

Analyz
er

NAMP

Total Mass Balance:

PIECE

FA FB F

Mass Balance on Component A

FA
x
FA FB

Solution: Notice that for this system, the two output variables
are F and x, and the input variable are FA and FB, from now
on, we will refer to the input variables as m1 and m2 for the
input feeds of A and B respectively.
Therefore, our Overall Mass Balance becomes

F m1 m2

(Eq 1)

(which is linear)

And the Component A Mass Balance becomes


linear)

m1
x
m1 m 2

(Eq 2)

(which is NON-

NAMP

PIECE

Since this is a 2x2 system, we only need to obtain the


(1,1) element of the RGA given by:
Recall:
F

m1

m1

both loops open

second loop closed

To calculate the numerator, take the derivative of the first


equation with both loops open with respect to m1 ,
yielding

m1

both loops open

NAMP

PIECE

In order to calculate the denominator, loop 2 must be closed,


and we will have to determine the value of m2 so that when a
change occurs in m1, x will return to its steady state value (x*).
To determine the value of m2 in this case, we must set x=x* in
Equation 2 and solve for m2 in terms of m1 and x*, the result is:

m1
m2=
-m1
x*

When loop 2 is closed, the mole fraction of the the component


A in the output at x*, m2 will respond to changes in m1, to
determine the relationship, we have to substitute the value of
m2 above into the Overall Mass Balance (Equation 1) yielding:

m1
F=m1+ -m1
x*

or

m1
F=
x*

NAMP

PIECE

The next step is to differentiate the expression of F obtained in


the last step with respect to m1 yielding:
F
1

m1 second loop x *
closed

If the numerator and denominator are substituted into the


statement for the relative gain (), we get:

x*
1/ x *

For a 2x2 matrix recall that the RGA is given by

Therefore the RGA of this system is:

x * 1 x *

1 x * x *

Where x* is
the desired
mole fraction
of A in the

NAMP

PIECE

Some things to consider about these results:


1. The RGA is dependent on the steady-state value of x*
desired for the composition of the blend; it is NOT
constant as it was in the linear systems we dealt with
before.
2. It is implied that the recommended loop-pairing will
depend on the steady-state operating point.
3. Due to the fact that x* is a mole fraction, it is bounded
between 0 and 1 (0 < x*< 1) and therefore, none of the
elements in the RGA will be negative. The implication of
this fact, is that in the worst possible scenario is that
there will be large interactions between the input
variables if the input and output variables are paired
improperly, but the system will not become unstable .

NAMP

PIECE

A loop pairing strategy for this system is as follows:


1. If x* is close to 1, the first implication is that m 1 is larger than
m2 . If we look at the RGA, the following pairing would be
recommended, F-m1, x-m2.(ie. The larger flow rate is used to
control the total flow rate out and the smaller flow rate is used
to control the composition.)
2. This is the most reasonable pairing because: when the product
composition is close to one (x* close to 1), we have almost pure
A coming out of the system, and so we can modify the flow rate
out quite easily by changing the flow rate of A into the blending
without changing the composition of the blend significantly.
Similarly if we alter the composition, the additional small
amounts of material B will not have a significant impact on the
flow rate of the blend out of the system. Thus, the flow
controller will not interact strongly with the composition
controller if the pairing : F-m1 and x-m2 is used, but if the
opposite pairing was used, the interaction would be severe.

NAMP

PIECE

3. When

the steady-state product composition is closer to


0, the RGA suggests that the loop pairing stated in point
2 should be switched, i.e. m2 (FB) should be paired with
the outgoing flow rate (F-m2) and m1(FA) should be
paired with the composition (x-m1). If you analyze the
effects that each variable has as done in point 2, you will
see that the physics of this system dictates such a
pairing.

4. An interesting situation arises when the composition


(x*) is equal to 0.5 (x*=0.5). In this case it does not
matter which input variable is used to control which
output variable. The observed interactions will be equal
and significant in either case.

NAMP

PIECE

LOOP PAIRING FOR PURE


INTEGRATOR MODES

Loop Pairing for Systems with Pure


Integrator Modes:
Since we have seen that interaction analysis using the
RGA is carried out using steady-state information, an
interesting situation occurs when dealing with systems
that contain pure integrator elements (i.e. if s was set
to zero, an element would become undefined), since
pure integrator elements show no steady-state. Several
suggestions are available to deal with this problem, but
we will use the industrial application of the a deethanizer to demonstrate one method to recommend a
loop pairing strategy.

NAMP

PIECE

Pure Integrator System Example 1 - The transfer function


for a 2x2 subsystem extracted from a larger system for an
industrial de-ethanizer is given below. Obtain the RGA and
use it to recommend loop pairings.

