Energy Conversion and Management 101 (2015) 126135

Contents lists available at ScienceDirect

Energy Conversion and Management

journal homepage: www.elsevier.com/locate/enconman

Irreversibility analysis for optimization design of plate n heat

exchangers using a multi-objective cuckoo search algorithm
Zhe Wang a, Yanzhong Li a,b,
a
School of Energy and Power Engineering, Xian Jiaotong University, Xian 710049, China
b
State Key Laboratory of Multiphase Flow in Power Engineering, Xian 710049, China

a r t i c l e i n f o a b s t r a c t

Article history: This paper introduces and applies an improved multi-objective cuckoo search (IMOCS) algorithm, a novel
Received 29 March 2015 met-heuristic optimization algorithm based on cuckoo breeding behavior, for the multi-objective opti-
Accepted 5 May 2015 mization design of plate-n heat exchangers (PFHEs). A modied irreversibility degree of the PFHE is sep-
Available online 2 June 2015
arated into heat transfer and uid friction irreversibility degrees which are adopted as two initial
objective functions to be minimized simultaneously for narrowing the search scope of the design. The
Keywords: maximization efciency, minimization of pumping power, and total annual cost are considered nal
Plate-n heat exchanger
objective functions. Results obtained from a two dimensional normalized Pareto-optimal frontier clearly
Multi-objective optimization
Improved multi-objective cuckoo search
demonstrate the trade-off between heat transfer and uid friction irreversibility. Moreover, a three
algorithm dimensional Pareto-optimal frontier reveals a relationship between efciency, total annual cost, and
Entropy generation minimization (EGM) pumping power in the PFHE design. Three examples presented here further demonstrate that the pre-
Entransy dissipation minimization (EDM) sented method is able to obtain optimum solutions with higher accuracy, lower irreversibility, and fewer
iterations as compared to the previous methods and single-objective design approaches.
2015 Elsevier Ltd. All rights reserved.

1. Introduction three categories. First is the conservation of energy, which is com-

Energy conservation and its effects on climate change is a cru-
cial consideration for social consciousness. The heat exchanger, a
piece of heat and cold transfer equipment widely used in various
aspects of industry, demands improvement in order to reduce its
energy consumption and increase energy efciency [1]. Among dif-
ferent types of heat exchangers, plate-n heat exchangers (PFHEs)
are regarded for excellent heat exchange capacity and compact
size. They are commonly used in air separation, aerospace, cryo-
genics, chemical and petrochemical processes industries, etc. [2]
Generally, the design of PFHEs is a trial-and-error process under
nite constraint conditions (structure and operation parameters),
which are selected to satisfy prescribed requirements and create
optimal design results. This is essentially a compromise between
desirable heat transfer efciency and low cost. Careful selection
of design parameters may improve performance, lower capital
and operating costs, and conserve energy and materials [3]. To this
effect, many analysis methods exist in order to suggest optimiza-
tion methods for heat exchangers. These can be separated into
Corresponding author at: School of Energy and Power Engineering, Xian
Jiaotong University, Xian 710049, China. Tel.: +86 29 82668725; fax: +86 29
82668789.
E-mail address: yzli-epe@mail.xjtu.edu.cn (Y. Li).
monly used in various thermodynamic analyses. Second is entropy
and exergy, which combine the rst and second laws of thermody-
namics [4,5]. The number of entropy generation for irreversibility
minimization has been used for modeling and optimization by
Bejan [6]. The irreversibility of heat exchanger is evaluated by
using entropy generation minimization, which is still an active
research eld [7]. However, a paradox may arise when this method
was used as a single objective function [8]. The third includes var-
ious new methods, such as thermal economy [9], entransy dissipa-
tion theory, and others. Notably, Guo et al. [10] have dened
entransy dissipation as the rate and efciency of heat exchanger
designs. Similar to the process of entropy generation, the more
entransy dissipates, the higher its degree of irreversibility is.
Thus the entransy dissipation becomes a quantitative index for
evaluating the performance of heat exchangers, and there is no
entropy generation paradox in applications of heat exchanger
design [11,12]. In general, there are two types of irreversible
energy loss in the heat exchangerheat transfer and uid friction
irreversibility degrees, which are competing functions. In this
study, a modied irreversibility degree was separated into two
objective functions, and both entropy generation minimization
(EGM) and entransy dissipation minimization (EDM) methods
were used for the optimization of heat exchanger for verifying each
other.

