Chemical Engineering Science, Printed in Great Britain.
_{V}_{O}_{I}_{.} _{4}_{5}_{,} NO.
THE
_{5}_{5} _{P}_{P}_{.}
13w1323,
1990
RELATIVE
GAIN
MULTIVARIABLE
FOR
NONSQUARE
SYSTEMS
OOG9 2509/90 53.00 + 0.00
0 1990 Pergamon Press plc
JINWEN 
CHANG 
and CHENGCHING 
YU+ 

Department 
of Chemical 
Engineering, National 
Taiwan Institute of Technology, 
Taipei, Taiwan 
10772, 

R.O.C. 

(First 
received 
9 
May 
1989; accepted in revised 
form 
25 
August 
1989) 
AbstractGenerally speaking, chemical processes in nature are nonsquare systems with unequal numbers of inputs and outputs. However, only limited tools, e.g. singularvalue decomposition (SVD), are capable of
analyzing nonsquare muItivariable systems directly. In this paper, Bristol’s relative gain array (RCA) is extended to nonsquare multivariable systems. The nonsquare relative gain array (NRC) is defined as the ratio of the openloop gain, when all loops are without any control, to the closedloop gain, when all loops other than the loop explored are under perfect control in the leastsquare sense. Properties of NRG are
rigorously 
derived. 
Similar 
to 
the square 
RCA, 
the NRC can be used 
to 
assess the performance 
of 
nonsquare 
control 
systems based on steadystate 
information. Futhermore, 
the NRG can be served as 
a criterion to choose a square subsystem from a nonsquare
Two distillation examples are presented to illustrate the use of the NRC.
system if a square control system is preferred.
Nonsquare
1. INTRODUCTION
with
systems,
systems
unequal
numbers
of inputs and outputs, arise naturally in chemical processes (Downs, 1984; Downs and Moore, 1981; Lourtie, 1985; Treiber, 1984). However, most multi variable control system analysis and/or design tools, e.g. the relative gain array (RGA), Niederlinski index etc. (McAvoy, 1983), deal with square systems only, except for the method of singularvalue decomposi tion (SVD). Mathematically, SVD has very good geo metric interpretation (Klema and Laub, 1980, Downs,
(1981) used
SVD for measurement selection and variable pairings in distillation processes. Despite being meaningful geometrically, the SVD method depends on input and
output scalings (Mijares et al., 1986; Nguyen et al.,
1984, Lau et al., 1985). Downs
and Moore
1988).
The RGA, known as an interaction measure for multivariable control systems (McAvoy, 1983), was
originally proposed by Bristol (1966). Morari 
and 
coworkers (Grosdidier et al., 1985; Skogestad 
and 
Morari, 1987) gave updated summaries of the RGA.
In the design of decentralized controllers (multiloop
SISO controllers), the RGA
workable variable pairings (Yu and Luyben, 1986). Manousiouthakis et al. (1986) generalized the concept of the RGA to block relative gain (BRG) which is capable of handling partially decentralized control systems. Both the RGA and BRG deal with square systems only. The purpose of this work is to extend the RGA to nonsquare systems. Systems having more outputs than inputs are considered in this paper. Obviously, it is not possible to control a system having more out puts than inputs perfectly, i.e. steadystate offsets in all outputs are zero (Goodwin and Sin, 1984). However,
is used to eliminate
un
in the leastsquare sense, the control objective can be stated as: to minimize the sum of square errors
(steadystate errors or steadystate offsets) of the out
puts with
_{l}_{a}_{w} _{“}_{p}_{e}_{r}_{f}_{e}_{c}_{t} _{c}_{o}_{n}_{t}_{r}_{o}_{l} _{i}_{n} the leastsquare sense”
throughout this paper. Therefore, the concept of per fect control, e.g. for deriving closedloop gain, can be constructed for nonsquare systems. Following the definition of the nonsquare RGA (NRG), algebraic properties of the NRG are derived. Generally, two approaches can be taken to design a control system for a nonsquare process. One approach is to design
fewer
inputs. We
call this kind
of control
a nonsquare controller to control all outputs
(Treiber, 1984; Lourtie, 1985). A quantitative relation
ship between the NRG and the ultimate performance
of a leastsquare perfect controller is derived. Another
approach is to square the system down, i.e. to select outputs such that the number of inputs and outputs are equal (Downs and Moore, 1981). A square subsys tem, e.g. measurements, selection criterion is also pro posed based on the NRG. Quantitative relationship can also be derived or conjectured from the NRG. Two distillation examples (Doukas and Luyben, 1978; Downs and Moore, 1981) are used to illustrate appli cations of the NRG.
2. DEFINITIONS
Consider
a process
transfer
y:
%,%
function
matrix
Ccs,
^{(}^{1}^{)}
function
with inputs u and outputs
Y(S)=
m x
_{w}_{h}_{e}_{r}_{e}
_{G}_{,}_{,}_{,}
matrix
y(., is an
an
UC,) is
is
an
M 2
with
m
n x
n
x
1 output
1 input
n
process
vector
vector.
transfer
The 
system 
(Fig. 
1) 
is 
under 
feedback control 
with 