1.318e 2.5 s
20 s 1

G ( s)

0.0385(182 s 1)

( 27 s 1)(10 s 1)(6.5s 1)

e 4 s

3s

0.36

Solution- Our usual course of action to determine the RGA


is to normally calculate the K matrix which is G(s) when
s=0. Unfortunately, we can see that elements (1,2) and
(2,2) contain pure integrator elements represented by 1/s,
which if we set s=0 would yield an undefined number.

NAMP

PIECE

1
I
s

Lets make the substitution,

If I is substituted into G(s), K becomes:

I
1.318

K lim lim
3
s 0
I

0.038 0.36I

The relative gain parameter ()

lim

0.038 x 0.333I
1
1.138 x 0.36I

NAMP

PIECE

We can see that in the term the Is cancel out, so we


obtain

=0.97
Therefore the resulting RGA is

0.97 0.03

0.03 0.97

It is quite obvious that it is desirable to pain in a 11,2-2 fashion.


If you encounter a system in which there the Is do not
cancel out, you will have to consult another reference.

NAMP

PIECE

LOOP PAIRING FOR NONSQUARE SYSTEMS

Loop Pairing for Non-Square Systems

In the previous slides, we have discussed how obtain


RGAs and how to use them for input/output pairings
when the process has an equal number of input and
output variables (square systems).
There are some cases, where multivariable systems do
not have the same number of input and output
variables, these are referred to as non-square systems.
The most obvious problem with non-square systems is
that after the input/output pairing, there will always be
either an input or an output that is not paired (a
residual ).

NAMP

PIECE

With non-square systems, we are faced


with two questions.
1) Which input/output variables should be
paired together?
2) Which variables are redundant and
which take an active role in control?

NAMP

PIECE

Classifying Non-Square Systems


We have 2 types of non-square systems,

1) Underdefined- there are fewer input variables than


output variables.

2) Overdefined- there are more input variables than


output variables.

Thus, a multivariable system with n output and m


input variables, whose transfer function matrix
will therefore be n x m in dimension is:
UNDERDEFINED if m<n and OVERDEFINED if
m>n

NAMP

PIECE

Underdefined Systems

n outputs

m inputs

As seen in the system above, there are less inputs m than


there are outputs n, thus is defined as an
underdefined system.
m=the number of inputs = 2

m<n

NAMP

PIECE

Underdefined Systems
The main issue with underdefined systems is that
not all outputs can be controlled, since we do not
have enough input variables.
The loop pairing is easier if we make the following
consideration
By economic considerations, or other such means,
decide which m of the n output variables are the
most important, these m output variables should
be paired with the m input variables; the less
important (n-m) output variables will not be
under any control.

NAMP

PIECE

Overdefined Systems
B4

m inputs

outpu
ts

As seen in the system above, there are less inputs m than


there are outputs n, thus is defined as an
underdefined system.
m=the number of inputs = 3
n=the number of outputs = 2

m>n

NAMP

PIECE

Overdefined Systems
Deciding the loop pairing of overdefined systems presents
a real challenge. In this case, there is an excess of input
variables, therefore we can achieve arbitrary control of
the fewer output variables in more than one way.
The situation we are faced with is as follows: since there
are m input variables to control n output variable (m>n),
there are many more input variables to choose from in
pairing the inputs and the outputs, and therefore, there
m
will be several different square subsystems from which

n
the pairing is possible. There are
possible square
m
m!
subsystems.
=
(m-n)!

Recall
that:

n!

NAMP

PIECE

The Variable Pairing Strategy for Overdefined


Systems is:
m

1. Determine all of the


n
model.

subsystems from a given

2.Obtain the RGAs for each of the square subsystems.


3.Examine the RGAs and chose the best subsystem on the
basis of the overall character of its RGA (in terms of how
close it is to the ideal RGA).
4. After determining the best subsystem, use its RGA to
decide which input variable within its subsystem to pair
with each output variable.

NAMP

PIECE

LOOP PAIRING IN THE


ABSENCE OF PROCESS
MODELS

Loop Pairing in the Absence of Process Models

Sometimes, situations arise where a process model is not


available, but it is still possible to determine their RGAs
from experimental data. There are 2 approaches as follows:
Approach 1- Experimentally determine the steady-state
gain matrix K, by implementing a step change in the
process input variables, one at a time, and observing the
ultimate change in each output variable.
Let y1j be the observed change in the value of the output
variable 1 in response to a change of mj in the jth input
variable mj ; then , by definition of the steady-state gain:

k1 j

y1 j

m j

NAMP

PIECE

In general, the steady-state gain between


the ith variable and the jth variable will be
given by

kij

yij

m j

Thus, the elements of the K matrix can be


calculated, and once the K matrix is known,
it is easy to calculate the RGA.

