You are on page 1of 9

A Controllability Index for Heat Exchanger Networks

Denis L. Westphalen,

Brent R. Young, and WilliamY. Svrcek*


Department of Chemical and PetroleumEngineering, University of Calgary,
2500 University DriveNW, Calgary, AB, Canada T2N 1N4
Heat exchanger networks are wi del y empl oyed i n the chemi cal processi ng i ndustri es to recover
energy, resul ti ng i n reduced operati ng costs. Several methodol ogi es can be found i n the l i terature
for the opti mal desi gn of heat exchanger networks. Typi cal opti mal cri teri a are maxi mum energy
recovery and mi ni mum heat-transfer area. However, the heat i ntegrati on of process streams
can l ead to process structures that are di ffi cul t to control . I n thi s work, a heat exchanger network
control l abi l i ty i ndex i s proposed as a measure of heat exchanger network control l abi l i ty. Thi s
control l abi l i ty i ndex can be cal cul ated easi l y, maki ng i t qui te appropri ate for use at the conceptual
desi gn stage of a chemi cal process. A case study i s presented i n whi ch the control l abi l i ty i ndex
i s used to compare the control l abi l i ty characteri sti cs of di fferent networks and al so to i denti fy
the tradeoffs between control l abi l i ty and heat i ntegrati on.
1. Introduction
The desi gn of heat exchanger networks (HENs) i s a
subject that has recei ved si gni fi cant attenti on duri ng
the past 3 decades. Furman and Sahi ni di s
1
publ i shed
a detai l ed revi ew of the l i terature on heat exchanger
network synthesi s and ci ted some 460 arti cl es. HEN
synthesi s i s not onl y an i mportant research area, as the
l i terature presents an extensi ve l i st of exampl es i n
whi ch new desi gn methodol ogi es have been successful l y
appl i ed i n actual i ndustri al cases. I n pl ai n words, the
synthesi s of an HEN can be descri bed as the desi gn of
the heat exchangers for a gi ven process i n whi ch al l
process hot and col d streams reach thei r speci fi ed outl et
temperatures, usi ng mi ni mum annual i zed i nvestment
and operati ng costs as the performance cri teri a.
The heat i ntegrati on of process streams can l ead to
process structures that are di ffi cul t to control , and i n
some cases, thi s i nhi bi ts retrofi t of exi sti ng processes.
Luyben et al .
2
presented a general procedure for pl ant-
wi de control , for the case i n whi ch energy i ntegrati on
dramati cal l y al ters the dynami c behavi or of the pl ant.
I n thi s si tuati on, speci al attenti on must be pai d to the
process-to-process heat exchangers, parti cul arl y i f they
are used for heat removal from exothermi c reactors.
Control l abi l i ty i s an i mportant i ssue that shoul d be
taken i nto account i n HEN synthesi s.
Kotjabasaki s and Li nnhoff
3
i ntroduced the concept of
sensi ti vi ty tabl es and descri bed how heat exchanger
areas shoul d be i ncreased wi th the ai m of i ncreasi ng
the networks fl exi bi l i ty. Thei r method i s i ntui ti ve and
i s parti cul arl y useful when there i s a need to i mprove
the control l abi l i ty of an exi sti ng HEN. However, i t i s
not possi bl e to i ncorporate the i ssue of fl exi bi l i ty duri ng
the topol ogi cal concepti on of the network.
I nvesti gati ng the process control of HENs, Mathi sen
et al .
4
proposed some heuri sti c rul es for bypass pl ace-
ment and the sel ecti on of mani pul ated vari abl es. Gl em-
mestad et al .
5-7
proposed a method for the opti mal
operati on of HENs, whereby the process i s peri odi cal l y
opti mi zed to defi ne new set poi nts for some key tem-
peratures.
However, better operati on can be achi eved usi ng
si mpl e control structures.
Ol i vei ra et al .
8
anal yzed i n detai l the i nteracti ons
between the process control and desi gn of a parti cul ar
HEN, usi ng steady-state opti mi zati on and the cal cul a-
ti on of the condi ti on number for the sel ecti on of ma-
ni pul ated vari abl es. No i nformati on on the dynami c
behavi or was used i n thei r probl em formul ati on, and
as a resul t, the desi gned network had poor control l abi l -
i ty. I n summary, the bul k of the l i terature i s concerned
wi th the probl em of the process control of exi sti ng
networks, that i s, networks that have al ready been
desi gned, and as a consequence, the proposed method-
ol ogi es cannot be used at the synthesi s stage.
The use of a resi l i ence i ndex (RI )
9-12
has been
proposed as a measurement of fl exi bi l i ty, operabi l i ty,
and control l abi l i ty of a gi ven process. Some attempts
have been made to use an RI i n the desi gn of HENs.
The resul ti ng methodol ogy generates HENs that can
handl e i nl et temperature vari ati ons. Because onl y
energy bal ances are used i n the cal cul ati ons and no
consi derati on i s gi ven to heat-transfer areas, the resul ts
of thi s methodol ogy are not useabl e.