+Author to whom 
correspondence 
should be addressed. 
a controller, 
K. 
The 
relative 
gain between the 
ith 
1309
1310
JINWEN CHANG
and CHENGCHING
Yu
2
2
Fig. 1. Concept of closedloop gain in a feedback system.
output (yi) and jth input (u,.), A:, is defined as the openloop gain over the closedloop gain. The open loop gain between yi and uj is described by the ijth
element of the process transfer function matrix. gii(.,. Generally, the openloop gain, gij(,,, is a function of
fore, we have
L 1
aYi
dui
Ykk+i
=
Bji
(5)
where gii is the jith element of Gi.
It should
be emphasized
that the only
assumption
made about the controller K is that the controller has integral action (can eliminate steadystate offset for asymptotically constant disturbances). No assump tion is made about the controller structure. Therefore,
Aij becomes 

Aj 
= 
Sijdji 

and the RGA, 
A, is defined 
as 
A = G@(G‘)=
(6)
_{(}_{7}_{)}
where @ denotes elementbyelement multiplication.
the frequency (let s = iw) and the definitions 
are the 

same for both square (m = n) and nonsquare 
(m > 
n) 
2B. Nonsquare system 
systems. However, the definition and interpretation 
of 
For a nonsquare system with more outputs than 
the closedloop gain are not that straightforward. The closedloop gain is defined separately for both square and nonsquare systems.
2A. Square systems
For
square systems, the closedloop
gain is defined
as the gain between yi and ui when all other outputs
are under perfect control, i.e. y,s are held constant for
all
1985). Figure 1 illustrates the concept of the closed loop gain provided that all other outputs are kept at set points. Generally, it is not possible to keep all other outputs, yks, under perfect control at all time if G,,, contains deadtime and/or RHPT zeros. There
fore, only the steadystate aspect of perfect control is meaningful for general systems. In this paper, only the
steadystate aspect (s = 0) of the closedloop
discussed. Mathematically, the relative gain can be defined as
k #
i (McAvoy,
1983; Grosdidier
and
Morari,
gain
is
Alternatively,
closedloop
ijj can be expressed
as
gain r
^{(}^{2}^{)}
As
steadystate
pointed
out
earlier,
the
openloop
is
transfer function:
gain
gain of the process
[
aY.
_.A
auj
1
uk,k+j
=
&j(O).
the
Since only the steadystate aspect is considered, the subscript (0) is dropped for clarity. The closedloop gain can be expressed as the element of Gi regard less of controller structure, e.g. multiloop SISO con trollers, blockdiagonal controllers or full multi variable controller, as shown in Appendix A. There
inputs (m > n), it is not possible to keep all outputs at their set points. Therefore, the sense of perfect control in the definition of closedloop gain should be modi fied. Perfect control in the leastsquare sense is pro
posed. That is, a controller
steadystate offsets are minimized in the sense of least square (i.e. the 2norm of the steadystate error vector is minimized). This type of controller is termed a least square perfect controller. From the IMC structure [Fig. 2 and Garcia and Morari (1982)] the output vector, y(,,. can be expressed as
such that the
K is designed
_{Y}_{(}_{s}_{i} = %,[I,,
and assuming no plantmodel mismatch, we have
+
Ge,,, tG,,,

&,)I
 ’ Gc,,,YE’
_{(}_{8}_{)}
where
Ge,,,
tion matrix
_{%}_{,}
an
is an n
is
x 
m IMC controller transfer func 

m 
x 
n nominal process 
transfer 
function 
matrix. 

The closedloop 
error 
becomes 
e(,) =
(1,

G,,, GM
~7:;
_{w}_{h}_{e}_{r}_{e}
_{e}_{,}_{,}_{,} is an
m x
1 error
vector
Since we are interested
in the steadystate
(10)
error, the
leastsquare perfect controller requires
=(a,
= (GT,, Go,)
’ GA,
=
G,.
_{(}_{1}_{1}_{)}
The
the MoorePenrose pseudoinverse, respectively
superscripts
T
and + denote
the transpose
and
n
Fig. 2. Nonsquare IMC
structure.
The relative gain for nonsquare multivariable
systems
1311
(Strang, 1980). Using the finalvalue theorem, the steadystate offsets become [subscript (0) is dropped for clarity]
_{(}_{1}_{2}_{)}
i5 =
(I

GG+),‘=‘.
_{w}_{h}_{e}_{r}_{e} _{t}_{h}_{e} _{s}_{u}_{p}_{e}_{r}_{s}_{c}_{r}_{i}_{p}_{t}
its nominal value at steady state. Geometrically, the
matrix
complement of the column space of G. Subsequently,
the sum of the square of the steadystate offsets, ErB, is minimized. Therefore, the steadystate closedloop re lationship, under the leastsquare perfect control, be tween ii and y”’ becomes
“”
denotes
g”’
the deviation
from
(I