http://dx.doi.org/10.1016/j.enconman.2015.05.009
0196-8904/ 2015 Elsevier Ltd. All rights reserved.
 Z. Wang, Y. Li / Energy Conversion and Management 101 (2015) 126135
Nomenclature
A heat exchanger surface area (m2)
Aow free ow area (m2)
b n height (m2)
C heat capacity rate (W/K)
Cr Cmin/Cmax
cp specic heat (J/kg K)
Cinv initial cost ($)
Cope operating cost ($)
Dh hydraulic diameter (m)
f Fanning friction factor
f(x) objective function
g(x) constraint
G entransy dissipation rate (W K)
h convective heat transfer coefcient (W/m2 K)
j Colburn factor
kel electricity unit price ($/MW h)
L heat exchanger length (m)
m mass ow rate (kg/s)
n n frequency (n/m)
n exponent of non-linear
Np number of hot side layer
NTU number of transfer units
N dimensionless number
Pr Prandtl number
P pressure (kPa)
Q heat duty (kW)
Previous optimization methods have failed to obtain functional
solutions over a broad range of common problem. The optimiza-
tion process for heat exchangers is complicated, and under numer-
ous constraints [13]. Considering these many existing
shortcomings, evolutionary algorithms (EAs) are often used to
address problems, generating solutions by use of a natural evolu-
tionary mechanism [14]. Because EAs do not need any assumptions
about the objective functions, the EAs are independent of the
mathematical model to be optimized and the initial values.
Recently, the application of EAs has received much interest from
researchers for the optimal design of heat exchangers. In a review
of relevant literature [15,16], different authors used different opti-
mization techniques for single as well as multi-objective optimiza-
tion of PFHEs.
Previous heat exchangers design effort have utilized traditional
mathematical methods, including simulated annealing [17] and
articial neural networks [18]. In 2004, Mishra et al. [19] exploited
a Genetic algorithm (GA) to minimize the total cost of optimal
design. Xie et al. [20] further added a pressure drop restriction to
a compact heat exchanger designed by GA. Mishra et al. [21] min-
imized the rate of entropy generation, based on the second law of
thermodynamics, to optimize cross-ow PFHEs. Sanaye and
Hajabdollahi [22] performed the simultaneous minimization of
total cost and maximization of efciency using a design which fea-
tured non-dominated sorting GA multi-objective PFHEs. 2010 also
saw the rapid development of certain meta-heuristic algorithms.
Rao and Patel [23] applied Particle Swarm Optimization (PSO) to
minimize entropy generation units and total volume for PFHEs.
Peng et al. [24] presented an improved PSO, which had a shorter
computational time and more favorable results compared to GA.
Imperialist competitive algorithm (ICA) is a social heuristic opti-
mization mechanism based on imperialism and colonial competi-
tion which Youse et al. [25] rst employed to optimize
cross-ow PFHEs under a constrained conditions, where the objec-
tive functions were EGM units, minimization of cost, and weight.
127
R specic gas constant (J/kg K)
R1 penalty parameter
Re Reynolds number
S entropy generation rate (W/K)
T temperature (K)
tf n thickness (m)
x a decision variable
X a set of each decision variables
Greek symbols
e effectiveness
l uid dynamic viscosity (kg/m s)
g pumping efciency
q density (kg/m3)
DT temperature difference (K)
DP pressure drop (kPa)
Subscripts
c cold uid
h hot uid
max maximum
min minimum
1 inlet
2 outlet
i, j variable number
Hadidi et al. [26] proposed to use the ICA to optimize
shell-and-tube heat exchangers from economic point of view,
which can help the manufacturer and engineers to reduce design
time in engineering applications. Babaelahi et al. [27] used
multi-objective optimization to make concession between thermal
efciency and pumping cost, achieving lower total cost in com-
pared to the original method. Wang et al. [28] used the dual tness
functions that act on Genetic Algorithm alternatively to obtain a
highly efcient automatic layer pattern arrangement. The thermal
performance showed good agreement with those of the layer
arrangement experiments. In the past two years, several optimiza-
tion strategies have been created to optimize the design of heat
exchangers, including teachinglearning-based optimization
(TLBO) [29], the harmony search algorithm [30], the Bees
Algorithm [31], and the Cuckoo search Algorithm [32,33].
However, no single algorithm is able to outperform the others for
all engineering applications, due to continuous improvements in
meta-heuristic algorithms. Any new methods which is introduced
to PFHEs design optimization must be submitted for further study.
On the other hand, with multiple objectives and constraints taken
into consideration, an optimum design problem must then be
solved, in which some of the objectives might be opposed or not.
So the results are more likely to fall into local optimum than single
objective. Multi-objective optimization is a difcult work. The
parameters are related and could inuence each other. In order
to solve this problem, we use a new method called improved
multi-objective cuckoo search algorithm, expecting more effective
results with good robustness.
This study investigates the irreversibility degree for optimiza-
tion geometric parameters of PFHEs using an improved
multi-objective cuckoo search algorithm (IMOCS). The primary
objectives of this work include: (1) Demonstrating the effective-
ness of the IMOCS algorithm for optimization of PFHEs. (2) Using
heat transfer and uid friction entropy generation/entransy dissi-
pation model as two separate evaluation function objectives to
128 Z. Wang, Y. Li / Energy Conversion and Management 101 (2015) 126135

Z 2     
optimize heat exchangers. (3) Comparing single objective and mcp dT T h;2 T c;2
SDT mh cp;h ln mc cp;c ln 18
multi-objective optimizations in the nal design results. The paper 1 T h;c T h;1 T c;1
is structured as follows: Section 2 describes thermal modeling and Z 2     
mRdP Ph;2 Pc;2
the formulation of objective functions of the PFHEs. Section 3 SDP  mh Rh ln  mc Rc ln 19
briey describes the basic Cuckoo search algorithm and details 1 P h;c Ph;1 Pc;1
the process of its improvement. To conclude, the optimization The total entropy generation rate in PFHEs can be written as:
results and relevant conclusions are presented in Sections 4 and
5, respectively. S SDT SDP 20

The entropy generation number is the dimensionless value for

2. Modeling formulation entropy generation rate, as follows:

2.1. Thermal and hydraulic formulation S

NS NS;DT NS;DP 21
maxC h ; C c
Detailed geometry of the PFHE and offset strip ns are shown in
 Entransy dissipation minimization (EDM)
Fig. 1. In this model, the e  NTU approach is utilized to predict the
Greater dissipation of entransy indicates a lower degree of
performance of a cross-ow PFHE. Notably, the PFHE is running
reversibility in the heat transfer of PFHEs. In order to obtain favor-
under a steady state, and the area distribution and heat transfer
able thermodynamic performance of PFHEs, entransy dissipation
coefcient are assumed uniform and constant. Thermal resistance,
must be reduced as much as possible. As noted by Gou in Ref.
heat loss, and longitudinal heat transfer of body in the PFHE are not
[10], the entransy dissipation rate, caused by heat conduction
considered. The number of cold layers is assumed to be one layer
and uid friction, for the ideal gas is expressed as follows:
greater than the hot layers. Table 1 shows the thermal and hydrau-
Z
lic model formulation. 2
1h i
GDT  mcp TdTh;c mh cp;h T 2h;1  T 2h;2 mc cp;c T 2c;1  T 2c;2
1 2
2.2. Objective functions 22