The goal of thi s work i s to present a new control l abi l -
i ty i ndex for HENs. Thi s i ndex i s i ntended to be a
pri mary functi on of the networks topol ogy that does not
depend on a parti cul ar control strategy or set of ma-
ni pul ated vari abl es. I t must be easi l y cal cul ated so that
i t can be used as a conceptual desi gn tool . I t i s expected
that the use of thi s control l abi l i ty i ndex wi l l enabl e
process engi neers to i ncl ude process control i ssues at
an earl y stage, namel y, conceptual desi gn.
2. Process Control of Heat Exchanger Networks
An HEN i s typi cal l y empl oyed to perform heat ex-
change, so that each process stream attai ns i ts speci fi ed
temperature. Every process i s subjected to random
di sturbances and operator upsets, whi ch necessi tates a
process control scheme that wi l l keep al l outl et temper-
atures near thei r set-poi nt val ues. As a fi rst step,
* To whom correspondence shoul d be addressed. E-mai l :
svrcek@ucal gary.ca.

Present address: Aspen Technol ogy, Cal gary, AB, Canada.


4659 I nd. Eng. Chem. Res. 2003, 42, 4659-4667
10.1021/i e020893z CCC: $25.00 2003 Ameri can Chemi cal Soci ety
Publ i shed on Web 07/30/2003
strategi es for the control of i ndi vi dual heat exchangers
are di scussed bel ow, and then strategi es for the control
of networks are anal yzed.
2.1. Process-to-Utility Heat Exchangers. Process-
to-uti l i ty heat exchangers are defi ned as those heat
exchange operati ons that exchange heat between a
process stream and a uti l i ty stream, as shown i n Fi gure
1. The outl et temperature of the process stream can be
easi l y control l ed usi ng the fl ow rate of the uti l i ty stream
as the mani pul ated vari abl e (e.g., Svrcek et al .
13
and
Dri edger
14
). General l y, the uti l i ty system of a chemi cal
compl ex i s desi gned to absorb l arge di sturbances i n the
process, maki ng process-to-uti l i ty heat exchangers rel a-
ti vel y easy to control .
2.2. Process-to-ProcessHeat Exchangers. Process-
to-process heat exchangers are defi ned as those heat
exchange operati ons that exchange heat between two
process streams, as shown i n Fi gure 2. The outl et
temperature of one process stream can be control l ed
usi ng the bypass fl ow of the other stream as the
mani pul ated vari abl e. Duri ng the operati on of a heat
exchanger, i f the bypass fl ow rate i s i ncreased, the mean
temperature di fference i s reduced, resul ti ng i n a smal l er
duty, as exempl i fi ed i n Fi gure 3 (where cal cul ati ons
were performed usi ng a constant UA val ue). As a resul t,
the actual mani pul ated vari abl e i s the heat l oad of the
heat exchanger. For that reason, i t i s not possi bl e to
control both outl et temperatures i n the same heat
exchanger, e.g., usi ng bypass streams on both si des.
Furthermore, a process-to-process heat exchanger that
wi l l be used to control an outl et temperature must be
desi gned wi th a nonzero bypass fl ow rate to handl e
di sturbances that requi re an i ncrease or decrease of the
heat l oad. Because the exi stence of a bypass stream
resul ts i n smal l er mean temperature di fferences, l arger
areas must be empl oyed, resul ti ng i n l arger capi tal
costs. Steady-state si mul ati ons show that, from an
economi c poi nt of vi ew, the bypass stream shoul d be
pl aced on the stream wi th the l argest heat fl ow rate
Figure 1. Process control of process-to-uti l i ty heat exchangers:
(A) use of col d uti l i ty, (B) use of hot uti l i ty.
Figure2. Process control of process-to-process heat exchangers.
Figure 3. I nfl uence of bypass stream on temperature di ffer-
ences: (A) 0% bypass fl ow rati o, (B) 20% bypass fl ow rati o.
Figure 4. Heat exchanger network from stream data shown i n
Tabl e 1.
4660 I nd. Eng. Chem. Res., Vol . 42, No. 20, 2003
capaci ty, because thi s stream can handl e l arger di stur-
bances wi th l ess addi ti onal capi tal cost, when compared
to the si tuati on i n whi ch the bypass i s pl aced on the
stream wi th the smal l est heat fl ow rate capaci ty. Seborg
et al .
15
stated that mani pul ated vari abl es that rapi dl y
affect the control l ed vari abl es shoul d be sel ected. Usi ng
thi s pri nci pl e, dynami c si mul ati ons i ndi cate that the
bypass stream must be pl aced on the exchanger stream
that has i ts outl et temperature control l ed, i ndepen-
dentl y of the heat fl ow rate capaci ty val ues. Steady-state
and dynami c si mul ati ons can l ead to confl i cti ng concl u-
si ons, and as a general pri nci pl e, dynami c resul ts shoul d
be used i n the desi gn of control schemes.
2.3. Heat Exchanger Networks. HENs contai n both
process-to-uti l i ty and process-to-process heat exchang-
ers. Stream spl i tti ng i s commonl y empl oyed i n HENs,
and as expl ai ned by Westphal en et al .,
16
stream di vi si on
i n a stream spl i tter shoul d not be used as a mani pul ated
vari abl e i n an HEN because i t mi ght not affect the
downstream outl et temperatures monotoni cal l y.