GG’)
projects
onto the orthogonal
,=G+y”“_
Between
of the closedloop
the ith output
and the jth
gain is
[
auj
aYi
1
CL
=
9ji
t
where gf is the jith entry of G+.
that in order
Notice
to achieve
_{(}_{1}_{3}_{)} 

input, the inverse 

(14) 

the control 
objec 
tive, controller structure, a full multivariable least square perfect controller, and steadystate gain of the controller are assumed implicitly. For a nonsquare system, G, the nonsquare rela
tive gain (NRG) is defined as
N
+
rayi
LauJoL=gij.g;
= rayi
(15)
Similarly,
and
where (g’
G+G.
the column 
sum vector 
(CS) is defined 

r 
n 
n 
n 
1T 

= 
[es(l), cs(2), 
, es(n)]’ 
as
_{(}_{2}_{0}_{)}
es(i)
=
(9+
9)ii
g)ii is the iith element
of
the
n x
(21)
n matrix
Property 1: The sum of the elements in each column of
the NRG for all js.
is always
equal to unity. That
is es(j)
=
1.0
Proof
nonsingular,
(GTG)i GrG
Since the jth column sum of the NRG, es(j), is the jjth element of G+ G, we have
For
an m
x
n matrix,
G, with
=
m >
I,.
n, if GrG
is
_{(}_{2}_{2}_{)}
we have
G+
G
=
es(j)
=
F
i=r
A;
=
1.0.
Property
1 says
that
the sum
of
NRG
QED
elements
along the longer side of the G matrix (the column sum
for m > n) is equal to unity. This property
with the result of square systems (RGA).
is consistent
Property 
2: The sum of the elements 
in each row 
of 

the 
NRG 
falls between 
zero 
and 
unity. 
That 
is: 

0 < 
rs(i) 
< 
1 
for 
all 
is. 
Proof: Appendix C.
This property points out the fundamental difference between the NRG and the RGA. Since it is not poss
ible to keep all outputs perfect for a nonsquare sys
less than unity
seems to indicate the deviation from perfect control for each output. The next two sections will discuss the row sum in greater details.
tem, the row sum of the NRG being
Thus, from the steadystate transfer matrix 
G and 
its 

_{p}_{s}_{e}_{u}_{d}_{o} _{i}_{n}_{v}_{e}_{r}_{s}_{e} _{G}_{+}_{,} the nonsquare 
relative 
gain 
ar 
_{r}_{a}_{y}_{,} _{A}_{N}_{,} can be promptly evaluated: 

AN = G@(G+)T. 
_{(}_{1}_{6}_{)} 
The definition is similar to the Bristol RGA’s. Despite the difference that the pseudoinverse, G+, is used instead of the normal inverse, Gl, the simplicity of the RGA remains.
3.
PROPERTIES
OF
THE
NONSQUARE
RCA
Corollary
1: Summation
of
all
row
sums,
rs(i)s,
is
(17)
equal
to summation
of all column
sums, cs(j)s,
and
From AN, we can compute
each row and in each column. The
the sums of all elements in
row
sum vector
(RS) is defined 
as 

RS= 
j$liyj, 
5 
, 
t 
A:jIT 
the 
summation 
is equal 
to 
the 
column 
dimension. 

That 
is 

i$l rs(0 = j$l 
es(j) 
= 
n 
(23) 

Proof! 
_{[} 
j1 
j= 
1 

= 
[m(l), 
is(2), 
.rs(m)]’ 
_{(}_{1}_{8}_{)}
where
rs(i)
is the sum
of the
ith row
of the NRG.
Furthermore, it can be shown 
that (Appendix 
C) 
This corollary indicates that the summation of row 

sums [rs(i)] is limited to 
the dimension 
of 
G 
(the 

us(i) = (99+ )ii 
(19) 
number of inputs). With n inputs, at best, we can keep 

where (gg + )ii is the iith element 
of the m x m matrix 
n 
output variables under 
perfect 
control. 
For 
an 

GG+. 
m 
x n nonsquare system, the n degrees of perfection 
1312
JINWEN CHANG
and CHENGCHINO
Yu
are distributed among m output variables. This corol lary clearly states the inherent limitation for systems with fewer inputs.
Property 3: The NRG is invariant under input scaling and is variant under output scaling. By input (output) scaling we mean post (pre)multiplication of G by a nonsingular diagonal matrix S,(S, ).
Proof:
(1) 
Input 
scaling: 
For 
an 
n x n diagonal 
scaling 

matrix 
S,, 
S, 
= 
diag 
(sl, 
sl,. 
, s,), it can be 
shown 

that 

(GS,)+ 
= S;‘G+C. 

From 
the definition, 
A? can be expressed 
as 

A: 
= (GS,)@(S;‘G+)r. 

The NRG 
for the system under input scaling 
is 

(A& 
= 
gijsj 
IIJ g; ; 
= gijg; 
= 
a& 

Therefore _{9} we have Af = 
AN. 
(2) Output
matrix
that
S,,
S,
scaling: For
=
the m x m diagonal
_{,} _{s}_{,}_{)}_{,}
scaling
diag (sls sa,
_{i}_{t} _{c}_{a}_{n} _{b}_{e} _{s}_{h}_{o}_{w}_{n}
(S,G)+
= (GTS,2G)1GTS:.
Proof:
Appendix
D.
Despite
being
variant
under
output
scaling,
the
physical meaning of output scaling is quite clear. With
respect
following
to
input/output
properties:
scalmgs,
the NRG
has the
(1) the NRG
(2)
is invariant
under input scaling
the NRG is variant under output scaling but the scaling factors weight the relative importance of the outputs.
Property 4: Any permutation
of rows and columns
in
a transfer function 
matrix 
G 
results 
in 
the 
same 
permutation in the NRG. Mathematically, 
let 
AN = NRG(G) and P, 
and P, 
are two square permu 

tation matrices. (A”),, = P,ANP,. 
If 
(AN), 
= NRG(P,GP,), 
then 
Proof: P, is an m x m permutation matrix and P, 
is an 
n x n permutation matrix. 
Since 
P, 
and 
P, 
thogonal matrices, e.g. P; 
’ = PT 
and Pi 
1 = 

have 

(P,GP,)+ 
= P;G+ 
P,‘. 