The heat transfer of heat exchangers is a series of irreversible  2 Z 2  2

T2  T1 dP T h;2  T h;1 P h;2
non-equilibrium thermodynamic process, which typically involves GDP  mR mh Rh ln
lnT 2 =T 1 h;c 1 P h;c lnT h;2 =T h;1 P h;1
the nite temperature difference between heat conduction tem-
 2
peratures, and the nite pressure drop of uid friction. Values from T c;2  T c;1 Pc;2
 mc Rc ln 23
an irreversibility evaluation function (IEF) such entropy generation lnT c;2 =T c;1 Pc;1
or entransy dissipation are divided into two parts, the IEF-related
heat transfer and uid friction. However, the IEF as related to these The total entransy dissipation rate for PFHEs is written as:
two parts are in conict. A decrease in one leads to an increase in G GDT GDP 24
the other, and vice versa. The multi-objective optimization method
used in this paper optimizes the conicting objectives simultane- The entransy dissipation number is the non-dimensionless
ously, in effort to create convergent optimal results, for example, value for the entransy dissipation rate as follows:
a minimum irreversible value for the thermodynamic process. G
 Entropy generation minimization (EGM) NG NG;DT NG;DP 25
e minC h ; C c T h;1  T c;1
By using EGM, it is possible to evaluate the irreversible losses in
PFHEs. According to the previous research [6], the ideal entropy  Cost estimation
generation rate for a gas can be obtained with nite temperature For cost functions, the investment cost, which is the annualized
difference and nite pressure drop as:

Lc
Lh

Ln

tf c
Cold flow

Hot flow
Fig. 1. Schematic diagram of cross ow PFHE and offset-strip n.
 Z. Wang, Y. Li / Energy Conversion and Management 101 (2015) 126135 129

Table 1
Calculating heat transfer rate and pressure drop.

Equations No. Remarks

Cr minC h ;C c 1 Heat capacity ratio

maxC h ;C c
h  i
e 1  exp 1
NTU 0:02 fexpC r NTU 0:78  1g 2 The efciency of PFHEs
Cr

T h;2 T h;1  e CCmin T h;1  T c;1 3 The outlet temperature of hot and cold uid
h

T c;2 T c;1 e CCmin T h;1  T c;1

c
h i
1 C min 1 1 4 The number of transfer units
NTU UAtot C min h A hc Ac h h

Ah Lh Lc N p 1 2nh bh  t f ;h  5 Heat transfer areas of hot and cold uid

Ac Lh Lc N p 11 2nc bc  t f ;c 
Atot Ah Ac 6 Total heat transfer areas of PFHE
Q e  C min T h;1  T c;1 7 The heat balance between hot and cold uid
Q mh cp;h T h;1  T h;2 mc cp;c T c;1  T c;2
c 1=n 8 Fin frequency
Aflow;h Lh N p bh  tf ;h 1  nh t f ;h 9 Free ow area of hot and cold uid
Aflow;c Lc N p 1bc  tf ;c 1  nh tf ;c
Re Dlh G 10 Reynolds number

Dh ct
2ct f bt f x 11 The hydraulic diameter of offset strip ns
2
f xbt f xt f bt f t f ct f t f

G Am
flow
12 Stream mass ux

j 0:6522Re 0:5403
a 0:1541
D0:1499 0:0678
c 5 1:34
 1 5:269  10 Re 0:504
a 0:456
D 1:055 0:1
c  13 The colburn coefcient and fanning factor for available
0:1 offset strip n.
f 9:6243Re0:7422 a0:1856 D0:3053 c0:2659  1 7:699  108 Re4:429 a0:92 D3:767 c0:236 
h j
2 Gcp 14 The heat transfer coefcient of hot and cold uid
Pr3

DP 4fLG
2
15 Pressure drop
2qDh
DP h;2 P h;1  DP h ; DP c;2 P c;1  DP c 16 The outlet pressure of hot and cold uid
W 1g m mc
q DP h q DP c
h 17 The total pumping power
h c

cost of the heat transfer area, and the operating cost, which combi-
nes with the electricity cost for the compressors, compose the total Table 2
annual cost (TAC). In the study, the economic model already con- Variation range of design parameters.
tains the detailed heat exchanger geometry to derive the heat Design parameters Min Max
transfer area. Any changes of design variables can lead to heat
Hot and cold ow n height bh, bc (m) 0.002 0.01
transfer area variations which affect the initial investment. Hot and cold ow n thickness tf,h tf,c (m) 0.0001 0.0002
Hot and cold ow n frequency nh, nc (n/m) 100 1000
TAC C inv C ope 26 Number of n layers Np 1 200
r 
Hot and cold ow length Lh, Lc (m) 0.1 1
C inv C A  Antot 27 Hot and cold ow n offset length xh, xc (m) 0.001 0.01
1  1 ry
   
DP h  m h DPc  mc effects of constraint violation. The multi-objective PFHE optimiza-
C ope kel s kel s 28
g  qh g  qc tion is represented as follows:

