## Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

**of shell-and-tube heat exchangers
**

José M. Ponce-Ortega

a,b

, Medardo Serna-González

a

, Arturo Jiménez-Gutiérrez

b,

*

a

Facultad de Ingeniería Química, Universidad Michoacana de San Nicolás de Hidalgo, Morelia, Mich. 58060, Mexico

b

Departamento de Ingeniería Química, Instituto Tecnológico de Celaya, Celaya, Gto. 38010, Mexico

a r t i c l e i n f o

Article history:

Received 9 October 2006

Accepted 25 June 2007

Available online 18 March 2008

Keywords:

Shell-and-tube heat exchangers

Bell–Delaware

Genetic algorithm

Optimization

a b s t r a c t

This paper presents an approach based on genetic algorithms for the optimal design of shell-and-tube

heat exchangers. The approach uses the Bell–Delaware method for the description of the shell-side ﬂow

with no simpliﬁcations. The optimization procedure involves the selection of the major geometric param-

eters such as the number of tube-passes, standard internal and external tube diameters, tube layout and

pitch, type of head, ﬂuids allocation, number of sealing strips, inlet and outlet bafﬂe spacing, and shell-

side and tube-side pressure drops. The methodology takes into account the geometric and operational

constraints typically recommended by design codes. The examples analyzed show that genetic algo-

rithms provide a valuable tool for the optimal design of heat exchangers.

Ó 2008 Elsevier Ltd. All rights reserved.

1. Introduction

The transfer of heat between process ﬂuids is an essential part of

most chemical processes. To carry out such heat transfer process,

shell-and-tube heat exchangers are widely used because they are

robust and can work in a wide range of pressures, ﬂows and tem-

peratures [1]. The traditional design approach for shell-and-tube

heat exchangers involves rating a large number of different exchan-

ger geometries to identify those that satisfy a given heat duty and a

set of geometric and operational constraints [2]. This approach is

time-consuming, and does not guarantee an optimal solution.

Jegede and Polley [3] reported a design approach based on sim-

pliﬁed equations that related the exchanger pressure drop, the sur-

face area and the heat transfer coefﬁcient; their model was based on

the Dittus–Boelter correlation for the tube-side ﬂow, and on the

Kern correlations for the shell-side ﬂow [4]. The combination of

the pressure drop relationships with the basic exchanger design

equation gave rise to a simple design algorithmthat avoids the iter-

ative procedure required to test different geometries. However, the

use of the Kern method may lead to signiﬁcant errors in the calcula-

tions because of its simpliﬁed ﬂow pattern model for the shell-side.

Polley et al. [5] developed an algorithm using the Bell–Delaware

method [6] to describe the ﬂow pattern of the shell-side ﬂuid. The

model accounts for leakage and bypass streams using the ﬂow

model proposed by Tinker [6]. Although the model by Polley

et al. [5] provides better estimations than the one by Jegede and

Polley [3], some shortcomings can be mentioned. In order to keep

the accuracy of the Bell–Delaware method, Polley et al. [5] devel-

oped a rather complex relationship for pressure drop estimation

on the shell-side, which requires an iterative procedure that in-

volves detailed estimations of exchanger geometries. The algo-

rithm also shows some lack of ﬂexibility for the shell-side,

because it is restricted by the assumptions that cross-ﬂow areas

are equal to window ﬂow areas, and that the spacings for end baf-

ﬂes are equal to those for the central bafﬂes. The second geometric

restriction ignores cases when large inlet and outlet nozzles make

it necessary to have higher inlet and outlet bafﬂe spacings than

central bafﬂe spacings [6]. Finally, the algorithm does not take into

account the end pressure losses on the tube-side due to contrac-

tions at the tube inlets, expansions at the exits, and ﬂow reversal

in the headers.

Recently, Serna and Jimenez [7] presented an algorithm for the

rigorous design of segmentally bafﬂed shell-and-tube heat

exchangers. The algorithm makes use of the maximum allowable

pressure drops of both streams without introducing geometric lim-

itations. In particular, the use of two compact formulations for

pressure drop estimations provides a simple algorithm with

remarkable convergences properties. The shell-side pressure drop

equation is based on the Bell–Delaware method, and the model

for the tube-side includes the estimation for end effects. However,

this algorithm does not explicitly take into account some of the

geometric and operational constraints regularly imposed for ex-

changer design, and it only considers the pressure drops as optimi-

zation variables. Therefore, sub-optimal design solutions are

typically obtained.

Chaudhuri and Diwekar [8] used simulating annealing for the

optimal design of heat exchangers, and developed a command

1359-4311/$ - see front matter Ó 2008 Elsevier Ltd. All rights reserved.

doi:10.1016/j.applthermaleng.2007.06.040

* Corresponding author. Tel.: +52 461 611 7575x139; fax: +52 461 611 7744.

