You are on page 1of 10

Ind. Eng. Chem. Res.

2003, 42, 4009-4018 4009

PROCESS DESIGN AND CONTROL

Mathematical Programming Model for Heat-Exchanger Network


Synthesis Including Detailed Heat-Exchanger Designs. 1.
Shell-and-Tube Heat-Exchanger Design
Fabio T. Mizutani,† Fernando L. P. Pessoa,† and Eduardo M. Queiroz†
Departamento de Engenharia Quı́mica, Escola de Quı́mica, Universidade Federal do Rio de Janeiro,
Caixa Postal 68542, Rio de Janeiro, RJ, 21949-900 Brazil

Steinar Hauan and Ignacio E. Grossmann*


Department of Chemical Engineering, Carnegie Mellon University, Pittsburgh, Pennsylvania 15213

This paper addresses the optimal design of shell-and-tube heat exchangers via a mathematical
programming approach. It is shown that it is possible to develop a design model for shell-and-
tube heat exchangers that takes into account some important construction variables: number
of tubes, number of passes, internal and external tube diameters, tube arrangement pattern,
number of baffles, head type, and fluid allocation (i.e., the allocation of the fluid streams to the
shell or tubes). The model is based on generalized disjunctive programming and is optimized
with a mixed-integer nonlinear programming reformulation to determine the heat-exchanger
design that minimizes the total annual cost accounting for area and pumping expenses. Examples
are presented to illustrate the model performance.

Introduction for detailed designs based on successive simulations. As


Polley and Panjeh Shahi4 have noticed, despite being
The transfer of heat to and from process fluids is an more accurate, the Bell-Delaware correlations are more
essential part of most chemical processes. The most complex and do not have a form that would permit their
commonly used type of heat-transfer equipment is the use in place of Kern’s correlations in the Polley et al.5
ubiquitous shell-and-tube heat exchanger.1 calculation strategy. Polley and Panjeh Shahi4 assumed
However, although heat-exchanger network synthesis the existence of a simple relationship between the
has been the subject of a significant research over the pressure drop and heat-transfer coefficient in exchang-
last 40 years,2 most formulations assume constant heat- ers that exhibit geometrical similarity in order to
transfer coefficients for all stream matches, which can develop a correlation similar to those used by Polley et
lead to nonoptimal results as these coefficients vary with
al.5 for the shell-side flow. The parameters of this
design.
correlation were estimated by rating heat exchangers
In the past decade, a set of papers focused on heat-
having this similarity and using the Bell-Delaware
exchanger design during the network synthesis.3-5 In
correlations.
the earlier paper,5 the authors developed a relationship
between the exchanger pressure drop, surface area, and Saffar-Avval and Damangir8 developed correlations
heat-transfer coefficient, based on the well-known Dit- for determining optimum baffle spacing for all types of
tus and Boelter correlation for the tube-side flow and shell-and-tube exchangers based on the design method
on the Kern6 correlations for the shell-side flow. These developed by Taborek,9 which is basically an updated
relationships make possible a direct calculation of the version of the Bell-Delaware method.
main exchanger parameters after setting of the tube Conventional shell-and-tube heat-exchanger design
diameter, the number of tube passes, and the tube algorithms proceed by systematically rating a large
arrangement. Thus, the traditional successive simula- number of possible exchanger geometries and identify-
tions method for heat-exchanger design are avoided. The ing those which successfully transfer the required heat
use of Kern’s correlations is the weakest assumption of load. Poddar and Polley10 presented a graphical tech-
this method used by Polley et al.5 and other authors. nique that identifies the space within which successful
Other available relationships for the shell-side flow designs are found and, therefore, reduces the large
are those presented by the Bell-Dellaware method as amount of computation effort. Maralikrishna and
described in Kakaç et al.7 These correlations are suited Shenoy11 extended this graphical technique and intro-
duced targets for minimum area and cost designs. Both
* To whom correspondence should be addressed. Tel.: 1-412- methods proposed by Poddar and Polley10 and Mara-
2683642. Fax: 1-412-2687139. E-mail: grossmann@cmu.edu. likrishna and Shenoy11 were based on Kern’s correla-

Fax: 55-21-25627425. tions.6
10.1021/ie020964u CCC: $25.00 © 2003 American Chemical Society
Published on Web 07/17/2003
4010 Ind. Eng. Chem. Res., Vol. 42, No. 17, 2003

Figure 1. Shell-and-tube heat exchanger.