NAMP

PIECE

Approach 2- It is possible to determine each element of


the RGA directly from experimentation.
As you may recall, each RGA element (ij) can be obtained by
performing two experiments. The first experiment determines
the open-loop steady-state gain by measuring the response of y i
to input mj , when all other loops are open. In the second
experiment, all other loops are closed using PI controllers to
ensure that there will be no steady-state offsets and the
response of yi to input mj is redetermined. By definition, the ratio
of these two gains is the desired relative gain element ( ij ).
The second approach is more time consuming, and involves too
many upsets to the process; for these reasons it is not desirable
in practice. Therefore, the first approach is preferred.

NAMP

PIECE

Final Comments on the RGA


1.The RGA requires only steady-state process information, it
is therefore easy to calculate and easy to use.
2. The main criticism of the RGA is that the RGA only
provides information about the steady-state interactions
within a process systems, and therefore, dynamic factors
are not taken into account by the RGA analysis.
3. The RGA only suggests input/output pairing such that the
interaction effects are minimized; it provides no
guidance about other factors which may influence the
pairing.

NAMP

PIECE

Other Factors Influencing the Choice of Loop Pairing


1.Constraints on the input variable: It is possible that
the best pairing obtained from the RGA will result in a
choice of input variable for y i that is severely limited by
some constraint (ex. maximum feed concentration) in a
way that it can not carry out the assigned control task.
2.The presence of a time-delay, inverse-response, or
other slow dynamics in the best RGA pairing: Since
the RGA is based on steady-state information, sometimes,
the best RGA pairing results can result in very slow
closed-loop response if there are long time delays,
significant inverse response or large time constants. If
this is the case, it would be more suitable to pair on more
unfavourable RGA elements if the slow elements could be
omitted to improve system performance.

NAMP

PIECE

Other Factors Influencing the Choice of Loop Pairing


3. Timescale Decoupling of Loop Dynamics : Often
timescale issues arise that can influence the choice of loop
pairing. For example, in a 2x2 system, it may be that for a
given pairing, the RGA indicates a serious loop interaction.
However, if at the same time, one of the loops responds a
great deal faster than the other, there can be a timescale
decoupling of the loops. This can occur if the fast loop
responds so fast that the effect on the slow loop seems to be
a constant disturbance, in opposition, the slow loop does not
respond at all to the high-frequency disturbances coming
from the fast loop. This indicates that loops with large
differences in closed-loop response times can be paired even
when the RGA indicates that the pairing is unfavourable.

NAMP

PIECE

Quiz#7
What system information is needed to construct
the RGA?
What is the difference between a underdefined
and overdefined system?
What is a difficulty in overdefined systems?

NAMP

PIECE

Controller Design Procedure-Multiloop Controller


Design
There are 2 stages in the design of multiple single-loop
controllers for multivariable systems:
Judicious choice of loop pairing
Controller tuning for each individual loop
We have discussed this first point a great deal in the past
slides, this should signify importance of the choice of loop
pairing in controller design.
Now, we must address the issue of tuning the individual
controllers.

NAMP

It should be obvious that when the RGA for a


process is close to ideal (ie. ij is very close to 1)
that the multiloop controllers are very likely to
function very well if they are designed properly.

However, when the RGA indicates strong


interactions for the chosen loop pairing (ie. ij is
very large or negative) the controller is not likely
to perform well even if it is tuned well.

PIECE

NAMP

PIECE

Controller Tuning for Multiloop Systems


The main challenge in controller tuning is the interactions between
the different control loops of a multi-loop system. Due to this fact,
it can be risky to adopt the obvious strategy of tuning each
controller individually without considering the other controllers and
hoping that when all the loops are closed that the overall system
performance will be adequate.
The procedure that is normally followed in practice is the following:
1.With the other loops on manual control, tune each control loop
independently until satisfactory closed-loop performance is
obtained.
2.Restore all the controllers to joint operation under automatic
control and readjust the tuning parameters until the overall closedloop performance is satisfactory in all the loops.

NAMP

PIECE

When the interactions between the control loops are not


too significant, the procedure mentioned before can be
quite useful. However, for systems with significant
interactions, the readjustment of the tuning in Step 2 can
be difficult and tedious. One can cut down on the amount
of guesswork that goes into such a procedure by noting
that in almost all cases, the controllers will need to be
made more conservative (ie. the controller gains will have
to be reduced and the integral times increased) when all
the loops are closed in comparison to when all of the
individual controllers are operating individually, with all of
the other loops open. The process of this changing of the
control parameters is referred to as detuning.