E-mail address: arturo@iqcelaya.itc.mx (A. Jiménez-Gutiérrez).

Applied Thermal Engineering 29 (2009) 203–209

Contents lists available at ScienceDirect

Applied Thermal Engineering

j our nal homepage: www. el sevi er . com/ l ocat e/ apt her meng

procedure to link the HTRI design program to the annealing algo-

rithm. The authors used simulating annealing as an optimization

technique because the HTRI design program is a black-box model,

and therefore explicit relationships for the geometric and opera-

tional constraints are not available.

Mizutani et al. [9] have recently presented an optimization pro-

cedure to design shell-and-tube heat exchangers using the

Bell–Delaware method to calculate the shell-side heat transfer

coefﬁcients and pressure drops. Mitzutani et al. [9] used disjunc-

tive programming techniques to get the optimal design and con-

sidered different construction alternatives. Because of the high

degree of non-linearity and possible non-convexities of the Bell–

Delawere method, gradient methods such as the one used by Mizu-

tani et al. [9] may get trapped into local optimal solutions.

This work presents an approach based on a genetic algorithm

(GA) for the optimal design of shell-and-tube heat exchangers.

The approach overcomes some of the limitations of earlier meth-

ods based on mathematical programming techniques, and uses

the Bell–Delaware model for the shell side of the exchanger.

2. Model formulation

The optimization procedure involves the selection of the major

geometric parameters such as the number of tubes passes, stan-

dard internal and external tube diameters, tube layout and pitch,

type of head, ﬂuid allocation, number of sealing strips, inlet and

outlet bafﬂe spacing, and shell-side and tube-side pressure drops.

The objective is to minimize the total annual cost for the exchan-

ger, including the capital costs for the exchanger and two pumps

(shell-side and tube-side), and the operating cost of such pumps.

The methodology considers explicitly major geometric and opera-

tional constraints.

Genetic algorithms are used to guide the search towards an

optimal solution. A shell-and-tube heat exchanger needs to be de-

signed for each set of values of the search variables; the design

method by Serna and Jimenez [10] was used for that purpose.

2.1. Heat exchanger model

The basic design equation is

A ¼

Q

F

T

DT

LM

1

h

S

þ R

ds

þ

D

t

2k

w

ln

D

t

D

ti

þ

D

t

D

ti

h

T

þ

D

t

D

ti

R

dt

ð1Þ

where all symbols are deﬁned in the nomenclature section.