Fi gure 4 shows an HEN wi th four process streams
and si x heat exchangers i n a gri d di agram. Tabl e 1
presents the stream data for thi s network. I f al l four
outl et stream temperatures are control l ed, then four
mani pul ated vari abl es shoul d be sel ected. Each heat
exchanger can contri bute one mani pul ated vari abl e (i ts
heat l oad), resul ti ng i n the sel ecti on of four heat
exchangers. Westphal en el al
16
proposed some heuri sti c
rul es for the sel ecti on of heat exchangers for process
control . Methodol ogi es
4,5,8,16
for the sel ecti on of the
mani pul ated vari abl es i n an HEN are wel l -documented
and are not di scussed i n thi s paper.
3. Relative Gain Array and Heat Exchanger
Networks
The rel ati ve gai n array (RGA) i s a tool that i s
commonl y used to determi ne the pai ri ngs of control l ed/
mani pul ated vari abl es i n a control scheme, more speci f-
i cal l y, el i mi nati ng the bad pai ri ngs. The fi rst step i n
the cal cul ati on of the RGA i s the cal cul ati on of the
steady-state process gai ns (K
ij
). These gai n val ues show
how a speci fi c mani pul ated vari abl e affects a control l ed
vari abl e. For i nstance, a bypass stream can be pl aced
Figure 5. Graphi cal user i nterface of the software tool devel oped i n thi s work.
Table 1. StreamData
stream
i nl et
temperature
(C)
suppl y
temperature
(C)
mean fl ow
rate capaci ty
(kW/K)
H1 170.0 60.0 30.0
H2 150.0 30.0 15.0
C1 20.0 135.0 20.0
C2 80.0 140.0 40.0
Table 2. RGAs for All Possible Combinations of
Manipulated Variables
set mani pul ated vari abl es i denti ty matri x
1 E1, E2, E3, E4 no
2 E1, E2, E3, E5 yes
3 E1, E2, E3, E6 yes
4 E1, E2, E4, E5 yes
5 E1, E2, E4, E6 yes
6 E1, E2, E5, E6 no
7 E1, E3, E4, E5 no
8 E1, E3, E4, E6 no
9 E1, E3, E5, E6 yes
10 E1, E4, E5, E6 yes
11 E2, E3, E4, E5 yes
12 E2, E3, E4, E6 yes
13 E2, E3, E5, E6 a
14 E2, E4, E5, E6 a
15 E3, E4, E5, E6 yes
a
Gai n matri x i s si ngul ar.
I nd. Eng. Chem. Res., Vol . 42, No. 20, 2003 4661
on the col d si de of heat exchanger E2. I f the bypass rati o
on heat exchanger E2 (Fi gure 4) i s i ncreased from 0.1
(nomi nal val ue) to 0.2 and al l other mani pul ated
vari abl es are kept constant, then the steady-state outl et
temperatures of streams H1, H2, C1, and C2 wi l l change
to 66.7, 22.9, 130.4, and 140.0 C, respecti vel y. The
steady-state process gai n i s cal cul ated as the rati o of
the change i n a control l ed vari abl e to that i n a mani pu-
l ated vari abl e. Thus, from the above resul ts, the fol l ow-
i ng process gai ns can be cal cul ated
The same procedure coul d be repeated for the other
heat exchangers assumi ng a speci fi c posi ti on (hot or col d
si de) of the bypass streams for each process-to-process
heat exchanger. Because each process-to-process heat
exchanger presents two possi bl e pl acements for the
bypass stream, the total number of al ternati ve K
matri ces that coul d be cal cul ated i s 8 (2
3
). However,
al though the posi ti on of the bypass stream wi l l affect
the dynami c behavi or of the network, the i nteracti ons
shown by the RGA wi l l not change. Because the actual
mani pul ated vari abl e i n each heat exchanger i s the
duty, the process gai n matri x (K) can be cal cul ated by
i ntroduci ng the step change i n the duty i nstead of i n
the bypass stream fl ow rate. Usi ng thi s procedure, the
fol l owi ng process gai n matri x can be cal cul ated for a
-10% change i n the duty of each heat exchanger
From the gai n val ues, one can concl ude that heat
exchangers E4, E5, and E6 affect onl y streams C2, C1,
and H2, respecti vel y, and that thei r l oads shoul d be
used as mani pul ated vari abl es to control the outl et
temperatures of those streams. Any change to heat
exchanger E1 i s propagated through the whol e network,
affecti ng al l outl et temperatures. Therefore, heat ex-
changer E1s duty shoul d not be used as a mani pul ated
vari abl e i n thi s network. The process gai ns hel p i n
el uci dati ng process i nteracti ons, and i n some compl ex
networks, i t i s i mpossi bl e to i nterpret al l of the i nterac-
ti ons just by i nspecti on.
The RGA () i s cal cul ated from the process gai n
matri x (K) usi ng eq 1, where the symbol means an
el ement-by-el ement mul ti pl i cati on
The i nterpretati on of the el ements (
ij
) of the RGA i s
wel l -descri bed i n the l i terature (e.g., Ogunnai ke and
Ray,
17
Svrcek et al .
13
), and i t can be summari zed as
fol l ows:
1. For
ij
) 1.0, there i s no i nteracti on wi th other
control l oops, and the pai ri ng i-j shoul d be used.
