C’ompurrrs them. Engng, Vol. 14, No. 3, pp. 319-335, 1990 009% I354/90 $3.00 + 0.

00
Printed in Great Britain. All rights reserved Copyright 6 1990 Pergamon Press plc

CAPITAL COST TARGETS FOR HEAT EXCHANGER

NETWORKS COMPRISING MIXED MATERIALS OF
CONSTRUCTION, PRESSURE RATINGS AND
EXCHANGER TYPES

S. G. HALL, S. AHMAD and R. SMITH

Centre for Process Integration, Department of Chemical Engineering, University of Manchester Institute
of Science and Technology, Manchester M60 IQD, U.K.

(Received 28 June 1988; final reaision received I1 July 1989; received for publication 30 September 1989)

Abstract-Recent developments in pinch technology allow the capital/energy tradeoff in heat exchanger
networks to be set prior to design. Targets for both energy and capital costs are required. The published
work on capital cost targets assumes all network exchangers obey a single cost law. This means in practice
that their materials of construction, pressure rating and design type must be assumed uniform. This paper
develops a method which allows capital cost targets to take account of differences in exchanger
specification. The enhancement applies cost factors to each stream, determined by the exchanger design
it requires. A case study demonstrates that variations in exchanger specification can considerably alter
the capital/energy tradeoff. As a result, the structure of the global cost-optimum network can change.
These structural changes for cost optimality are thus caused by differences in exchanger specification. In
many cases, without correct initialization of the design to allow for these differences, the global
optimum-cost network cannot be found.

1. INTRODUCTION and reduced capital cost (Fig. I). In doing so, the
minimum utility requirements and their cost increase.
Significant progress in understanding heat exchanger
At some value of A7’,,,i,, there will be a minimum in
networks has been made in recent years, particularly
the total cost. This locates the minimum temperature
through the introduction of pinch technology
difference AT,,, which should be used to initialize the
(Linnhoff et al., 1982; Linnhoff and Hindmarsh, network design.
1983; Linnhoff and Vredeveld, 1984). Given the Significant benefits arise from such pre-design
material and energy balance for a process, the heating optimization (Ahmad, 1985; Ahmad and Linnhoff,
and cooling requirements of the process streams are 1986; Linnhoff and Ahmad, 1986, 1990; Smith and
known. Pinch technology allows the minimum utility
Linnhoff, 1988):
requirements to be calculated directly from the
stream data subject to an assumed minimum (a) the design task itself is simplified. Designs
temperature difference AT,,,,, . These energy targets based on an optimized initialization require
are predicted before the heat exchanger network is virtually no evolution or optimization to
designed. reduce their total cost;
More recent work by Ahmad and Linnhoff (1984), (b) in many problems, correct initialization is
Townsend and Linnhoff (1984), Ahmad (1985) and essential to locate the global optimum cost
Linnhoff and Ahmad (1990) show that the stream design structure;
data, together with the assumed value of AT,,,i,, also (c) design options in the material and energy
define a limit for the minimum capital cost of a heat balance, for example changes in reactor designs
exchanger network. This can be calculated directly or separation system design, can be effectively
from the process streams ahead of design. For the screened without performing repeated heat
engineer contemplating new design, this ability is exchanger network designs. This screening can
useful in several ways. The network cost target be performed using equipment costs for the
provides a reference against which the cost of any reaction and separation system together with
design can be compared. As a result, designs with a target costs for the heat exchanger network and
capital cost significantly above target can easily be utilities;
identified. (d) different utility options such as furnaces, gas
The capital cost targets can also be calculated over turbines and different steam levels can be
a range of AT,,,,, values. As AT,,,,, increases, all assessed more easily and with greater con-
temperature differences throughout the heat fidence knowing the capital cost implications
exchanger network increase, resulting in reduced area on the heat exchanger network.

CACE 141--F 319

320 S. G. HALL et al.

Annual
cost

Fig. I. The trade-off between capital and energy costs.

The work to date on captial cost targets assumes where:

that individual heat exchangers in a network all obey
the same cost law. In practical terms this would mean ATLM,= the logarithmic mean temperature difference
a network of exchangers all made in the same of enthalpy interval i,
material of construction, having the same pressure (q,), = enthalpy change of stream .j in interval i,
rating and of the same type. The application of such (h,), = heat transfer coefficient (including film, wall
capital cost targeting procedures is somewhat limited. and fouling contributions) of stream j in
Ahmad (1985) and Ahmad er al. (1990) suggest a interval i.
method to allow for non-uniform exchanger specifi-
This formula assumes heat transfer to be “vertical”
cations in capital cost targets. The present paper de-
against the composite curves (Fig. 3a). If all stream
velops the method in details and applies it in partic-
heat transfer coefficients (/z-values) are identical, the
ular to deal with capital cost targets and for non-
area predicted by equation (1) is a rigorous minimum.
uniform materials of construction, pressure ratings
When h-values vary significantly, deliberate
and equipment types. It will be shown that account-
“non-vertical” heat transfer, as shown in Fig. 3b, can
ing for variations in exchanger specification alters the
exploit the available temperature driving forces to
capital/energy tradeoff and this in turn can lead to
achieve an overall area less than that of the vertical
changes in network structure for cost optimality.
model used in equation (I). Providing h-values differ
by less than one order of magnitude, equation (1) has
Z. CAPITAL COST TARGETS TO DATE
been found to approximate network area to within
Given the material and energy balance of a process, 10% of the actual minimum (Townsend and
the stream temperatures and enthalpies are known. Linnhoff, 1984). Methods based on linear program-
By combining hot or cold stream enthalpies within ming for minimum area (Saboo et al., 1986) can
given temperature ranges, Linnhoff et al. (1982) show exploit h-value differences in this way to predict an
how “composite curves” can be constructed for the area lower than that resulting from the vertical model
process (Fig. 2). The relative position of the two used here.
curves is determined by the chosen minimum Recently, Ahmad et al. (1990) have compared the
temperature difference AT,,,. This sets the maximum various approaches for the calculation of area targets
possible heat recovery and the energy target. and suggest that area targets based on both the
In order to predict the capital cost of a heat vertical model and linear programming should be
exchanger network, the number of exchangers and compared before deciding how to proceed with
their surface area requirement must be determined. finding a near-optimum design. The designer will not
Total area is most easily evaluated by dividing the necessarily try and achieve the minimum area
composite curves into “enthalpy intervals” (Fig. 2). predicted by linear programming because a large
Townsend and Linnhoff (1984) give a simple formula number of matches above the minimum is usually
which allows the minimum network area to be necessary to approach the minimum area. This is not
predicted. usually the case with the area target from the vertical
model, which is often observed to be closely
A ml” =
“@” -L-~g~)],, (1) approached in a near-minimum number of matches
(Ahmad, 1985; Hall, 1986; Linnhoff and Ahmad,
 Capital cost targets for heat exchanger networks 321

Tempemture

fiat Utility
r ; 1
Hot Streams, I

Enthalpy Interval i
Stream population j

Cold Streams
qj thj
I
Maximum Energy 1
Recovery 1
I

b Enthdpy
Fig. 2. Enthalpy intervals in the composite curves.