In the investment cost function, CA, n, r and y are the surface area of X
m
Min/f i xi  R1 g j x2 29
cost per unit (90$/m2), non-linear exponent (0.6), interest rate (0.1) j1
and depreciation time (10) respectively. The Atot is the total heat
transfer area, which is expressed as Atot = Ah + Ac. For the operating It is subject to the following set of constrains:
cost, kel, s and g are the electricity unit price (20$/MW h), operation T
hours (5000 h) and compressor efciency (0.6), respectively. It is f i xi 2 f 1 x1 ;f 2 x2 ;  ;f n xn  ; xi 2 X 1 ;X 2 ;   ; X n T ; i 1;2; .. .; n
undeniable that manufacturing the two parts of the heat exchang- g j x 6 0; j 1;2;. .. ; m
ers considering a different n frequency may cause additional cost, X i;min 6 X i 6 X i;max
and the model in this study is only a simplied assumption of the
actual heat exchanger. In order to validate the effectiveness of this
30
new method and make a comparison between the present study where fi(xi) stand for the n objective functions, xi and Xi respectively
and the previous researches, the model of PFHE in this paper adopts represent the vector of decision variables and the column vector
the same assumption in [18,20]. search space of the i independent variables, and gj(xi) are stated
the constraint conditions. R1 is a static penalty function parameter
with a relative large value compared to fi(xi). A similar approach
2.3. Design parameters and constraints
was previously considered by Ref. [24], and the code of the
improved algorithm for the model of cross ow plate-n heat
There are eleven variables which affect the design parameters
exchanger in this research has been developed in MATLAB2008a.
of PFHEs, as follows: (i) n height; (ii) n thickness; (iii) n fre-
quency; (iv) number of n layers; (v) dimensions of the heat
exchanger; (vi) n dimensions. The design parameters variation 3. Optimization method
ranges are shown in Table 2.
Multi-objective constrained optimization is used in the present 3.1. Multi-objective Cuckoo search (MOCS) algorithm
work, in order to simultaneously optimize several objectives under
given constraints. During the optimization process, a penalty func- The Cuckoo search (CS) method, a nature-inspired,
tion is added to the objective functions, which investigate the meta-heuristic algorithm, was developed in 2009 by Yang et al.
130 Z. Wang, Y. Li / Energy Conversion and Management 101 (2015) 126135
[34]. It is based on cuckoo breeding parasitic behavior and Lvy
ights. Parasitic breeding behavior is characterized by a process
in which cuckoo birds lay their eggs in other host birds nests, in
order to increase their egg hatching rate. If the host bird nds these
foreign eggs, however, it either discards them or abandons its nest
entirely and rebuilds it. The Cuckoo search algorithm uses this
behavior as a model, traversing the search space to nd optimal
solutions. Lvy ights describe the foraging behavior of many ani-
mals and insects, where a variable step size is punctuated by 90
turns, leading to an intermittent scale free search. This is a more
efcient search method than Brown movement, which is suitable
for the algorithm global search. For further details, refer to Ref.
[35].
The following idealized rules provide a simple description of the
MOSC method. (1) Each cuckoo chooses a random nest in which to
lay K eggs, representing the K objective functions. (2) In the ran-
dom selection of a set of host nests, the most favorable nests with
highest quality eggs are preserved for the next generation. (3) The
host bird discovers the foreign eggs with a probability pa e [0, 1],
then a new nest with K eggs is built. The path and the position iter-
ative formula of the cuckoo random search nest is as follows:
X it1 X ti a  Lk
a a0 X tj  X ti
k
Lk u t ; 1 < k 6 3
where X t1
i is candidate position of t + 1 generation and i nests, X ti is
the t generation and i nests position. a is the step size and a0 is a
constant. The  indicates entry-wise multiplications, Lk is a Lvy
distribution function describing randomly walked steps. For details,
refer to Ref. [36]. Based on these preconditions, the basic steps of
MOCS can be summarized as the pseudo code shown in Fig. 2.
3.2. Improvement of algorithm
A signicant disadvantage to solving multi-objection optimiza-
tion problems is that the convergence rate and the quality of solu-
tions for the original MOCS algorithm deteriorate due to
interference between multiple objective functions. To solve with
the low convergence speed and low solution quality typical of
the latter period, an improved multi-objective Cuckoo Search
Algorithm (IMOCS) is proposed here. The improvement strategy
follows two steps. First, a non-uniform mutation operator is
embedded in the original MOCS to create a perfect balance
between exploration and exploitation. Next, a differential evolu-
tion operator is employed to boost cooperation and information
Fig. 2. Pseudo-code of the MOCS algorithm.
exchange among the groups and enhance the quality of
convergence.
 Non-uniform mutation operator (NMO)
NMO corresponds to evolutional generation, providing adjust-
ments to the range of variation dynamically. The variation range
is relatively large in early stages, ensuring favorable global search
capability for the algorithm. The variation range then shrinks as
local optimization is gradually introduced. To this effect, introduc-
ing NMO to MOCS creates a balance between global exploration
and local exploitation in order to improve the accuracy and rate
of convergence. This paper adopts the Michalewicz [37] design of
NMO. Its specic operation is as follows:

xk Dt; U k  xk ; random0; 1 0
xk 32
xk  Dt; xk  Lk ; random0; 1 1
Subject to:
Dt; y y  1  r 1t=Tr ; Lk 6 xk 6 U k 33
Assuming the solution vector X x1 ; x2 ; . . . xk ; . . . xn , variance com-
ponent is pb. X k is the kth component of the non-uniform mutation,
31 and r, T, and r are a random number between 0 and 1, total gener-
ation number, and a constant, respectively.
 Differential evolution operator (DEO)
The original MOCS algorithm mainly utilizes Lvy ight to
update the populations of the host nest location. Because the indi-
viduals do not communicate effectively, they cannot share knowl-
edge and experience and do not form a collaborative advantage.
These defects must be remedied by specic improvement strate-
gies. DEO, one of these, uses mutation, crossover, and selection
operation to promote competition and cooperation between popu-
lations with an effective information sharing mechanism. This has
proven a very effective global optimization tool [38]. Its specic
operation is as follows:
Mutation:
V ti X ti F  X tr1  X tr2 34
where V ti , X ti , X tr1 , X tr2
are the variation, male parent, and randomly
selected individuals, respectively. F is the variation factor.
Mutated individuals retain their parent and information from any
two individuals, completing the information transfer.
Crossover:
(
V tij ; r < CR or j p
U tij 35
X tij ; others
where U tij , CR, r and p are the crossover results, crossover probabil-
ity, a random number either [0, 1] and a randomly selected dimen-
sion, respectively. The crossover operation that essentially
recombines a parent and variation to generate a candidate, which
ensures the contribution of at least one component value.
Selection:
8 t t
< U i ; ; U i dominant
>
X t1
i
t t
X i ; ; X i dominant 36
>
:
randomU ti ; X ti ; others
where randomU ti ; X ti is an equal probability random selection.
Selection operation inherits the dominant individuals to the next
generation, an elitist strategy.
Based on the above theory and relevant improvements, the
owchart of the improved MOCS algorithm for PFHE optimization
is shown in Fig. 3. Firstly, input design condition, perform design
procedure and compute design variables of the exchanger.
 Z. Wang, Y. Li / Energy Conversion and Management 101 (2015) 126135 131

Secondly, determine the objective function and generate initial 3000

position of nests and eggs randomly. Thirdly, perform Levy search
2700 A-G using IMOCS A* A
using NMO method and probabilistically eliminate inferior eggs
and nests using DEO strategy. The suitability of one solution is con- A'-G' using Modified TLBO[29] A'
2400
rmed, i.e., each egg must satisfy the specied constraints of the A*-E* using NSGA2[22]
2100