The exchanger area is related to the ﬁlm heat transfer coefﬁ-

cients and the allowable pressure drops through two compact for-

mulations recently development by Serna and Jimenez [7]. For

turbulent tube-side ﬂow in the tube-side, the compact formula is

Nomenclature

A heat transfer surface area

A

f

annualization factor for capital cost

c cost coefﬁcient (exponent) in capital cost law for ex-

changer

C

1

to C

5

coefﬁcients in detailed capital cost laws for exchangers

C

a

, C

0

b

cost coefﬁcients in capital cost law for heat exchangers

C

b

bafﬂe cost assuming 1/2 in thickness, based on weight

C

ba

base cost to cover those overheat and labor cost

C

e

, C

f

cost coefﬁcient in capital cost law for pumps

C

exc

capital exchanger cost

C

p

ﬂuid speciﬁc heat at constant pressure and average tem-

perature

C

pow

cost per unit of power

C

pump

capital pump cost

C

sh

shell cost including fabrication, based on weight

C

tb

cost of tubes, based on outside heat transfer surface

C

td

tube-sheet and bafﬂe drilling, and bundle tubing cost,

based on number of tubes

C

ts

tube-sheet cost based on weight, including cutting but

not drilling

D

s

inside shell diameter

D

ti

tube inside diameter

D

t

tube outside diameter

e cost coefﬁcient (exponent) in capital cost law for pump

ﬁtness ﬁtness function

F

T

correction factor to logarithmic mean temperature dif-

ference for non-countercurrent ﬂow

g vector of feasible constraints

h clean heat transfer coefﬁcient

H

Y

annual plant operation time

k

s

, k

t

, k

w

thermal conductivity of shell-side ﬂuid, tube-side ﬂuid,

and tube wall

K

S

shell-side constant for pressure drop relationship

K

T

tube-side constant for pressure drop relationship

L

to

overall tube length

L

tt

total tube length

m exponent for shell-side heat transfer coefﬁcient in pres-

sure drop relationship

_ m ﬂuid mass velocity

n exponent for tube-side heat transfer coefﬁcient in pres-

sure drop relationship

N

s

number of shells

N

tt

total number of tubes

penalty penalty factor in the ﬁtness function

Q heat duty

r penalty term

R

bs

ratio bafﬂe spacing to inside shell diameter

R

ds

shell-side fouling factor

R

dt

tube-side fouling factor

R

dw

combined resistance of tube wall and fouling factors

S

m

cross-ﬂow area near shell centerline

S

w

net cross-ﬂow area through one bafﬂe window

T stream temperature

TAC total annual cost

t

s

shell thickness

t

t

tube-sheet thickness

U overall heat transfer coefﬁcient

v velocity for ﬂuid stream

x vector of optimization variables

x

1

to x

10

search optimization variables

DP pressure drop for ﬂuid stream

DT

LM

log-mean temperature difference

g pump efﬁciency

q density of ﬂuid stream

l viscosity of ﬂuid stream

Subscripts

c cold

h hot

max maximum value

mat tubes material

min minimum value

s, S shell-side

t,T tube-side

w at wall temperature

204 J.M. Ponce-Ortega et al. / Applied Thermal Engineering 29 (2009) 203–209

DP

T

¼ K

T

Aðh

T

Þ

n

ð2Þ

Eq. (2) accounts for pressure drops in straight tubes and in tube

ends. For turbulent ﬂow on the shell-side ﬂow

DP

S

¼ K

S

Aðh

S

Þ

m

ð3Þ

Eq. (3) is based on the Bell–Delaware method. The compact formu-

lation was obtained after an analytical treatment of the original

equations, and therefore it has the same degree of applicability as

the original Bell–Delaware method. The deﬁnitions of K

S

, K

T

, m

and n, and how these parameters depend on the geometric param-

eters of the exchanger and the ﬂuid physical properties, along with

the design algorithm used in this work, are shown by Serna and

Jimenez [10]. The main steps of the design algorithm are included

in Appendix A.

2.2. Constraints for a feasible design

The design of a heat exchanger involves a number of constraints.

For convenience, the constraints may be classiﬁed into operating

constraints and geometric constraints. Some of the operating con-

straints are maximum allowable pressure drops and velocities for

both sides of the exchanger. Geometric constraints include maxi-

mum shell diameter, maximum tube length, minimum and maxi-

mum ratio of bafﬂe spacing to shell diameter, and minimum and

maximum ratio of cross-ﬂow area to area in a window.

The maximum pressure drops (DP

T,max

, DP

S,max

) depend on the

external pumps, and set upper bounds for the operating pressure

drops

DP

T

6 DP

T;max

ð4Þ

DP

S

6 DP

S;max

ð5Þ

Upper bounds on the velocities for both the tube-side and the shell-

side prevent erosion and ﬂow-induced tube vibration, while lower

bounds prevent fowling. These constraints are considered in the

model,

v

t;min

6 v

t

6 v

t;max

ð6Þ

v

s;min

6 v

s

6 v

s;max

ð7Þ

Sinnott [1] recommends velocities for liquids from 1 to 2.5 m s

À1

on

the tube-side and 0.3–1 m s

À1

on the shell-side.

Upper limits of the shell diameter and the tube length are part

of the primary geometrical constraints

D

s

6 D

s;max

ð8Þ

L

TT

6 L

TT;max

ð9Þ

As shown in Example 3, the algorithm can also be restricted to the

use of standard dimensions for these variables.

For bafﬂes, close spacings leads to higher heat transfer coefﬁ-

cients but at the expense of higher pressure drops. On the other

hand, wide bafﬂe spacings result in bypassing and reduced cross-

ﬂow, with a decrease in the heat transfer coefﬁcient. Therefore,

constraints on R

bs

are set

R

bs;min

6 R

bs

6 R

bs;max

ð10Þ

Typical values of R

bs,min

and R

bs,max

are 0.2 and 1.0, respectively.

Bounds on the ratio of cross-ﬂow area to area in a window, S

m

/

S

w

, are also considered

ðS

m

=S

w

Þ

min

6 S

m

=S

w

6 ðS

m

=S

w

Þ

max

ð11Þ

Typical values are 0.8 and 1.4 for lower and upper bounds,

respectively.

2.3. Objective function

The objective function consists of minimizing the total annual

cost of the exchanger. The total cost consists of ﬁve components:

the capital cost of the exchanger, the capital costs for two pumps,

and the operating (power) costs of the pumps. The expression for

the total annual cost is of the form

TAC ¼ A

f

ðC

exc

þ C

pump;T

þ C

pump;S

Þ þ C

pow

H

Y

g

_ m

t

DP

T

q

t

þ

_ m

s

DP

S

q

s

ð12Þ

where C

exc

, C

pump,T

and C

pump,S

are the capital cost for the exchanger,

tube-side and shell-side pumps, respectively. The capital costs for

the pumps are given by the following equations:

C

pump;T

¼ C

e

þ C

f

_ m

t

DP

T

q

t

e

ð13Þ

C

pump;S

¼ C

e

þ C

f

_ m

S

DP

S

q

s

e

ð14Þ

The capital cost for the heat exchanger is typically calculated by an

equation of the form

C

exc

¼ C

a

þ C

0

b

A

c

ð15Þ

Alternatively, one can incorporate a more detailed estimation by

splitting the capital cost into the cost of component parts and man-

ufacturing costs. Purohit [11] suggested the following expression

for that purpose:

C

exc

¼ C

ts

þ C

sh

þ C

b

þ C

td

þ C

tb

þ C

ba

ð16Þ

where C

ts

is the tube-sheet cost based on weight, including cutting

but not drilling; C

sh

is the shell cost including fabrication, based on

weight; C

b

is the bafﬂe cost assuming 1/2-in thickness, based on

weight; C

td

is the tube-sheet, bafﬂe drilling and bundle tubing cost,

based on number of tubes; C

tb

is the cost of tubes, based on outside

heat transfer surface; and C

ba

is the base cost to cover overhead and

labor costs, which is independent on the type of material.

2.4. Search variables

A vector x of search variables was manipulated as part of the

optimization algorithm. The vector contains 10 components,

according to the degrees of freedom of the problem; the deﬁnition

of variables is given in Table 1. Once the search vector is deﬁned,

the algorithm by Serna and Jimenez [10] is used to obtain the de-

sign of the exchanger. It should be clear that a major difﬁculty is

the selection of design values that satisfy all of the geometric

and operational constraints.

3. Optimization model using genetic algorithms

To generate an efﬁcient optimization method, genetic algo-

rithms are used. Genetic algorithms search for an optimum solu-

tion based on the mechanics of natural selection and genetic

[12,13].

Table 1

Search optimization variables

Variable Deﬁnition

x

1

Tube-side pressure drop

x

2

Shell-side pressures drop

x

3

Bafﬂe cut (between 15% and 45%)

x

4

Number of tube passes (1, 2, 4, 6 or 8)

x

5

Standard inside, outside tube diameters and pitch (80 standard

possible combinations given by the TEMA)

x

6

Tube pattern arrangement (triangular, square or rotated square)

x

7

Hot ﬂuid allocation (tubes or shell)

x

8

Number of sealing strings (0, 1, 2, 3 or 4)

x

9

Tube bundle type (ﬁxed-tube plate, packed-tube plate, ﬂoating head,

pull-through ﬂoating head or U-tube bundle)

x

10

Ratio inlet and outlet bafﬂe spacing

J.M. Ponce-Ortega et al. / Applied Thermal Engineering 29 (2009) 203–209 205

A population of chromosomes is formed initially from a random

set of solutions. New generations are then produced, and some

measure of ﬁtness for the chromosomes is used to guide the selec-

tion. Offsprings are formed by either (a) merging two chromo-

somes from the current generation using a crossover operator, or

(b) modifying a chromosome using a mutation operator. A new

generation is formed by selecting some of the parents and off-

springs, based on the ﬁtness values, and rejecting others to keep

the population size constant. Fitter chromosomes have higher

probabilities of being selected. After several generations, the algo-

rithm converges to the best chromosome, which represents the

optimal solution to the problem provided by the GA. Fig. 1 shows

a general representation of the operations behind a GA.

3.1. Handling the constraints

The original problem can be set as,

min TACðxÞ

s:t: g

i

ðxÞ P0 8i ¼ 1; 2; . . . ; m ð17Þ

where x is the vector of optimization variables. The set of con-

straints g(x) correspond to the inequalities given by (4)–(11).

For the implementation of the genetic algorithm, we used a

penalty function in the objective function, to provide the following

ﬁtness function to be minimized:

fitnessðxÞ ¼ TACðxÞ þpenaltyðxÞ ð18Þ

The penalty function accounts for the violation of the constraints,

such that

penaltyðxÞ ¼

0 if x is feasible

¸

m

i¼1

r

i

g

2

i

ðxÞ otherwise

ð19Þ

where r

i

is a variable penalty coefﬁcient for the ith constraint; r

i

var-

ies according to the level of violation. To provide an efﬁcient algo-

rithm, the value of each r

i

was increased proportionally as a

function of the number of generations.

4. General approach

The main steps of the approach are shown in Fig. 2. The algo-

rithm begins generating a set of random initial populations, that

is, a set of values for the optimization variables according to the

population size. Each of these individuals (set of design or search

variables) is then fed to the design algorithm for heat exchangers

to obtain a set of designs. The ﬁtness function for each individual

of the population is evaluated; from those values, the algorithm se-

lects the best individuals of the current generation as the parents

to a new generation. Then, new generations are created through

the mutation and crossover operations. The procedure is repeated

until the optimal design is detected.