1990). If there is a large difference between the area predicted location of AT,,, hardly changes between
targets predicted by the vertical and linear program- the two methods.
ming models then the designer may wish to try and In summary, the area target predicted by equation
exploit differences in h-values in the design. (1) is not a rigorous minimum, although the accuracy
As far as pre-design optimization is concerned, in most cases is good enough for pre-design optimiza-
Ahmad et al. (1990) further demonstrate that while tion. Furthermore, practical considerations in design
the linear programming method leads to a lower area typically lead to an area requirement larger than
and capital cost target, the slope of the capital cost minimum. The increased complexity of more sophis-
profile against AT,,,,, is often little affected compared ticated methods is often not justified. Capital cost
with the use of the vertical model in equation (l), targets developed in this paper will therefore be based
even for large variations in h-values. As a result, the on the vertical model assumed in equation (1).

Temperature Tel perature

Low h-value

1 C

A=zzEnth
Enthalpy
A----Y-_
1 “I-, WA2
a) Vertical heat transfer b) Non-vertical heat transfer

Fig. 3. Heat-transfer models

322 S. G. HALL et al.

A target for the number of heat exchange matches expression which sums the area contribution of each
or units in a network N,,,, is well established stream:
(Linnhoff et a/., 1982). If exchangers are pure
countercurrent devices then the number of units will
be equal to the number of exchangers. If, however,
equipment is used which is not truly countercurrent,
for example l--2 shell-and-tube exchangers, then:

N si,cib2 Nunila (2) where Aj = contribution to area target from stream j.

Recent work by Ahmad (1985) allows a target for Equation (5) shows that each stream makes a
the number of l-2 shells Nsni. to be evaluated from contribution to total heat transfer area defined only
the composite curves. The area target, equation (I), by its position in the composite curves and its
can also be modified to account for the increase in h-value. This contribution to area means also a
surface area observed when heat transfer in the contribution to capital cost. If, for example, a
equipment is not truly countercurrent. corrosive stream requires special materials of
To predict the capital cost of a network, it must construction, it will have a greater contribution to
first be assumed that a single heat exchanger with capital cost than a similar non-corrosive stream. If
surface area A can be costed according to a simple only one cost law is to be used for a network
relationship such as: comprising mixed materials of construction, the
area contribution of streams requiring special
installed capital cost of exchanger = a + bA’, (3) materials must somehow increase. One way this
may be done is by weighting the heat transfer
where:
coefficient to reflect the cost of the material the stream
a, b, c = cost law coefficients which vary according requires.
to materials of construction, pressure rating To develop the approach, first consider a single
and type of exchanger. exchanger whose cost may be represented as:

When cost targeting, the distribution of the installed capital cost of reference exchanger
targeted area between network exchangers is
= a, + b, A”, (6)
unknown. Thus to cost a network using equation (3),
some area distribution must be assumed, the simplest where:
being that all exchangers have the same area:
a,, b,, c, = cost law coefficients for the reference
exchanger.
network capital cost = N,,,, a + b &+%
[ (L]]. (4) If instead the heat exchanger is made to a different
At first sight the assumption of equal area exchang specification, its cost may be represented as:
ers used in equation (4) might seem crude. However, installed capital cost of special exchanger
from the point of view of predicting capital cost the
assumption turns out to be a remarkably good one =a,+b,Acz, (7)
(Ahmad, 1985; Hall, 1986; Ahmad et al., 1990). At
where:
the targeting stage, no given distribution can be
judged consistently better than another since the a2, bz, c2 = cost law coefficients for a special
network is not yet known. Evidence will be presented exchanger.
later to demonstrate that the equal area assumption
gives good enough capital cost predictions for the In the approach to be developed, the cost of the
purposes of optimizing targets. special exchanger can be determined from the refer-
So far, the resulting network capital cost targets ence cost law by using a modified area A *:
assume no variation in exchanger specification. In installed capital cost of special exchanger
other words, all exchangers are made from the same
materials of construction with the same pressure = a, + b, A *‘I. (8)
rating and are the same type. This limits the applica- Hall (1986) shows that exchanger cost data can
tion of such targets. usually be manipulated such that fixed costs, repre-
sented by the coefficient n in equations (6-8) do not
3. CAPITAL COST TARGETS TO ACCOUNT FOR
vary with exchanger specification. Equations (7) and
NON-UNIFORM EXCHANGER COSTS (8) can now be rearranged to give the modified
exchanger area A * as a function of actual area A and
The area target, equation (1) sums the area contri- the cost law coefficients:
butions from each enthalpy interval. Ahmad (1985)
and Ahmad et al. (1990) have shown that this
equation can be rearranged to an equivalent
 Capital cost targets for heat exchanger networks 323

The relationship between heat exchanger area and The same philosophy of weighting area
overall heat transfer coefficient U is given by: contributions in a single exchanger can be extended
to weighting stream area contributions for a whole
A=--------- Q (10) network. Some additional error in the targets is
AT,, . CT’ incurred by this extension resulting from the fact that
where Q = exchanger heat load. a stream may pass through several exchangers all
The ratio Q/AT,, is constant for a given heat with different areas. At the targeting stage, exchang-
exchanger and hence the modified U-value U* can be ers are assumed to be all the same size. The special
related to actual U-value: stream cost-weighting factor is then expressed as:

$, = (t!>‘;“(&!J +‘. (17)