2. For
ij
) 0, mani pul ated vari abl e j has no effect on
control l ed vari abl e i.
3. For
ij
) 0.5, there i s a hi gh degree of i nteracti on
wi th other control l oops.
4. For 0.5 <
ij
< 1.0, there i s i nteracti on wi th other
control l oops; however, the i-j pai ri ng woul d be prefer-
abl e as i t woul d mi ni mi ze i nteracti ons.
5. For
ij
> 1.0, the i nteracti on wi th other l oops
reduces the effect of the control l oop.
6. For
ij
< 0.0, the pai ri ng i-j mi ght l ead to an
unstabl e operati on.
Because the RGA can be cal cul ated onl y for a square
process gai n matri x, the sel ecti on of mani pul ated vari -
abl es i s performed a pri ori , and the RGA i ndi cates the
best possi bl e pai ri ng. For i nstance, the use of heat
exchangers E3, E4, E5, and E6 resul ts i n the fol l owi ng
RGA
Obvi ousl y, heat exchangers E3, E4, E5, and E6 shoul d
be used to control the outl et temperatures of streams
H1, C2, C1, and H2, respecti vel y. Thi s concl usi on can
al so be achi eved by i nspecti on or by usi ng the heuri sti c
rul es presented i n the l i terature.
I f, for i nstance, heat exchangers E2, E4, E5, and E6
are sel ected as mani pul ated vari abl es, the fol l owi ng gai n
matri x i s obtai ned
Gi ven thi s gai n matri x, i t i s not possi bl e to cal cul ate
the RGA because the gai n matri x i s si ngul ar, and
therefore, no useabl e pai ri ngs exi st. Upon further
anal ysi s of the network, i t can be seen that no heat
exchanger affects stream H1; that i s, i t i s i mpossi bl e to
control al l outl et temperatures usi ng thi s choi ce of
mani pul ated vari abl es. I t can al so be seen that al l
el ements i n the RGA correspondi ng to stream H1 are
equal to zero, meani ng that no mani pul ated vari abl e
affects thi s control l ed vari abl e.
A computer program was devel oped to cal cul ate the
RGA of an HEN and i mpl ement fol l owi ng al gori thm:
K
H1,E2
) 67, K
H2,E2
) -71,
K
C1,E2
) -46, K
C2,E2
) 0
K )
E1 E2 E3 E4 E5 E6
[
-0.0167 0 -0.0333 0 0 0
-0.0179 -0.0667 0.0357 0 0 -0.0667
-0.0116 0.0500 0.0232 0 0.0500 0
0.0250 0 0 0.0250 0 0
]
H1
H2
C1
C2
) [K
-1
]
T
K (1)
)
E3 E4 E5 E6
[
1 0 0 0
0 0 0 1
0 0 1 0
0 1 0 0
]
H1
H2
C1
C2
K )
E2 E4 E5 E6
[
0 0 0 0
-0.0667 0 0 -0.0667
0.0500 0 0.0500 0
0 0.0250 0 0
]
H1
H2
C1
C2
(1) Defi ne the HEN
(stream data, uti l i ty data, and topol ogy).
(2) Sol ve the materi al , energy, and
heat-transfer equati ons of the network.
(3) Store al l resul ts
(duti es, areas, and outl et temperatures)
for the base case.
4662 I nd. Eng. Chem. Res., Vol . 42, No. 20, 2003
The fol l owi ng assumpti ons were made i n the HEN
model : (1) The heat fl ow rate capaci ti es for al l process
streams are constant. (2) I ndi vi dual heat-transfer coef-
fi ci ents are constant for al l process and uti l i ty streams.
(3) Onl y countercurrent or 1-2 shel l s are sel ected. (4)
Each heat exchanger can consi st of many shel l s i n
seri es. (5) Heat-transfer equati ons are sol ved i ncl udi ng
the F
T
correcti on factor.
A nonsequenti al equati on sol ver was devel oped
whereby each i ndi vi dual process-to-process heat ex-
changer can have i ts duty or area speci fi ed. I n some
cases, where al l of the areas of the heat exchangers i n
a l oop are speci fi ed, i t i s not possi bl e to sol ve the
equati ons of the network. For those cases, the sol ver
automati cal l y sel ects the set of heat exchangers that wi l l
be used i n an i terati ve procedure. Fi gure 5 shows a
screen shot of the output of the program, whi ch was
wri tten i n C++.
Al though the RGA i s an i mportant tool i n the sel ec-
ti on of the best mani pul ated/control l ed vari abl e pai r-
i ngs, i t provi des no gui dance i n the sel ecti on of the best
set of mani pul ated vari abl es. The number of di fferent
possi bl e combi nati ons of mani pul ated vari abl es i n a
network can be cal cul ated usi ng
For i nstance, for the HEN shown i n Fi gure 4, there
are four process streams and si x heat exchangers,
resul ti ng i n a total of 15 possi bl e combi nati ons. The
RGAs were cal cul ated for al l 15 combi nati ons, and Tabl e
2 shows the combi nati ons that resul ted i n the RGA
bei ng si mi l ar to the i denti ty matri x, I (contai ni ng onl y
0s and 1s). Ni ne di fferent sets of perfect pai ri ngs are
shown; however, there i s no i nformati on on the control
performance of each set of these pai ri ngs.