The permutated NRG becomes 
are or
P,‘,
we
(24)
(AN), = CP,GP,18 CV’oGP,)+I=
= CPoGP,l
@ CP,(G+)‘P,l
The scaling matrix (S,) cannot be factored previous case except for the following
out as the case. For
s1=s2= 
=s,=si,wehave 

A: 
= (S,G)O[(S,G)+]7T 

= s,G@ 
[(c)‘l+ 
= 
AN. 

1 

Generally 
A,N # AN.
QED
Similar 
to the RGA, 
the NRG 
is invariant 
under 
input scaling. However, the NRG is variant 
under 
= PoCW46.3(G+)=lP,
= P,ANP,.
QED
Similar to the RGA, any permutation of rows and columns in a transfer function matrix G results in the same permutation in the NRG.
Property 5: The definition 
of the NRG is consistent 

with the definition of the RGA 
when the number 
of 

input variables is equal 
to 
the number 
of output 

variables. Mathematically, then AN = A. 
if 
G is a square matrix, 
output scaling. It should be emphasized that property 3 holds only for systems with more outputs than inputs (for systems with more inputs than outputs, the opposite is true). Since the concept of leastsquare perfect control is used in the definition of the closed loop gain, the output scaling is equivalent to the weighted least square [Lawson and Hanson (1974, p. 183)]. The magnitude of an output scaling factor, i.e. si, weights the relative importance of the ith output. The next corollary gives the interpretation of the scaling factor, si, under asymptotic conditions.
:Corollary 
2: 
If 
the ith scaling 
factor 
of 
the output 

scaling matrix 
is zero, 
i.e. 
sI = 
0, then 
the ith 
con 
trolled variable is disregarded
square perfect control
sum, rs(i), is zero. If the ith scaling factor, si, goes to
in the context
the
of least
row
and
corresponding
ProoE
For
Property
a square and nonsingular
G+
= (G=G)‘G=
= G‘(G=)lGr
= G‘.
matrix
5 shows that the RGA
is a special
G:
QED
case of
the NRG. Properties l5 state fundamental proper
ties of the NRG.
Property 
6: 
For 
an 
m 
x 
1 system, 
G 
= 
[gii, 
gsi, 

_{,} g,ilT, 
the NRG 
is 

AN 
= 
[I,,, 
A,,, 
, &,7= 
with
infinity, i.e. si + 
co, then the ith controlled 
variable is 
always held constant and the corresponding approaches unity. 
row sum 
The relative gain for nonsquare multivariable systems
1313
For a onecolumn
matrix, the NRG
is a vector
with
all positive elements. Each NRG element is the ratio
of the square of the openloop gain to the sum of square of all elements.
Property 7: The elements of the NRG approach infin
ity as the nonsquare singular.
system matrix G becomes nearly
Proof: From eqs (C3) and (C4) of Appendix be expressed as
C, Ji”,can
gij 
5 
det G’j[I’J 
det G [Z’ti,] 

A$ 
= 
1’ 
= 
1 

det (GTG) 

As G becomes 
nearly 
singular 
[det (GrG) 
4 
01, the 
elements of the NRG approach infinity.
QED
Proof:
Appendix
B.
It seems that the analytical expression, eq. (25), does
not directly
.gr,, since 3,;s appear in both the numerator and the
and
imply
relative
sensitivity
between
1:
denominator
term in the RHS
of the RHS
of eq. (25). However,
the first
to the
of eq. (25) is implicitly
related
1;
as we can see from eq. (BS) of Appendix
B. Let
us
case as an
example, the numerical results for the first (the
take the illconditioned
G in the previous
element in brackets)
of eq. (25) for all ijs with 6 = 0.001 are
and the second
term of the RHS
[
With
[ZOOO] + 
( 
 
1000) 
[ 
 
20001+ 
(1000) 

[2000] 
+ 
( 
 
1000) 
[ 
 
20001 + (1001) 
[ 
 
lOOO] + (2001) 
6 = 
0.0001, we have 
[999]
+
(

2000) 1
Consider
is singular
a nearly singular nonsquare
if and only 

a 

G= 
a 

[ 
a 
a(1 

becomes 

r 1+b 26 

A”= 
I 
 l+S 

2s 

1 

 