The overall heat transfer coefficient in a single Hall (1986) has shown that shell-and-tube heat
exchanger comprises resistance contributions from exchanger cost laws can often be adjusted with little
both streams. Each contribution contains allowances loss in accuracy such that c is constant for different
for film, wall and fouling resistances. In practice the materials of construction:
overall heat transfer coefficient will depend to some
c, = c2 = c. (18)
extent on the exchanger flow arrangement. It is not
possible to specify such details at the targeting stage, Equation (17) then simplifies to:
hence the overall heat transfer coefficient must be
assumed independent of the flow arrangement: (19)
1 1 1
I (12) If the cost laws permit equation (19) to be used
U h”OT &Xn
then the calculations are somewhat simplified.
Equation (11) may be split stream-wise to obtain However, either equations (I 7) or (19) can be used in
an expression for the modified h-value h,* of either the procedure to follow.
stream in the match: Once the #~-factor has been evaluated for each
stream, a weighted network area target A,$, can be
calculated:

A stream-specific cost-weighting factor 4, to apply

to the h-value of a special stream j can now be
defined. This is the ratio of weighted to actual stream When the weighted h-values (+h) vary appreciably,
h-values: say by more than one order of magnitude, an
improved estimate of A in can be evaluated by linear
(14) programming (Saboo et al., 1986).
The weighted area A,& when applied using
The application of such a factor to cost a special equation (4) returns an essentially correct capita1 cost
heat exchanger but using the reference cost law is for a network of exchangers involving mixed
therefore: specification.
It is possible to approach the problem of capital
installed cost of special exchanger =

c’
cost targeting using other methods. For example, the

11
1 1 cost of the total area requirement can be calculated
~ (15) for each of the materials being considered. Simple
‘#+IOT ho-r + &OLD hco,n
weighting methods can then be used to interpolate
where between these pure specification costs. Weighting
according to the number of streams or relative duty
4 HOT- &OLD. (16)
of the streams corresponding to each specification is
For a single heat exchanger made entirely to the quite successful if the streams are fairly similar to one
special requirements, the capital cost resulting from another in their heat transfer coefficients and their
equation (15) is as accurate as the cost laws. This temperature differences for heat exchange on the
must be the case from the way in which the 4-factor composite curves. If there are large differences in heat
was derived. If the requirements of both streams are transfer coefficients and temperature differences,
different leading to a mixed specification, then the these must be accounted for in the interpolation by
capital cost target will be less accurate. The error weighting according to the relative area contributions
involved depends upon whether the special of the groups of streams (Rippin, 1989) or by the
requirements are caused by differences in materials of +-factor approach described in this paper.
construction, pressure rating or design type. These The procedure proposed here to target the capital
errors will be discussed later. cost of a network comprising mixed exchanger
324 S. G. HALL et al.

Table I. Stream and utiliry data for example Table 2. Installed shell/tube heat exchanger cost laws for
Heat different materials of constractmn
capacity Carbon-steel (CS) Cost (S) = 30,800 + 750A08’
Temperature (“C) flowrate h-value Stainless-steel (SS) Cost (6) = 30,800 + I6444 ‘*’
stream SUPPlY Target (kW “C- ‘) (kW m~“C’) CS!SS or ssics Cost (S) = 30,800 + 1339A”*’
I-Hot 120 65 50 0.50 Titanium (TI) Cost (S) = 30,800 + 4407A’”
2.Hot 80 50 300 0.25 CS!TI or TI/CS Cost (S) = 30,800 + 3349A08’
3-Hot 135 110 290 0.30 SS/TI or TIISS Cost (S) = 30,800 + 3749A08’
4-Hot 220 95 20 0.18 where A = shell/tube exchanger area (m’)
5-Hot 135 105 260 0.25 Plant life = 6 yr, capital interest = 10% per annum
6-Cold 65 90 150 0.27
‘I-Cold 75 200 140 0.25
B-Cold 30 210 100 0.15 capital cost targets. Consider first the application of
9-Cold 60 I40 50 0.45 the method to networks requiring mixed materials of
Steam 250 0.35
Cooling I5 - 0.20 construction. An example will be used to illustrate the
water procedure and assess its accuracy. The stream and
utility data for the example are presented in Table 1.
Some of the streams will later be assumed corrosive,
specification can be summarized as follows: requiring special materials of construction. Appropri-
1. Choose a reference cost law for the heat ate shell-and-tube heat exchanger cost equations are
exchangers. As will be discussed later, this is presented in Table 2. These cost laws from Hall
normally taken to be the cost law which applies (1986) were evaluated using the detailed exchanger
to the category of streams which make the costing procedure of Purohit (1983).
largest contribution to capital cost. As a reference, first consider how the target and
2. Calculate &factors for those streams which design capital costs compare for a network made in
require a specification different from that of the a single material. Assuming all heat exchangers to be
reference using equations (17) or (19). If carbon-steel shell-and-tube, the stream data of Table
equation (17) is to be used then the actual area 1 give the targets shown in Table 3 at AT,,,,, = 20°C.
A,i, must first be calculated from equation (1). A network designed at this AT,,,,, is shown in Fig. 4a.
3. Calculate the weighted area A$, from equation Figure 4b shows the same network after exchanger
(20). If the weighted h-values (4jih, for each loads have been redistributed to minimize capital cost
stream i) differ by more than one order of but keeping the energy consumption fixed (see the
magnitude, A& can, if necessary, be more Appendix). The design capital costs are compared in
accurately evaluated using a linear program- Table 3 and are each within 3% of target.
ming technique (Saboo et al., 1986). The results The good agreement between targets and design in
can be compared to define the incentive for Table 3 for the network in a single material
using the more complex linear programming demonstrates the accuracy of the assumptions made
model for the minimum (weighted) area, in targeting, particularly that of equal exchanger
bearing in mind that the nearer the design areas made in equation (4). The errors in Table 3 are
approaches this area the more units above min- typical for a wide range of problems (Ahmad, 1985;
imum that will be required (Ahmad ef al., 1990). Hall, 1986).
4. Calculate the target capital cost for the mixed- Being close to area target, the design in Fig. 4a is
specification heat exchanger network from a good candidate upon which to judge minimum
equation (4) using the cost law coefficients for capital cost when using different or mixed materials
the reference specification. of construction. In the first case. suppose streams 3,
4, 6 and 9 require stainless-steel while the remaining
This procedure will now be applied to different streams require carbon-steel. For greatest accuracy,
situations with mixed exchanger specifications. the reference cost law should be set primarily by that
material whose streams contribute most to capital
cost. It is recommended that this is normally based on
4. CAPITAL COST TARGETS FOR MIXED MATERIALS
OF CONSTRUCTION those streams which contribute most to total heat
transfer area. In the example, these are the carbon-
A method has been proposed which enables mixed steel streams. Since the exchanger cost laws have the
exchanger cost laws to be considered in network same area exponent, the +-factor for stainless-steel