For the GA implementation, a population size of 100 individuals

was used, with an elite count of three individuals. A crossover frac-

tion value of 0.8 was used. The algorithm stopped when no further

improvement in the ﬁtness function in 30 successive generations

was observed. As an alternative termination step, a maximum of

300 generations was imposed.

5. Results and discussion

Three case studies are used to show the application of the pro-

posed algorithm.

Example 1. This example is taken from Mizutani et al. [9] and

involves two ﬂuids with the properties shown in Fig. 3. The tube

thermal conductivity is 50 W/m K.

A summary of the results obtained with the proposed model is

given in Table 2, where a comparison with the results reported by

Mizutani et al. [9] is included. The results for Design A in Table 2

were obtained keeping the constraints imposed by Mizutani et al.

[9] of a tube length of 4.88 m and a bafﬂe cut of 25%. The solution

Solutions

encoding

chromosomes

1100101010

1011101110

0011011001

1100110001

corssover

1100101010

1011101110

1100101110

1011101010

mutation

0011011001

0011001001

evaluation

Offspring

1100101110

1011101010

0011001001

decoding

Solutions

fitness

computation

selection

Roulettewheel

new population

↓

↓

↓

↑

Fig. 1. General structure of genetic algorithms.

Initial population

Set optimization variables

Shell and tube heat

exchanger designs

Tests constraint

Evaluate penalty term

Fitness function

TAC+penalty term

Optimum?

Yes

End

No

New generation

Mutation

Crossover

Selection

New generation

TAC and implicit variables

for each design

Fig. 2. Solution strategy for the optimum design of heat exchangers.

Fig. 3. Stream data for Example 1.

206 J.M. Ponce-Ortega et al. / Applied Thermal Engineering 29 (2009) 203–209

was found in 121 generations, which required the simulation of

12,100 heat exchangers, and consumed 51 s of CPU time. One can

notice that the design obtained with the proposed algorithm has a

lower total annual cost than the one obtained by Mizutani et al. [9].

This observation seems to indicate that the solution obtained by

Mizutani et al. [9] got trapped into a local optimal point within

their search algorithm. The main difference between both results is

in the pumping costs. Our solution shows a reduction in pumping

costs of about 60% that, in spite of an increment in the area cost of

11%, provides a reduction in the total annual cost of 22%. It is

important to notice that the ﬁlm heat transfer coefﬁcients for

Design A are closer to each other than the ones obtained by

Mizutani et al. [9], thus providing a more efﬁcient design. Large

differences in ﬁlm coefﬁcients are linked to an inefﬁcient use of

pressure drops, which raises the pumping costs needed for the

exchanger.

The GA was also used for this problemwithout the constraints in

tube length and bafﬂe cut imposed by Mizutani et al. [9]. The results

are reported as Design B in Table 2. One can notice a signiﬁcant

reduction in the total area required by the exchanger. This is the

result of the number of passes being reduced to one, and of smaller

tube diameters being selected. This arrangement produces higher

stream velocities with better heat transfer coefﬁcients, which

provide a smaller area. Another issue worth of mention is that the

relationship L

tt

/Ds is higher than for the other two designs. Design B

has a total annual cost 49.88% lower than the one obtained by

Mizutani et al. [9], and 17.13% lower than Design A.

Example 2. This example was previously analyzed by Serna and

Jimenez [7]. A shell-and-tube heat exchanger must be designed

to cool down oil using cooling water. Fig. 4 shows the design data.

The tube wall thermal conductivity was neglected.

The solution was obtained after 90 generations using a CPU

time of 71 s. Table 3 shows a summary of the results obtained with

the proposed algorithm, as well as the design reported by Serna

and Jimenez [7]. Their design was based on gradient methods, and

they did not optimize the geometry of the exchanger; the main

design variables such as bafﬂe and tube characteristics were

speciﬁed. From Table 3 it can be seen that the design obtained

using the algorithm proposed in this work meets all the geometric

and operational constraints. On the other hand, the design by Serna

and Jimenez [7] shows a shell-side stream velocity 65% higher than

the maximum recommended value, which can lead to erosion in

the bafﬂes and tube vibrations.

The new design provides the geometric conﬁguration (tubes,

bafﬂes, shell) needed as part of the optimal solution. A proper use

of the pressure drops for each side of the exchanger provides a high

heat transfer coefﬁcient, thus optimizing the area and the cost of

the exchanger. The design obtained using GA has total pumping

costs 10.7% lower than the one reported by Serna and Jimenez [7],

along with reductions in the exchanger area of 10% and in total

annual cost of 6.1%.

Example 3. In this example, the data from Example 2 were taken,

but three major aspects were changed. First, only standard sizes for

the tube length and the shell diameter were considered. Second, a

different economic environment was assumed, in which higher

capital investment is required for heat exchangers. And third, the

economic model involved a more detailed description for the cost

of the exchanger.