L 
6 

500.5 
=
[
500.5
1000
a
a
+
if det(GTG)
6)
1
2s
1
~
26
~ 1+6
6


5000.5
5000.5

10,000
=
O]:
system G [G
[20,000] 
+ 
(  
10,000) 
[ 
 
2O,ooo] + (10,001) 
[2O,M30] + 
(  
10,000) 
[  2O,OOO] + (10,001) 
[
[

lO,OOO] + (20,001)
L99991+ ( 
20,001)
1
The results show that the premultiplication factor in
the RHS of eq. (25) grows in the magnitude
proportional to the size of dc. Therefore, we can conjecture that the relative sensitivity between ,%gand gij is proportional to the size of 1;. For a system with large value of nzs, it becomes more sensitive to errors in gvs.
which is
The NRG
I.
500
1001 1
500
4. NONSQUARE
CONTROL
SYSTEM
PERFORMANCE
As pointed out earlier, two approaches can be taken to handle nonsquare systems. The first ap proach is to design a nonsquare controller for a given nonsquare system (Treiber, 1984, Lourtie, 1985). Im
plications of the NRG to the ultimate performance of
nonsquare controllers are discussed in this section.
Since it is not possible to keep every output at its set
point with fewer inputs (notice that the process trans
n) in the face
of disturbances, the ultimate steadystate performance measure is, therefore, the steadystate offset (error) in each output. In most cases, a leastsquare perfect controller [eq. (1 l)] is designed such that the 2norm of the steadystate offset is minimized. The NRG measures, quantitatively, the ultimate steadystate performance of a nonsquare control system.
fer function
Let
S =
0.001:
A”
and
for
6 =
0.0001:
AN=
is an m
x
n matrix
with
m >
_{}_{5}_{0}_{0}_{0}
10,001 1
5cOo.
For
This special case implies a nearly singular system
matrix results in large elements in the NRG.
special
n: are. However, the degree of singularity for non
square systems is less critical than square systems,
to have a singular (or nearly
since it is not very likely
singular) system, especially for cases with m + n.
this
case, the smaller
6 is, the larger the values of
Theorem 1: Consider a stable nonsquare system with
more outputs than inputs. If a leastsquare perfect
controller
system is stable, then the deviation
of
steadystate
set point
and the closedloop of the ith row sum
[eq.
(1 l)]
from
is employed
unity
[l

the NRG
rs(i)]
is equal
to the
offset in the ith output
is made
when a unit step
2),
if
_{(}_{2}_{6}_{)}
Property 
8: Consider 
an m x n transfer matrix G with 

its pseudoinverse 
G+ 
and 
its associate 
NRG, 
AN. 

Relative 
changes in the gijs and 25s are related 
by the 

following 
expressions. 

_{d}_{a}_{:} nc 
 
g; det [ (G”)T(Gij)] lgdet(GTG) 
+ 
(1  
2a;) 
1 _{2}_{.} 
_{(}_{2}_{5}_{)} 
change
in the ith output.
(I

structure
GG+
),*“.
(Fig.
the
leastsquare perfect controller (Section 2B) is used, the closedloop relationship becomes
Proof: Based on the IMC
B =
1314
JINWEN CHANG
and CHENGCHING
Yv
a
output,
If
unit
step set point i.e.
change
is made
on the
ith
_{Y} Set = (0, . 
. 
. 
, 0, 
1, 0, 
, o)= 

the steadystate error 
of the 
ith output 
is 

ei = 
1  
(gg+)ii 

where (gg+)ii is the iith entry 
of GG+. 

From 
eq. (19), we have 

ci = 
1 
 
w(i). 

QED 

The closeness of each row 
sum, 
n(i), 
to 
unity has 
in nonsquare control system
performance. In the case when one output is more
important than the other, we can simply weight that
particular
a strong implication
output heavier than the other, e.g. rescale
the process transfer function matrix with output scaling matrix provided with a larger scaling factor on that output. The row sums of output resealed NRG change accordingly and the row sum of that particu
the
physical meaning of output scaling on the NRG be
lar output becomes closer to unity. Therefore,
comes more transparent.
4A. Example: DL column (Doukas and Luyben, 1978) Consider a sidestream distillation example (Fig. 3), separating benzene, toluene and xylene, originally studied by Doukas and Luyben (1978). Table 1 gives the process transfer function matrix. The concentra tions of four impurities in three product streams are controlled by three manipulated variables: reboiler duty, reflux ratio and side stream flow rate. Doukas and Luyben (1978) squared up the system by adding
40
.
Raflux

87


Sidestrram
Feed
Fig.
3. DL
distillation
column.
Steam
Bottoms
Table 1. Process transfer function matrix for the DL 
column 

_{R}_{e}_{b}_{o}_{i}_{l}_{e}_{r} _{d}_{u}_{t}_{y} _{(}_{u}_{,}_{)} 
Reflun ratio (21,) 
Side draw (I(~) 
Toluene 
in bottom 
(yl) 
Toluene 
in bottom 
(y2) 
Benzene in side draw (ys)
Benzene in side draw (yl)

g g*le
11.36s
+
1.59s
1
5.984e2.24’
14.298
+
1
2.38e0,42”
(1.43s
+
1)2

11.67e‘~9’S
12.19s
+
1
0.374e 7.7’S
(22.2s
+
1)2

1.986e0~7’”
l)*
(66.67& +
0.0204e0~5g"
(7.145
+
l)*

0.176e0.485
(6.9,s +
l)*

11 3e3.79’
(21.74s
+
1)”
5.24eC60’
(400s
+
I)2

0.33e0.68”
I)*
(2.388
+
4.48e0.525
(11.11s
+
1)2
The
relative gain for nonsquare multivariable systems
1315
a fourth manipulated variable, the sidedraw location. Treiber (1984) treated the DL column as a 4 x 3
system. A nonsquare controller
the nonsquare system. The NRG can be used to assess control system performance a priori.
should
be design for
The 
NRG 
for 
the 4 
x 
3 system 
is 