Table 3. Target/design comparison for a carbon-steel network

Energy Installed
consumption Number of Area capital cost
(MW) units (m’) (S)
Target (AT,,, = 2VC) 20.95 I3 A m,n= 9524 2,440,009
Pinch design
Fig. 48 (AI,,, = ZO’C) 20.95 13 9659 2,374,733
Minimum cost
design (Fig. 4b) 20.95 I3 9545 2.357.337
 Capital cost targets for heat exchanger networks 325

Exchanger duties -MW

strenm CP’S -nwr’c
S,reom h-valuer -kWli&C

xchonger duties-HW
trwmn CP’S -MWl’C
tram h-mlues -kWh&

Fig. 4. Initial pinch design and minimum cost network for example stream data at fixed energy
consumption (AT,,,,, = 20°C).

streams relative to carbon-steel is determined using from those in Fig. 4b. The actual capital costs of the
equation (19): designs compared in Table 4 are within 2% of target.
It should be noted that this network comprises some
dss = & = & = &, = & = g ‘!08’= 0.3795. (21) exchangers made solely in carbon-steel, some solely in
( > stainless-steel and some exchangers in a mixture of
The target capital cost is compared with the design the two materials. Prior to this paper, the designer
in Fig. 4a, both before and after exchanger load would probably have used cost targets based on
redistribution to minimize capital cost, in Table 4. carbon-steel throughout and stainless-steel through-
Some exchangers now obey different cost laws, hence out as bounds on the actual capital cost. In fact, the
the exchanger loads will be redistributed differently minimum network capital cost is 22% higher than a

Table 4. Target/design comparison for carbon-stainless-steel network

Energy Installed
consumption Number of Area capital cat
(MW) units (Ill? (16)

Target (AT,,, = 20°C) 20.95 13 A :,. = 12,438 2.932.450

Pinch design
Fig. 4a (AT,,, = 20°C) 20.95 13 Y659 2,986.560
Minimum cost design 20.95 13 9739 2.971.438

Streams 3. 4. 6 and 9 reouires stainless-steel

326 S. G. HALL er al.

Table 5. Target/design comparison for carbon-steel-titanium network

Energy I”Stalled
consumption Number of caDhal c”st
(MC) units cm’) 0)
Target (AT,, = ZO’C) 20.95 13 A:,, = 23,616 4656,459
Pinch design
Fig. 4a (AT,,, = 20°C) 20.95 13 9659 5,015,361
Minimum cost design 20.95 13 10.744 4.791.104
Streams 3, 4, 6 and 9 require titanium

Table 6. Target/design comparison for network in carbon- and stainless-steels and titanium
Energy installed
consumption Number of Area capital cost
(MWI units (“l2) (S\
Target (AT,,, = 20-C) 20.95 13 A:,, = 18,456 3,886,064
Pinch design
Fig. 4a (AT,,. = 2o’C) 20.95 13 9659 3,938,727
Minimum cost design 20.95 13 11.007 3.x4.609
Streams 3, 4 in stainless-steel, streams 6 and 9 in titanium

target cost based on carbon-steel throughout and derived using $I-factors and design costs when mixed
is 39% lower than one based on stainless-steel materials of construction are required. Similar
throughout. This wide band represents a considerable accuracy has been observed for other networks by
degree of uncertainty. Hall (1986).
Now consider a second case in which streams 3, 4, The reason cost-weighting works so well for mixed
6 and 9 require titanium and the remainder require materials of construction is because shell-and-tube
carbon-steel. Again, the reference material is chosen exchangers are able to accommodate different materi-
to be carbon-steel. The relative 4-factor of titanium als within a single unit. The actual cost of such an
is given by: exchanger is somewhere between the costs of the
exchanger made entirely in either material, as is the
&, = & = qJ = & = 49 = = 0.1123. (22) target cost.

Target and design capital costs for the exchangers 5. CAPITAL COST TARGETS FOR MMED PRESSURE
in this case are compared in Table 5 using the RATINGS
network in Fig. 4a before and after optimization for
capital cost. The design capital costs are less than 5% Accurate capital cost targets for heat exchanger
above the target. It should be noted that this is an networks with mixed pressure ratings can also be
determined using cost-weighting factors. A shell-and-
extreme case and the accuracy can be highlighted
tube heat exchanger can accommodate streams of
when we try to bound the problem using uniform cost
different pressure and the problem is similar to that
laws. The minimum design capital cost is 96% higher
of materials of construction.
than a target capital cost based on carbon-steel
throughout and 61% lower than a target based on To show the accuracy of the procedure, a capital
cost target is determined for the stream data given in
titanium throughout. These bounds are so wide that
Table 1 at AT,, = 20°C. For the purpose of this
they would probably be of no value at all to the
designer. example, streams 4 and 9 are rated at 35 bar and
streams 3 and 7 at 60 bar. The remainder are rated
Consider as a third and final case, one in which
streams 3 and 4 require stainless-steel, streams 6 and at IO bar or below and these set at the reference cost
law. Heat transfer coefficients are assumed consistent
9 require titanium and the remainder require carbon-
with these pressures. Table 7 shows the exchanger
steel. Again, carbon-steel is chosen as reference. Here,
cost laws used. These have also been evaluated using
& = &Q= 0.3795 and & = I$~= 0.1123. Target and
design capital costs are compared in Table 6. They information given by Purohit (1983).
differ by less than 3%. For comparison, the minimum
design cost is 54% higher than a capital cost target
Table 7. Installed carbon-steel shell/tube heat exchanger
based on carbon-steel throughout, 23% below a cost laws for different pressure ratings
target based on stainless-steel throughout and 70% Pressure (bar)
lower than a target based on titanium throughout. shell/tube
With three materials, bounds have less significance lo/lo cost (S) = 30,800 + 750Aa8’
since it is unclear which bounds on network capital IO/35 Cost (S) = 30 800 + 890A O.”
35135 Cost ($) = 301800 + 1089A ‘*I
cost are now relevant. IO/60 Cost (S) 2 30,800 + 983A 08’
In all the examples presented, close agreement 60160 Cost (5) = 30,800 + l438A ‘*’
is observed between network capital cost targets 35/4a cost (%)= 30,800 + l20lA On’
 Capital cost targets for heat exchanger networks 327