The exchanger cost was calculated from the cost of component

parts plus manufacturing costs. The following relations, proposed

by Purohit [11], were used for Eq. (16)

C

ts

¼

pq

mat

C

1

ðD

s

þ2t

s

Þ

2

t

t

3456

ð20Þ

C

sh

¼

pq

mat

C

2

D

s

L

to

t

s

144

ð21Þ

C

0

b

¼

pq

mat

C

1

D

2

s

N

b

13824

ð22Þ

C

td

¼ C

4

N

tt

ð23Þ

C

tb

¼ C

3

A ð24Þ

C

ba

¼ C

5

ð25Þ

The constants for the cost equations were taken from Purohit [11]:

C

1

= 0.5 $/lb, C

2

= 1.0 $/lb, C

3

= 75 (A)

À0.4

$/ft

2

, C

4

= 2.0 $/tube,

C

5

= $30,000. For the exchanger, q

mat

= 486.954 lb/ft

3

. Values for t

s

and t

t

were calculated according to TEMA [14] standards.

To get convergence, 391generationwere needed, witha comput-

ing time of 182 s. Table 4shows a summaryof the results obtained. A

heat exchanger with an area of 220 m

2

was obtained. One can notice

how the economic scenario inﬂuences the resulting optimum

design. Tube diameters of mediumsize were selected, thus reducing

the number of tubes. This provided a proper compromise between

the material and manufacturing costs associated with the tubes.

Given the economic scenario considered, the minimumtotal annual

Table 2

Results for Example 1

Concept Mizutani et al. [9] This work

Design A Design B

Area (m

2

) 202 242.88 161.34

h

T

(W/m

2

K) 6,480.00 1,628.94 4,493.71

h

S

(W/m

2

K) 1,829.00 2,991.26 2,003.71

v

T

(m/s) – 0.83 1.00

v

s

(m/s) – 0.37 0.40

U (W/m

2

K) 860.00 714.51 873.62

Number of tubes 832 653 739

Tubes arrangement Square Triangular Square

Number of tube-passes 2 6 1

D

ti

(mm) 12.60 22.918 10.92

D

t

(mm) 15.90 25.40 12.70

Number of bafﬂes 8 8 13

Heat kind Fixed Floating pull Floating pull

Hot ﬂuid allocation Shell Tube Shell

F

Ã

T

DT

LM

24.90 25.01 30.79

D

s

(m) 0.687 1.105 0.639

Total tube length (m) 4.88 4.88 5.602

Bafﬂe spacing (m) 0.542 0.516 0.391

Bafﬂe cut (%) 25 25 31.85

DP

T

(Pa) 22,676.00 10,981.30 7,748.18

DP

S

(Pa) 7,494.00 4,714.28 6,828.43

Pumping cost ($/yr) 2,424.00 960.36 1,033.98

Area cost ($/yr) 2,826.00 3,142.59 2,468.77

Total annual cost ($/yr) 5,250.00 4,102.95 3,502.75

Fig. 4. Data for Example 2.

J.M. Ponce-Ortega et al. / Applied Thermal Engineering 29 (2009) 203–209 207

cost for this problem was $105,067 $/yr, with the highest contribu-

tion due to the capital cost for the exchanger.

One of the major aspects to highlight from this problem is that

the use of detailed capital costs for the heat exchanger of the type

of Eqs. (20)–(25) introduces additional non-linearities and possibly

non-convexities to the problem. Such items can affect the perfor-

mance of standard methods based on mathematical programming

techniques. On the other hand, the GA can handle these types of

models without potential convergence problems. The use of GA

also provides an interesting potential for the synthesis of heat ex-

changer networks, as recently reported by Ravagnani et al. [15]. In

such work, the best placement of the heat exchangers to accom-

plish some degree of energy integration was detected. As it is com-

mon in the network synthesis stage, the work by Ravagnani et al.

[15] did not incorporate the design details of the exchangers. The

model presented in this work could be incorporated into a heat ex-

changer network formulation to provide both the network struc-

ture and the equipment sizing with the type of details shown in

this work.

6. Conclusions

An algorithm for the optimal design of shell-and-tube heat

exchangers based on genetic algorithms has been presented. The

model uses the Bell–Delaware correlations for a proper calculation

of heat transfer coefﬁcients and pressure drops in the shell-side.

The use of GA together with the Bell–Delaware method allows sev-

eral design factors, typically speciﬁed from experience and later

subject to a rating test, to be calculated as part of the optimum

solution. Also, the objective function can accommodate any type

of information available for the cost of equipment; highly non-lin-

ear functions that arise from a detailed cost model for a heat ex-

changer can be handled without the convergence problems

typically encountered in mathematical programming techniques

based on gradient methods. Also, because of their nature, genetic

algorithms provide better expectations to detect global optimum

solutions than gradient methods, in addition to being more robust

for the solution of non-convex problems. The solution to examples

taken from the literature show how previously reported designs

can be improved through the use of the approach presented in this

work.