NAMP

PIECE

One method of detuning for a 2x2 system is as follows:


1.Use any of the single-loop tuning rules (Ziegler-Nichols,
Cohen and Coon, etc) to obtain starting values for the
individual controllers; let the controller gains be K ci*.
2. These gains should be reduced using the following
expressions that depend on the relative gain parameter :

( 2 ) K * 1.0
ci

K ci

K ci * 1.0
2

It may still be necessary to retune these controllers after


they have been put in operation; however, this will not
require as much effort as if one were starting from scratch.

NAMP

PIECE

DESIGN OF MULTIVARIABLE
CONTROLLERS-Introduction

Design of Multivariable Controllers


In the next section, we will discuss the design of true
multivariable controllers that utilize all of the available
process output information jointly to determine what the
complete input vector u should be. Thus each control
command from the multivariable controller will be based on
all of the output variables, not just based on one. In
principle, it will be possible to eliminate all of the
interactions between the process variables. The objective of
the next section is to present some of the principles and
techniques used for designing multivariable controllers, as
designing multivariable controllers is one of the more
challenging problems faced in industrial process control. We
will start by addressing loop decoupling, the most widely
used multivariable controller technique. We will then address
Singular Value Decomposition (SVD) which is a means of
determining when it is structurally unstable to apply

NAMP

yd

PIECE

gc1

v1

u1

+
+

g11

y1

+
+

gI1
g12
Please consider the following
system:
g21

gI2
yd

2
2

gc1

v2

u2

Figure 1-D

g22

y2

NAMP

PIECE

Lets assume that the input/output variable pairing has


been determined to be: y1-u1, y2-u2 yn-un pairings.
Under the multiple, independent, single-loop control
strategy, each controller gci operates according to:

The controller
transfer
function multiplied by
the difference in the
set point of yi(ydi) and
the actual yi output

ui=gci(ydi-yi)
OR
ui=gcii

The
difference
between the
desired yi and
the actual yi
output.

The output error

NAMP

PIECE

However, a true multivariable controller must decide on


ui, not using only i, but using the entire set of 1, 2 n.
Thus, the controller actions are obtained by:
u1=f1 (1, 2 , n)
u2=f2 (1, 2 , n)
u3=f3 (1, 2 , n)

un=fn(1, 2 , n)
The design problem is to find the f1(.),f2(.)fn(.) so that
each of the output variable errors is driven to zero.

NAMP

PIECE

DECOUPLING
INTRODUCTION

Decoupling:

In Decoupling, as seen in the Figure on Slide 132,


additional transfer function blocks are introduced between
the single-loop controllers and the process, functioning as
links between the otherwise independent controllers. The
actual control action experienced by the process will now
contain information from all of the controllers. For
example, a 2x2 system, whose individual controller
outputs are gc11 and gc22 if the decoupling blocks for each
loop have transfer functions of gI1 and gI2 respectively,
then the control equations will be given by:

u1=gc11+gI1 (gc22)
u2=gc22+gI2 (gc11)

NAMP

PIECE

Decoupling Introduction
We know from our discussion of input/output pairing that
the pairing of y1-u1, y2-u2,yn-un couplings are desirable; it
is however the yi-uj cross-couplings, by which yi is
influenced by uj (for all i and all j with ij), that are
undesirable: they are responsible for the control loop
interactions.
It is clear that any technique that eliminates the
undesired cross-coupling will improve the performance of
control systems. It is however NOT possible to ELIMINATE
the cross-couplings; that is a physical impossibility since it
will require altering the physical nature of the system.
Consider an example of this on the following slide.

NAMP

Cold flow
rate
Hot flow rate

PIECE

It is not possible to stop


the hot stream from
affecting the
temperature of the
stirred tank, even though
the main objective of this
stream is to maintain the
tank level. It is also true
that we can not prevent
the cold stream from
affecting the tank level
even though controlling
the temperature is its
main responsibility.

NAMP

yd

PIECE

+
-

Gc
Single
Loop
Controller

GI

Interaction
Compensati
on

The main objective in decoupling is to compensate for the


effect of interactions as a result of cross-coupling of the process
variables. As shown in the figure above, this can be achieved
by introducing an additional transfer function block( the
Interaction Compensator) between the Single Loop Controllers
and the process. This Interaction Compensator, together with
the Single Loop Controllers now form the multivariable
decoupling controller. In the ideal case, the decoupler causes
the control loops to act as if they are totally independent of
each other, reducing the tuning task so that it will be possible

NAMP

PIECE

The design problem is to find the element G I (the


compensator) to satisfy one of the following
objectives.
Dynamic Decoupling- To eliminate interactions from all
control loops, at every instant in time
Steady-State Decoupling- To only eliminate steady-state
interactions from all loops; in this case dynamic interactions
are tolerated. Although this type of decoupling is less
rigorous than this dynamic decoupling, it leads to much
simpler decoupler designs.
Partial Decoupling- To eliminate dynamic or steady-state
interactions in a subset of the control loops. This focuses only
on the critical loops with the strongest interactions, leaving
those with weak interactions to act without decoupling.