0.241 
 
0.103 
0.861 

0.006 
1.094 
 
0.100 

0.028 
0.000 
0.002 

0.726 
0.008 
0.237 

with 

cs 
= [l.OOo 
1.000 
l.Ooo]~ 
RS
=
CO.998
1.000
0.030
0.971]=.
cc
5
1.2
0.8
0.6
0.4
0.2
0.0
0.2
I
O
260
460
Time
SbO
_{a}
From 
the row sum of the NRG, 
it is clear that the 
Fig. 
5. Closedloop 
responses 
for 

third output will be controlled poorly if a nonsquare 
change in yz with a nonsquare 

controller is used. A nonsquare 
IMC controller 
is 

designed 
to verify this prediction 
by the NRG. Since 

our objective is to illustrate the steadystate offsets in 

the controlled variables, a simple nonsquare IMC controller is designed: 

The controller is equivalent to the nonsquare version of SPMC [simplified model predictive _{c}_{o}_{n}_{t}_{r}_{o}_{l} (Arulalan and Deshponde, 1987)]. Simulation results 

(Figs 
47) show that little steadystate offsets in the 

first, second and fourth output when unit step set point changes are made in these variables. However, a significant offset is observed in y3 for a unit step set point change in yy’ (Fig. 6). Table 2 summarized the steadystate offsets in all outputs which are exactly the results predicted by the NRG. Treiber (1984) also 
200 
400 

pointed out that the third output, y3, is poorly con trolled from the results of closedloop simulation. 
Time 

Clearly, the ultimate nonsquare control system per 
Fig. 
6. Closedloop 
responses 
for 

formance can be assessed from the NRC directly. 
change in y, 
with a nonsquare 
02C 
_{o}_{c} 

Time 
4 

Fig. 
4. Closedloop 
responses 
for 
a 
unit 
step 
set 
point 
change in y, 
with a nonsquare 
IMC 
controller. 
co
s
Fig.
1 .o
0.8
0.6
0.4
0.2
0.0
0.2
0.4
f
2 0
1
200
1
400
Time
7. Closedloop change in y,
responses
for
a 
unit 
step 
set 
point 

IMC 
controller. 

600 
1 
_{0} 

a 
unit 
step 
set 
point 

IMC 
controller. 

1 

500 
e 
0 

a 
unit 
step 
set 
point 

IMC controller. 
with a nonsquare
1316
JINWEN
CHANG
and CHENGCHING
Yu
Table 2. Steadystate
step set point change in each variable with the static IMC leastsquare controller
offsets of the DL
column
for
a
unit
SCl
Yl
Se‘
Y2
Se,
Y3
SC‘
Y4
_{Y}_{l}
_{Y}_{2}
_{Y}_{3}
_{Y}_{4}
0.0016
0.0001
 0.0388
0.0067
 o.ooo1 
 0.0388 
0.0067 
0.0000 
_{}_{0}_{.}_{0}_{0}_{2}_{5} 
 0.0004 
0.0025 
0.9699 
 0.1664 
 0.0004 
 0.1664 
0.0286 
5. CONTROL
STRUCTURE
SELECTION
An alternative approach to designing a control system for a nonsquare process is to square the system down. That is to select the same number of outputs as inputs and a square controller can be designed for the square subsystem (Fig. 8). This is
a common practice in chemical process control
(Joseph and Brosilow, 1978; Downs and Moore, 1981;
Downs, 1984; Morari and Stephanopoulos, 1980) and
is generally termed selection of secondary measure
ments. However, only the method of SVD proposed
by Downs and Moore
a nonsquare system in a direct fashion. Despite hav
ing good geometric interpretation, the SVD method is
both input and the output scaling dependent and the secondary measurement selection method proposed
by Downs
sions, e.g. the largest element in the different column vector of U may happen to be the same output. The NRG provides an alternative for the selection of
a square subsystem
(1981) is capable of handling
and Moore (1981) may fail in some occa
(secondary
measurements).
Consider a nonsquare process, G, with a square
controller as shown in Fig. 8. The nonsquare
G can be partitioned into a square subsystem G, and
the complementary (remaining) subsystem G,. The control objective is to minimize the sum of square
error (SSE) of uncontrolled outputs when the square subsystem is under perfect control. In the leastsquare sense, this is equivalent to choosing n out of m (with
m > n) data sets and solve the n linear algebraic equa
tions directly for the coefficient such that SSE for the complete m data sets is minimized. Notice that there are C(m, n) combinations to choose from. In terms of tubular reactor control, this means to choose some temperature measurement location(s) such that when the chosen temperatures are under integral control the temperature profile in the reactor can be held most constant for any variation in the controlled temperature. In terms of distillation control, this is to select temperature control tray(s) such that the tem perature profile within the distillation column can be kept most constant. The row sum of the NRG pro vides some information in this regard.
system
Consider an m x n process transfer function matrix
G with m > n. If we choose n outputs for control, 
the 