Chaudhuri and Diwekar12 presented a strategy that


uses simulated annealing for the optimal design of heat
exchangers. The algorithm is supported by the HTRI
design program, which is a computer software marketed
by companies. Other examples of these kinds of pro-
grams are HTFS and B-JAC. The authors used the
simulated annealing as an optimization technique be-
cause the HTRI design program is a black-box model;
hence, explicit relationships for the constraints are not
available.
This work presents an optimization procedure in Figure 2. Tube pattern arrangements.
order to design individual heat-exchanger units. The
and-tube heat-exchanger kind of equipment with the
optimization model is developed for shell-and-tube units
considerations of nonphase changes and constant physi-
and uses Bell-Delaware correlations to calculate the
cal properties. Figure 1 illustrates a shell-and-tube heat
heat-transfer coefficient and the pressure drop in the
exchanger and Figure 2 two different tube pattern
shell-side flow.
arrangement design options.
It is shown that it is possible to develop a design
In this work, the optimal design of heat exchangers
model for shell-and-tube exchangers that takes into
is viewed as a tradeoff between area and pumping costs.
account different construction alternatives, such as the
High turbulence inside the shell and tubes results in
number of tubes, number of tube passes, tube internal
higher heat-transfer coefficients, which results in a
and external diameters, tube arrangement pattern,
lower exchange area at the expense of a higher ∆P and
number of baffles, head type, and fluid flow allocation
higher pumping cost. Basically, the model consists of
(i.e., the allocation of the fluid streams to the shell or
determining a unit geometry that enables the exchange
tubes). The model is based on generalized disjunctive
of a specified heat load. Hence, it is necessary to
programming and is optimized with a mixed-integer
calculate the heat-transfer coefficients for each stream.
nonlinear programming (MINLP) reformulation to de-
Thus, the problem includes the following decisions:
termine the heat-exchanger design that minimizes the
tube diameters, tube length, tube pattern arrangement,
total annual cost accounting for area and pumping
number of tube passes, number of tubes, head type,
expenses. Therefore, the tradeoff between investment
number of baffles, and fluid allocation. Fluid allocation
and operational costs, basically area and pumping
is the decision of which fluid should be allocated in the
expenses, is considered during the unit design.
tube side and which fluid should be allocated in the shell
Problem Statement side. These design decisions are represented by Boolean
Given are hot and cold streams for heat exchange with variables.
their specified supply and target temperatures, as well Although internal and external diameters of the tubes
as their fluid flow rates. Also given for each stream are can take continuous values, they are restricted to
its physical properties: viscosity, density, thermal certain commercial standards and, consequently, in this
conductivity, and thermal capacity. The problem then work, they are related to integer variables. On the other
consists of determining the optimal shell-and-tube heat- hand, the number of tubes takes discrete values, but it
exchanger design options: number of tubes, number of is considered as a continuous variable because this
tube passes, internal and external tube diameters, tube number is commonly large.
arrangement pattern, number of baffles, head type, and Essentially, the model is based on the following
fluid allocation. The objective is to minimize the total assumptions:
annualized investment and operating costs. (i) Bell-Delaware correlations, as described in
Taborek,9 are used for the shell-side Fanning factor and
Heat-Exchanger Design Model heat-transfer coefficient.
Problem Formulation. In this section, we consider (ii) The turbulent Sieder-Tate equation is used for
the heat exchange between two fluids using the shell- the tube-side heat-transfer coefficient.
Ind. Eng. Chem. Res., Vol. 42, No. 17, 2003 4011

(iii) The Blasius equation is used for the tube-side mh ) mh1 + mh2 (8)
Fanning factor.
c
m ) mc1 + mc2 (9)
(iv) A single-phase flow regime is used for both shell
and tube fluids. mt ) mh1 + mc1 (10)
The Bell-Delaware method has different correlations
ms ) mh2 + mc2 (11)
to the shell-side heat-transfer coefficient and the pres-
sure drop depending on the Reynolds number range. mh1 e mupperyf1 (12)
This feature gives a particular characteristic during the
modeling that is to implement these correlations through mc1 e mupperyf2 (13)
disjunctions.
mh2 e mupperyf2 (14)
Because the heat-exchanger model depends on design
decisions and, on the other hand, the shell-side Reynolds mc2 e mupperyf1 (15)
number imposes which Bell correlation should be used, t
the following generalized representation form can be µ ) yft µh + yfs µc (16)
applied:13
µs ) yf2 µh + yf1 µc (17)
Cpt ) yf1Cph + yf2Cpc (18)
min f(x) s
Cp ) yf2Cph + yf1Cpc

[ [ ]]
(19)
g(x) e 0
h(x) ) 0 kt ) yf1kh + yf2kc (20)

Ydesign
i
ks ) yf2kh + yf1kc (21)
gdesign
i (x) e 0 F ) t
yf1Fh + yf2Fc (22)
hdesign (x) ) 0

i
Fs ) yf2Fh + yf1Fc (23)
i∈D YReo
j
Reo
∨ gj (x) e 0 yf1 + yf2 ) 1 (24)
j∈Reo Reo
hj (x) ) 0
where m ∈ TP ) {1, 2, 4, 6, 8}, yntp is the set of binary
Ay e a variables that decides the number of tube passes, and
x ∈ X, y ∈ {0, 1}, Y ∈ {true, false} (PHx) Ntp is a continuous variable that represents the number
of tube passes; n ∈ NB ) {7, 8, 9, ..., 25}, ynb is the set
of binary variables that decides the number of baffles,
and Nb is the number of baffles; o ∈ TD ) {1, 2, 3, 4,
The first and second sets of constraints represent 5}, ytema is the set of binary variables that decides the
global inequalities and equalities that hold irrespective tube diameters, dtout and dtin are the external and
of the discrete choices. Ydesign and YReo are Boolean internal diameters, respectively; Ain and Aout are pa-
variables that correspond to a given unit geometry and rameters with the Tubular Exchanger Manufactures
shell-side Reynolds number regime, respectively. Fi- Association (TEMA14) tube diameters.
nally, the last linear inequality represents the logic To calculate pn (tube pitch normal to flow) and pp
relations for the Boolean variables Y in terms of the 0-1 (tube pitch parallel to flow), the following disjunction
variables y. is used:

[ ]
The following equations represent the number of
tubes passes, number of baffles, tube diameters, fluid arrangement ) triangular
allocation, and physical properties: pn ) 0.5pt ∨
pp ) 0.866pt


m
myntp
m ) N


tp

yntp
m ) 1
(1)

(2)
pn ) pt
pp ) pt [
arrangement ) square
]
m
where pt is the tube pitch. This disjunction can be
∑n nynbn ) Nb (3) described by the following set of equations:

∑n ynbn ) 1 (4) pn1 ) 0.5pt1 (25)


pn2 ) pt2 (26)
∑o ytema
o Ain t
o ) din (5)
pp1 ) 0.866pt1 (27)
∑o ytema
o Aout
o ) dtout (6) pp2 ) pt2 (28)
pp ) pp1 + pp2 (29)
∑o ytema
o )1 (7)
pp1 e ppupperyarr
tr (30)
4012 Ind. Eng. Chem. Res., Vol. 42, No. 17, 2003

pp2 e ppupperyarr (31) parameters Ahead ) {1.007, 1.025, 1.0, 1.010} and Bhead
sq
) {0.087, 0.044, 0.039, 0.008} are empirical coefficients.
pt ) pt1 + pt2 (32) The bundle (Dotl) is calculated by the Sinnott empiri-
cal equation:1
pt1 e ptupperyarr
tr (33)
pt2 e ptupperyarr
sq (34) Dotl ) dtout(Nt/k1)(1/n1) (46)
pn ) pn1 + pn2 (35)
where Nt is the number of tubes and dtout is the tube
pn1 e pnupperyarr (36) external diameter. Because the variables k1 and n1
sq
depend on the tube pass number as well as on the tube
pn2 e pnupperyarr
tr (37) pattern arrangement, it is necessary to introduce a
binary set of variables yarrntp, which represents both the
yarr
tr + yarr
sq )1 (38) number of tube passes and pattern arrangement. This
set of binary variables is related to the sets yntp and yarr
where pn1, pn2, pt1, pt2, pp1, and pp2 are disaggregated with the following logic proposition:
variables from the Convex Hull representation.15
Sm is the reference normal area for shell-side flow, (yntp arr arrntp
m ∧ yn ) w ym,n
which is equal to the normal free area between baffles
at the center line in the shell cross-flow and is repre-
or in the conjunctive normal form16
sented by the following disjunction:

[ )]
arrangement ) triangular ¬yntp arr arrntp
m ∨ ¬ yn ∨ ym,n

(
Sm ) ls Ds - Dotl +
(pt - dtout)(Dotl - dtout) ∨
that is expressed by the constraints

[ )]
pt
arrangement ) square 1 - yntp arr arrntp
m - yn + ym,n

(
g0 (47)
(pt - dtout)(Dotl - dtout)
Sm ) ls Ds - Dotl +
pn ∑∑
m n
arrntp
ym,n )1 (48)

where Ds is the inside shell diameter and Dotl is the


shell outer tube limit. This disjunction can be described Therefore, the variables k1 and n1 are calculated by
by the subsequent equations:
∑ ∑ arrntp k
Am,n ) k1
( )
ym,n 1
(49)
(pt - dtout)(Dotl - dtout) m n
Sm ) ls Ds - Dotl + (39)
aux
∑ ∑ arrntp n
ym,n Am,n ) n1
1
(50)
aux ) pt1 + pn1 (40) m n

k1 n1
To calculate the inside shell diameter, the following where the parameters Am,n and Am,n are empirical
disjunction, based on typical values (TEMA14), is used: coefficients presented in the appendix.
ji (Colburn factor) and f si (shell-side Fanning factor)

[
head kind ) pull - through
Ds ) 1.007Dotl + 0.087
∨ ] are given by the following empirical correlations:

[ ]
head kind ) slit - ring ji ) a1 × 1.064a(Res)a2 (51)

Ds ) 1.025Dotl + 0.044

[ ]
a3
head kind ) outside packed a) (52)
∨ 1 + 0.14(Res)a4
Ds ) Dotl + 0.039

[
head kind ) Fixed or U tube
Ds ) 1.010Dotl + 0.008 ] f si ) b1 × 1.064b(Res)b2 (53)

represented by the subsequent equations:


b3
b) (54)
1 + 0.14(Res)b4
D ) s
∑q Dq (41)
where the coefficients a1, a2, a3, b1, b2, and b3 assume
Dotl ) ∑q Dotlq (42) different values depending on the Reynolds number in
the shell side as well as the tube pattern arrangement.
These values are presented in the appendix.
Dq ) Ahead
q Dotlq + Bhead
q yhead
q (43) It is important to point out that, in eqs 51 and 53,
the constant 1.064 comes from the original Bell-
Dotlq e Dotlupperyhead
q (44) Delaware equation part [1.33/(dtout/pt)]a and the as-
∑q yhead
q )1 (45) sumption that pt is equal to 1.25 times dtout. This
assumption can be relaxed in a more detailed procedure.
Consequently, it is necessary to introduce a disjunc-
where q ∈ TH ) {pull-through, split-ring, outside tion, which represents the shell-side Reynolds number
packed, U tube} represents the type of head set. The range, as
Ind. Eng. Chem. Res., Vol. 42, No. 17, 2003 4013