NAMP

PIECE

SIMPLIFIED DECOUPLING

Design of Ideal Decouplers - Simplified Decoupling


First we will consider some important aspects of the block
diagram in Figure 1-D (found on slide 132)
1. There are two compensator blocks gI gI , one for each
loop.
1

2. There is a new notation: the controller outputs are now


v1 and v2, while the actual control action implemented
on the process remains as u1 and u2. This distinction is
necessary because the output of the controllers and
the control action to be implemented on the process
no longer have to be the same.

NAMP

PIECE

SIMPLIFIED DECOUPLING

3. Without the compensator, u1=v1 and u2=v2 and the


process model remains
y1=g11u1+g12u2
y2=g12u1+g22u2
The interactions persist, as u2 is still cross-coupled with
and affecting y1 through the g12 element, and u1 affects
y2 by cross-coupling through g21.
4. With the interaction compensator, Loop 2 is informed
of changes in v1 through gI , so that u2 what the process
actually feels is adjusted accordingly. The same process
is preformed by Loop 1 by gI which adjusts u1 from
information about v2.
2

NAMP

PIECE

The design question is now posed:


What should gI1 and gI2 be if the effects of loop
interactions are to be completely neutralized?
To answer this:
Lets consider Loop 1 in Figure 1-D where the process
model is :
y1=g11u1+g12u2
y2=g12u1+g22u2
Because of the compensators, the equations governing
the control action are:
u1=v1+gI v2
u2=v2+gI v1
1

NAMP

PIECE

If we substitute the expressions for u1 and u2 into the


expressions for y1 and y2 seen on the previous slide the
system is defined as:
y1= g11(v1+gI v2) + g12(v2+gI v1 )
y2=g12(v1+gI v2) +g22(v2+gI v1 )
1

Which Yields

y1=(g11+g12gI )v1+(g11gI +g12)v2 (Eq.1-D)


y2=(g21+g22gI )v1+(g22+g12gI )v2 (Eq.2-D)
2

NAMP

PIECE

In order to only have v1 affect y1 and to eliminate the


effect of v2 on y1, we must choose a value of gI1 so that
the coefficient of v2 in Eq.1-D will disappear i.e.:
g11gI1+g12=0
Then solving for gI1
g12
gI1 =g11
A similar procedure can be done for Loop 2, which
eliminates any influences of v1 on y2, with the
g21 a value of:
manipulation of Eq 2-D we
obtain
gI 2 =g22

NAMP

PIECE

The transfer functions seen on the previous slide are the


decouplers needed to exactly compensate for the effect
of loop interactions in the 2x2 system shown in Figure 1D.
If we now substitute our expressions for gI1 and gI2 into
Equations 1-D and 2-D respectively we will yield:

g g
y1= g11- 12 21 v1
g22

g12g21
y2= g22 v2
g11

NAMP

PIECE

Now the system is completely decoupled with only v 1


affecting y1, and v2 affecting y2.
We can see in the figure below the equivalent block
diagram where the loops appear to act independently and
therefore, can be individually tuned.
yd1 +

yd2 +
-

gc1

gc
2

v1

g12g21
g11g22

v2

g g
g22 - 12 21
g11

y1

y2

NAMP

PIECE

Lets consider that the closed loop system is under steady


state. If the steady state gain for an element gij =Kij,
observe how the system is expressed at steady-state.

K K
y1= K 11 - 12 21 v1
K 22

K 12K 21
y2= K 22 v2
K 11

Recall the definition of for a 2x2 system:

Then the system simplifies to:


K
y1= 11 v1

K
y2= 22 v2

1
K 12K 21
1
K 11K 22

NAMP

PIECE

When we examine the simplified decoupling , the effective


closed-loop steady-state gain in each loop is the ratio of the
open-loop gain and the relative gain parameter ().
Note that when is very large, the effective closed-loop gains
become very small, and control system performance may be
jeopardized.
It is important to note that when dealing with systems with
dimensions larger than 2x2, the simplified decoupling method
can become very tedious. For an N x N system there are (N 2-1)
compensators. The same principles as used for a 2 x 2 system
are applicable, but the work becomes very cumbersome.
On the next slide we will see an example of a 3 x 3 system,
which has 6 compensator blocks, it is clear that using
simplified decoupling in this situation would be very tedious.

