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 webbased 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: OpenEnded 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 SingleInput SingleOutput
(SISO) systems
Understand interaction between control loops in
MultipleInput MultipleOutput (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 closedloop 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 + an1sn1 + ... + 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(1Kc)/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 righthand
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 lefthand 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 (n1)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 closedloop system is unstable if the Frequency Response of the
openloop 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 openloop phase angle is 1800.
Thus,
The Bode Stability criterion provides information on the
closedloop stability from openloop 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 closedloop 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 openloop transfer function, GOL(s) that is proper and
has no unstable polezero 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 closedloop 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
; = tan1 ( ) = 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 singleloop
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
steadystate 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 #2Unit 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.
3The 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 steadystate 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 closedloop 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 steadystate gains:
ij
y i
all loops
m
j
open
y i
all loops closed
m
j
exceptfor
the mj loop
openloopgain
ij
closedloopgain
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
steadystate 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 steadystate 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 steadystate 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 steadystate, openloop 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 steadystate 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 elementbyelement 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 steadystate 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 K1 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 steadystate 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 nonzero 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 oneway interaction (which is
explained on the next slide).
NAMP
PIECE
If the G(s) indicates that there is a oneway 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 11/22 pairing.
NAMP
PIECE
Now, consider the 2x2 system whose transfer matrix is:
1.5 0.5
0.5 1.5
In this case, a 11/22 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 11/22
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 12/21 pairing. (as the
elements 12,21 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
11/22/33 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 y1u1, y2u2..ynun,
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.Openloop 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 n1 loops are open.
Under closedloop 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 timedelays. However, since Eq.N
depends on Steady State gains (s=0, therefore, the gains
are independent of timedelays). Due to this fact, the results
of this theorem also provide important information about
timedelay 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 steadystate
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 steadystate 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 11,22,33 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 11,23,32,
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 22 has been
interchanged with the element 23 and the
element 33 has been interchanged with the old
element 22.
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 11,22,33 pairing, but you
will notice that this pairing violates Rule 1, as the RGA
element 11 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 steadystate 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 Nonlinear systems.
Example 1 RGA and Loop pairing of nonlinear 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 steadystate 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 looppairing will
depend on the steadystate 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, Fm1, xm2.(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 : Fm1 and xm2 is used, but if the
opposite pairing was used, the interaction would be severe.
NAMP
PIECE
3. When
the steadystate 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 (Fm2) and m1(FA) should be
paired with the composition (xm1). 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 steadystate 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 steadystate. 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 deethanizer 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 11,22 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 NonSquare 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 nonsquare systems.
The most obvious problem with nonsquare 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 nonsquare 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 NonSquare Systems
We have 2 types of nonsquare 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 (nm) 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.
=
(mn)!
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 steadystate
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 steadystate gain:
k1 j
y1 j
m j
NAMP
PIECE
In general, the steadystate 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 openloop steadystate 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 steadystate 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 steadystate 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 steadystate 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 timedelay, inverseresponse, or
other slow dynamics in the best RGA pairing: Since
the RGA is based on steadystate information, sometimes,
the best RGA pairing results can result in very slow
closedloop 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 highfrequency disturbances coming
from the fast loop. This indicates that loops with large
differences in closedloop 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 ProcedureMultiloop Controller
Design
There are 2 stages in the design of multiple singleloop
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 multiloop 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 closedloop 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 singleloop tuning rules (ZieglerNichols,
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
CONTROLLERSIntroduction
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 1D
g22
y2
NAMP
PIECE
Lets assume that the input/output variable pairing has
been determined to be: y1u1, y2u2 ynun pairings.
Under the multiple, independent, singleloop 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(ydiyi)
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 singleloop 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 y1u1, y2u2,ynun couplings are desirable; it
is however the yiuj crosscouplings, 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 crosscoupling will improve the performance of
control systems. It is however NOT possible to ELIMINATE
the crosscouplings; 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 crosscoupling 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
SteadyState Decoupling To only eliminate steadystate
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 steadystate
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 1D (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 crosscoupled with
and affecting y1 through the g12 element, and u1 affects
y2 by crosscoupling 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 1D 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.1D)
y2=(g21+g22gI )v1+(g22+g12gI )v2 (Eq.2D)
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.1D 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 2D 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 1D and 2D 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 steadystate.
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
closedloop steadystate gain in each loop is the ratio of the
openloop gain and the relative gain parameter ().
Note that when is very large, the effective closedloop 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 21)
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 1D 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 1D 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=G1 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 1D 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 timedelay structure in G(s) must be such
that the smallest timedelay 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 GR1GD 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 offdiagonal 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=G1GR 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
STEADYSTATE
DECOUPLING
SteadyState 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 steadystate
decoupling only uses the steadystate gain portion of each
of the transfer elements.
Therefore, if each transfer function element gij(s), has a
steadystate gain term Kij, and if the gain matrix is defined
as K, the steadystate decoupling results in the same way
as it did for a 2x2 system that we discussed earlier.
NAMP
PIECE
STEADYSTATE
DECOUPLING FOR A 2X2
SYSTEM
Simplified steadystate 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
STEADYSTATE
DECOUPLING FOR A 2X2
SYSTEM
Simplified steadystate decoupling for a 2x2 system
In this case, the decoupler matrix is given by:
1
GI =K KR
Where KR is the steadystate 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 steadystate 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 steadystate 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
W1=WT
VTV=I=V VT
V1=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 )225]=0
Ie. 1,2, 3= 10,5,0
NAMP
PIECE
SINGULAR VALUE
DECOMPOSITION EXAMPLE
Note that the nonzero 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 nonzero 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
STEADYSTATE DECOUPLING BY
SINGULAR VALUE
DECOMPOSITION
SteadyState Decoupling by Singular Value
Decomposition
The Singular Value Decomposition (SVD) of the steadystate
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
STEADYSTATE 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
STEADYSTATE 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
STEADYSTATE 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 crosscoupling
among the process variables) to a system that has
no crosscoupling. The openloop 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
(steadystate) 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)
Гораздо больше, чем просто документы.
Откройте для себя все, что может предложить Scribd, включая книги и аудиокниги от крупных издательств.
Отменить можно в любой момент.