TAC ($/year)
problem. Fourthly, approve the suitability of this solution and go B
B'
to the next iteration. Finally, stop computation loop after a prede- 1800
ned number of iterations or when a specied condition has been B*
1500 C'
satised.
C* C
1200
E* D*
4. Results and discussion 900
D' D
F' E'
600 G'
4.1. Case 1: IMOCS algorithm effectiveness E
G F
300
0.50 0.55 0.60 0.65 0.70 0.75 0.80 0.85 0.90 0.95
Fig. 4 represents the multi-objective optimization Pareto opti-
mal curve using the IMOCS algorithm, as compared to both Efficiency
Sanaye and Hajabdollahi [22] using the NSGA-2, and Rao and
Fig. 4. The distribution of Pareto-optimal points solutions for PFHE using IMOCS,
Patel [29] using a modied TLBO approach. Under the same oper- NSGA2 [22] and modied TLBO [29] algorithms.
ating conditions, the results shown in Fig. 4 clearly demonstrate
the conict between two objectives, TAC and heat exchanger ef- algorithm are considered highly accurate compared to other meth-
ciency. Specications of Sanaye (AE), Rao (A0 G0 ) and this study ods. The computing time of the Pareto optimal curve is much lower
with sample points (AG) in Pareto optimal fronts are shown. for IMOCS than the NSGA-2, about 20%. In order to test the conver-
Maximum efciency occurs at point A (0.9085), where the TAC is gence accuracy of IMOCS algorithm, MOCS algorithm and IMOCS
highest ($ 2609.75). The corresponding minimum TAC exists at algorithm were run 20 times respectively. The number of times
design point G ($ 441.68), where the efciency is at its lowest value that the Pareto-optimal solution was obtained and the correspond-
(0.5256). The Pareto optimal design points of efciency, using the ing average run time were recorded. The results indicated that
IMOCS algorithm, are 1.18% and 0.68% higher compared to the IMOCS algorithm could obtain the Pareto-optimal solution with a
NSGA-2 and modied TLBO approach, under the same total annual higher probability (90%) in the same condition, while the probabil-
cost. Similarly, the optimal design points of TAC are 4.88% and ity of the original MOCS algorithm was only 65%. Besides, the aver-
3.59% lower than the NSGA-2 and modied TLBO approach, under age run time of original MOCS algorithm was 81 s, which is much
the same efciency. Thus, the optimization results of the IMOCS longer than the 69 s consumed by IMOCS algorithm. It revealed

Start

Input design condition

Initialize modeling

Design constraints
Improvement
Original MOCS strategies
Objective f 1 (x)  Objective f k (x)  Objective f n (x) 

Random Initialization
Penalty function
(N nests and K eggs)
NMO
Optimal balance solution as initial population

Levy search
Random(n,m)
(lay eggs in different nests)

Global search Local search

Objective calculation, fitness evaluation
( Determining noninferior nest location and eggs )

Balance conflicting Eliminated inferior solution DEO

objectives (Kill eggs or abandon nests)
Mutation

Crossover
Generation is max? No
Pareto optimal or stop criterion
Generation+1

Selection

Yes

End

Fig. 3. Schematic diagram of an improved MOCS algorithm for PFHE optimization.

132 Z. Wang, Y. Li / Energy Conversion and Management 101 (2015) 126135

that, compared to the original MOCS algorithm, IMOCS algorithm where i, j represent the normalization solution space values before
has a higher accuracy and a faster convergent rate. and after, and Min and Max are the minimum and maximum values
in the population. This method maps all solution on the Pareto
4.2. Case 2: minimum irreversibility optimization fronts in a 01 distribution. A design point is considered acceptable
when it is the shortest distance between the ideal point and the
The irreversibility losses in heat exchangers are typically caused curve. Fig. 5(IIIIV) shows the normalized Pareto front curves using
by their temperature difference and pressure drop. However, heat the above method, and sections D and F are the acceptable design
transfer and uid friction are two conicting objectives in the sections. Among them, the closest distance from zero is the
design of PFHEs, any parametric change that increases heat trans- pseudo-ideal point. In addition, the normalized Pareto fronts t a
fer leads to a decrease in the uid friction, and vice versa. In order polynomial curve, so the relationship between NS;DT and N S;DP is as
to balance the results appropriately, Fig. 5 demonstrates the two follows:
irreversible evaluation functions used to balance these conicting 5 4 3
objectives. For detailed design parameters of their application, NS; DP 3:891NS; DT 12:2NS; DT  14:67NS; DT
refer to Ref. [21]. 2
8:907NS; DT  3:446NS; DT 0:891 39
As previously mentioned, Fig. 5(III) represents the Pareto opti- 5 4 3
mal curve of irreversibility caused by temperature difference and NS; DT 2:128NS; DP 6:718NS; DP  8:469NS; DP
pressure drop, using entropy generation and entransy dissipation 6:139NS; DP
2
 3:213N S; DP 0:9504 40
as objective functions, obtained by the IMOCS algorithm. All solu-
tions in the Pareto frontier are potentials of optimum design value, The relationship between NG;DT and NG;DP is as follows:
therefore, this study uses a normalization method to determine the
5 4 3
nal solution. First, the ideal point is dened as zero, theoretically NG; DP 5:054NG; DT 15:44NG; DT  18:21NG; DT
the minimum point of the two objective functions. Second, the 2
10:86NG; DT  4:031NG; DT 0:9836 41
Pareto front curves are divided into nine sections (AI) after nor-
5 4 3
malized processing, making all curves comparable. The normalized NG; DT 3:052NG; DP 9:02NG; DP  10:7NG; DP
formula is as follows: 7:143NG; DP
2
 3:401NG; DP 0:9806 42
NDT i  NDT min
NDT j 37 The irreversibility of temperature differences is far larger than the
NDT max  NDT min
pressure drop in the heat exchangers; a lower entropy generation
NDP i  NDP min
NDP j 38 related to the heat transfer led to a higher thermal efciency, while
NDP max  NDP min entropy generation related to uid friction was increased. Thus,
these two types of entropy generations were considered as two

I 0.010 A II 0.028 A
Noninferior solution
Noninferior solution
B 0.024 B
0.008
C 0.020 C
Search space

0.006 D Search space D

NS, P

NG, P

0.016
E E

F 0.012 F
0.004
G G

H 0.008 H
I
Pareto frontier I Pareto frontier
0.002
0.004
0.030 0.035 0.040 0.045 0.050 0.055 0.090 0.105 0.120 0.135 0.150 0.165 0.180
NS,T NG,T