[ ] [ ][ ]
yres yres ∑n ynarr Aan ) a4 (61)
4

yres
1
2 3
∨ Res e 104 ∨ Res e 103 ∨
Res > 104
∑l ∑n yl,n
rearr b
Al,n ) b1

[ ][
Res > 103 Res > 102
1
(62)

]
yres
4
Res e 102 ∨
yres
5 ∑l ∑n yl,n
rearr b
Al,n ) b2
2
(63)
Res e 10
Res > 10
∑n yarr
n An
b 3
) b3 (64)

represented by the subsequent set of equations


∑n yarr
n An
b 4
) b4 (65)

Res ) ∑l Resl
where the parameters Aal,n and Abl,n are empirical coef-
∑l yres
l ) 1
ficients presented in the appendix.
The following model equations hold irrespective of the
discrete choices.
Resl g 0
Res1 g 104yres
1
Tube-side Fanning factor:

Res1 e 106yres fit ) 0.079/(Ret)0.25 (66)


1
Total number of tubes:
Res2 g 103yres
2 Q 1
Nt ) (67)
Res2 e 104yres U × Ft × LMTD πdt Lt
2 out

Res3 g 102yres Tube pitch:


3
pt ) 1.25dtout (68)
Res3 e 103yres
3
Baffle spacing:
Res4 g 10yres
4
Lt
Res4 102yres ls ) (69)
e 4 Nb + 1
Res5 e 10yres (55) Tube-side Prandtl number:
5
µtCpt
Prt ) (70)
Like the variables k1 and n1, it is necessary to introduce kt
a set of binary variables yrearr, which represents both Tube-side Nusselt number (turbulent Sieder-
the Reynolds number regime and pattern arrangement.
Tate equation):
This set of binary variables is related to the sets yres
and yarr by the following logic proposition: Nut ) 0.027(Ret)0.8(Prt)1/3 (71)
Tube-side heat-transfer coefficient:
(yres arr rearr
l ∧ yn ) w yl,n ht ) Nutkt/dtin (72)
Number of tube rows crossed in one cross-flow
or in the conjunctive normal form:16 section:
Nc ) 0.5Ds/pp (73)
¬yres arr rearr
l ∨ ¬ yn ∨ yl,n Fraction of total tubes in cross-flow:
1
that is expressed by the constraints Fc ) {π + 2λ sin[arccos(λ) - 2 arccos(λ)]} (74)
π
λ ) 0.5Ds/Dotl (75)
1 - yres arr
l - yl
rearr
+ yl,n g0 (56)
Number of effective cross-flows in each window:
∑l ∑n yl,n
rearr
)1 (57) Ncw ) 0.2Ds/pp (76)
Fraction of cross-flow area available for bypass
flow:
Thus, the coefficients a and b can be calculated by
ls[(Ds)2 - Dotl + 0.5np × wp]
Fsbp ) (77)
∑l ∑n yl,n
rearr a
Al,n ) a11
(58) Sm
Shell-to-baffle leakage area:
∑l ∑n yl,n
rearr a
Al,n ) a22
(59) Ssb ) 0.667πDsδsb/2 (78)
Tube-to-baffle leakage area for one baffle:
∑n yarr
n An
a 3
) a3 (60)
Stb ) 0.25πdtoutδtbNt(1 + Fc) (79)
4014 Ind. Eng. Chem. Res., Vol. 42, No. 17, 2003

Area flow through the window: Rb pressure drop correction factor for
1.038(D ) s 2
πdtoutNt(1 - Fc) bundle-bypassing effects:
Sw ) - (80) Rb ) exp(-1.3456Fsbp) (94)
4 8
Shell-side Reynolds number: Shell pressure drop:
s
Re ) msdtout/µtSm (81) Ncw
Shell-side heat-transfer coefficient for an ideal
(
∆Ps ) 2∆Pbi 1 +
Nc b )
R + (Nb + 1)∆PbiRbRl +
Nb∆PwiRl (95)
tube bank:
ji × Cptmt kt 2/3
hoi ) ( ) Pumping cost:

( )
(82)
Sm Cptµt ∆Ptmt ∆Psms
Pcost ) ccost + (96)
Correction factor for baffle configuration effects: Ft Fs
Jc ) Fc + 0.54(1 - Fc)0.345 (83) An important design recommendation by TEMA14 for
Correction factor for baffle-leakage effects: shell-and-tube heat exchangers is that baffles spacing
Ssb + Stb ls should be at least 20% of the shell diameter, which
Jl ) R + (1 - R) exp -2.2
Sm ( ) can be represented by the following constraint included
in the model:
Ssb
R ) 0.44 1 -
Stb + Ssb ( (84) ) ls g 0.2Ds (97)
Correction factor for bundle-bypassing effects:
Finally, the objective function determines the total
Jb ) exp(-0.3833Fsbp) (85) annual cost for the design heat exchanger by summing
Shell-side heat-transfer coefficient: heat-exchanger area and pumping expenses:
hs ) hoi × Jc × Jl × Jb (86) min annual cost ) acost(Q/UFt∆Tln)0.6 + Pcost
Overall heat-transfer coefficient: (98)
1

( )
U) Constraints (1)-(97), along with the objective function
dtout rindtout dtout log(dtout/dtin) 1 in eq 98, define the generalized MINLP model for the
+ + + rout + s optimal shell-and-tube heat-exchanger design.
t t
(h d ) d t 2k h Remarks. It is important to point out that other
in in
(87) bounds, such as maximum baffle spacing, tube- and
Tube-side Reynolds number: shell-side pressure drops, tube-side velocity, shell-side
velocity, and maximum shell diameter among others,
4mtNtp can be included in the model. Additionally, recom-
Ret ) (88) mended limits in correction factors such as the correc-
πdtin µtNt
tion factor for baffle-leakage effects Jl can be also
Tube-side velocity: included.
It is important to note that the turbulent regime is
Retµt
vt ) (89) imposed as an operational condition through the tube
Ftdtin side; hence, it’s Reynolds number lower bound is used
as 4000. Additionally, some assumptions were intro-
Tube-side pressure drop including ∆P in the duced in order to simplify the use of the original Bell-
unit heads:

[ ]
Delaware model as follows:
2fitNtpLt(vt)2 1. pt is equal to 1.25 times dtout, which influences
∆Pt ) Ft + 1.25Ntp(vt)2 (90) constraints in (68), (51), and (53).
dtin 2. There is a 25% baffle cut, which influences con-
Shell-side pressure drop for an ideal tube bank: straints (73) and (78).
3. The number of sealing strips is equal to Nc/6, which
2fis × Nc(mt)2 influences constraints (85) and (94).
∆Pbi ) (91)
FtSm2 4. The correction factors for low-shell-side Reynolds
number effects and for nonuniform baffle spacing are
assumed to be unity.
Pressure drop in the window: 5. No correction is considered for the fluid viscosity
(mt)2 temperature dependence.
∆Pwi ) (2 + 0.6Ncw) (92) 6. Only the possibilities of one shell pass and one, two,
2Sm × SwFt four, six, and eight tube passes are considered.
7. The number of tubes is treated as a continuous
Rl pressure drop correction factor for variable, which can lead to some error when this
baffle-leakage effects: number is small.
Ssb Stb + Ssb κ According to Polley and Panjeh Shahi,4 the two
[
Rl ) exp -1.33 1 + (
Stb + Ssb Sm )( )] engineering activities, flowsheet synthesis, and detailed
exchanger design are usually conducted by different
Ssb
κ ) -0.15 1 +
Stb + Ssb ( + 0.8 (93) ) engineers, working at different times for different
managers and sometimes even in different geographical
Ind. Eng. Chem. Res., Vol. 42, No. 17, 2003 4015

Table 1. Example Results


example 1 example 2 example 3
area (m2) 202 227 217
U (w/m2‚K) 860 857 803
tout
c (K)
316 319
Mc (kg/s) 58 46
Tout
h (K)
316
no. of tubes 832 777 746
tubes arrangement square square triangular
no. of passes 2 4 4
dtin (mm) 12.6 14.8 14.8
dtout (mm) 15.9 19.1 19.1
no. of baffles 8 7 7
head kind fixed or U tube outside packed head fixed or U tube
hot fluid allocation shell tubes shell
LMTD 24.9 22.3 23.3
Ft temperature correction factor 0.812 0.75 0.75
shell diameter (m) 0.687 0.854 0.754
tube length (m) 4.88 4.88 4.88
baffle spacing (m) 0.542 0.610 0.610
Dotl (m) 0.6723 0.815 0.739
tube-side ∆P (Pa) 22 676 18 335 42 955
shell-side ∆P (Pa) 7494 7719 5814
tube-side coefficient (w/m2‚K) 6480 2632 6577
shell-side coefficient (w/m2‚K) 1829 4110 1627
Jl 0.837 0.869 0.809
cold fluid cost ($/year) 14 980 11 409
auxiliary heat load cost ($/year) 3960
pumping cost ($/year) 2424 1638 2868
area cost ($/year) 2826 3023 2943