NAMP

yd1

PIECE

gc1

v1

u1

gI12

yd3

gc3

gI23
gI31

g31

v2

gc2

u3 u2 u1

g21
g22
g23

gI21

u2

v3

gI32

+
+

g12
g13

gI13
yd2

g11

u3

g32
g33

y1

+
+

+
+

y2

+
+

y3

NAMP

PIECE

GENERALIZED
DECOUPLING

Generalized Decoupling
Please refer to Figure 1-D which we will use this figure to
outline a more generalized procedure for decoupler
design.
yd

gc1

v1

gI1

u1

gc1

v2

g12

gI2
yd

g11

g21
+

u2

g22

y2

NAMP

PIECE

GENERALIZED
DECOUPLING

1. We can observe from Figure 1-D that:


y=Gu

u=GIv

So that:

y=GGIv

NAMP

PIECE

GENERALIZED
DECOUPLING

2. In order to eliminate all interactions, y must be related


to v through a diagonal matrix, let us call it G R(s), now we
must chose GI such that

GGI=GR(s)
And the compensated input/output relation becomes:

y=GR(s)v
Where GR represents the equivalent diagonal process that
the diagonal controllers GC are required to control.

NAMP

PIECE

3. Therefore, the compensator (GI) must be given by:

GI=G-1 GR
4. The compensator obtained depends on what GR is
selected. The elements of GR should be chosen to
provide the desired decoupled behaviour with the
simplest possible decoupler. A common choice for G R is:

GR=Diag[G(s)]
Ie. The diagonal elements of G(s) are retained as the
elements of the diagonal matrix GR, however, other
choices have been used.

NAMP

PIECE

The Relationship between Generalized and


Simplified Decoupling
Generalized decoupling may be related to simplified
decoupling, by noting that for simplified decoupling applied to
a 2x2 system, the compensator transfer function matrix is
given by:

GI =

gI 2

gI1

While for a 3x3 system, the


1compensatory
gI12 gI13 matrix GI takes the
form:

GI = gI 21

gI 31

gI 32

gI 23

NAMP

PIECE

Quiz #8
What is the main objective of decoupling?
What is a downfall of simple decoupling?
Is it often easy to achieve perfect decoupling?

NAMP

PIECE

LIMITATIONS OF
DECOUPLING

Some Limitations of the Application of Decoupling


There are some limitations to the application of
decoupling, and we must keep these in mind in order to
maintain a proper perspective when designing
decouplers.
Perfect decoupling is only possible if the process model is
perfect, which is hardly ever the case, so perfect
decoupling in practice is impossible.
Perfect dynamic decouplers are based on model inverses.
As such, they can only be implemented if such inverses
are both causal and stable.

NAMP

PIECE

LIMITATIONS OF
DECOUPLING

To illustrate the idea of stable and casual, please consider the


2x2 compensators we saw in Figure 1-D whose transfer
functions are GI1 and GI2 must be casual (no e+s terms) and
stable.
To satisfy causality for the 2x2 system, any time delays in g 11
must be smaller than the time delays in g 12 and a similar
condition must hold for g22 and g21.
To satisfy stability, a second condition that g11 and g22 must
not have any right hand plane zeros and also g 12 and g21 must
not have any right hand plane poles. This leads to the
following general conditions that must be satisfied in order to
implement simplified dynamic decoupling for N x N systems.

NAMP

PIECE

LIMITATIONS OF
DECOUPLING

1.Causality: In order to ensure causality in the compensator


transfer functions the time-delay structure in G(s) must be such
that the smallest time-delay in each row occurs on the diagonal.
For simplified decoupling, this is an absolute requirement, but it
is possible to add delays to the inputs u 1,u2un, to satisfy the
requirement if the original process G does not comply. This is
equivalent to defining a modified process as G m:

Gm=GD
Where D is a
diagonal matrix
of time delays

ed11s

ed22s

D(s)=

O
0

dnns

NAMP

PIECE

LIMITATIONS OF
DECOUPLING

The simplified decoupler is then designed by using the


elements of Gm rather than G, and the matrix D must be
inserted into the control loop as shown below:
modified process Gm

+
yd

Gc

Single
Loop
Controller
s

GI

Decoupler

D
Delay
s

G
Process

In the case of generalized decoupling, one may use the


modified process Gm as above, or alternatively, the time
delays in the diagonal matrix GR can be adjusted, in order
that the elements of GI=(GD)-1GR are casual. This is
equivalent to requiring that GR-1GD have the smallest

NAMP

PIECE

LIMITATIONS OF
DECOUPLING

2. Stability- In order to ensure the stability of the

compensator transfer functions, the causality


condition must be satisfied and there are no
Right Hand Plane zeros of the process G(s). This is
an absolute requirement for simplified decoupling
and reduces to the condition that there are no
Right Hand Plane zeros in the diagonal elements
of G and that the off-diagonal elements of G are
stable. For generalized decoupling, this may be
performed by adjusting the dynamics of G R in
order that the elements of GI=G-1GR be stable.