4. Condition Number and Heat Exchanger
Networks
As shown i n Tabl e 2, for a typi cal HEN, several
di fferent control schemes can be sel ected that yi el d
acceptabl e control l abi l i ty. As an addi ti onal step, i t i s of
i nterest to compare these di fferent acceptabl e control
schemes i n a qui ck and rel i abl e way. Furthermore, the
control l abi l i ty characteri sti cs of the di fferent networks
shoul d be compared, so that control performance can
be i ncl uded i n the desi gn of new networks i n the earl y
desi gn stage.
Ogunnai ke and Ray
17
suggested that the condi ti on
number of the process gai n matri x coul d be used as a
di mensi onl ess measure of control l abi l i ty. The condi ti on
number of a matri x A i s defi ned as the rati o of the
l argest of the w
j
s to the smal l est of the w
j
s, where the
w
j
s are obtai ned usi ng a techni que cal l ed si ngul ar val ue
decomposi ti on. A matri x i s si ngul ar i f i ts condi ti on
number i s i nfi ni te, and i t i s i l l -condi ti oned i f i ts condi -
ti on number i s very l arge.
18
Several case studi es were performed, and they showed
that, i ndeed, the condi ti on number provi des an excel l ent
measure of control l abi l i ty for HENs. For the tested
confi gurati ons, the smal l est condi ti on numbers al ways
resul ted i n the most control l abl e process (from common
sense poi nt of vi ew and dynami c si mul ati ons) and
corresponded to rel ati ve gai n arrays that were cl ose to
the i denti ty matri x.
Tabl e 3 shows the condi ti on numbers for al l possi bl e
combi nati ons of mani pul ated vari abl es i n the network
shown i n Fi gure 4. The l argest condi ti on number i s
obtai ned when heat exchangers E2, E4, E5, and E6 are
sel ected as the mani pul ated vari abl es. As di scussed
earl i er, thi s combi nati on of mani pul ated vari abl es can-
not be used because i t does not al l ow for the si mul ta-
neous control of al l outl et temperatures. I t can be seen
that the condi ti on numbers of sets 7, 9, 11, and 15 are
of the same order of magni tude; therefore, any of them
coul d be sel ected for the control scheme. Dynami c
si mul ati on shoul d be used to sel ect the best control
scheme.
5. Controllability Index
The condi ti on number appears to be a useful measure
of control l abi l i ty; that i s, i t coul d be used as a control -
l abi l i ty i ndex (CI ) to compare di fferent HENs. However,
the condi ti on number suffers from the fact that di fferent
val ues can be obtai ned for the same network (eq 2),
dependi ng on the choi ce of mani pul ated vari abl es. From
a conceptual desi gn poi nt of vi ew, each new HEN shoul d
be desi gned usi ng the best control scheme. Hence, the
CI of an HEN i s defi ned as the smal l est of the condi ti on
numbers obtai ned for al l possi bl e combi nati ons of
mani pul ated vari abl es.
(4) Cal cul ate the K matri x:
(4.1) Sel ect a heat exchanger j
(process-to-process or process-to-uti l i ty).
(4.2) Speci fy i ts duty as 90% of the nomi nal val ue.
(4.3) Sol ve the equati ons of the network
keepi ng the areas of al l other process-to-
process heat exchangers and the duti es of
al l other process-to-uti l i ty heat exchangers
constant and equal to the val ues used i n the
base case.
(4.4) Cal cul ate al l K
ij
val ues for heat exchanger j.
(4.5) Repeat steps 4.1-4.4 unti l al l heat
exchangers have been sel ected.
(5) Cal cul ate the RGA:
(5.1) Sel ect a set of mani pul ated vari abl es.
(5.2) Cal cul ate K
-1
.
(5.3) Cal cul ate [K
-1
]
T
.
(5.4) Cal cul ate the RGA usi ng eq 1.
C )
m!
n!(m- n)!
(2)
Table 3. Condition Numbers of All Combinations of
Manipulated Variables
set mani pul ated vari abl es condi ti on number
1 E1, E2, E3, E4 5.5
2 E1, E2, E3, E5 6.9
3 E1, E2, E3, E6 8.1
4 E1, E2, E4, E5 8.6
5 E1, E2, E4, E6 9.8
6 E1, E2, E5, E6 1.8 10
16
7 E1, E3, E4, E5 4.2
8 E1, E3, E4, E6 6.6
9 E1, E3, E5, E6 4.8
10 E1, E4, E5, E6 6.4
11 E2, E3, E4, E5 4.5
12 E2, E3, E4, E6 5.3
13 E2, E3, E5, E6 2.5 10
23
14 E2, E4, E5, E6 1.0 10
50
15 E3, E4, E5, E6 3.2
I nd. Eng. Chem. Res., Vol . 42, No. 20, 2003 4663
Another i mportant i ssue i n the control l abi l i ty anal ysi s
of HENs i s the i denti fi cati on of subnetworks. A subnet-
work i s defi ned as an i ndependent set of streams that
are heat-i ntegrated. Fi gure 6 shows an HEN i n whi ch
four di fferent subnetworks can be i denti fi ed. Because
the heat exchangers of one parti cul ar subnetwork can
never be sel ected to pai r wi th an outl et temperature of
a stream l ocated i n a di fferent subnetwork, the RGA
and condi ti on number shoul d be cal cul ated for each
subnetwork separatel y. As proposed previ ousl y, the CI
of each subnetwork i s cal cul ated as the smal l est of the
condi ti on numbers obtai ned for al l possi bl e combi na-
ti ons of mani pul ated vari abl es. I f, i n a gi ven network,
al l but one subnetwork show good control l abi l i ty (smal l
condi ti on numbers), the subnetwork wi th poor control -
l abi l i ty i mpacts the control performance of the whol e
HEN, and therefore, the CI of the whol e network i s
defi ned as the l argest of the CI s obtai ned for al l
subnetworks.