locations. The proposed model takes into account some


aspects for shell-and-tube heat-exchanger designs. Be-
sides the fact that simplifications assumed in the Bell-
Delaware model described above and its own original
errors related to experimental data need special atten-
tion by the designer, additional detailed design issues
such as vibration problems may be included in order to
extend the proposed model. However, the present work
aims at introducing some industrial reality during the
flowsheet synthesis and, therefore, bridging parts of the
gap between HENS and heat-exchanger designs while, Figure 3. Stream data for example 1.
at the same time, using mathematical programming
Table 2. Summary of Solver Results
tools in a simultaneous synthesis and design framework.
It should be noted that heat balance constraints have example 1 example 2 example 3
not been included for both exchanging fluids. This is due equations 132 155 156
to the fact that it is assumed that supply and target continuous variables 155 170 172
temperatures, as well as fluid flow rates, are specified; discrete variables 64 67 67
therefore, the heat to be exchanged has a fixed value. OA iterations 4 32 17
CPU time (s)a 0.8 14.0 5.6
Another important point is that the LMTD, logarithm NLP solver CONOPTT2 CONOPT2 CONOPT2
mean temperature difference, can be calculated before MILP solver CPLEX CPLEX CPLEX
the optimization, as well as the temperature correction a Pentium III 1 GHz.
factor Ft. At last, new constraints that include heat
balances can be easily included in the model, as will be
shown in examples 2 and 3. under fixed mass flow rates and target and supply
stream temperatures.
Examples The summary of the results with the proposed model
is present in Tables 1 and 2. As seen from the results,
Example 1. The first example consists of using the the proposed optimization model finds the optimal
model described in the previous section to design a shell- solution in four major iterations with DICOPT++.17
and-tube heat exchanger. Consider cold and hot fluids Because of the nonconvexities in the model, the heuristic
A and B with the properties presented in Figure 3. termination of DICOPT++, which stops after no im-
Because all temperatures, as well as the mass flow provement, had to be used.
rates, are specified, the heat load is a fixed parameter. Example 2. This example consists of designing a
Consider also the tube thermal conductivity as 50 shell-and-tube heat exchanger for the same two fluids
W/m‚K and the tube and shell side deposit resistance as those used in example 1. Also assuming the same
as 1.7 × 10-4 W/m‚K. Assuming the shell-tube and fluid properties data of example 1, as well as the fouling
shell-baffle clearances are constant and equal to 2.5 factor, clearances, and tube length, it is assumed that
and 8 mm, respectively, and the tube length is set as the cold fluid target temperature and its mass flow rate
4.88 m, it is desired to design a heat exchanger that are both variables. Also, it is considered that the
satisfies the heat balances for the cold and hot streams refrigerant has as cost of $7.93/1000 tons, and this cost
4016 Ind. Eng. Chem. Res., Vol. 42, No. 17, 2003

is added to the objective function (98). Figure 4 il-


lustrates the proposed problem.
Because the cold fluid target temperature is treated
as a variable in this example, it is necessary to introduce
into the model the LMTD equation; in this work, the
Chen18 (1987) approximation is used:

LMTD ) [t1t2(t1 + t2)/2]1/3 (99)

t1 ) thin - tcout (100)


Figure 4. Stream data for example 2.
t2 ) thout - tcin (101)

For the same reason, the energy balance equation to


the cold fluid is necessary:

Q ) mcCpc(tcin - tcout) (102)

Additionally, it is necessary to determine the Ft


temperature correction factor, which is given by the
following equations:

thin - thout
R) (103)
tcout - tcin

tcout - tcin
S) (104) Figure 5. Stream data for example 3.
thin - tcin
Table 3. Sinnott Empirical Equation Parameters for
Equations 47 and 48
xR2 + 1 ln (11--RSS ) K1 n1

[ ]
Ft ) f1(R,S) ) tube passes triangular square triangular square
2 - S(R + 1 - xR2 + 1)
(R - 1) ln 1 0.319 0.215 2.142 2.207
2 - S(R + 1 + xR2 + 1) 2
4
0.249
0.175
0.156
0.158
2.207
2.285
2.291
2.263
for R * 1 (105) 6 0.0743 0.0402 2.499 2.617
8 0.0365 0.0331 2.675 2.643
It is important to point out that the Ft function has a
discontinuity when R is equal to 1 and its limit is given Example 3. This example consists of designing a
by shell-and-tube heat exchanger for the same two fluids
as those used in example 1. Also, assuming the same
Sx2

[ ]
Ft ) f2(S) ) for R ) 1 fluid properties data of example 1, as well as the fouling
2 - S(2 - x2) factor, clearances, and tube length, like in example 2,
1 - S ln the cold fluid target temperature and its mass flow rate
2 - S(2 + x2) are both variables and it is considered that the refriger-
(106)
ant has as cost of 7.93$/1000 ton, and this cost is added
To introduce the temperature correction factor Ft to the objective function. In this example, the hot fluid
calculation into the model and, simultaneously, to avoid target temperature is also considered as a variable, so
problems with the discontinuity, it is necessary to define the heat load may vary but with a cost of $20/kW‚yr to
two auxiliary variables R1 and R2 with their limits: 0.1 the hot fluid energy not exchanged in the designed unit
e R1 e 0.99 and 1.01 e R2 e 20. At last, use the in order to achieve a total heat load of 4339 kJ as in
following disjunction: examples 1 and 2. Figure 5 illustrates the proposed