III 1.0 A IV 1.0 A

Normalized solution Normalized solution

0.8 B 0.8 B

C Pseudo Ideal point C Pseudo Ideal point

0.6 0.6
D
NG,P

D
NS, P

*
*

0.4 E 0.4 E

F F
0.2 G 0.2 G
H H
I I
Normalized Pareto frontier Normalized Pareto frontier
0.0 0.0
Ideal point Ideal point

0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
N*S,T N*G,T

Fig. 5. The Pareto fronts for PFHE with xed design conditions: (I) N S;DT vs. N S;DP ; (II) N G;DT vs. N G;DP ; (III) N S;DT vs. N S;DP ; (IV) N G;DT vs. N G;DP .
 Z. Wang, Y. Li / Energy Conversion and Management 101 (2015) 126135 133

separate conicting objective functions. In addition, the irreversibil- of the evaluation function for the irreversibility of a heat exchanger
ity of temperature difference is far larger than that of the pressure [39], entropy generation or entransy dissipation, is decided by the
drop in the heat exchangers for the plate n heat exchanger model. heat capacity ow ratio of the cold and hot uid, the temperature
So, in a single objective optimization, the irreversibility of pressure variation, etc. For this case, the maximum entransy dissipation of
drop is always submerged in a large irreversibility of temperature the heat exchanger is corresponding with the maximum heat
difference, and does not reect the optimization of the pressure transfer rate when the pumping power is a given constraint; while
drop. Table 3 shows the multi-objective optimized values of the the entropy generation is more suitable to reect the irreversibility
design variables using the MOCS algorithm and compares them to of the heat-work conversion process in the heat exchanger when
results obtained by Ref. [21,23] in a single-objective optimization. the pumping power is adopted as a variable. Therefore, if the heat
The main difference is the separating irreversibility evaluation func- exchanger optimization is focused on the heat transfer efciency,
tion, which IMOCS considers heat transfer and uid friction. The entransy dissipation minimization (EDM) is the better method to
values for NS;DT and N S;DP are 0.0376 and 0.0043, and the total evaluate the irreversibility. When the heat-work conversion ef-
entropy generation number was only 33.8%, 21.1% less than the ciency is the key point of the optimization, entropy generation
Ref. [21,23]. Furthermore, relative to previous studies, the PFHE minimization (EGM) is much more suitable for the irreversibility
pressure drop decreased by 3.9% and 1.6%, and total annual cost evaluation.
reduced by 3.15% and 0.98%.
Table 4 displays a comparison results between entropy genera- 4.3. Case 3: optimization of three objectives
tion minimization (EGM) and entransy dissipation minimization
(EDM). In this case study, optimization results from the two irre- Refer to Ref. [25] for an example case. A gas-to-air cross-ow
versible evaluation function methods are different. Under mini- PFHE possesses heat equal to 1069.8 kW. Operating conditions
mum conditions for PFHE irreversibility, the latter area is 1.04% are shown in Table 5. This PFHE must be designed for minimum
less than the former, the total pressure drop is increased by irreversibility, and must balance efciency, total annual cost and
1.72%, the efciency is increased by 0.26%, and the total cost pumping power. The three objective functions, alongside mini-
reduced by just 0.46%. In addition, the minimum entropy genera- mum irreversibility, are studied by multi-objective MOCS
tion method plays a signicant role in optimizing the heat exchan- optimization.
ger pressure drop, while minimum entransy dissipation creates Fig. 6(IIV) represents the Pareto optimal curve in a three
more advantages in the optimization of the heat exchanger area. dimensional objective space obtained by IMOCS algorithm. The
Although the cost difference between these two methods is fairly projection of solution regions in the Pareto optimal curve are
negligible for the total annual cost (including the investment and divided into six sections (AG sampling points), where each region
operation), the minimum entransy dissipation is favorable when has a counterpart in the optimal solution domain. As shown in
the maximum heat transfer efciency is the design goal. The choice Fig. 6(IIII), the projection of solution regions clearly reveal a con-
ict between the heat transfer entropy generation N S;DT and pres-
sure drop entropy generation N S;DP . In Fig. 6(I), the TAC
Table 3
Comparison of the single-objective and multi-objective optimization results for PFHE. maximum occurs at point A ($3134.25), where the entropy gener-
ation is lowest (0.12 + 0.009). Conversely, minimum TAC exists at
Parameters Ref. [21] Ref. [23] Present
point G ($545.05), where the entropy generation is highest
Cold Hot Cold Hot Cold Hot (0.18 + 0.002). After balancing between cost and entropy genera-
ow ow ow ow ow ow tion, point D in the Pareto fronts indicates the optimal design
b (mm) 9.53 9.53 9.80 9.80 10 9.9 region, where the total annual cost is $2008.03, and the entropy
tf (mm) 0.146 0.146 0.10 0.10 0.09 0.09 generations (N S;DT and N S;DP ) are 0.131 and 0.003. Fig. 6(II) shows
n (n/m) 534 534 442 442 419 501
x (mm) 6.30 6.30 9.80 9.80 9.9 10
the D point of the pumping power as 5244.72 W and entropy gen-
L (m) 0.994 0.887 0.925 0.966 0.902 0.912 erations of 0.130 and 0.003. The efciency is 0.827, and the entropy
DP (kPa) 7.100 4.120 1.750 11.10 3.723 6.062 generations (N S;DT and N S;DP ) are 1.28 and 0.004, where are shown
NS,DT 0.0376 in D point of Fig. 6(III).
NS,DP 0.0043 Fig. 6(IV) shows the three objectives which maximize of ef-
NS 0.0633 0.0531 0.0419 ciency, and minimize total annual cost and pumping power.
e 0.9161 0.9161 0.9161
Further, the optimization search space for the three objectives is
Ctotal ($) 1856.8 1816.2 1798.3
based on the Pareto optimal objective space of the minimum heat
transfer and pressure drop entropy generation, which reduce the
initial population and number of iterations. As observed in
Table 4
Comparison of the EGM and EDM results.
Fig. 6(IV), the maximum efciency, pumping power and total
annual cost exist at design point A (0.85, 10.91 kW, and $3160.1,
Parameters Entropy generation Entransy dissipation
respectively). The minimum is found at design point G (0.55,
Cold ow Hot ow Cold ow Hot ow
b (mm) 10 9.9 9.8 10
tf (mm) 0.09 0.09 0.1 0.1 Table 5
n (n/m) 419 501 492 426 Operating parameters of application case 3.
x (mm) 9.9 10 9.9 9.7 Parameters Hot uid Cold uid
L (m) 0.902 0.912 0.918 0.89
Mass ow rate (kg/s) 1.66 2
DP (kPa) 9.619 9.785 Inlet temperature (K) 900 200
Atot (m2) 1.845 1.826 Inlet pressure (kpa) 160 200
NS,DT 0.0376 Max pressure drop (kpa) 9.5 8.0
NS,DP 0.0043 Specic heat (J/kg K) 1122 1073
NG,DT 0.114 Density (kg/m3) 0.6296 0.9638
NG,DP 0.012 Dynamic viscosity (Ns/m2) 401  107 336  107
e 0.9161 0.9185 Prandtl number 0.731 0.694
Ctotal ($) 1798.3 1790.1
134 Z. Wang, Y. Li / Energy Conversion and Management 101 (2015) 126135

