0% found this document useful (0 votes)

103 views

Design of Heat Exchanger Using Falcon Optimisation Algorithm PDF

Uploaded by

Anchuri Lokesh

Available Formats

Download as PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

103 views

Design of Heat Exchanger Using Falcon Optimisation Algorithm PDF

Uploaded by

Anchuri Lokesh

Available Formats

Download as PDF, TXT or read online on Scribd

You are on page 1/ 63

Accepted Manuscript

Design of heat exchangers using falcon optimization algorithm

Emerson Hochsteiner de Vasconcelos Segundo, Viviana Cocco Mariani,

Leandro dos Santos Coelho

PII: S1359-4311(19)30164-4
DOI: https://doi.org/10.1016/j.applthermaleng.2019.04.038
Reference: ATE 13628

To appear in: Applied Thermal Engineering

Received Date: 7 January 2019

Revised Date: 8 April 2019
Accepted Date: 12 April 2019

Please cite this article as: E.H.d. Vasconcelos Segundo, V.C. Mariani, L.d. Santos Coelho, Design of heat exchangers
using falcon optimization algorithm, Applied Thermal Engineering (2019), doi: https://doi.org/10.1016/
j.applthermaleng.2019.04.038

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers
we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and
review of the resulting proof before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
DESIGN OF HEAT EXCHANGERS USING FALCON
OPTIMIZATION ALGORITHM

Emerson Hochsteiner de Vasconcelos Segundoª*,

Viviana Cocco Marianiª,b, Leandro dos Santos Coelho b,c

a
Department of Mechanical Engineering, Pontifical Catholic University of Paraná
(PUCPR), Curitiba, PR, Brazil
b
Department of Electrical Engineering, Federal University of Parana (UFPR), Curitiba,
PR, Brazil
c
Industrial and Systems Engineering Graduate Program (PPGEPS), Pontifical Catholic
University of Paraná (PUCPR), Curitiba, PR, Brazil

*Corresponding author
E-mail addresses: [email protected] (E. H. Vasconcelos Segundo),
[email protected] (V. C. Mariani), [email protected] (L. S. Coelho).

Abstract: This paper proposes a novel metaheuristic optimizer based on the hunt

behavior of falcons called Falcon Optimization Algorithm (FOA). FOA is a robust and

powerful stochastic population-based algorithm that needs the adjustment of few

parameters for its three-stage movement decision. Simulation results based on well-

known twelve benchmark single-objective functions demonstrate the efficiency,

effectiveness and robustness of the proposed method in comparison to other algorithms.

Furthermore, the results of its single-objective application in heat exchangers shell-and-

tube and plate-fin types allowed better results than previous works for the objective

functions total cost for shell-and-tube heat exchanger (28% and 57.8% of reduction for

cases 1 and 2, respectively) and number of entropy generation units (15.42% of

reduction for case 1) and effectiveness (10% of increasing for case 2) for plate-fin heat
exchanger type, alongside with a thermal-hydraulic discussion. Moreover, the FOA

reached some solutions better than those previously reported in the literature.

Keywords: Metaheuristics, Optimization, Falcon Optimization Algorithm, Shell-and-

Greek letters
effectiveness
viscosity (N/m² s)
density (kg/m³)
overall pumping efficiency

Subscripts
best best result achieved so far
cold stream
hot stream
inlet
iter current iteration
iter-1 last iteration
minimum
maximum
outlet
shell side
tube side
wall

1. INTRODUCTION

Optimization is about the search for the best possible solution (or a set of solutions in
multi-objective optimization), or trade-off solutions, for a given problem with or
without constraints. Over the last decades several researchers have focus their attention
to develop new and efficient optimization techniques for several needs [1, 2]. The major
challenge of the global continuous optimization remains about those problems that
present many local optima. Mathematical optimization techniques were the only
methods for optimizing problems before the proposal of the metaheuristic techniques,
where the mathematical optimization techniques are mostly deterministic, suffering
from local minima entrapment problem. Some of them, such as gradient-based methods
that require derivative information of the search space. The referred problems make this
kind of approach non-efficient algorithms for solving real problems related to global
optimization [3]. However, adjoint based optimization techniques can present high
efficiency for solving computationally intensive problems, such as CFD-based and even
single-objective optimization problems [4]. Yet, the combination of metaheuristic
optimization with gradient-based methods may lead to high efficiency techniques as
well [5].

Among the wide universe of algorithms, the global optimization field based on
metaheuristics has been active, generating every year different kinds of stochastic
algorithms. Differently of the deterministic algorithms, stochastic techniques may find
distinct solutions from the same initialization population. Metaheuristic algorithms uses
the observation of nature in physical, chemical and biological forms, basing its
conception in natural phenomena or in the behavior of some species [6-8]. Actually, the
modern nature-inspired methods are called metaheuristics [9]. Metaheuristic algorithms
have shown to be promising alternatives for solving optimization problems, including
those which presents non-linearity and multimodality. In the last decades many
algorithms were proposed and presented to the scientific community, where some of
them become well-known such as Differential Evolution (DE) [10], Genetic Algorithm
(GA) [11] and Particle Swarm Optimization (PSO) [12].

This growing field enhanced the interest of many researchers around the globe to apply
such techniques in engineering problems, where heat exchangers are present. The heat
exchangers most common to the investigated in the literature are the shell-and-tube and
plate-fin heat types. Shell-and-tube heat exchangers is widely used in industrial
processes such as condensers in nuclear power stations, steam generator in water reactor
plants under pressure and to feed water heaters among other alternative energy
applications in ocean, thermal and geothermal areas [13]. Plate-fin heat exchangers is
also used in several industrial processes - specially in gas-to-gas applications - in
cryogenics, micro-turbines, automobiles, chemical process plants, naval and
aeronautical applications [14-16].

Taking into account the works developed and presented to the scientific community in
the last decades, the shell-and-tube heat exchangers optimization aim mainly the
minimization of the system cost [17-32]. In reference of single-objective optimization
of shell-and-tube heat exchangers, several objective functions have been investigated
through many metaheuristics over the last two decades. Chauduri et al. [19] investigated
this type of heat exchanger using Simulated Annealing (SA), in that study was
considered an optimization algorithm for large scale combinatorial problems, where the
aim was the minimization of the heat transfer area e total cost. The major contribution
of that work could be considered the identification of better results with the inclusion of
more decision variables and configurations alternatives to the design.

Almost a decade later, Selbas et al. [20] studied the minimization of total cost utilizing
GA adopting several variables such as outside diameter of the tubes, flow arrangement
of the tubes, number of tube passages, outside shell diameter and baffles spacing,
reaching better results than previous studies for the same cases. Caputo et al. [21], also
using GA and same objective function, evaluated three different cases where inside
shell diameter, outside tube diameter and baffles spacing were adopted as design
variables, achieving significant reductions in the value of total cost.