NAMP

PIECE

PARTIAL
DECOUPLING

Partial Decoupling
If some loop interactions are weak or if some of the loops
do not need to achieve high performance, the partial
decoupling is a method one should consider. If this is the
case, only a subset of the control loops where the
interactions are important and high performance is
important are focused on.
Typically partial decoupling is considered for 3x3 or higher
dimension systems. The main advantage is the reduction of
dimensionality. Partial decoupling is also applicable to 2x2
systems, in this case, one of the compensator blocks is set
to zero for the loop that is to be excluded from decoupling.

NAMP

PIECE

STEADY-STATE
DECOUPLING

Steady-State Decoupling
The difference between dynamic decoupling and steadystate decoupling is that dynamic decoupling uses the
complete, dynamic version of each transfer function
element to obtain the decoupler, and steady-state
decoupling only uses the steady-state gain portion of each
of the transfer elements.
Therefore, if each transfer function element gij(s), has a
steady-state gain term Kij, and if the gain matrix is defined
as K, the steady-state decoupling results in the same way
as it did for a 2x2 system that we discussed earlier.

NAMP

PIECE

STEADY-STATE
DECOUPLING FOR A 2X2
SYSTEM

Simplified steady-state decoupling for a 2x2 system

Here:

K12
gI1 =K11

and

K21
gI 2 = K22

These expressions to describe the transfer function of


the compensator block are simple, constant, numerical
values so they will always be realizable and can be
implemented.

NAMP

PIECE

STEADY-STATE
DECOUPLING FOR A 2X2
SYSTEM

Simplified steady-state decoupling for a 2x2 system


In this case, the decoupler matrix is given by:
-1

GI =K KR
Where KR is the steady-state version of GR(s). The
inversion indicated is a matrix of numbers, and
therefore, the inversion will always be realizable and
easily implemented.
The main advantages of steady-state decoupling are that
the design involves simple numerical computations
and that the resulting decouplers are always realizable.

NAMP

PIECE

Quiz #9
What 2 conditions must a system satisfy to
achieve perfect dynamic decoupling?
What is the main advantage of partialdecoupling?
Why is steady-state decoupling a favorable
method if applicable?

NAMP

PIECE

SINGULAR VALUE
DECOMPOSITION

Singular Value Decompostion

Any real n x m matrix K, it is possible to find orthogonal


(unitary) matrices W and V such that

WTAV=
Here is the m x n matrix described below:
s 0
0 0

where

1 0 0
0
0
2

s
L L L

0 0 0

L
L
L
L

0
0
L

Where, for p=min(m,n), the diagonal elements of S:


1> 2> > r> 0,(r > p), together with r+1=0, p=0 are
called the singular values of A; these are the positive square
roots of the eigenvalues of ATA; r is the rank of A .

NAMP

PIECE

SINGULAR VALUE
DECOMPOSITION

W is the m x m matrix

W= w1 Mw2 M
L Mwm
Whose columns wi, i=1,2,,m are called the left singular
vectors of A; these are normalized (orthonormal)
eigenvectors of AAT.
V is the n x n matrix:

V= v1 Mv2 M
L Mvn

Whose n columns vi, i=1,2,,n are called the right singular


vectors of A; these are normalized (orthonormal)
eigenvectors of ATA.

NAMP

PIECE

SINGULAR VALUE
DECOMPOSITION

Because they are composed of orthonormal vectors, the


matrices W and V are orthogonal (or unitary) matrices i.e.

WTW=I=WWT
So that

Also
So that

W-1=WT
VTV=I=V VT
V-1=VT

NAMP

PIECE

SINGULAR VALUE
DECOMPOSITION

By applying these properties of unitary matrices, we can


obtain the relationship:

A=W VT
Analogously to the eigenvalue/eigenvector expression for
square matrices, we have the more general pair of
expressions

Avi= iwi
ATiwi= ivi

NAMP

PIECE

SINGULAR VALUE
DECOMPOSITION

The ratio of the largest to the smallest singular


value is designated the condition member of A:
ie.

1
( A)
p
This gives the most reliable indication of how close
A is to being singular. Note that for a singular
matrix, (A)=, thus nearness to singularity is
indicated by excessively large (but finite) values
for (A)

NAMP

PIECE

SINGULAR VALUE
DECOMPOSITION EXAMPLE

Example - Singular Value Decomposition of a 3x2


matrix

1 2

A 2 1
2 1

Therefore,

9 2
A A=

2 6
T

NAMP

PIECE

SINGULAR VALUE
DECOMPOSITION EXAMPLE

The eigenvalues are obtained as 10 and 5 , thus


the singular values of A are :
1, 2=10 and 5
Ordered so that 1>2 as required for SVD
analysis, the next step is to determine the 3x2
matrix .