A computer program that cal cul ates the CI of an HEN
was devel oped, and i ts cal cul ati on steps can be sum-
mari zed as fol l ows:
Figure 6. Exampl e of a network wi th four i ndependent subnetworks (A, B, C, and D).
(1) Defi ne the HEN
(stream data, uti l i ty data, and topol ogy).
(2) Cal cul ate the gai n matri x.
(3) I denti fy al l subnetworks.
(4) For each subnetwork
(4.1) I denti fy al l combi nati ons of mani pul ated
vari abl es.
4664 I nd. Eng. Chem. Res., Vol . 42, No. 20, 2003
As an exampl e, Tabl e 4 presents the stream data for
a refi nery crude col umn preheat trai n consi sti ng of 19
streams. The heat-transfer coeffi ci ents for al l process
streams were speci fi ed as 0.4 kW/(m
2
C). Tabl e 5 shows
the speci fi cati ons of the process-to-uti l i ty heat exchang-
ers, and Tabl e 6 shows the speci fi cati ons of the process-
to-process heat exchangers (1-2 shel l s) of the network
presented i n Fi gure 7.
The resul ts are summari zed i n Tabl e 7. I t i s possi bl e
to construct 1, 495, 1, 1, and 1 di fferent combi nati ons
of mani pul ated vari abl es for the subnetworks 1-5,
respecti vel y. The condi ti on numbers of subnetworks 1
and 3-5 (each of whi ch has onl y one possi bl e combi na-
ti on) are 8.2, 2.8, 9.6, and 3.2, respecti vel y. However,
the smal l est condi ti on number of subnetwork 2 (495
possi bl e confi gurati ons) i s 60. Accordi ng to the proposed
methodol ogy, then, the CI of the network shown i n
Fi gure 7 i s 60. Thi s l arge number woul d suggest that
onl y poor control l abi l i ty can be achi eved for thi s HEN
and that other topol ogi es shoul d be i nvesti gated to
i mprove the overal l control l abi l i ty of the HEN. Note that
l ess than 2 s was requi red to cal cul ate the CI of thi s
network usi ng a Penti um I I 500-MHz PC.
Figure 7. Refi nery crude col umn preheat trai n.
(4.2) Obtai n the cal cul ated gai n val ues and
defi ne a new gai n matri x for the
combi nati on of mani pul ated vari abl es.
(4.3) Cal cul ate the condi ti on number of
thi s new gai n matri x.
(4.4) Cal cul ate the CI of the subnetwork as the
smal l est of the condi ti on numbers obtai ned
for al l combi nati ons of mani pul ated vari abl es.
(5) Cal cul ate the CI of the network as the
l argest of the CI s obtai ned for al l subnetworks.
I nd. Eng. Chem. Res., Vol . 42, No. 20, 2003 4665
6. Case Study
Tabl e 8 shows stream data for a process wi th two hot
and two col d process streams. One network was de-
si gned usi ng the pi nch desi gn method (Fi gure 8, net-
work A), and a di fferent network was desi gned usi ng
the Recommended Designs feature (an MI NLP method)
of the software HX-Net
19,20
(Fi gure 9, network B).
Tabl es 9 and 10 summari ze the resul ts of the control -
l abi l i ty anal yses performed on networks A and B,
respecti vel y. The resul ti ng CI s are 3.95 and 2.50 for
networks A and B, respecti vel y. Therefore, from a
process control l abi l i ty poi nt of vi ew, network B i s
preferred.