Appendix A. Design algorithm

The parameters K

S

, K

T

, m and n from the compact formulas de-

pend on the exchanger conﬁguration, which is not known until the

problem is solved for a given set of design variables. To develop an

efﬁcient algorithm, Serna and Jimenez [10] proposed to use the

parameters of the compact formulations as search variables, and

decoupled the equations with the unknown variables h

T

, h

S

and

A. Rearranging the simultaneous systemof Eqs. (1)–(3), one obtains

two sequential algebraic equations for h

T

and h

S

:

h

T

À

DP

T

F

T

DT

LM

K

T

Q

K

S

DP

T

K

T

DP

S

h

n

T

þ R

ds

þ

Dt

2kw

ln

Dt

D

ti

þ

Dt

D

ti

h

T

þ

Dt

D

ti

R

dt

¸

¸

1

n

¼ 0 ð26Þ

h

S

À

K

T

DP

S

h

n

T

K

S

DP

T

1

m

¼ 0 ð27Þ

To solve this set of equations, the following steps are used:

1. Specify the design data. These data include: (1) Mass ﬂowrates,

physical properties, inlet and outlet temperatures and fouling

factors; (2) the search optimization variables: Pressure drops

for each stream, inside and outside tube diameters, tube pitch,

number of tube passes, tube layout, tube bundle type, the ﬂuids

allocation, the number of sealing strips, the clearances and the

bafﬂe cut. The heat duty and the correction factor F

T

are calcu-

lated from the speciﬁcation (1).

2. Guess initial values for K

S

, K

T

, m and n. Kern method can be used

to get initial values.

3. Solve Eq. (26) numerically to obtain h

T

.

4. Determine h

S

and A sequentially from Eqs. (27) and (1).

5. Obtain sequentially geometric parameters for the exchanger,

once h

T

, h

S

, and A have been calculated (for further details, see

Serna and Jimenez [10]).

6. Obtain the constants for the Bell–Delaware method, the ideal

heat transfer factor, the ideal friction factor, as well as the leak-

age and bypass area and correction factors for the shell-side.

7. Calculate the new values for the tear variables K

S

, K

T

, m, n (for

details, see Serna and Jimenez [10]).

Table 3

Results for Example 2

Serna and Jimenez [7] This work

Area (m

2

) 200.02 180.49

h

T

(W/m

2

K) 5,567.92 6,062.84

h

S

(W/m

2

K) 1,009.17 1,247.17

v

T

(m/s) 1.00 1.26

v

s

(m/s) 1.32 0.78

U (W/m

2

K) 483.06 535.32

Number of tubes 838 1,545

Tubes arrangement Square, 90° Square, 45°

Number of tube-passes 4 4

D

ti

(mm) 16.60 10.92

D

t

(mm) 19.10 12.70

Number of bafﬂes 21 9

Heat kind Fixed tubes Floating pull

Hot ﬂuid allocation Shell Shell

F

Ã

T

DT

LM

(°C) 51.06 46.97

D

s

(m) 0.899 0.769

Total tube length (m) 4.156 3.080

Bafﬂe spacing (m) 0.181 0.301

Bafﬂe cut (%) 16.75 22.82

DP

T

(Pa) 15,822.0 30,543.15

DP

S

(Pa) 61,354.0 36,467.05

Pumping cost ($/yr) 3,874.90 3,501.52

Area cost ($/yr) 27,311.70 25,902.32

Total annual cost ($/yr) 31,186.60 29,403.84

Table 4

Results for Example 3

Variable Value

Area (m

2

) 220.72

h

T

(W/m

2

K) 4,977.26

h

S

(W/m

2

K) 835.18

v

T

(m/s) 1.11

v

s

(m/s) 0.76

U (W/m

2

K) 437.75

Number of tubes 1,058.18

Tubes arrangement Triangular, 30°

Number of tube-passes 6

D

ti

(mm) 17.27

D

t

(mm) 19.05

Number of bafﬂes 6

Heat kind Fixed tubes

Hot ﬂuid allocation Shell

F

Ã

T

DT

LM

(°C) 46.97

D

s

(m) 0.863

Total tube length (m) 3.658

Bafﬂe spacing (m) 0.456

Bafﬂe cut (%) 26.94

DP

T

(Pa) 27,451.96

DP

S

(Pa) 13,876.96

Pumping cost ($/yr) 2,662.22

Area cost ($/yr) 102,405.49

Total annual cost ($/yr) 105,067.71

208 J.M. Ponce-Ortega et al. / Applied Thermal Engineering 29 (2009) 203–209

8. Check for convergence. The algorithm converges when all of the

following deviation functions are smaller than some given tol-

erance. If convergence is not achieved, calculate the values of

the tear variables for the next iteration and go back to step 3.