Fesanghary et al. [22], on their turn, proposed the investigation of shell-and-tube heat
exchangers by Harmony Search Algorithm (HSA) minimizing total cost. In that work, a
sensitivity analysis was performed to identify those geometrical parameters that less
influence in the final result, reducing the amount of decision variables for the
optimization. Then, Patel & Rao [23] and Sahin et al. [24] applied Particle Swarm
Optimization (PSO) and Artificial Bee Colony (ABC), respectively, for cost
optimization of shell-and-tube heat exchanger in the same case that Caputo et al. [21],
obtaining enhancements in the results of the simulations.
A few years later, Hadidi & Nazari [25], Hadidi et al. [26] and Asadi et al. [13] studied
some of the cases presented earlier in references [21] and [23] using Biogeography-
Based Optimization (BBO), Imperialist Competitive Algorithm (ICA) and Cuchoo
Search Algorithm (CSA), respectively, obtaining better results than the ones achieved
by GA and PSO previously. Amini & Bazargan [27] applied a technique already used,
GA, for cost and heat transfer rate, demonstrating the contradictory effect that some
parameters have over both objective functions aimed. Lately, Mohanty [28, 29],
utilizing Firefly Algorithm (FA) and Gravitational Search Algorithm (GSA), and
Vasconcelos Segundo et al. [30], using a variant of Differential Evolution, investigated
shell-and-tube heat exchangers obtaining 29%, 22.4% and 26.99% of reduction,
respectively, in the values of total cost in comparison to the original design. Recently,
Barros et al. [31] presented a sustainability optimization of shell-and-tube heat
exchanger based on Integrated Value Model for Sustainability Assessment, Petinrin et
al. [32] adopted the minimization of entropy generation for shell-and-tube heat
exchanger in crude oil preheat train using Firefly Algorithm.

About plate-fin heat exchangers and its optimization with metaheuristics techniques we
can cite the work of Mishra et al. [16] with GA and focusing the minimization of
entropy generation units. In that study the decision variables were several geometrical
parameters and a direct relation between a reduction in the entropy generation units and
the pressure drops was obtained. For the same cases, Rao & Patel [14, 33] applied PSO
and Teaching-Learning-Based Optimization (TLBO) and enhanced the results obtained
previously with GA. Yousefi et al. [34 - 36], similar to previous authors, utilized ICA,
GA, PSO and HSA, respectively, obtaining progressive enhancements in the results for
cases studies explored previously. Recently, Zarea et al. [37] and Banooni et al. [38]
investigated plate-fin heat exchangers with the use of Bees Algorithm (BA), focusing
maximization of efficiency, minimization of entropy generation units and total cost,
while Vasconcelos Segundo [39] performed similar evaluation for entropy generation
units through an Adaptive Differential Evolution version called JADE. Other studies
such as Peng et al. [40] approached PSO for minimization of total cost, Zhang et al. [41]
adopted the optimization of minimum entropy generation by a distributed parameter
model, and Ghosh et al. [42] proposed a GA approach applied to maximum heat load.
The main motivation of this study is to propose a novel nature-inspired algorithm to
compete with the current optimization algorithms with few adjustment parameters that
can be posteriorly tuned and to be useful as basis for other algorithms that may emerge
from the observation of birds of prey. The main inspiration of the proposed algorithm is
the hunt behavior of the falcons.

The remainder of this paper is organized of the following form. Section 2 presents the
hunt behavior of falcons and an algorithm implementation procedure. Section 3 presents
the experimental setup and performance analysis tools. The results of the proposed
algorithm for several single-objective benchmark functions and engineering problems of
heat exchangers and their analysis are then provided in section 4 and section 5
concludes the work and suggests directions for future studies.

2. FALCON OPTIMIZATION ALGORITHM

The main inspiration for the proposed algorithm and details about its implementation
are presented in this section.

2.1 Hunt behavior of falcons

The inspiration of this algorithm was the hunt behavior of falcons when they are
pursuing a prey in flight. Falcons are solitary and their strategy of hunt are based on its
needs, however these strategies follow some patterns being the principal of them the
rules of flight. According to several works of Tucker [43 - 46] the birds of prey, in
which the falcon is included, have distinct paths, curved path, to reach their prey. The
path has two segments: firstly, a logarithmic spiral along which the falcon keeps its
head straight while looking sideways at the prey with maximum visual acuity, and
secondly, a straight segment along which the falcon flies straight to the prey, dives
when it is close enough to view the prey with binocular vision.
So, the falcons basically perform a movement that can be divided in three moments:
first (1st Stage), the seek for preys, secondly (2nd Stage), the logarithmic spiral to adjust
its dive and finally the third (3rd Stage), the dive itself, which can result in good result,
i.e. catch a prey, if not, where the falcon simply return its movement based on its own
experiences. Important to notice that the phenomena of klepto-parasitism between
falcons or birds of prey are common and has received scant attention of the scientific
community so far [47].

2.2 Algorithm

The step-wise procedure for the implementation of FOA is given in this subsection,
where special attention is given to step 4, which holds the movements presented in the
previous steps.

Step 1: Initialize problem data and adjustable control parameters

The optimization problem, decision variables and constraints are defined. Then, the
adjustable parameters of FOA such as quantity of falcons ( ), maximum allowed
speed ( ), values of cognitive ( ), social ( ), following ( ) constants and
awareness and dive probabilities ( ) are provided.

Step 2: Initialize position and velocity of the falcons

The falcons are randomly positioned (using generation of random values with uniform
distribution) in a D-dimensional space respecting the boundary conditions, generating
the matrix:

(1)
where is the falcon position, respecting the quantity of candidates in all its
dimensions.

The velocities are randomly generated between the and the restrictions, where
both are determinate respecting:

(2)

(3)

where and are the maximum and minimum velocities allowed, respectively,
and is the upper bound limit range of the problem for each dimension.

Step 3: Evaluate fitness and find global and individual best positions

For each falcon the fitness value ( ) for the problem is obtained, generating the vector:

(4)

Then, the best individual is set to be the position and for each falcon its best
individual position is set as . These positions will be used to generate the new
positions according to the logic that rules the movement behind the awareness and the
dive probabilities.

Step 4: Generate new positions and update the falcon positions

At first, two random numbers ( ) are generated with uniform distribution for
each falcon for the comparison with awareness and dive probabilities. Then the first
probability compared is the awareness probability, where if is lower than the
falcon performs a movement of seeking for preys according to the its experience and the
other falcon experiences:

(5)

where and are the current position and velocity of the falcon,
respectively. The procedure described is similar to the search performed by the PSO
algorithm.

If is higher than , then the dive probability is compared with . If is

higher than , then the falcon targets one chosen prey ( ) and performs its
initial movement for hunting, the logarithmic spiral given by:

(6)

where is a constant defining the shape of the logarithmic spiral, equals to 1, and is a
random number in the range [-1,1] that defines how much the next position of the falcon
would be close to its actual target.

If is lower than , then the fitness of the chosen prey is compared to the fitness of
the falcon where if the prey is fittest it will be followed by the falcon, similar to a dive
movement:

(7)

If not, the falcon performs a movement according to its own individual best position:

(8)
The representation of the falcon’s movement, according to the previous description, is
presented in Fig. 1.

Figure 1. Falcon hunt behavior representation where can be observed the 1 st Stage (seek
for preys), the 2nd Stage (adjust for dive) and 3rd Stage (proper dive condition).

The new position obtained are then checked concerning velocities and boundary
conditions. After, its new fitness values are computed and the new values of and
are obtained. Important to notice that in all procedure descripted in Step 4, while
analyzing one candidate solution it is considered the phenomena of klepto-parasitism
between falcons. This consideration allows that at each generation one falcon may
consider all others as potential targets, even preys, for the several movements
performed.