Table 8. Target/design comparison for a network of different stream pressures

Energy Installed
consumption Number of Area capital cost
(MW) units (m’) ($)
Target (AT,,, = 20 ‘C) 20.95 13 A :,. = 12,597 2.956,532
Pinch design
Fig. 4a (AT,,. = ZO’C) 20.95 I3 9659 2,792,341
Minimum cat design 20.95 I3 9652 2,776,141
Streams 4, 9 rated at 35 bar; streams 3, 7 at 60 bar; remainder at IO bar

A comparison between target and designs for the between those of the reference and special exchanger
network is presented in Table 8. To minimize individ- types. Although some accuracy is lost, the targets still
ual exchanger cost in the design, the high-pressure give capital costs which are better than otherwise
stream is assumed to be placed on the tube side. obtained assuming a single exchanger-type cost law
When the network in Fig. 4a is optimized for the throughout.
exchangers in this case, the minimum capital cost of To demonstrate this, consider the example stream
the design is only 6% below the target capital cost. data of Table 1 and assume the tabulated h-values to
If it is assumed that all exchangers are rated for the be appropriate now for flow through plate heat
same pressure, a capital cost target at 10 bar is 12% exchangers. Assume streams 3 and 7 require spiral
below this network capital cost, one at 35 bar is 21% heat exchangers and all others require plate exchang-
above and a target assuming 60 bar is 55% above. ers. The plate and spiral exchanger cost laws are
The bounding range is impractically large and does presented in Table 9 and are based on cost laws given
not help to locate the network cost target with any by Kumana (1984). Note that the spiral cost law takes
confidence. account of the increase in area required to perform a
As with mixed materials of construction, the cost- given heat exchange task in a spiral relative to a plate
weighting approach to mixed pressure ratings gives exchanger (Hall, 1986).
close agreement between target and design capital Table 10 shows a comparison between targets and
costs for reasons similar to those outlined for mixed the design from Fig. 4a. When this network is
materials. Again, mixed pressure ratings can be optimized for the exchangers in this case the
accommodated in a shell-and-tube exchanger. As minimum design capital cost is 3% below target. If
such, design practice is consistent with the targeting the network is assumed to be made entirely in plate
assumptions. heat exchangers, the design capital cost is 46% over
target. Similarly, the design capital cost is 50% below
6. CAPITAL COST TARGETS FOR MIXED EQUIPMENT the target assuming all spiral exchangers to be
TYPES present.
In this particular example, close agreement has
Another situation which causes differences in cost
been obtained between actual and target capital costs.
laws is for varying design types. This is different in
This is not always the case. Further examples of
principle from mixed materials and pressure ratings
capital cost targets for networks of mixed exchanger
(Hall, 1986). A match between, say a clean stream
types have been presented by Hall (1986). The
requiring a reference type exchanger such as shell-
accuracy of such cost targets is often good. However,
and-tube and a slurry stream requiring a spiral
due to the inability to mix different exchanger types
exchanger, forces the match to be a spiral. In cost
in one unit as assumed in the targets, the cost-weight-
targeting, a weighting factor would be applied solely
ing factors should be applied judiciously. Results are
to the slurry stream resulting in a cost somewhere
not consistently reliable.

Table 9. Installed cap&d cost data for different heat 7. AN ANALYSIS OF THE ERRORS ARISING FROM
exchanger types made in stainless-steel. ENHANCED CAPITAL COST TARGETING
Plate-and-frame cost (E) = 1950AO’”
Spiral Cost {Sj = 19,687/l OS9 The assumptions made at the targeting stage lead
where .4 = eqwalent plate-and-frame surface area (m2) to a number of errors in network cost targets. Some
Plant life = 6 yr. capital interest = 16% per annum
of these errors are common to all capital cost targets

Table IO. Target/design comparison for a network of diffenng exchanger types

Energy Installed
consumption Number of Ared caoital cost
CM% units (m? (8
Target (AT,,, = 20 C) 20.95 I3 A:,, = 16,175 6,424,9 I7
Pinch drsiwn
Fig. 4a (AT,,. = 20’C) 20.95 I3 9659 6,552,263
Minimum cost design 20.95 I3 10,421 6,223,490
Streams 3, 7 spiral exchangers; remainder plate exchangers
328 S. G. HALL et al.

gable 1I. Effectthat the relative magnitude of stream h-values has on cost target accuracy for the network in Fig. 4a at extreme conditions
Taraet installed Ootimized design installed Error in
h, (7.x h,,,, ‘~h4.6.9 cap,tdcost cap:tal cost target
Materials h, (kWm? C-‘) (kW mm2 Cm’) ($) ($) (“A)
cs-ss 0.1 0.1510 I.5100 4,52 1,965 3,836,029 fl8
IO 0 2500 0.2500 3,705,095 3,685,994 fl
10.0 I .2400 0. I240 2.639.470 3,015,317 -~ I2
CS-TI 0. I 0.1510 I Sloe I I .266.021 9.153.349 +23
I.0 0.2500 0.2500 8,079,235 8,183,646 -I
10.0 I .2400 0.1240 3.671.912 5,755,183 -36
Streams 3, 5, 7, 8 are corrosive. Area large:et constant at 9022 ml. AT,,,,, = 20’C