[ ][ ][ ]
problem.
R e 0.99 R g 0.99 R g 1.01 Like in example 2, the logarithm mean temperature
R1 ) R ∨ R e 1.01 ∨ R2 ) R and the cold fluid energy balance equations are neces-
Ft ) f1(R1,S) Ft ) f2(S) Ft ) f1(R2,S) sary. Because the hot fluid target temperature is a
variable in this example, the hot fluid energy balance
which can be introduced into the MINLP problem by equation is necessary.
using the “big-M” constraints; for this specific case, the
upper limit of the functions f1 and f2, which is equal to
Q ) mhCph(thin - thout) (107)
1, is known beforehand, and it is recommended to use
a lower bound to the design such as 0.75.
The summary of the results is presented in Tables 1 The summary of the results with the proposed model is
and 2. As is seen from the results, the proposed present in Tables 1 and 2. As is seen from the results,
optimization model finds the optimal solution in 32 the proposed optimization model finds the optimal
major iterations with DICOPT++.17 solution in 17 major iterations with DICOPT++.17
Ind. Eng. Chem. Res., Vol. 42, No. 17, 2003 4017

Table 4. Empirical Coefficients for Equations 56-63


tubes Reynolds
pattern number a1 a2 a3 a4 b1 b2 b3 b4
triangular 105-104 0.321 -0.388 1.450 0.519 0.372 -0.123 7.00 0.500
104-103 0.321 -0.388 0.486 -0.152
103-102 0.593 -0.477 4.570 -0.476
102-10 1.360 -0.657 45.100 -0.973
<10 1.400 -0.667 48.000 -1.000
square 105-104 0.370 -0.395 1.187 0.370 0.391 -0.148 6.30 0.378
104-103 0.107 -0.266 0.0815 0.022
103-102 0.408 -0.460 6.0900 -0.602
102-10 0.900 -0.631 32.100 -0.963
<10 0.970 -0.667 35.000 -1.000

Table 5. Tube Internal and External Diameters Used in Fc ) fraction of total tubes in cross-flow
Ain and Aout Fsbp ) fraction of cross-flow area available for bypass flow
Ain (mm) Aout (mm) Ain (mm) Aout (mm) Ft ) temperature correction factor
15.875 12.573 38.100 32.563 f si ) tube-side Fanning factor
19.050 14.834 38.100 33.884 hoi ) shell-side heat-transfer coefficient for an ideal tube
25.400 21.184 50.800 45.263 bank
31.750 27.534 50.800 46.584 ht ) tube-side heat-transfer coefficient
hs ) shell-side heat-transfer coefficient
Conclusion Jb ) correction factor for bundle-bypassing effects
Jc ) correction factor for baffle configuration effects
In this paper, an optimization model has been pro- ji ) Colburn factor
posed for the design of a shell-and-tube heat exchanger. Jl ) correction factor for baffle-leakage effects
The main contribution of this work is an optimization kh ) hot fluid thermal conductivity
model that includes correlations from the Bell-Dela- kc ) cold fluid thermal conductivity
ware method in order to calculate the heat-transfer kt ) tube-side fluid thermal conductivity
coefficient and the pressure drop in the shell-side flow. ks ) shell-side fluid thermal conductivity
Therefore, design options are included in the model with k1 ) Sinnott correlation parameter
the aim of determining the best design configuration. LMTD ) logarithm mean temperature difference
The model is based on generalized disjunctive program- ls ) baffle spacing
ming and is optimized with a MINLP formulation to Lt ) tube length
determine the heat-exchanger design that minimizes mh ) hot stream flow rate
the total annual cost accounting for area and pumping mc ) cold stream flow rate
expenses. mt ) tube-side fluid flow rate
The application and usefulness of the proposed method ms ) shell-side fluid flow rate
have been shown in three example problems. The n1 ) parameter in eq 44
Nb ) number of baffles
results indicate that the methodology can properly
Nc ) number of tube rows crossed in one cross-flow section
account for the tradeoffs between area and pumping
Ncw ) number of tube columns effectively crossed in each
costs. Finally, different design options such as flexible window
target temperatures or mass flow rates can be easily Nt ) number of tubes
included in the model by using heat balances and Ntp ) number of tube passes
logarithm mean temperature equations as constraints. Nu ) Nusselt number
Pcost ) pumping cost
Acknowledgment pn ) tube pitch normal to flow
pp ) tube pitch parallel to flow
The authors acknowledge financial support provided Pr ) Prandtl number
by CNPq (National Council of Science and Technological pt ) tube pitch
Development) and NSF Equipment Grant CTS-0094407. Res ) shell-side Reynolds number
Ret ) tube-side Reynolds number
Appendix Rl ) pressure drop correction factor for baffle-leakage
effects
Tables 3-5 are given.
Rb ) pressure drop correction factor for bundle-bypassing
effects
Notation rin ) tube-side deposit resistance
acost ) area cost constant rout ) shell-side deposit resistance
Sm ) reference normal area for shell-side flow
Ain
o ) set of TEMA tube internal diameters Ssb ) shell-to-baffle leakage area
Aout
o ) set of TEMA tube external diameters Stb ) tube-to-baffle leakage area for one baffle
ccost ) pumping cost constant Sw ) area flow through the window
Cph ) hot fluid heat capacity U ) overall heat-transfer coefficient
Cpc ) cold fluid heat capacity vt ) tube-side fluid velocity
Cpt ) tube-side fluid heat capacity yarr ) binary variable which defines tube pattern arrange-
Cps ) shell-side fluid heat capacity ment
dtin ) tube internal diameter yarrntp ) binary variable which represents yarr and yntp
dtout ) tube external diameter yf ) binary variable which defines the fluid allocation
Ds ) shell diameter yhead ) binary variable which defines the exchanger head
Dotl ) tube bundle diameter ynb ) binary variable which defines the number of baffles
4018 Ind. Eng. Chem. Res., Vol. 42, No. 17, 2003