Step 5: Finally, after all the evaluations the process is continued until the maximum
number of iterations ( ) thus repeating the Step 4. The pseudo code of the Falcon
Optimization Algorithm (FOA) is shown in Fig. 2.

Falcon Optimization Algorithm

Step 1. Initialize parameters (control hyperparameters):
Population size, ;
Maximum number of generations, ;
Maximum allowed speed, ;
Values of cognitive, , social, , and following, , constants;
Values of awareness ( ) and dive probability ( );
Insert boundaries restrictions , upper, and , lower bounds, and;
Insert objective function.
Step 2. Initialization of the population with random positions and velocities.
Step 3. In for each individual set and its corresponding , and and its corresponding .
Step 4. for to do:
Step 4.1. for to do:
Generate random numbers and .
Select one random individual, .
Step 4.2. if
Step 4.3 if

else
Step 4.4. if

else

end if

end if
end if
end for
Step 4.5. Check velocity and boundaries restrictions;
Step 4.6. Evaluate the fitness for each individual, ;
Step 4.7. if for any particle:

end if
Step 4.8. if any :

end if
end for
Step 5. Output results.
Figure 2. Pseudo code for the Falcon Optimization Algorithm.
3. EXPERIMENTAL SETUP AND HEAT EXCHANGERS MODELS

It has been proved that no search algorithm is the best on average for all problems of
optimization [48]. An algorithm may solve some problems better and some problems
worst in comparison to other algorithms depending to the characteristics of the problem.
The evaluations of the optimization performance of the proposed FOA were made
considering a set of benchmark functions, twelve functions for single-objective
optimization, and four engineering problems of heat exchangers regarding shell-and-
tube and plate-fin heat types.

All the evaluated problems have different natures of objective functions, range of the
decision variables and engineering problems that presents boundary constraints. Falcon
Optimization Algorithm has been executed in the MATLAB® computational
environment on a computer with i7-5500U 2.4 GHz processor with 16 GB RAM
(Random Access Memory). For the benchmark functions simulations were performed
30 independent runs, with initial population of 30 individuals and maximum number of
generations of 1000. However, for the engineering problems 50 independent runs were
performed, initial population of 10 , with generations maximum number of 200. In
FOA were adopted , and equals to 2, and for all simulations
and case studies investigated.

3.1 Benchmark Functions

The benchmark functions are shown in Table 1 and each have different characteristics
such as continuous, separable, differentiable, scalable, unimodal or multimodal. The
unimodal functions are suitable for evaluating the exploitation of the optimization
techniques since they have one global optimum and no local optima, while multimodal
functions are suitable for evaluating exploration and local optima avoidance as they
have many local spots [3].
Table 1. Single-objective benchmark test functions.
Test function Search space Optimal
[lb,ub]D value
Ackley [-32,32] 30 0

Alpine [-10,10] 30 0

Griewank [-600,600]30 0

Levy [-10,10] 30 0
;

Powell [-4,5] 30 0

Rastrigin [-5.12,5.12]30 0

Rosenbrock [-2.048,2.048]30 0

Salomon [-100,100]30 0

Sphere [-100,100]30 0

Styblinski- [-5,5]30 ~-39D

Tang

Sum of [-1,1]30 0
Different
Powers

Sum [-5.12,5.12]30 0
Squares

3.2 Heat exchangers models

The following subsection provides the mathematical models for the heat exchangers
shell-and-tube and plate-fin, respectively.

3.2.1 Shell-and-tube heat exchanger

The thermal modeling of the shell-and-tube heat exchangers were taken from [30] and
its design and geometrical parameters are presented in Figs. 3a and 3b, respectively.
(a)

(b)
Figure 3. (a) Typical configuration of the shell-and-tube heat exchanger. (b) Triangular
tube pitch arrangement of shell-and-tube heat exchanger.

Depending of the flow regime, the tube side heat transfer coefficient and friction factor
are calculated by,
(9)

(10)

where and are the tube thermal conductivity and friction factor, respectively.

The shell side heat transfer coefficient is given as,

(11)

where is the shell hydraulic diameter, or equivalent diameter, given by [49]

(12)

where .

The Reynolds number for tube ( ) and shell ( ) sides can be computed by,

(13)

(14)

Also, the Nusselt Number for tube ( ) and shell ( ) sides can obtained by,

(15)
(16)

The flow velocities for the tube ( ) and shell ( ) sides are given by,

(17)

(18)

where is the number of tube passes, is the tube inside diameter ( ) and
is the number of tubes, determined by,

(19)

where and are coefficients determined based on the flow arrangement (shown in
Fig.3b) and number of passes, where for the studied case, the values are equals to 0.249
and 2.207, respectively, while is the cross-sectional area normal to flow direction
given by,

(20)

The tube side Prandtl number for tube ( ) and shell ( ) sides are determined by,

(21)

(22)

The overall heat transfer coefficient can be determined by,

(23)

The logarithmic mean temperature is obtained from,

(24)

The correction factor is given by,

(25)

where is the correction coefficient, and is the efficiency, are determined,

respectively, by,

(26)

(27)

The heat exchanger surface area is given by,

(28)

For sensible heat transfer, the rate of heat transfer is determined by,

(29)

The tube length is obtained from,

(30)

The total annual cost Ctot is considered as the objective function, including capital
investment ( ), energy cost ( ), annual operating cost ( ), and the total discounted
operating cost ( ).

(31)
The capital investment cost is determined as a function of the heat exchanger surface
area according to,

(32)

for heat exchangers made with stainless steel for both, shell and tubes.

The total discounted operating cost is determined as a function of the pressure drops for
shell and tube sides as,

(33)

(34)

(35)

where , , , and are the energy cost, the amount of annual work hours, the
annual discount rate and the pumping efficiency, respectively.

The shell and the tube pressure drops are given by,

(36)

(37)

where is the friction factor. Important to notice that the tube side pressure drop is
computed as the sum of the distributed pressure drop along the tube length and
concentrated pressure losses in elbows as well as in the inlet and outlet nozzles.

3.2.2 Plate-fin heat exchanger

The thermal modelling for the plate-fin heat exchanger was taken from [36], where the
Fig. 4a presents the typical arrangement of this type of heat exchanger and Fig.4b shows
its geometrical parameters.
(a)

(b)

Figure 4. (a) Arrangement of the plate-fin heat exchanger; (b) Geometrical parameters
of plate-fin heat exchanger.

Considering a cross-flow plate-fin heat exchanger with both fluids unmixed, the heat
transfer rate is calculated by,

(38)

where is the effectiveness, is the minimum heat capacity rate and and are
the inlet temperatures of the hot and cold fluids, respectively.

The effectiveness is given by [50],

(39)

where and is the number of transfer units determined from,

(40)

where is the free flow area, is the heat transfer area, is the Colburn factor, is
the specific heat and is the Prandtl number for each fluid, hot and cold.

The free flow areas can be obtained from,

(41)
(42)

where is the height of the fin, is the fin thickness, is the fin frequency, is the
heat exchanger length and is the number of fin layers for each fluid, remembering
that .

The heat exchangers areas are given by,

(43)
(44)

where the total heat exchanger area .

The Colburn and Fanning friction factors are determined by [51],

(45)
(46)

where is the Reynolds number, , , the fin spacing

for hot and cold fluids.

The Reynolds number for each fluid of the system can be obtained from,

(47)

where is the mass flow rate and is the viscosity of each fluid, remembering that
.