and sxne are. specific to the use of cost-weighting a coincidence, it should nevertheless be examined to
factors. Ahmad (1985) describes the following causes appreciate the errors possible. Table 11 demonstrates
for the errors common to all capital cost targets: the effect that this kind of difference in h-values
has, based on the example stream and cost data in
. Total heat transfer area is assumed to be divided
Tables 1 and 2. Corrosive streams 3, 5, 7 and 8
equally between network exchangers.
require either stainless-steel or titanium. The heat
l The area target is usually less than the area
transfer coefficients of the corrosive streams are set
observed in design.
to one value whilst all the other non-corrosive stream
These effects respectively give small positive and h-values are set to another value. In satisfying the
negative errors in the cost targets and partially cancel. ratios of h-values in Table 11, the actual h-values of
The errors arising directly from cost-weighting corrosive and non-corrosive streams have been set to
factors have been discussed by Hall (1986). Consider give the same overall area target between the cases.
first, enhanced cost targets for mixed materials of Because equation (4) is used for capital cost
construction. There is no error when g-factors are predictions, maintaining the same area target is
applied to the h-values of streams in a single important for the analysis to isolate those errors
exchanger made in one material. In practice, attributed to h-value differences.
exchangers can have differing shell-side and tube-side Table 11 compares the resulting capital cost targets
materials, where QHoT# &oLD. Some error does with the capital cost of the network in Fig. 4a after
occur when predicting the cost of a single mixed- optimization in each case. It is seen that the errors in
material exchanger and this requires further capital cost can be large if h-values differ markedly.
examination. However, the large errors here represent extreme
Reconsider equation (8) to target the cost of a cases which are unlikely to be met in practice. When
single heat exchanger in mixed materials. In a single the h-values of corrosive and non-corrosive streams
exchanger, a relatively small h-value causes a stream have similar values, the errors are quite small. In
to dominate total exchanger area and hence target many practical situations stream h-values do not
cost. If the non-corrosive stream h-value is compara- differ by more than one order of magnitude. Further-
tively small, the target cost approaches the actual cost more, when seeking lower overall area in the network,
of a heat exchanger made solely in the reference matches are very often made between streams of
material. Thus, exchanger or network capital cost similar h-value. Thus, the errors in such cases are
depends to some extent on the magnitude of the typically smaller than those shown in Table 1 I. This
difference in stream h-values. In the more usual cases is demonstrated in Table 12 where the h-value
found in practice, heat transfer coefficients of special differences distribute among streams regardless of
and reference streams share a common range of which are corrosive and non-corrosive, and the errors
values. However. the largest errors for a network of involved are appreciably smaller.
mixed-material exchangers are expected when the In the case of a network with mixed pressure
h-values of streams requiring one material all vary ratings, trends similar to those of Tables 11 and 12
significantly from the h-values of streams requiring are expected when using cost-weighting factors. This
another material. Although this would in practice be again is due to the ability to have different pressure

Table 12. Effect that the relative magnitude of stream h-values has on cost target accuracy for the network in Fig. 4a at typical conditions
Target inscalled Optimized design installed Error in
h 2 1.x v caoital cent caoital cost tarnet
Materials h,.w (kW m ’ C ‘) (kW mm* Cm’) 0) (S) (G)
cs-ss 0. I 0.1513 I.5130 3.544.687 3.633.422 -2
1.0 0.2500 0.2500 3.705.095 3,6X5:994 fl
10.0 I .2370 0.1237 3.898.125 3.574.547 f9
CS-TI 0.1 0.1513 I.5130 7.442,973 8.007.957 -7
1.0 0.2sOc 0.2500 8,079,235 8,183,646 -I
10.0 1.2370 0.1237 8.844.995 8.061.43 I +10
SIrearns 3. 5. 7. 8 are corrosive. Area tarect constant at 9022 m2. AT-._ = 20°C
 Capital cost targets for heat exchanger networks 329

fable 13. Utility cost data reductions do occur by exploiting significantly

Cost of hot utility = 120 S LW-’ yr-’ different (weighted) h-values, the slope of the capital
Cost of cold utility = IO S kW-’ yr-’
cost profile against AT,, is little affected and the
optimum value of AT,,,, is hardly changed compared
ratings on the shell-side and tube-side of a heat with the use of equation (20).
exchanger. The following example uses the stream data of
For networks comprising different exchanger Table 1 and cost data of Tables 2 and 13 to optimize
types, it is not possible to mix types in a single unit. a network requiring mixed materials of construction
Although a cost-weighting factor may be applied to ahead of design.
one stream in targeting, this assumes that different To initialize the network we must base capital cost
exchanger types can be mixed. In practice, such a targets on a single exchanger cost law. As a reference
match is forced to be a special-type exchanger. Thus case, first consider a network in which only carbon-
there may be some discrepancy between design and steel is required. This produces the total cost plot in
target costs. Greatest cost accuracy is expected using Fig. 5. Minimum total cost occurs at AT,,, cs = 9°C
the procedure if all special stream h-values are very and a corresponding network design purely in
much smaller than reference stream h-values. In this carbon-steel is given in Fig. 6. Continuous optimiza-
case special streams control network area and hence tion of this network for total annual cost (see the
cost. Appendix) is represented in Fig. 5. The network
In summary, the observed accuracy of the optimization profile is with respect to energy. The
proposed procedure for targeting the capital cost of AT,,,,, on the energy axis simply refers to the value on
networks is remarkable. It is representative of a wider the composite curves to give the corresponding
range of different networks which have been studied energy consumption as a target. The network
(Hall, 1986). optimization profile in Fig. 5 confirms the results for
uniform exchanger cost laws presented by Ahmad
(1985) and Ahmad and Linnhoff (1986), that a design
8. OPTIMIZATION PRIOR TO DESIGN
based on an optimized initialization is usually close
The capital/energy tradeoff for heat exchanger to the global cost-optimum design.
networks comprising a single material has already Consider now the case where streams 4, 5, 6 and 7
been discussed using Fig. 1. There is a cost-optimal require titanium and the remainder require carbon-
value of AT,,,, for design. Cost-weighting factors steel. Three total cost target profiles for networks in
should now enable the optimization ahead of design carbon-steel only, titanium only and a mixture of the
of networks requiring non-uniform exchanger two are compared in Fig. 7. The mixed materials
specifications. The stream h-values normally show profile is derived on the basis of cost weightings for
larger variations after their adjustment by cost- titanium relative to carbon-steel being applied to the
weighting factors and, as discussed earlier, these h-values of streams 4, 5, 6 and 7. The optimum
can be usefully exploited to reduce the weighted area initialization for a network of only titanium exchang-
A,& by using linear programming. The results ers is A To,, T, = 27°C and the corresponding network
of Ahmad et al. (1990), however, reveal that such design is shown in Fig. 8. The optimum initialization
refinements are often unnecessary towards obtaining for the network comprising mixed materials,
the optimum ATmi, location. While capital cost however, is AT,,,, cslT, = 18.2”C. The corresponding

($1&I Total Cost Target

4.1 -

1
2.9 - Gpt.C s
I
I jaqs I 20;95 I 23;85 i 26;75 EnergyW)
10 20 30 40 ATmin (‘C)
Fig. 5. Total cost target plot for a network of carbon-steel shell-and-tube heat exchangers
330 S. G. HALL et al.