I II
Optimal solution domain Optimal solution domain
Solution regions projection Solution regions projection

Search Search
space space

III IV

Optimal solution domain Optimal solution domain

Solution regions projection Solution regions projection

Search
space

Search
space

Fig. 6. Distribution of solutions in three dimensional objective space by IMOSC algorithm: (I) N S;DT and N S;DP vs. Cost; (II) N S;DT and N S;DP vs. W; (III) N S;DT and N S;DP vs. e; (IV) W
and e vs. Cost.

59.9% and 0.52%, increase the efciency 0.6%, decrease the entropy
Table 6
generation rate by 5.29% and 2.41%, and decrease total annual cost
Comparison results between the Ref. [25] and present study.
by 4.31% and 1.82%. Furthermore the convergence iterations were
Parameters Ref. [25] Ref. [25] Present decreased to 33.3% and 50%, respectively, compared to the Ref.
Hot/cold ow length (m) 0.95/0.44 1/0.88 0.943/0.957 [25].
Hot/cold n height (mm) 7.2/7.2 5/5 9.12/8.35
Hot/cold n frequency (n/m) 417/417 240/240 301/258
Hot/cold thickness (mm) 0.1/0.1 0.19/0.19 0.16/0.19
5. Conclusions
Hot/cold n offset length (mm) 7.2/7.2 9.6/9.6 8.98/9.52
Number of hot side n layers 57 77 56 In this study, an improved MOCS algorithm is successfully
Pressure drop at hot/cold side (kPa) 4.2/0.52 1.23/0.67 1.16/0.73 introduced for the multi-objective optimization of PFHEs, based
No-ow length (m) 0.87 0.87 0.963
on e  NTU model. To improve the efciency and convergence rate
Effectiveness 0.821 0.821 0.826
Number of entropy generation units 0.1416 0.1374 0.1341 of MOCS, a non-uniform mutation operator and a differential evo-
TAC ($) 2101.4 2048.1 2010.7 lution operator were applied to the MOCS algorithm to decrease
Convergence iterations 30 40 20 search time and increase the robustness of selections. Based on this
algorithm, minimum heat transfer and uid friction irreversibility
were considered to be two objective functions. Efciency, total
2.98 kW, and $541.6, respectively). Point D indicates an ideal bal- annual cost, and pumping power of the PFHEs were also analyzed
ance of these, where minimum pumping power, maximum ef- as multi-objective functions. Eleven design parameters were con-
ciency and moderate cost are obtained. Moreover, these results sidered optimized variables, and a penalty function was added to
which provide minimum entropy generation are obtained under negotiate constraints. Three previous studies were selected as
dened conditions. A detailed comparison of these results against examples to demonstrate the proposed algorithms accuracy and
the Ref. [25] is shown in Table 6. An increase in n height and a performance. The main conclusions of this study are as follows.
decrease in the number of layers and n offset length lead to an
increase in the cross section of hot and cold ow. Additionally, 1. The Pareto optimal curve of efciency and total annual cost
the pressure drop at the cold side increases due to increasing the using the improved MOCS reveal conict between the two
cold uid length. The opposite occurs for the hot uid. A combina- objectives. Design results were more favorable compared to
tion of these factors causes the total pressure drop to decrease to previous approach.
 Z. Wang, Y. Li / Energy Conversion and Management 101 (2015) 126135
2. Optimal balance between the heat transfer and uid friction
irreversibility was obtained. The multi-objective optimization
results of separating entropy generation into two conicting
objectives produced a lower pressure drop, lower entropy gen-
eration, and lower total annual cost compared to a
single-objective optimization.
3. The three objective optimization search space, based on Pareto
optimal objective space for minimum heat transfer and pres-
sure drop entropy generation, reduced the initial population
and number of iterations. From the Pareto front, the optimal
shape of plate-n heat exchanger can be determined according
to the actual requirements, i.e. the optimal heat transfer, the
lowest pressure drop, or both weigh.
Acknowledgements
This research was supported by the Chinese State Science and
Technology Support Program (Grant No. 2012BAA08B03), and the
Specialized Research Fund for the Doctoral Program of University
by MOE China (Grant No. 20130201110069).
References
[1] Cornelissen RL, Hirs GG. Exergetic optimisation of a heat exchanger. Energy
Convers Manage 1997;38(15):156776.
[2] Hesselgreaves JE. Compact heat exchangers: selection, design and
operation. Gulf Professional Publishing; 2001.
[3] Muralikrishna K, Shenoy UV. Heat exchanger design targets for minimum area
and cost. Chem Eng Res Des 2000;78(2):1617.
[4] Hardy RJ, Binek C. First law of thermodynamics. Thermodyn Stat Mech: Integr
App 2014:1926.
[5] Hardy RJ, Binek C. Second law of thermodynamics. Thermodyn Stat Mech:
Integr App 2014:91108.
[6] Bejan A. Fundamentals of exergy analysis, entropy generation minimization,
and the generation of ow architecture. Int J Energy Res 2002;26(7):043.
[7] Huang S, Ma Z, Cooper P. Optimal design of vertical ground heat exchangers by
using entropy generation minimization method and genetic algorithms.
Energy Convers Manage 2014;87:12837.