The hydraulic diameter can be determined for both fluids by,

(48)

Also, the frictional pressures drop for the two streams is given by,

(49)

for each one of the fluids.

The convective heat transfer coefficient for both fluids can be determined by,

(50)

The no flow length was computed by,

(51)

where is the plate thickness, considered constant equals to 0.005 m.

The outlet pressures and temperatures for both fluids are given by,

(52)
(53)

(54)

(55)

The objective function is defined as the entropy generation units, that is given by Bejan
[52],

(56)

valid for and , where is the specific gas constant.

4. RESULTS AND DISCUSSION

The following subsections present the results obtained with the simulations for both
single-objective benchmark functions and heat-exchanger cases. For the single-
objective function the comparison algorithms were set as follows: Genetic Algorithm
(GA) (crossover and mutation probabilities equals to 0.9 and 0.001, respectively),
Particle Swarm Optimization (PSO) (cognitive and social constants equals to 2 and
inertia weight equals to 0.5), Bat Algorithm (BA) (both loudness and pulse rate equals
to 0.5) and Firefly Algorithm (FA) ( , and equals to 0.5, 0.2 and 1,
respectively). For better understanding of the comparison algorithms the authors suggest
consulting the references [11, 12, 53, 54].
4.1 Single-objective benchmark problems

The results for the benchmark single-objective optimization are shown in Table 2
alongside to its average and standard deviations values and average processing time.

Table 2. Results obtained for FOA in comparison to GA, PSO, BA and FA for the
benchmark functions for best, average (avg) and standard deviation (std) values for
fitness, average time processing and p-value Wilcoxon ranksum test.
Criteria
Best 1.3852 1.5110 0.9968 0.0168 0.1750 20.5875
Avg. 1.8924 2.9225 1.0459 0.0497 0.4002 36.6592
Std. 0.3452 0.8502 0.0162 0.0307 0.1625 9.9996
GA

Avg. time(s) 2.5885 0.5818 0.6250 1.0927 0.6339 0.5911

p-value 3.7215e-07 1.3352e-09 1.2118e-12 1.2118e-12 1.2118e-12 1.5064e-05
Best 0.8612 0.8215 1.0533 0.0545 0.1899 21.0174
Avg. 1.6623 1.8706 1.0983 2.0823 0.4050 41.1642
PSO

Std. 0.4534 0.4575 0.0380 1.4746 0.1501 13.6616

Avg. time (s) 7.9276 5.8609 5.8734 6.3901 5.7896 5.8557
p-value 3.7215e-07 1.1863e-06 1.2118e-12 1.2118e-12 1.2118e-12 2.8291e-06
Best 12.3141 23.4478 100.6721 7.2178 0.0275 49.7505
Avg. 16.1136 30.4655 184.0465 17.9712 32.8453 91.4395
Std. 1.3517 2.9119 46.6673 6.0254 44.9121 30.7990
BA

Avg. time (s) 2.5604 0.5818 0.6130 1.1401 0.5328 0.4740

p-value 3.7215e-07 8.8290e-04 1.2118e-12 1.2118e-12 1.2118e-12 8.1611e-06
Best 0.0093 0.5398 0.0017 0.0000 0.1473 21.8912
Avg. 0.0134 0.9503 0.0039 0.0061 1.7444 38.4072
Std. 0.0020 0.2912 0.0011 0.0227 1.7165 14.5257
FA

Avg. time (s) 4.7297 2.8078 2.8344 3.2958 2.8036 2.7552

p-value 3.7215e-07 2.2623e-11 1.2118e-12 1.2118e-12 1.2118e-12 3.5656e-11
Best 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
Avg. 2.6619 0.6183 0.0000 0.0000 0.0000 18.8216
FOA

Std. 6.9024 0.9076 0.0000 0.0000 0.0000 14.8799

Avg. time (s) 8.7828 7.0042 6.7995 7.3380 6.8266 6.8370

Best 25.3860 16.9261 2.4142 -1174.7962 0.0000 0.0862

Avg. 27.9151 22.5981 4.7844 -1174.4607 0.0000 0.1557
Std. 0.8640 3.2763 1.9540 0.2332 0.0000 0.0453
GA

Avg. time (s) 0.5969 0.6281 0.6005 0.7833 0.6833 0.5385

p-value 2.8646e-11 1.2118e-12 1.2118e-12 1.0712e-10 1.2118e-12 1.2118e-12
Best 26.3517 23.2967 3.4187 -1089.8305 0.0000 0.1493
Avg. 28.5154 28.9801 11.9498 -1006.2565 0.0000 0.3710
PSO

Std. 0.7806 4.1177 5.3295 41.4020 0.0000 0.1261

Avg. time (s) 5.7698 5.9719 5.7375 5.9052 6.1714 5.7349
p-value 2.8646e-11 1.2118e-12 1.2118e-12 2.9543e-11 1.2118e-12 1.2118e-12
Best 23.0936 39.2966 9120.2529 -999.6799 0.0000 0.0787
Avg. 28.3941 66.9899 18756.4165 -924.5327 0.0000 58.7502
Std. 9.7760 12.8177 5891.0471 47.5878 0.0000 66.2120
BA

Avg. time (s) 0.4771 0.6224 0.4479 0.6052 0.6677 0.4531

p-value 2.8646e-06 1.2118e-12 1.2118e-12 2.9543e-11 1.2118e-12 1.2118e-12
Best 25.8389 4.8910 0.0013 -1132.5747 0.0000 0.0036
Avg. 28.1003 8.6860 0.0025 -1081.2113 0.0000 0.1133
Std. 0.9219 2.1352 0.0006 32.4674 0.0000 0.1289
FA

Avg. time (s) 2.7609 2.8563 2.7313 2.8755 2.9073 2.7250

p-value 2.8646e-11 1.2118e-12 1.2118e-12 2.9542e-11 1.2118e-12 1.2118e-12
Best 0.0007 0.0000 0.0000 -1174.4743 0.0000 0.0000
FOA Avg. 1.0435 0.0000 0.0000 -1170.6883 0.0000 0.0000
Std. 0.6704 0.0000 0.0000 1.1506 0.0000 0.0000
Avg. time (s) 6.7094 6.9047 6.7172 6.8750 6.8948 6.6750
Best results are presented in bold font.

It can be observed in Table 2 that the proposed algorithm obtained best values for
eleven of the twelve benchmark functions tested ( to , and ), outcoming GA,
PSO, BA and FA. The function was better optimized by GA followed by FOA and
the function was optimized for all comparison algorithms alongside FOA. However,
despite the performance of the proposed technique, FOA reached processing time as
high as PSO. The pairwise comparison between FOA and each one of the other
algorithms evaluated for the benchmark functions using the Wilcoxon ranksum test over
all simulations suggest that the proposed algorithm has statistically significant
superiority and that the results do not came from the same population.

Another situation that may be emphasized is the possible reason why FOA did not
performed better in all multimodal function (the exception is the function). This may
be associated to the mechanism of selection of random chosen candidates to perform the
logarithmic spiral and dive motions, that mostly promotes exploration of the search
space and take more generations to local optima avoidance. Considering the results
obtained for some functions, new simulations ( , , ) were
performed where the results were evaluated for the search history, convergence rate and
average fitness for each generation population (Fig. 5).
Figure 5. Function example, search history for best candidate solution per generation
(red points are the best value obtained in final generation), convergence rate and
average (avg) fitness value for every multimodal function.