system can be partitioned 
into 

[,;I 
= 
[ ;I]u 
_{(}_{2}_{7}_{)} 
where
(controlled)
vector
Clearly,
is
yn,
Notice
closedloop
scribed
ys
for
is
an
n x
1 output
and
vector
(m 
for
the selected
1 output
outputs
the
yR is an
n)
(uncontrolled)
x
outputs.
there are C(m, n) choices for ys. The objective
outputs,
remaining
to minimize
for
any
the SSE for the uncontrolled
in the controlled
variation
outputs,
ys.
The
are de
that only steadystate
by
gains
for
error is considered.
subsystem
the square
6 =
G,
1r:“_
_{(}_{2}_{8}_{)}
When the square subsystem is under perfect control, the steadystate error for all outputs is
e
=
(I,
x n 
GG;
‘)y:=’
_{(}_{2}_{9}_{)}
where 
I, 
X,, is 
an m x IZ matrix 
with 
unity 
in 
the 

diagonal 
zero 
elsewhere, 
and 
e 
is 
an 
111x 1 steady 
state 
error 
vector. For 
a particular choice 
of 
the 

square subsystem, G,, 
the SSE is defined 
as 

SSE 
= 
k 
IIs II: = 5 
II(ImX,,  GG, ‘)y:‘: 
)I: 
(30) 

i=1 
i=l 

where ytft 1s. an n x 1 vector with unity in 
the ith entry 
and zero elsewhere,
corresponding to the specific input 7:::. The row sum
of the NRG provides optimal (exact) solution to the
problem for two special cases, namely, case of n = 1 and case of m = n + 1, and suboptimal (approxi mated) solution for other cases.
and e(i)
is an
m
x
1 error
vector
5.4. Case ofn = 1
This
case
is quite
common
in chemical
process
control: to choose a measurement location for single loop control, e.g. to choose a temperature control tray in a distillation control. In this case G is a onecolumn matrix. From property 5, the row sum is
ts(i) =
m
2
9il
^{=}
22.
(31)
&
g5
From
eqs (30)
choice
and (31)
of G,,
it can
be shown
that, for
m)
a particular
(Appendix
i.e. the ith subset of S(l,
B), the SSE becomes
m
SSE(i)
=
c
sj:
j=1
L Sfl
_{(}_{3}_{2}_{)}
and
m
c si:
=
+
SSE(i)
_1=11.
n(i)
_{(}_{3}_{3}_{)}
system.
_{9}_{i}_{l}
Fig.8. Square control
structure for a nonsquare
The relative gain for nonsquare multivariable systems 1317
Notice that SSE(i) is for the case when the ith output is chosen as the controlled variable. Equation (33) shows that choosing the largest row sum of the NRG can lead to the smallest SSE subject to any perturba tion in the controlled variable. From property 6, we know that a large of giI corresponds to a large value of AZ and rs(i). Therefore, choosing the element with the largest magnitude from a transfer function matrix
is
the largest row sum from the
NRG in this case. This is physically understandable for the singleloop control, since the heuristic in com
mon practice is to choose a system that the manipu lated variable has a large effect on the controlled variable.
1, a small
row
sum in the RHS
of eq. (39) indicates
a small SSE in the corresponding square subsystem. Now we can justify that the small row sum of the NRG in the complementary (remaining) system indi cates a small SSE in the square subsystem. Therefore, the criterion for the selection of a square subsystem is to eliminate the controlled variable with the smallest row sum in the NRG.
SC.
Besides the two special cases, the cases with two or more input variables and many output variables also occur frequently. For example, we often try to keep the temperature profile in a distillation column with reflux flow, steam flow, etc. Unfortunately, no direct link between the SSE and row sum can be derived analytically. The closest relationship between the SSE and the row sum of the NRG for two input variables is (Appendix E)
Other
cases
det(GTG) 
{ 
ii1 
*s(i)) 
 
2 
(40) 

SSE(j) 
= 
(det 
G[ 
j])z 
where
S(2, m).
i is the ith element
of the jth
2tuple
subset of
the subsystem selection criterion generated
the square subsystem
two cases, choosing
outputs with smaller row sums leads to
equivalent
to choosing
SB.
Caseofm=n+
For this almost
1
square system, we can square down
the first n outputs to form
the system
a square
between
C(n
k [k
ntuple
the
similarly
by choosing
subsystem,
the
+
from
SSE
1, n) =
n +
“(k