[8] Ogulata RT, Doba F, Yilmaz T. Irreversibility analysis of cross ow heat
exchangers. Energy Convers Manage 2000;41(15):158599.
[9] Ghanei A, Assareh E, Biglari M, et al. Thermal-economic multi-objective
optimization of shell and tube heat exchanger using particle swarm
optimization (PSO). Heat Mass Transfer 2014:110.
[10] Guo ZY, Zhu HY, Liang XG. Entransy-a physical quantity describing heat
transfer ability. Int J Heat Mass Transfer 2007;50(13):254556.
[11] Xu M, Guo J, Li X. Thermodynamic analysis and optimization design of heat
exchanger. Advances in transport phenomena 2011. Springer International
Publishing; 2014. p. 63167.
[12] Wang WH, Cheng XT, Liang XG. Entropy and entransy analyses and
optimizations of the Rankine cycle. Energy Convers Manage 2013;68:828.
[13] Lee IS. Heat exchanger optimization. MS Thesis New South Wales Univ.,
Kensington (Australia). School of Mechanical and Industrial Engineering; 1987,
1.
[14] Zang H, Zhang S, Hapeshi K. A review of nature-inspired algorithms. J Bionic
Eng 2010;7:2327.
[15] Liu G. Sustainable feasibility of solar photovoltaic powered street lighting
systems. Int J Electr Power Energy Syst 2014;56:16874.
135
[16] Li Y, Liao S, Liu G. Thermo-economic multi-objective optimization for a solar-
dish Brayton system using NSGA-II and decision making. Int J Electr Power
Energy Syst 2015;64:16775.
[17] Reneaume JM, Pingaud H, Niclout N. Optimization of plate n heat exchangers:
a continuous formulation. Chem Eng Res Des 2000;78(6):84959.
[18] Jia R, Sunden B. Optimal design of compact heat exchangers by an articial
neural network method. In: ASME 2003 heat transfer summer
conference. American Society of Mechanical Engineers; 2003. p. 65564.
[19] Mishra M, Das PK, Sarangi SK. Optimum design of cross ow plate-n heat
exchangers through genetic algorithm. Int J Heat Exchang 2004;5(2):379401.
[20] Xie GN, Sunden B, Wang QW. Optimization of compact heat exchangers by a
genetic algorithm. Appl Therm Eng 2008;28(8):895906.
[21] Mishra M, Das PK, Sarangi S. Second law based optimisation of crossow plate-
n heat exchanger design using genetic algorithm. Appl Therm Eng
2009;29(14):29839.
[22] Sanaye S, Hajabdollahi H. Thermal-economic multi-objective optimization of
plate n heat exchanger using genetic algorithm. Appl Energy
2010;87(6):1893902.
[23] Rao RV, Patel VK. Thermodynamic optimization of cross ow plate-n heat
exchanger using a particle swarm optimization algorithm. Int J Therm Sci
2010;49(9):171221.
[24] Peng H, Ling X, Wu E. An improved particle swarm algorithm for optimal
design of plate-n heat exchangers. Ind Eng Chem Res 2010;49(13):61449.
[25] Youse M, Darus AN, Mohammadi H. Second law based optimization of a plate
n heat exchanger using Imperialist Competitive Algorithm. Int J Phys Sci
2011;6(20):474959.
[26] Hadidi A, Hadidi M, Nazari A. A new design approach for shell-and-tube heat
exchangers using imperialist competitive algorithm (ICA) from economic point
of view. Energy Convers Manage 2013;67:6674.
[27] Babaelahi M, Sadri S, Sayyaadi H. Multi-objective optimization of a cross-ow
plate heat exchanger using entropy generation minimization. Chem Eng
Technol 2014;37(1):8794.
[28] Wang Z, Li Y, Zhao M. Experimental investigation on the thermal performance
of multi-stream plate-n heat exchanger based on genetic algorithm layer
pattern design. Int J Heat Mass Transfer 2015;82:51020.
[29] Rao RV, Patel V. Multi-objective optimization of heat exchangers using a
modied teaching-learning-based optimization algorithm. Appl Math Model
2013;37(3):114762.
[30] Youse M, Enayatifar R, Darus AN, et al. Optimization of Plate-n heat
exchangers by an improved harmony search algorithm. Appl Therm Eng
2013;50(1):87785.
[31] Kefayat M, Ara AL, Niaki SAN. A hybrid of ant colony optimization and articial
bee colony algorithm for probabilistic optimal placement and sizing of
distributed energy resources. Energy Convers Manage 2015;92:14961.
[32] Khosravi R, Khosravi A, Nahavandi S, et al. Effectiveness of evolutionary
algorithms for optimization of heat exchangers. Energy Convers Manage
2015;89:2818.
[33] Berrazouane S, Mohammedi K. Parameter optimization via cuckoo
optimization algorithm of fuzzy controller for energy management of a
hybrid power system. Energy Convers Manage 2014;78:65260.
[34] Yang XS, Deb S. Cuckoo search via Lvy ights. In: World congress on nature &
biologically inspired computing, 2009. NaBIC 2009. IEEE; 2009. p. 2104.
[35] Fister Jr I, Yang XS, Fister D, et al. Cuckoo search: a brief literature review.
Cuckoo search and rey algorithm. Springer International Publishing; 2014.
p. 4962.
[36] Yang XS, Deb S. Multi-objective cuckoo search for design optimization.
Comput Oper Res 2013;40(6):161624.
[37] Michalewicz Z. Genetic algorithms data structures evolution programs. 3rd
ed. Springer; 1996. p. 2857.
[38] Das S, Suganthan PN. Differential evolution: a survey of the state-of-the-art.
IEEE Trans Evol Comput 2011;15(1):431.
[39] Cheng X, Zhang Q, Liang X. Analyses of entransy dissipation, entropy
generation and entransy-dissipation-based thermal resistance on heat
exchanger optimization. Appl Therm Eng 2012;38:319.