In this case the first metric has qualitative character, showing the history of sampled
points over the generations progress (black points) until the best value obtained (red
point). The plots show that FOA may explore the search space and exploit the
surroundings of the global optima with high accuracy. The second and third metrics
have quantitative character, where the average fitness over the generation progress is
clearly improved and the convergence rate shows high acceleration to the global optima.
4.2 Shell-and-tube heat exchanger optimization

Two case studies to shell-and-tube heat exchanger were used. The lower and upper
bounds adopted in both studies were the shell internal diameter ( ) ranging from 0.2 m
to 1.0 m, the tubes outside diameter ( ) ranging from 0.008 m to 0.051 m and baffles
spacing ( ) ranging from 0.2 m to 0.5 m. The objective function was computed for the
total discounted operating costs with years, , ,
, and .

The first case study for shell-and-tube heat exchanger was presented by Sinnott [55]
where a system for thermal exchanging with two tube passages and one shell passage
for methanol and brackish water, with 4.43 MW of heat duty, have to be optimized for
minimum total cost. The second case, described in Table 3, consists in the application
from Kern [56] where a heat exchanger with two tube passages and one shell passage
for distilled water and raw water, with 0.46 MW heat duty, have to be designed also for
minimum total cost. The parameters inputs and physical properties of the fluids are
presented in Table 3 for both cases.

Table 3. Shell-and-tube inputs and physical properties of the fluids for two case studies.
Parameter Case 1 Case 2
Shell side Tube side Shell side Tube side
(kg/s) 27.80 68.90 22.07 35.31
Ti (ºC) 95 25 33.9 23.9
To (ºC) 40 40 29.4 26.7
ρ (kg/m³) 750 995 995 995
Cp (kJ/kg.K) 2.84 4.20 4.18 4.18
µ (Pa.s) 0.00034 0.0008 0.0008 0.00092
µw (Pa.s) 0.00038 0.00052 - -
k (W/m.K) 0.19 0.59 0.62 0.62
Rf (m²k/W) 0.00033 0.0002 0.00017 0.00017

The results, in the first case study, for the shell-and-tube optimization aiming the
minimization of the total annual cost were compared with the works of Sinnott et al.
[55], Caputo [21] (GA), Patel & Rao [23] (PSO), Hadidi et al. [26] (ICA), Hadidi et al.
[25] (BBO), Sahin et al. [24] (ABC) and Asadi et al. [13] (CSA). It can be observed in
Table 4 in comparison to the original study [54] that a reduction of 28% was obtained,
while decreases for the variables and and an increase in were achieved. Also,
increasing in heat transfer coefficients for both fluids were obtained. Considering a
more recent case [13], a reduction about 2.5% was obtained for total cost and the same
behavior about the decision variables were observed in relation to the original case
study (reduction of 17.41% in , 33.11% in and increase of 20.77% in ).
However, both heat transfer coefficients and pressure drop for tube side increased their
respective values.

Even if FOA achieved better total cost value, it presented higher operational cost, due to
the pressure drops obtained. It is important to notice that reduction in the heat exchanger
length lead to a decreasing in the heat exchanger surface transfer area, notwithstanding
the number of tubes and the outside diameter of the tubes present increase and decrease,
respectively. Yet, increasing in the number of tubes and decreasing in the shell diameter
provide lower flow velocities, directly linked to the pressure drops.

Table 4. Optimal geometries and thermal properties of the shell-and-tube heat

exchanger case study 1.
Original
GA PSO ICA ABC BBO CSA
Parameter design FOA
[21] [23] [26] [24] [25] [13]
[55]
(m) 0.894 0.830 0.810 0.879 1.3905 0.801 0.826 0.6822
(m) 4.830 3.379 3.115 3.107 3.963 2.040 2.332 2.269
(m) 0.356 0.500 0.424 0.500 0.4669 0.500 0.414 0.500
(m) 0.020 0.016 0.015 0.015 0.0104 0.010 0.0151 0.0101
918 1567 1658 1658 1528 3587 1754 2715
(m/s) 0.75 0.69 0.67 0.699 0.36 0.77 0.65 0.995
14925 10936 10503 10429 - 7643 10031 10000
5.7 5.7 5.7 5.7 - 5.7 5.7 5.7
(W/m² K) 3812 3762 3721 3864 3818 4314 6104 9961
(Pa) 6251 4298 4171 5122 3043 6156 4186 6093
(m) 0.014 0.011 0.0107 0.011 - 0.007 0.0107 0.0072
(m/s) 0.58 0.44 0.53 0.42 0.118 0.46 0.56 0.54
18381 11075 12678 9917 - 7254 13716 8677
5.1 5.1 5.1 5.1 - 5.1 5.1 5.1
(W/m² K) 1573 1740 1950.8 1740 3396 2197 2083 2381
(Pa) 35789 13267 20551 12367 8390 13799 22534 17496
(W/m² K) 615 660 713.9 677 832 755 848.2 888.5
(m²) 278.6 262.8 243.2 256.6 - 229.95 209.1 195.5
(€) 51507 49259 46453 48370 44559 44536 40343.70 39513
(€) 12973 5818 6778.2 5995 6233.8 6046 7281.40 6906
(€) 64480 55077 53231 54366 50793 50582 47625.10 46419
Best result is in bold font.

In the second case study the optimization also aimed the minimization of the total cost
and the results presented in Table 5 are compared with the original design [56], Caputo
et al. [21] (GA), Patel & Rao [23] (PSO), Sahin et al. [24] (ABC) and Hadidi & Nazari
[25] (BBO). Comparing the results obtained with FOA against those ones presented in
the original case study [56] a reduction of 57.8% was achieved alongside increasing in
shell diameter and baffles spacing and for the tube side heat transfer coefficient, while
decreasing were reached for both tube and shell pressure drops and for the shell side
heat transfer coefficient. About a newer work [25], the difference between the values for
the total cost differs 3.9% favoring FOA. Also, FOA presented with a lower value
(6.8%) and variable with a higher value (20%, respectively) while did not
changed. In this case, shell side heat transfer coefficients and pressure drop reached
lower values than the ones presented by [25] (4.65% and 6.7% respectively) while for
tube heat transfer coefficients and pressure drop achieved higher values (1.5% and 9.9%
respectively).

For the second case of shell-and-tube heat exchanger, the increase in the shell diameter
and number of tubes lead to a reduction of the investment cost, since they provide
changes in the heat exchanger length and heat exchangers surface transfer area. It is
important to notice that, similarly to first case study, reductions in the flow velocities
culminate in impacts over the pressure drops, making the operational cost increase, even
if compensated later with the decreasing in the investment cost.