10

0
0

NAMP

PIECE

SINGULAR VALUE
DECOMPOSITION EXAMPLE

Right Singular Values


The first eigenvector or ATA corresponding to 1 is
obtained from adj(ATA- 1 I)
-4 2
T
adj(A A- 1 I) =

A possible choice for the eigenvector is the second


column. Normalizing this with 22+12= 5, the norm of the
vector, we obtain the first right singular vector v1
corresponding to 1= 10 2

5
v1 =
1

5

NAMP

PIECE

SINGULAR VALUE
DECOMPOSITION EXAMPLE

In the same way, the second normalized eigenvalue


corresponding to 2=5 is: 1

5

v2 =
2

5
Therefore:

v=
1

5
2

NAMP

PIECE

SINGULAR VALUE
DECOMPOSITION EXAMPLE

From V we can determine VT to be:

5
5
1
2

5
5

VT =

You can verify that V is a unitary matrix by


evaluating VTV and confirming that the product
is I.

NAMP

PIECE

SINGULAR VALUE
DECOMPOSITION EXAMPLE

Left Singular Values


For the given matrix:

5 0 0

T
AA 0 5 5
0 5 5

The Eigenvalues of this 3x3 matrix are obtained from the


characteristic equation which in this case is:
(5-) [(5- )2--25]=0
Ie. 1,2, 3= 10,5,0

NAMP

PIECE

SINGULAR VALUE
DECOMPOSITION EXAMPLE

Note that the non-zero eigenvalues of AAT are identical to


the eigenvalues of ATA.
5 0 0
For 1=10
(AAT - 1 I) = 0 5 5
0 5 5
To find the adjoint of this matrix, we first find the
cofactors and take the transpose of the matrix of
cofactors. In this case,
0
0 0
adj(AAT - 1 I) = 0 25 25
0 25 25

NAMP

PIECE

SINGULAR VALUE
DECOMPOSITION EXAMPLE

And by normalizing any of the non-zero columns, we


obtain the first left singular value of A, and by a similar
procedure the second and third eigenvectors can be
determined using values of 2, =5 and 3=0

w1 =

2
1

w2 = 0

w3 =

2
1

NAMP

PIECE

SINGULAR VALUE
DECOMPOSITION EXAMPLE

When the 3 eigenvalues are combined:

W =

2
1

1
0
0

2
1

Now we have all of the elements desired to decompose


the matrix. You can verify that A=W VT by multiplying
the elements we have determined.

NAMP

PIECE

STEADY-STATE DECOUPLING BY
SINGULAR VALUE
DECOMPOSITION

Steady-State Decoupling by Singular Value


Decomposition

The Singular Value Decomposition (SVD) of the steady-state


gain matrix of a process is another approach to steadystate decoupling.
The SVD of a process gain matrix K can be written as:
K=W VT
then applying the SVD of K, the steady state model becomes:

y= W VT u

NAMP

PIECE

STEADY-STATE DECOUPLING BY
SINGULAR VALUE
DECOMPOSITION

We will multiply both sides by WT (recall the orthogonality


properties of W), our expression becomes:

WTy= VT u
Recall that when the matrix K is a square matrix is a
diagonal matrix of singular values. This allows us to
define a new output variables and new input
variables where:
= WTy
And
= VT u

NAMP

PIECE

STEADY-STATE DECOUPLING BY
SINGULAR VALUE
DECOMPOSITION

Now, the process model becomes


=

Because is diagonal, this indicates that the system is


completely decoupled at steady state.

yd

WT

d +

Gc

WT

NAMP

PIECE

STEADY-STATE DECOUPLING BY
SINGULAR VALUE
DECOMPOSITION

The implication of this is the following: instead of


controlling y with u, the transformed variables will
convert the original system (with cross-coupling
among the process variables) to a system that has
no cross-coupling. The open-loop gain of each loop
of the transformed system is indicated clearly by
the singular values and conditioning is
automatically accessed from the condition number.
A controller can now be designed for the equivalent
(steady-state) system which controls by using .
If this controller is designated Gc then the scheme
would be implemented as seen in the previous
slide.

NAMP

PIECE

REFERENCES

References:
Ogunnaike,B.,Ray,W. Process Dynamics,
Modeling, and Control. Oxford University Press,
New York (1994)
Seborg, D., et al. Process Dynamics and Control.
John Wiley & Sons, Inc, United States of America
(2004)
Thibault, Jules. Courses Notes, CHG 3335: Process
Control. University of Ottawa, Ottawa (July 2004)