h Exchanger
Stream CP’s
duties
- MWI’C
-MW
CP

so .O!i

-25 .30

.30 .29

-18 .02

.25 -26

Fig. 6. Pinch design for example stream data, AT,,,,, = 9°C.

design is shown in Fig. 9 comprising mixed worth 8% in total cost. Also shown in Fig. 10, a
materials. design initialized at AT,,,. TI in mixed materials
The total cost optimization paths for the mixed- optimizes to a total cost of $4.57 x lOh yr-‘, this
material networks designed at these three optima are being 4% higher than the best achievable from
all presented in Fig. 10. This shows that the design A To,, CS/TI
initialized at AT,,, cs but costed in mixed materials The cause of these cost differences between
changes significantly on optimization and obtains a optimized designs lies in the fact that networks based
minimum total cost of $4.74 x lo6 yr-‘. The design on different AT,,,,, initializations can have different
optimized from ATOP,CS,TI does not change structures. This feature of network optimization has
substantially during optimization. Its optimized total been observed and explained previously for networks
cost is $4.39 x 106yr-‘, within 1% of target. The constructed in a single material by Ahmad (1985) and
difference between these two optimized designs is Ahmad and Linnhoff (1986). Their explanation is

40 AT,;” (‘Cl
Fig. 7. Total cost target plots for networks of carbon-steel (CS), titanium (Tl) and carbon-steel-titanium
(CS/TI) shell-and-tube heat exchangers.
 Capital cost targets for heat exchanger networks 331

h Exchanger duties -MW 1350 cp

Stream CP’s -MWI’C
.50 Stream h-values -kWlm2’C .OS

.25 .30

30 .29

.02

.25 .26

.27 .I5

.I4

.lO

.05
l-6
I

Fig. 8. Pinch design for example stream data, Arm,,, = 27-C.

that some networks must experience a structural is not guaranteed. This example serves to emphasize
change during optimization if they are to obtain total the importance of correct initialization prior to design
cost optimahty in their capital/energy tradeoff. to locate the global optimum structure when the
Conventional continuous optimization (described in network comprises mixed materials of construction.
the Appendix) seeks to minimize total cost for a given Similar results would be expected when optimizing
network structure. The method cannot systematically networks for streams having different pressure
add and delete matches. Hence, global cost optimality ratings.

Exchanger duties -MW

h Stream CP’s - MWI’C CP

.05

25 .30

.29

.02

.26

Fig. 9. Pinch design for example stream data, AT,,, = 18.2 C

332 S. G. HALL et al.

10% r)Total Cost Target

25

CSlTI

,
aJpi.cs ATo;t.cS/TI AX.&tJ1

1520 I q45 I 20;95 I 23;85 I 26;75Energy (MW1

4.0;)
10 20 30 40 ATmin ('C)
Fig. 10. Total cost target plot and optimization paths for a network in carbon-steel and titanium.

Capital cost target predictions for mixed exchanger Fig. 12. If all network exchangers are assumed to be
types are less reliable. Consider optimization for spiral types, the minimum total cost target occurs at
networks of different exchanger types using the same AT,,,.,, = 28°C. This corresponds with the same
stream data and the cost data of Tables 9 and 13. structure as the design in Fig. 8. When cost-weighting
Consider first a reference case in which all factors are used to predict the capital component of
streams require only plate-and-frame exchangers. total cost more accurately for mixed-type networks,
The total cost target plot, assuming the network to be the resulting total cost target plot indicates the
made entirely in the reference plate exchangers is minimum cost to be at AT_,, .PF,SP= 25°C.
shown in Fig. Il. The minimum cost occurs at The total cost optimization paths of the designs of
AT,,, PF = I8.2”C. The corresponding design has the the mixed-type networks designed at the three optima
same structure as shown in Fig. 9. It is seen in are superimposed onto the mixed-type total cost
Fig. 11 that the cost of this design does not substan- target plot in Fig. 13. The mixed exchanger type
tially improve upon optimization. Again, this is in networks at AT,,, sP and AT,,, PF,SP have the same
lint with previous studies using only one exchanger structure and optimize to a similar minimum total
cost law as examined by Ahmad (1985) and Ahmad cost of $5.20 x 106yr-‘, being within 1% of target.
and LinnhofT (1986). However, the design in mixed-types at AT,,, pF has a
Now assume streams 3 and 7 require spiral minimum total cost of $5.79 x lo6 yr-‘, being 11%
exchangers and the remainder require a plate design. above the best achieveable. Again, the reason for this
The three total cost target plots for plate-only, discrepancy is that the network has a structure
spiral-only and plate-spiral networks are compared in different from that at the global optimum (Ahmad,

($106/y~l
Total Cost Target

I 2675 Energy (MW)

40 ATmin (‘Cl

Fig. I I. Total cost target plot for a network of plate heat exchangers.
 Capital cost targets for heat exchanger networks 333

0 IO 20 30 40 AT,,,;"
('Cl
Fig. 12. Total cost target plot for networks of plate-and-frame (PF) exchangers, spiral (SP) exchangers
and a mixture of these two (PF/SP).

1985; Ahmad and Linnhoff, 1986; Ahmad et aI., first suggested by Ahmad (1983, to networks
1990). Also, the design initialized at AT,,, sP did comprising mixed materials of construction, pressure
actually find the global optimum solution because it ratings and equipment types. It has been shown that
had the correct structure. capital cost targets for networks of exchangers in
It must be emphasized again that although cost- mixed materials of construction and with mixed
weighting improves the accuracy of cost targets for pressure ratings can be predicted with good accuracy
networks of mixed exchanger type, it is not as reliable providing heat transfer coefficients do not vary by
as the cases for mixed materials of construction and orders of magnitude in the matches made. Indeed,
pressure ratings. even for such extreme cases, the approach can be
The examples of total cost optimization prior to incorporated into a more accurate area targeting
design given here illustrate the importance of the method based on linear programming such as that
correct initialization of networks comprising mixed suggested by Saboo et al. (1986).
exchanger specifications if approach to the global The approach may be further extended, but less
optimum design is to be achieved. reliably, to obtain capital cost targets for networks
comprising different exchanger types. Application
9. CONCLUSIONS
of cost-weighting factors to such cases has been
found to improve the accuracy of capital cost targets
This paper has applied a method of dealing with and hence AT,,, initialization over conventional
mixed-exchanger specifications in capital cost targets, methods.