Table 5. Optimal geometries and thermal properties of the shell-and-tube heat

exchanger case study 2.
Original
Parameter GA [21] PSO [23] ABC [24] BBO [25] FOA
design [56]
(m) 0.387 0.62 0.59 1.0024 0.55798 0.520
(m) 4.88 1.548 1.45 2.4 1.133 1.2316
(m) 0.305 0.44 0.423 0.354 0.5 0.500
(m) 0.013 0.016 0.0145 0.0103 0.01 0.0120
160 803 894 704 1565 1017
(m/s) 1.76 0.68 0.74 0.36 0.898 0.9580
36400 9487 9424 - 7804 10000
6.2 6.2 6.2 - 6.2 6.2
(W/m² K) 6558 6043 5618 4438 9180 9317.3
(Pa) 62812 3673 4474 2046 4176 4591
(m) 0.013 0.011 0.0103 - 0.0071 0.0086
(m/s) 0.94 0.41 0.375 0.12 0.398 0.4265
16200 8039 4814 - 3515 4568
5.4 5.4 5.4 - 5.4 5.4
(W/m² K) 5735 3476 4088.3 5608 4911 4682.6
(Pa) 67684 4365 4721 2716 5917 5523
(W/m² K) 1471 1121 1177 1187 1384 1369.5
(m²) 46.6 62.5 59.15 - 55.73 47.65
(€) 16549 19163 18614 17893 18059 16723
(€) 27440 1671 1696 1584.2 1251.5 1837.3
(€) 43989 20834 20310 19478 19310 18560.3
Best result is in bold font.

The convergence rate for cases 1 and 2 applying FOA are shown in Figs. 6a and 6b,
respectively, where the best values are reached after 120 generations for case 1 and 180
generations for case 2, respectively.

(a) (b)
Figure 6. Convergence rate of the best simulation obtained by FOA for the shell-and-
tube heat exchanger study (a) case 1 and (b) case 2.

For both shell-and-tube heat exchangers optimization aiming total cost minimization,
despite the best results obtained by FOA higher values for the operational cost (linked
to the pressure drops) were observed, compensated by lower investment costs related
directly to the heat exchanger area.

Yet, a thermal-hydraulic analysis was performed for each variable for both cases,
considering the variation of one variable on the limit range while the remaining
variables were fixed to the optimization value obtained. The Figs. 7 to 12 presents the
results regarding the convection heat transfer coefficient, pressure drops and Reynolds
and Nusselt numbers.
Figure 7. Variation of the shell diameter in limit range and its influence in
convective heat transfer coefficients, pressure drops, Reynolds and Nusselt numbers for
the shell-and-tube heat exchanger study case 1 (red lines for shell side and blue line for
tube side). The discontinuity of the curves for the convective heat transfer coefficient
and Nusselt number for the tube side are consistent to the flow regime changes for the
shell-and-tube heat exchanger model adopted.
Figure 8. Variation of the shell diameter in limit range and its influence in
convective heat transfer coefficients, pressure drops, Reynolds and Nusselt numbers for
the shell-and-tube heat exchanger study case 2 (red line for shell side and blue line for
tube side). The discontinuity of the curves for the convective heat transfer coefficient,
friction coefficient and Nusselt number for tube side and pressure drop for shell side are
consistent to the flow regime changes for the shell-and-tube heat exchanger model
adopted.
Figure 9. Variation of the baffle spacing in limit range and its influence in convective
heat transfer coefficient, pressure drop and Reynolds number for the shell side for the
shell-and-tube heat exchanger study case 1.
Figure 10. Variation of the baffle spacing in limit range and its influence in
convective heat transfer coefficient, pressure drop and Reynolds number for the shell
side for the shell-and-tube heat exchanger study case 2.
Figure 11. Variation of the outside tube diameter in limit range and its influence in
convective heat transfer coefficients, Reynolds and Nusselt numbers for the shell side
for the shell-and-tube heat exchanger study case 1 (red line for shell side and blue line
for tube side). The discontinuity of the curves for the convective heat transfer
coefficient and Nusselt number for tube side are consistent to the flow regime changes
for the shell-and-tube heat exchanger model adopted.
Figure 12. Variation of the outside tube diameter in limit range and its influence in
convective heat transfer coefficients, Reynolds and Nusselt numbers for the shell side
for the shell-and-tube heat exchanger study case 2 (red line for shell side and blue line
for tube side). The discontinuity of the curves for the convective heat transfer
coefficient and Nusselt number for tube side are consistent to the flow regime changes
for the shell-and-tube heat exchanger model adopted.
From the Fig. 7 and 8 it can be observed that higher values for the shell diameter results
in lower Reynolds numbers for shell, influenced by the shell diameter in the
determination of the cross-sectional area normal to flow direction, and tube, influenced
by the shell diameter in the determination of the flow velocity once this parameter is
dependent to the number of tubes calculation. This behavior contribute to the decay
curves of the convective heat transfer coefficients for shell and tube fluids, which both
presents dependency with the Reynolds numbers, Nusselt numbers (directly dependent
of the convection heat transfer coefficients) and pressure drops, where for the shell exist
explicit dependency with the shell diameter (where there is also contribution by the
Reynolds number and length of the heat exchanger, this last by the inverse relation of
the number of tubes). Yet, the decay in the Reynolds number for tube side results in an
increase on the respective friction factor.

Regarding the baffles spacing (Figs. 9 and 10), the model reveals that this variable
presents influence in the shell side about the Reynolds number (higher values lead to
lower Reynolds numbers), by the determination of the cross-sectional area normal do
the flow direction, resulting in a decay behavior of the convective heat transfer
coefficient and in the pressure drop (where exist also an inverse explicit dependency).

About the outside diameter of tubes (Figs. 11 and 12) there is a dependency of the that
variable with the Reynolds numbers for both shell and tube sides, the first by the
determination of the internal diameter of tubes and the second by the equivalent
diameter. An increase in the outside diameter of tubes lead to higher Reynolds
Numbers, impacting in the convective heat transfer coefficient where a decay behavior
is obtained both shell and tube sides, including the variation in the convective heat
transfer coefficient for the tube side regarding the mathematical model and its
dependency with the Reynolds number and the flow regime. This behavior also impacts
the Nusselt numbers of the fluids.

In general, for case study 1 and 2, despite the decay curve for the Reynolds numbers for
the shell diameter and baffles spacing the flow regime tend to be maintained turbulent,
leaving this situation only for values of shell diameter and baffle spacing very close to
the upper limit range. For the outside tube diameter also corroborates the previous
affirmation about the flow regime once its increase only causes augmentation of the
Reynolds numbers values (even in the lower limit range, the Reynolds numbers
obtained are already close to the turbulent flow regime). Also, the results for the Nusselt
numbers indicates that the process in the entire range of the variables is based on
convective heat transfer process. Yet, the discontinuity of the curves for the convective
heat transfer coefficient, friction coefficient and Nusselt number for tube side and
pressure for hot side are consistent to the flow regime changes for the shell-and-tube
heat exchanger model adopted.

4.3 Plate-fin heat exchanger optimization

Two plate-fin heat exchangers were tested, and the physical properties of the fluids are
presented in Table 6. The first case study for plate-fin heat exchanger was taken from
Shah [57], where a gas-to-air single pass cross-flow heat exchanger, with 1069.8 kW of
heat duty, have to be designed for minimum entropy generation units with maximum
pressure drops of 9500 and 800 Pa for hot and cold fluids, limited to 1x1x1 m. The
second case study investigated is from Kakaç and Liu [58] where a gas-to-air cross flow
plate-fin heat exchanger with a requirement heat duty of 3300 kW and air as fluid on
both sides has to be designed for maximum effectiveness. The heat exchanger is limited
to 2×3×2 m and the maximum pressure drops for hot and cold fluids are 7.5 kPa and 4.5
kPa, respectively.