yntp ) binary variable which defines the number of tube (7) Kakaç, S.; Bergles, A. E.; Mayinger, F. Heat Exchangerss
passes Thermal-Hydraulic Fundamentals and Design; Hemisphere Pub-
yres ) binary variable which defines the shell-side Reynolds lishing Corp.: Bristol, PA, 1981.
number regime (8) Saffar-Avval, M.; Damangir, E. A. General Correlation for
Determining Optimum Baffle Spacing for all Types of Shell and
yrearr ) binary variable which represents yres and yarr Tube Exchangers. Int. J. Heat Mass Transfer 1995, 38-13, 2501-
ytema ) binary variable which defines tubes diameter 2506.
∆Pt ) tube-side pressure drop (9) Taborek, J. Shell and Tube Heat Exchangers: Single Phase
∆Pbi ) shell-side pressure drop for ideal cross-flow Flow. Heat Exchanger Design Handbook; Hemisphere Publishing
∆Ps ) shell-side pressure drop Corp.: Bristol, PA, 1983; section 3.3.
∆Pwi ) pressure drop in the window (10) Poddar, T. K.; Polley, G. T. Heat Exchanger Design
µh ) hot fluid viscosity Through Parameter Plotting. Chem. Eng. Res. Des. 1996, 76, 849-
852.
µc ) cold fluid viscosity
(11) Maralikrishna, K.; Shenoy, U. V. Heat Exchanger Design
µt ) tube-side fluid viscosity Targets for Minimum Area and Cost. Chem. Eng. Res. Des. 2000,
µs ) shell-side fluid viscosity 78, 161-167.
Fh ) hot fluid density (12) Chaudhuri, P. D.; Diwekar, U. M. An Automated Approach
Fc ) cold fluid density for the Optimal Design of Heat Exchangers. Ind. Eng. Chem. Res.
Ft ) tube-side fluid density 1997, 36, 3685-3693.
Fs ) shell-side fluid density (13) Raman, R.; Grossmann, I. E. Modelling and Computational
δsb ) shell-baffle clearance Techniques for Logic Based Integer Programming. Comput. Chem.
Eng. 1994, 18, 563-578.
δtb ) tube-baffle clearance (14) TEMA. Standards of the Tubular Heat Exchanger Manu-
factures Association, 7th ed.; Tubular Heat Exchanger Manufac-
tures Association: New York, 1988.
Literature Cited (15) Balas, E. Disjunctive Programming and a Hierarchy of
Relaxations for Discrete Optimization Problems. SIAM J. Alg.
(1) Sinnott, R. K. Coulson & Richardson’s Chemical Engi- Discuss. Methods 1985, 6, 466-486.
neeringsChemical Engineering Design, revised 2nd ed.; Butter- (16) Raman, R.; Grossmann, I. E. Symbolic Integration of Logic
worth-Heinemann: Oxford, U.K., 1996; Vol. 6. in Mixed-Integer Programming techniques for Process Synthesis.
(2) Furman, K. C.; Sahinidis, N. V. Computational Complexity Comput. Chem. Eng. 1993, 17, 909-927.
of Heat Exchanger Network Synthesis. Comput. Chem. Eng. 2001, (17) Viswanathan, J.; Grossmann, I. E. A Combined Penalty
25, 1371-1390. Function and Outer-Approximation Method for MINLP Optimiza-
(3) Jegede, F. O.; Polley, G. T. Optimum Heat Exchanger tion. Comput. Chem. Eng. 1990, 14, 769-782.
Design. Trans. Inst. Chem. Eng. 1992, 70 (A2), 133-141. (18) Chen, J. J. Letter to the Editor: Comments on improve-
(4) Polley, G. T.; Panjeh Shahi, M. H. M. Interfacing Heat ment on a replacement for the logarithmic mean. Chem. Eng. Sci.
Exchanger Network Synthesis and Detailed Heat Exchanger 1987, 42, 2488.
Design. Trans. Inst. Chem. Eng. 1991, 69, 445-457.
(5) Polley, G. T.; Panjeh Shahi, M. H. P.; Jegede, F. O. Pressure Received for review December 3, 2002
Drop Considerations in the Retrofit of Heat Exchanger Networks. Revised manuscript received May 27, 2003
Trans. Inst. Chem. Eng. 1990, 68, 211-220. Accepted May 29, 2003
(6) Kern, D. Q. Process Heat Transfer; McGraw-Hill: New York,
1950. IE020964U

You might also like