(.$l&&) Total Cos

5.6-

5.4-

5.2-
I
1520 18;45 ;,20;95
A~piPFlSb,3~,, 1 2975 Energy (MW)
I I 1 1,
0 10 20 30 40 AT,,,;"
('Cl
Fig. 13. Total cost target plot and optimization paths for a network of plate and spiral heat exchangers.
334 S. G. HALL et al.

Variations in exchanger specification can consider- APPENDIX

ably alter the capital/energy tradeoff. As a result, the
Continuous Optimization of’ a Gicen Heat Exchanger
structure of the global cost optimum network can Network Structure
change. Without sufficiently accurate initialization of
The continuous optimization of heat exchanger networks
the design, finding the global optimum network can be based on redistribution of the exchanger duties.
structure cannot be guaranteed. Given a network structure, it is possible to identify for it
In conclusion, this paper has applied a simple “loops” and “paths” (Linnhoff and Hindmarsh, 1983) such
modification to the existing capital and total cost as those shown in Figs Al and A2. Duty can be “shifted”
along loops (for example, by adjusting exchanger duties by
targeting procedures. This modification considerably
amount W or X in Figure Al) to affect only the network
increases the accuracy of the targets when applied to area and therefore capital cost. It is seen that loops can exist
networks having non-uniform exchanger specifica- over utility exchangers, and their duties can be adjusted
tions. In practice, this should improve the scope of without affecting the overall energy consumption of the
application of such targets. network (for example, the two coolers in Fig. Al are
connected via a single cold utility stream). Shifting duty
along paths (by adjusting exchanger duties by Y or Z in
REFERENCES Fig. A2) affects both the energy consumption and area of
the network. Furthermore, stream-splits may exist in the
Ahmad S., Heal exchanger networks: cost tradeoffs in energy design, and variation of their branch flowrates can be
and capital. Ph.D. Thesis, Univeristy ofManchester (1985). superimposed on the use of loops and paths.
Ahmad S. and B. Linnhoff, Overall cost targets for heat Thus, loops, paths and stream-splits offer the degrees of
exchanger networks. IChemE I Ifh Annual Research freedom for manipulating the network cost. The problem is
Meeting on Heat Transfer, Bath (1984). one of multivariable non-linear continuous optimization.
Ahmad S. and B. Linnhoff, Supertarget: optimization of a Suitable implementations for this task are described by
chemical solvents plantAifferent process structures for Abjarnson and Haug (1985) and Ptacnik and Klemes
different economics. ASME Winter Meetzng. AES, Vol. 1, (1988). The constraints are only those of feasible heat
pp. 15-20. Anaheim, California (1986). transfer: positive temperature difference and non-negative
Ahmad S., B. Linnhoff and R. Smith, Cost optimum heat heat duty for each exchanger. Furthermore, if stream-splits
exchanger networks- Part 2. Targets and design for exist then positlvc branch flowrates are additional
detailed capital cost models. Computers them. Engng, constraints. The apparent constraints of obtaining the
submitted for publication (1990). required target temperature from the specified supply
Abjsrnson 0. A. and M. Haug, Choice of variables and temperature, or if the stream is a constant temperature
approximations in the optimization of heat exchanger phase change, then its required enthaipy need not be added
networks. PSE ‘85, IChemE Synp. Ser. No. 92, as constraints. This is because the initial network should
pp. 197-208 (1985). satisfy these requirements and the use of loops and paths
Hall S. G., Capital cost targets for heat exchanger networks: maintains the heal change of each stream.
differing materials of construction and different heat If the network is optimized at fixed energy consumption
exchanger types. M.Sc. Dissertation. University of then only loops and stream-splits are used. The objective
Manchester (1986). function can be to minimize overall network area or, as in
Kumana J. D., Cost update on speciality heat exchangers. Sections 47 of this paper, the overall capital cost. The
Chem. Engng 91, 169-172 (1984).
LinnhoR B. el al.. User Guide on Process Integratiotzfor the
Eficient Use of Energy. Institution of Chemical
Engineers, Rugby, U.K. (1982).
Linnhoff B. and S. Ahmad, SUPERTARGETING-
optimum synthesis of energy management systems.
ASME Winter Meetings, AES, Vol. 1. pp. I-14. Anaheim,
California (I 986).
Linnhoff B. and S. Ahmad. Cost optimum heat exchanger
networks-Part I. Minimum energy and capital using
simple models for capital cost. Computers them. Engng,
submitted for publication (1990).
Linnhoff B. and E. Hindmarsh, The pinch design method for
heat exchanger methods. Chem. Engng 38,745-763 (1983). -X
Linnhotf B. and D. R. Vredeveld, Pinch technology has
Fig. A I. Heat ( W, X) can be shifted around loops in a
come of age. C/tern Engng Prog. 80, 3340 (1984).
network.
Ptacnik R. and J. Klemes, An application of mathematical
optimization methods in heat-exchange network synthe-
sis. Cornpuwrschrm. Engng 12, 231-235 (1988).
Purohit G. P., Estimating costs of shell and tube heat
exchangers. Chem. Engng 90, 5647 (1983).
Rippin D. W. T.. Private Communication (1989).
Saboo A. K.. M. Morari and R. D. Colberg, RESHEX-an
interactive software package for the synthesis and analysis
of resilient heat exchanger networks. Computers them.
Engng 10, 591-599 (1986).
Smith R. and B. Linnhoff, The design of separators in the
contex of overall processes. Chew. Engng Res. Des. M$
195-228 (1988).
Townsend D. W. and B. Linnhoff. Surface area targets for
heat exchanger networks. IChemE I lth Annuul Re.warch Fig. A2. Heat (Y, Z) can be shifted along paths in a
Meeting on Hew Transfir. Bath (I 984). network.
 Capital cost targets for heat exchanger networks 335

network in Fig. 4 is shown before and after such an fixed network topology. The profiles for networks in these
optimization for carbon-steel exchangers throughout and figures are with respect to energy. The AT,,,,, on the energy
subject to the cost data in Table 2. axis simply refers to the composite curves setting which
When energy consumption is allowed to vary (Section 8) would give the corresponding energy consumption as a
paths must also be included. As the network energy target.
consumption increases, the overall capital cost decreases. It is possible that after optimization, whether for capital
The objective function now becomes minimum total annual or total cost, some exchanger duties become zero, in which
cost. The network optimizations in Figs 5, 10, 11 and 13 case the network structure is obviously interpreted to
refer to such total cost minimization subject to the exclude such exchangers.