Table 6. Plate-fin heat exchanger inputs and physical properties of the fluids for two
case studies.
Parameter Case 1 Case 2
Hot fluid Cold fluid Hot fluid Cold fluid
(kg/s) 1.66 2 25.4 25
(ºC) 900 200 460 300
(kPa) 160 200 100 900
(J/kg K) 1122 1073 1060 1060
(kg/m³) 0.6296 0.9638 0.54 4.86
(Ns/m²) 0.0000401 0.0000336 0.000032 0.000032
0.731 0.694 0.69 0.69

For the first case study the lower and upper boundary of the optimization variables were
heat exchanger length for hot and cold fluids ( ) ranging from 0.1 m to 1.0 m, the height
of fin ( ) from 0.002 m to 0.01 m, the fin frequency ( ) from 100 to 1000, the fin
thickness ( ) from 0.0001m to 0.0002 m, the lance length of the fin ( ) from 0.001 m to
0.01 m, and the number of fin layers for the hot fluid ( from 1 to 200.

The results for case 1 are presented in Table 7 and were compared with the original
design [57] and the works of Yousefi et al. [34] (GA and ICA) and Zarea et al. [37]
(BA). It can be observed in comparison to the original study [57] that a reduction of
15.42% was obtained for the objective function, with increases in all decision variables,
except and , pressure drops also presented significant reduction. Considering a
more recent work [37], a better value was achieved alongside increases in the decision
variables and , while decreases were observed for the variables , , , and .
About the pressure drops an increase of 28% and a decrease of 11.47% for cold and hot
fluid streams were obtained, respectively. The optimization achieved better distribution
between the pressure drops for hot and cold streams with a lower difference between
both results. Also, in comparison to the last recent study the heat exchanger the
optimization reached even lower volume (0.8784 m³ for FOA against 0.9343 m³ for
BA) and better effectiveness.

Table 7. Optimal geometries and thermal properties of the plate-fin heat exchanger case
study 1.
Parameter Original GA [34] ICA [34] BA [37] FOA
design [57]
(m) 0.3 0.95 1 0.997 0.900
(m) 0.3 0.44 0.88 0.940 1
(mm) 2.49 7.2 5 8.33 8.6
(fins/m) 782 417 240 257.02 265.2
(mm) 0.10 0.1 0.19 0.166 0.1
(mm) 3.18 7.2 9.6 9.51 7.2
167 57 77 56 53
(kPa) 9.34 4.2 1.23 0.741 0.656
(kPa) 6.90 0.52 0.67 0.460 0.589
(m) 1 0.87 0.87 0.997 0.976
- 0.821 0.821 0.826 0.827
0.1576 0.1416 0.1374 0.1341 0.1333
Best result is in bold font.

For the second case study the results for the gas-to-air plate-fin optimization aiming the
maximization of effectiveness were compared with the original design [58] and Zarea et
al. [37] (BA), like presented in Table 8. The lower and upper bounds for the
optimization variables were the heat exchanger length for hot fluid ( ) ranging from
0.1 m to 2.0 m, the heat exchanger length for cold fluid ( ) ranging from 0.1 m to 3.0
m, and the other parameters have the same values that the previous case study. In this
case, restrictions about the maximum allowable pressure drops and no flow length were
adopted.

Comparing the results, presented in Table 8, obtained with the original case study
design [58] an enhancement about 10% for the effectiveness was reached with increases
in all the decision variable, except and . The pressure drops for hot and cold fluid
streams also presented significant decreases while the no flow length reached value
closer to the constraint. Now, concerning the recent work [37], alongside the better
value obtained for the effectiveness by FOA, increases for the decision variables , ,
and were achieved while for and decreases were observed. About the
pressure drops a slightly reduction is presented for the hot fluid stream and for the cold
fluid stream 10.27% of increasing. For the second plate-fin heat exchanger case, it can
be observed that increases in the length of the streams lead to increasing in the heat
exchangers transfer area, also rising the NTU, impacting the effectiveness of the thermal
transfer process.

Table 8. Optimal geometries and thermal properties of the plate-fin heat exchanger case
study 2.
Parameter Original design [58] BA [37] FOA
(m) 0.9 1.654 1.525
(m) 1.8 2.99 3.00
(m) 0.0057 0.00572 0.01000
(m) 0.00015 0.000169 0.000100
(fins/m) 500 388.9 514
(mm) 0.006 0.00857 0.0097
149 175 94
(kPa) 15 7.5 7.491
(kPa) 10 3.38 3.727
(m) 1.79 1.99 1.983
0.778 0.83 0.8558
Best result is in bold font.

The convergence rate for FOA is illustrated in Figs. 13a and 13b, for case 1 and 2,
respectively, where can be observed that the optimal value was achieved after 50
generations for both cases.
(a) (b)
Figure 13. Convergence rate of the best simulation obtained by FOA for the plate-fin
heat exchanger study (a) case 1 and (b) case 2.

For the plate-fin heat exchangers case studies, despite the distinct difference in the
objective functions adopted, better values were achieved in comparison to the previous
studies and original design. Also, fact to be emphasized was the variation of all the
decision variables promoted by FOA, suggesting the potential for the space search
exploration that the algorithm possesses.

Similarly, to the cases for the shell-and-tube heat exchangers, a thermal-hydraulic

analysis for the plate-fin heat exchanger was evaluated considering the all de decision
variables. Each variable was variated within the limit range while the remaining
variables continued with the optimization value reached. These results are presented in
Figs. 14 to 21 regarding the influence of each variable in the convective heat transfer
coefficients, Colburn and Fanning friction factors, Reynolds number, pressure drops and
some comments regarding the constraints violation for the pressure drops are presented.
Finally, Fig. 22 presents a comparison of the no flow length constraint violation.
Figure 14. Variation of the lengths for hot ( ) and cold fluids ( ) and number of hot
layers ( ) in limit range and its influence in convective heat transfer coefficients,
pressure drops, Colburn and Fanning friction factors and Reynolds numbers for the
plate-fin heat exchanger study case 1 (continuous line for left y axis and dashed line for
right y axis). The dotted lines and arrows indicate the fraction of the curve that are
violating the constraint for the cold stream pressure drop for and for the hot stream
pressure for and .
Figure 15. Variation of the lengths for hot ( ) and cold fluids ( ) and number of hot
layers ( ) in limit range and its influence in convective heat transfer coefficients,
pressure drops, Colburn and Fanning friction factors and Reynolds numbers for the
plate-fin heat exchanger study case 2 (continuous line for left y axis and dashed line for
right y axis). The dotted lines and arrows indicate the fraction of the curve that are
violating the constraint for the cold stream pressure drop for and for the hot stream
pressure for and .
Figure 16. Variation of the fin geometrical parameters ( , , and ) in limit range and
its influence in Colburn and Fanning friction factors and Reynolds numbers for the
plate-fin heat exchanger study case 1 (red line for hot fluid and blue line for cold fluid).
Figure 17. Variation of the fin geometrical parameters ( , , and ) in limit range and
its influence in Colburn and Fanning friction factors and Reynolds numbers for the
plate-fin heat exchanger study case 2 (red line for hot fluid and blue line for cold fluid).
Figure 18. Variation of the fin geometrical parameters ( , , and ) in limit range and
its influence in convective heat transfer coefficients for the plate-fin heat exchanger
study case 1 (red line for hot fluid and blue line for cold fluid).
Figure 19. Variation of the fin geometrical parameters ( , , and ) in limit range and
its influence in convective heat transfer coefficients for the plate-fin heat exchanger
study case 2 (red line for hot fluid and blue line for cold fluid).
Figure 20. Variation of the fin geometrical parameters ( , , and ) in limit range and
its influence in pressure drops for the plate-fin heat exchanger study case 1 (red line for
hot fluid and blue line for cold fluid). The blue dotted lines and arrows indicate the
fraction of the curve that are violating the constraint for the cold stream pressure drop,
while the red dotted lines and arrows indicate the fraction of the curve that are violating
the constraint for the hot stream pressure drop.
Figure 21. Variation of the fin geometrical parameters ( , , and ) in limit range and
its influence in pressure drops for the plate-fin heat exchanger study case 2 (red line for
hot fluid and blue line for cold fluid). The blue dotted lines and arrows indicate the
fraction of the curve that are violating the constraint for the cold stream pressure drop,
while the red dotted lines and arrows indicate the fraction of the curve that are violating
the constraint for the hot stream pressure drop.
(a)