This algorithm was used to design each of the exchangers that

were needed as part of the genetic algorithm application.

References

[1] R.K. Sinnot, Coulson & Richarson’s Chemical Engineering-Chemical Engineering

Design, second ed., vol. 6, ButterWorth–Heinemann, Oxford, UK, 1996.

[2] K. Muralikrishna, U.V. Shenoy, Heat exchanger design targets for minimum

area and cost, Chemical Engineering Research and Design 78 (2000) 161–167.

[3] F.O. Jegede, G.T. Polley, Optimum heat exchanger design, Chemical Engineering

Research and Design 70 (A2) (1992) 133–141.

[4] D.Q. Kern, Process Heat Transfer, McGraw-Hill, New York, 1950.

[5] G.T. Polley, M.H.P. Panjeh Shahi, F.O. Jegede, Pressure drop considerations in

the retroﬁt of heat exchanger networks, Chemical Engineering Research and

Design 68 (1990) 211–220.

[6] J. Taborek, Shell and Tube Heat Exchangers: Single Phase Flow, Heat Exchanger

Design Handbook, Hemisphere Publishing Corp., Bristol, PA, 1983. Section 3.3.

[7] M. Serna, A. Jimenez, A compact formulation of the Bell–Delaware method for

heat exchanger design and optimization, Chemical Engineering Research and

Design 83 (A5) (2005) 539–550.

[8] P.D. Chaudhuri, U.M. Diwekar, An automated approach for the optimal design

of heat exchangers, Industrial Engineering and Chemistry Research 36 (1997)

3685–3693.

[9] F.T. Mizutani, F.L.P. Pessoa, E.M. Queiroz, S. Hauan, I.E. Grossmann,

Mathematical programming model for heat exchanger network synthesis

including detailed heat-exchanger design. 1. Shell-and-tube heat exchanger

design, Industrial Engineering and Chemistry Research 42 (2003) 4009–

4018.

[10] M. Serna, A. Jimenez, An efﬁcient method for the design of shell and tube heat

exchanger, Heat Transfer Engineering 25 (2) (2004) 5–16.

[11] G.T. Purohit, Estimating costs of shell and tube heat exchangers, Chemical

Engineering 90 (17) (1983) 57–67.

[12] D.E. Goldberg, Genetic Algorithms in Search, Optimization and Machine

Learning, Addison-Wesley, Reading, MA, 1989.

[13] M. Gen, R. Cheng, Genetic Algorithms and Engineering Design, John Wiley &

Sons, New York, 1997.

[14] TEMA. Standards of the Tubular Heat Exchanger Manufactures Association,

seventh ed., Tubular Heat Exchanger Manufactures Association, New York,

1988.

[15] M.A.S.S. Ravagnani, A.P. Silva, P.A. Arroyo, A.A. Constantino, Heat exchanger

network synthesis and optimization using genetic algorithm, Applied Thermal

Engineering 25 (2005) 1003–1017.

J.M. Ponce-Ortega et al. / Applied Thermal Engineering 29 (2009) 203–209 209

- AN IMPLEMENTATION OF.pdf
- An Improvement Algorithms
- DamageIdentification_OSP_inSHM_GA.pdf
- Design Aids for Air Vessels for Transient Protection of Large Pipe (Thesis)
- A Survey of Multi Objective Optimization
- Genetic Alg
- ICEE-013
- Genetic Agent Approach for Improving on-the-fly Web Map Generalization
- torsional oscillations in alternators5
- Organizing Engineering Research Papers(40)
- tselnw74
- Performance of Evolutionary Algorithms on NK Landscapes with Nearest Neighbor Interactions and Tunable Overlap
- Automated Software Testing with Evolutionary Algorithms
- SNP Optimizer
- VRP
- Abstract
- Economic Thermal Power Dispatch With Emissin Constraint and Valve Point Effect Loading Using Improved Tabu Search Algorithm
- lec18
- Allocation of FACTS usin Genetic algorithm.pdf
- HIS_GA
- Lecture2 genetic algorithmAlgorithm
- Tlbo Codes
- II CM _SY_110112115544
- 9A02709 Optimization Techniques
- Research on the Batch Production Process Planning Problem of Assembly Line Based on the Virtual Enterprises
- Genetic Algorithms
- Print
- 7. Optimla Capacitor - Full
- Strategic Placement and Sizing of Passive Filters in a Power System for Controlling Voltage Distortion
- 1-s2.0-S0378779605002695-main

Close Dialog## Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

Loading