G,,
and
rs.
to derive
Since
the relation
there
are
of
a set
and
1)”
=
n +
1 possible
l]
choices
elements,
to select
the results for
can
choice,
also
he
we have
remaining
combinations
G= this particular
derived.
G,
For
G,
[l[
9ml
4m2
= 9.19nz
911
912
‘
.
91.
1
9nn.(34)
.
Despite
by previous
9mn
by eliminating
the minimum SSE [as defined in eq. (30)]. Equation
(40) bears no direct link to this criterion. Counter examples can be established such that by choosing the “n” largest of row sums is not equivalent to the minim
ization of the SSE.
test the
heuristic generated from previous two cases. Ten thousand random matrices were generated for a given dimension. The optimal and suboptimal square sub systems are identified (by optimal we mean that the particular square subsystem results in the minimum SSE and by suboptimal we mean that the subsystem results in the one next to the minimum SSE) and the subsystem selected by the NRG criterion is checked against the optimal and suboptimal subsystem. The results (Table 3) show that the NRG subsystem selec tion criterion leads to the optimal subsystem in almost 77% of the cases tested and results in the op timal or/and a suboptimal subsystem in almost 88% of all cases tested. The reason for the NRG subsystem selection criterion fails to produce the op timal subsystem can be seen from eq. (40). The row sum (which the criterion based on) only accounts for part of the contribution to the SSE. The other factor comes from the determinant of the selected square system. Generally, the NRG subsystem selection criterion en sures a large determinant, but it does not guarantee the largest determinant. However, for a square sub system with competitive row sums, determinants of square subsystems play a deciding role for the values of the SSE. This explains the observation that, when the NRG subsystem selection criterion fails to pro duce the optimal square subsystem, the second largest row sum generally resultsin the minimum SSE. From the
A computer
experiment
is conducted
to
This particular choice corresponds to the first subset from S(n, m). Therefore, SSE( 1) can be calculated from eq. (30):
SSE(l)
=
Since G,
5
i=l
=
n+l
11e(i) 11:=
5
i=l
G[l]
(Appendix
I/I

GG,
‘y:‘:
B), we have
II;
_{(}_{3}_{5}_{)}
SSE(l) = ,g2(detGci’)’= _{d}_{e}_{t}_{(}_{G}_{=}_{G}_{)} _{} _{(}_{d}_{e}_{t}_{G}_{[}_{l}_{]}_{)}_{Z}
(det G[l])’
(det G[lJ)’
and from
Appendix
C
rs(m)
=
det (GTG) 
@etG Cll)’
det(GTG)
From
eqs (36) and
(37), we get
(36)
(37)
SSE(l)
=
Mm)
1 
w(m)
^{(}^{3}^{8}^{)}
or
in a more
general
SSE(j)
where thej
in SSE(j)
form
=
rs(n +
2 
j)
1 
rs(n +
2 
j)
denotes the sum of square error
when 
the jth 
subset 
of 
S(n, m) 
is chosen 
to form 
a square subsystem. Since the value of the row sum is between 
zero and 
1318 
JINWEN CHANG 
and CHENGCHING 
Yu 

Table 
3. Numbers 
of optimal 
and suboptimal 
cases generated 
by the 

NRG 
square subsystem selection criterion’ 

m 

n 
4 
5 
6 
7 
8 
9 

2 
9532: 
906 1 
9307 
8830 
8350 
7185 

46X0 
530 
582 
479 
818 
1353 

3 
9144 
8356 
7593 
7564 
7353 

420 
1176 
1091 
1081 
1518 

4 
7974 
6302 
5478 
6182 

1228 
923 
2060 
1485 

5 
7785 
6446 
6357 

1461 
1696 
541 

6 
7788 
7362 

762 
1158 

7 
7x10 

1042 

+Number of cases tested for a eiven dimension = 
10,000. 

*Number of optimal cases. 

$Number of suboptimal cases. 

results 
of 
the 
computer 
experiment, we 
can further 
conjecture
that
the NRG
subsystem
selection
cri
terion
subsys
tem in most cases. After having selected the variables to square down the system, we can go on to design
controllers for the square subsystem.
will
result
in the optimal
or suboptimal
5Cl. Example: DL column.From the RS informa
tion (Section 4), we can eliminate the third output to form a square subsystem. Thus the rest of AN becomes
(AN)’ =
[
0.241
0.006
CO.7261

0.103
Cl.0947
0.008

If the variables are paired according
CO.8611
0.100
0.237
_{I}
.
to the NRG
as
shown in the (A”)’ (those in brackets), the controller structure becomes (u,, y,; u2, y2; z+, y,}, The RGA, A, for the square subsystem is
Fig.
0
200
400
Time
600
9. Closedloop
responses
for
a
unit
change in Yt CyBis the uncontrolled
step
set
output).
6
point
A=
C
0.245
0.006
co.7491

0.103
[ 1.0941
0.009
[0X58]

0.100
0.242
_{I}
.
Notice that there is little difference between A and A” for this example. A multiloop PI controller is used to control the square subsystem. The controller gains and reset times (Table 4) are determined used the BLT method (Luyben, 1986). Steadystate offsets exist for the uncontrolled variable, y,, for unit step set point changes in the controlled variables (Figs 9l 1). Sum mation of the square of these steadystate errors (SSE)
Table
4.
Controller
{UI_Y,I loop {WY, 1 loop {UaY, I loop
gains
BLT
and
method
reset times from
the
Controller
_{K}
gain



0.0986
10.2330
0.1159
Reset time
*,
(min)
25.33
107.48
143.73
0.8 
0.6 
_{D} _{Y}_{I}
Molto più che documenti.
Scopri tutto ciò che Scribd ha da offrire, inclusi libri e audiolibri dei maggiori editori.
Annulla in qualsiasi momento.