The resul ts i n Tabl e 10 suggest that a more thorough
anal ysi s of the probl em i s warranted. I t can be seen that
the CI of subnetwork 1 i s 2.00; however, the overal l
control l abi l i ty of the whol e network i s due to the
subnetwork 2. The heat l oad of heat exchanger E3 i s
very smal l when compared wi th the l oads of the other
heat exchangers. For that reason, thi s heat exchanger
cannot handl e the l arge di sturbances i n the remai nder
of network, and the overal l control l abi l i ty coul d be
i mproved i f thi s heat exchanger were desi gned wi th a
Table 4. StreamData
stream
i nl et
temperature
(C)
suppl y
temperature
(C)
mean fl ow
rate capaci ty
(kW/K)
H1 361.1 252.5 151.9
H2 292.0 251.7 637.1
H3 288.6 214.9 405.8
H4 279.8 54.9 99.7
H5 251.7 181.2 100.3
H6 221.5 160.5 302.8
H7 221.1 91.6 49.7
H8 204.9 55.0 101.6
H9 168.1 43.9 21.5
H12 139.8 41.9 57.4
H13 136.5 43.0 51.2
H14 120.0 77.6 638.9
C1 136.1 348.9 803.6
C2 140.0 175.1 69.4
C3 140.1 149.6 331.2
C4 140.0 149.5 330.5
C5 15.0 143.2 298.1
C6 15.0 143.1 298.3
C10 50.0 105.0 58.3
Table 5. Process-to-Utility Heat Exchangers
uni t duty (kW) uti l i ty type
E18 98 910 fi red heat
E19 1586 col d water
E20 800 col d water
E21 4000 col d water
E22 6930 col d water
E23 8230 col d water
Table 6. Process-to-Process Heat Exchangers
uni t area (m
2
) uni t area (m
2
) uni t area (m
2
)
E1 1048.10 E7 173.50 E13 549.17
E2 1609.64 E8 414.46 E14 549.17
E3 601.99 E9 1258.24 E15 415.66
E4 260.93 E10 2531.95 E16 349.14
E5 1033.69 E11 268.08 E17 290.21
E6 1644.79 E12 756.34
Table 7. Controllability Index Results
subnetwork streams
number of
combi nati ons CI
1 H1, C1, H3, H2 1 8.2
2 H12, C5, H14, C6, H9, H4 495 60
3 H13, C10 1 9.6
4 H5, C4, C3 1 9.6
5 H7, C2 1 3.2
Table 8. StreamData
stream
i nl et
temperature
(C)
suppl y
temperature
(C)
mean fl ow
rate capaci ty
(kW/K)
H1 300.0 80.0 30.0
H2 300.0 71.1 45.0
C1 40.0 180.0 40.0
C2 140.0 240.0 60.0
Figure 8. Case study, network A.
Figure 9. Case study, network B.
Table 9. Controllability Analysis of Network A
set
mani pul ated
vari abl es
condi ti on
number
RGA si mi l ar
to I?
1 E1, E2, E3, E4 24.1 no
2 E1, E2, E3, E5 10.3 no
3 E1, E2, E3, E6 4.05 yes
4 E1, E2, E4, E5 142 yes
5 E1, E2, E4, E6 12.6 no
6 E1, E2, E5, E6 5.01 yes
7 E1, E3, E4, E5 13.0 no
8 E1, E3, E4, E6 3.95 no
9 E1, E3, E5, E6 5.71 no
10 E1, E4, E5, E6 7.97 no
11 E2, E3, E4, E5 14.4 yes
12 E2, E3, E4, E6 14.1 no
13 E2, E3, E5, E6 7.36 yes
14 E2, E4, E5, E6 9.04 yes
15 E3, E4, E5, E6 6.23 10
14
no
Table 10. Controllability Analysis of Network B
set
mani pul ated
vari abl es
condi ti on
number
RGA si mi l ar
to I?
subnetwork 1
1 E1, E2 2.00 yes
2 E1, E5 2.62 yes
3 E2, E5 4.27 yes
subnetwork 2
1 E3, E4 2.50 yes
4666 I nd. Eng. Chem. Res., Vol . 42, No. 20, 2003
l arger duty, resul ti ng i n l ess heat i ntegrati on. Thi s
si mpl e exampl e shows that a tradeoff between energy
savi ngs and control l abi l i ty i s the i ssue and that i t can
be resol ved easi l y usi ng the tool s proposed i n thi s work.
7. Conclusions
I n thi s work, a new HEN control l abi l i ty i ndex, based
on the condi ti on number, i s proposed as a measure of
heat exchanger network control l abi l i ty. Thi s control -
l abi l i ty i ndex can be cal cul ated easi l y, maki ng i t ap-
propri ate for use duri ng conceptual desi gn and duri ng
pl ant revamps.
Operati ng and capi tal costs are the usual vari abl es
used i n the sel ecti on of the most sui tabl e HEN for a
process. Someti mes, the resul ti ng networks are di ffi cul t
to operate because process control aspects were not
taken i nto account duri ng the desi gn stage. Moreover,
several networks mi ght present comparabl e costs, and
the control l abi l i ty i ndex proposed i n thi s work coul d be
used as an addi ti onal deci si on vari abl e.
For a gi ven network, the control l abi l i ty i ndex provi des
i nsi ght i nto ways i n whi ch the control l abi l i ty can be
i mproved and cl earl y i denti fi es the tradeoffs bei ng made
between control performance and energy savi ngs.
Thi s methodol ogy al so anal yzes al l possi bl e pai ri ngs
between mani pul ated and control l ed vari abl es i n an
HEN and suggests the best al ternati ves. The fi nal
synthesi s of the control structure shoul d be checked wi th
the ai d of dynami c si mul ati on tool s.
Nomenclature
C ) number of combi nati ons
CI ) control l abi l i ty i ndex
i ) i ndex for the control l ed vari abl e (outl et temperature)
j ) i ndex for the mani pul ated vari abl e (heat exchanger
duty)
K ) steady-state process gai n matri x
K
ij
) i ndi vi dual el ement of the process gai n matri x
m) number of heat exchangers i n the network
n ) number of process streams i n the network
) rel ati ve gai n array (RGA)

ij
) i ndi vi dual el ement of the rel ati ve gai n array
Acknowledgment
Fi nanci al support from The Paci fi c I nsti tute for the
Mathemati cal Sci ences (PI MS), Hyprotech Ltd., and the
Department of Chemi cal and Petrol eum Engi neeri ng i s
grateful l y acknowl edged. Dr. Hi ren Shethna of Hypro-
tech Ltd. i s thanked for frui tful di scussi ons.
Literature Cited
(1) Furman, K. C.; Sahi ni di s, N. V. A Cri ti cal Revi ew and
Annotated Bi bl i ography for Heat Exchanger Network Synthesi s
i n the 20th Century. I nd. Eng. Chem. Res. 2002, 41, 2335.
(2) Luyben, M. L.; Tyreus, B. D.; Luyben, W. L. Pl antwi de
Control Desi gn Procedure. AI ChE J . 1997, 43, 3161.
(3) Kotjabasaki s, E.; Li nnhoff, B. Sensi ti vi ty Tabl es for the
Desi gn of Fl exi bl e Processes. (1) How Much Conti ngency i n Heat
Exchanger Networks i s Cost-Effecti ve. Chem. Eng. Res. 1986, 64,
199.
(4) Mathi sen, K. W.; Skogestad, S.; Wol ff, E. A. Bypass Sel ec-
ti on for Control of Heat Exchanger Networks. Presented at the
Fi rst European Symposi um on Computer Ai ded Process Engi neer-
i ngsESCAPE 1, El si nore, Denmark, May 24-28, 1992.
(5) Gl emmestad, B.; Mathi sen, K. W.; Gundersen, T. Opti mal
Operati on of Heat Exchanger Networks Based on Structural
I nformati on. Comput. Chem. Eng. 1996, 20, S823.
(6) Gl emmestad, B.; Gundersen, T. A Systemati c Procedure for
Opti mal Operati ons of Heat Exchanger Networks. AI ChE Symp.
Ser. 1998, 320, 94.
(7) Gl emmestad, B.; Skogestad, S.; Gundersen, T. Opti mal
Operati on of Heat Exchanger Networks, Comput. Chem. Eng.
1999, 23, 509.
(8) Ol i vei ra, S. G.; Li porace, F. S.; Araujo, O. Q. F.; Quei roz,
E. M. The I mportance of Control Consi derati ons for Heat Ex-
changer Network Synthesi s: A Case Study. Braz. J . Chem. Eng.
2001, 18, 195.
(9) Lenhoff, A. M.; Morari , M. Desi gn of Resi l i ent Processi ng
Pl ants I . Process Desi gn under Consi derati on of Dynami c Aspects.
Chem. Eng. Sci. 1982, 37, 245.
(10) Marsel l e, D. F.; Morari , M.; Rudd, D. F. Desi gn of Resi l i ent
Processi ng Pl ants I I . Desi gn and Control of Energy Management
Systems. Chem. Eng. Sci. 1982, 37, 259.
(11) Saboo, A. K.; Morari , M. Desi gn of Resi l i ent Processi ng
Pl ants I V. Some New Resul ts on Heat Exchanger Network
Synthesi s. Chem. Eng. Sci. 1984, 39, 579.
(12) Saboo, A. K.; Morari , M.; Woodcock, D. C. Desi gn of
Resi l i ent Processi ng Pl ants VI I I . A Resi l i ence I ndex for Heat
Exchanger Networks. Chem. Eng. Sci. 1985, 40, 1553.
(13) Svrcek, W. Y.; Mahoney, D. P.; Young, B. R. A Real-Time
Approach toProcess Control; John Wi l ey & Sons Ltd.: Chi chester,
U.K., 2000.
(14) Dri edger, W. Control l i ng shel l and tube exchangers.
Hydrocarbon Process. 1998, 77.
(15) Seborg, D. E.; Edgar, T. F.; Mel l i champ, D. A. Process
Dynamics and Control; John Wi l ey & Sons: New York, 1989.
(16) Westphal en, D. L.; Young, B. R.; Svrcek, W. Y. Strategi es
for the Operati on and Control of Heat Exchanger Networks.
Accepted for presentati on at the Foundati ons of Computer-Ai ded
Process Operati ons, Coral Spri ngs, FL, 2003.
(17) Ogunnai ke, B. A.; Ray, W. H. Process Dynamics, Modeling
and Control; Oxford Uni versi ty Press: New York, 1994.
(18) Press: W. H.; Fl annery, B. P.; Teukol sky, S. A.; Vetterl i ng,
W. T. Numerical Recipes in CsThe Art of Scientific Computing;
Cambri dge Uni versi ty Press: Cambri dge, U.K., 1988.
(19) HX-Net, versi on 5.0; Hyprotech Ltd., AEA Technol ogy
Engi neeri ng Software: Cal gary, Al berta, Canada, 2001.
(20) Shethna, H. K.; Jezowski , J. M.; Casti l l o, F. J. L. A New
Methodol ogy for Si mul taneous Opti mi zati on of Capi tal and Op-
erati ng Cost Targets i n Heat Exchanger Network Synthesi s. Appl.
Therm. Eng. 2000, 20, 1577.
Received for review November 11, 2002
Revised manuscript received May 28, 2003
Accepted May 29, 2003
I E020893Z
I nd. Eng. Chem. Res., Vol . 42, No. 20, 2003 4667

You might also like