(b)
Figure 22. Violation of the constraint for the no flow length for plate-fin study case 1
(a) and case 2 (b). The dotted lines and arrows indicate the fraction of the curve that are
violating the constraint for the no flow length.
First of all, we have to consider that the length of the plate-fin heat exchanger for hot
fluid impacts the parameters for the cold one and vice-versa, while the number of hot
and cold layers impacts their respective fluid. Explained that we can observe (from Figs.
14 and 15) that greater heat exchanger lengths, no matter if its hot or cold, lead to lower
Reynolds numbers once exist dependency of that parameter with the free flow areas.

This fact causes an increasing behavior for the Colburn and Fanning friction factors,
where the first influence the convective heat transfer coefficient. Despite this influence,
the variation caused by the Colburn factor is weaker of the that caused by the free flow
area, culminating in a decay curve for the convective heat transfer coefficient. The
Fanning friction factors contributes for the pressure drops, but also not enough to
overcome the influence of the free flow areas inverse relation for this parameter.

Similarly, the number of hot layers directly contribute for the free flow area for the hot
fluid, decreasing the Reynolds number as its value increases influencing the Colburn
and Fanning friction factors, the convective heat transfer coefficients and the pressure
drops in an equivalently to the analysis for the lengths of the plate-fin heat exchanger.

The variables related to the fins (Figs. 16 and 17) work primarily on the hydraulic
diameter, which influence directly the Reynolds numbers and consequently the Colburn
and Fanning friction factors. We may not forget the relations for the fin spacing and
geometrical parameters , and (where higher decreases , higher decreases ,
higher decreases that decreases , and higher decreases but increases , and
).

Considering the results for the Colburn factor and those obtained for the convective heat
transfer coefficient (Figs. 18 and 19) we can infer that the thickness of the fin has weak
of low influence on the convective heat transfer coefficient for both cases.

Similar behavior is observed again for the thickness of the fin for the pressure drops
(from Figs. 20 and 21), considering its influence for the Fanning friction factor.

From the analysis presented, some observations may be highlighted, such as that the
increases in the lengths of the heat exchanger lead to non-turbulent flow regime. Yet,
from the four variables related to the fin geometry those that has greater influence on
the convective transfer coefficient are the fin height and density of fins for plate-fin heat
exchanger case study 1, and for the plate-fin case study 2 the lance length of the fin may
be included in this matter. Also, the variables linked to the fin geometry that present
higher variation for the pressure drops are the height of the fin and the density of fins.
The thickness of the fin, in both cases, presented low influence on the Colburn factor
and Fanning friction factor.

Taking account the constraint for pressure drops and no flow length, for the plate-fin
case study 1, the pressure drop for cold stream constraint begin to be violated for
lower than 0.712 m, lower than 0.0072 m, lower than 0.0031 m and higher than
343 fins/m, while the pressure drop constraint for the hot stream present this issue for
lower than 0.145 m, for a very few layers for the hot fluid (less than 9 layers) and
values closer to the lower limit range (0.002 m).

About the plate-fin case study 2, the pressure drop for cold stream constraint begin to be
violated for lower than 1.335 m, lower than 0.0088 m, for values close to 0.006
m and lower, close to 0.0017 and higher and close and superior to 577 fins/m, while
the pressure drop constraint for the hot stream present this issue for lower than 94
and closer and superior to 514 fins/m (for , , and the constraint belong to the
limit range achieved in the optimization).

Regarding the no flow length constraint, the case study 1 presents violation for very
high values for the height of the fin (superior to 0.0088 m) and for a number of hot
layers superior of 55 layers, approximately. For the case study 2 this violation occurs
only for values of closer and superior to 93 layers.

5. CONCLUSION AND FUTURE RESEARCH

Based on the hunt behavior of falcons, a novel metaheuristic algorithm, called Falcon
Optimization Algorithm (FOA), was proposed in this paper. FOA is a population-based
algorithm that needs the adjustment of few parameters. The results obtained showed that
the proposed algorithm reached good results both single-objective, achieving better
values against the other algorithm tested (GA, PSO, BA and FA) with competitive time
processing. Considering the results achieved in the single-objective optimization the
novel proposed technique it was applied in heat exchangers shell-and-tube and plate-fin
types, obtaining better results than previous works for the objective functions total cost
for shell-and-tube heat exchanger (28% and 57.8% of reduction for cases 1 and 2,
respectively) and number of entropy generation units (15.42% of reduction for case 1)
and effectiveness (10% of increase for case 2) for plate-fin heat exchanger type. Also,
the simulations suggested the great potential that FOA has to explore de space search
for distinct variables, continuous or discrete. Also, the thermal-hydraulic analysis was
able to provide the influence of each variable about convective heat transfer
coefficients, pressure drops, Reynolds and Nusselt numbers and friction coefficient for
the shell-and-tube heat exchangers, and for the plate-fin heat exchangers about the
convective heat transfer coefficients, pressure drops, Colburn and Fanning friction
factors and Reynolds numbers and constraints violation.

For future research multiobjective optimization could be explored under capacity,

convergence and diversity point of views for an algorithm with few adjustment
parameters, or even self-adjustable parameters, for the Pareto front achievement.

ACKNOWLEDGEMENTS

The authors would like to thank National Council of Scientific and Technologic
Development of Brazil - CNPq (grant: 152503/2018-8-PDJ) for the scholarship
financial support of this work, and grants: 303908/2015-7-PQ, 303906/2015-4-PQ,
404659/2016-0-Univ, 405101/2016-3-Univ) and Araucaria Foundation (PRONEX –
042/2018) for financial support of this work. Furthermore, the authors wish to thank the
Editor and anonymous referees for their constructive comments and recommendations,
which have significantly improved the presentation of this paper.

REFERENCES

[1] I. Boussaid, J. Lepagnot, P. Siarry, A survey on optimization metaheuristics,

Information Sciences 237 (2013) 82-117.
[2] S. Mahdavi, M. E. Shiri, S. Rahnamayan, Metaheuristics in large-scale global
continues optimization: A survey, Information Sciences 295 (2015) 407-428.

[3] S. Mirjalili, Moth-flame optimization algorithm: A novel nature-inspired heuristic

paradigm, Knowledge-Based Systems 89 (2015) 228-249.

[4] K.T. Gkaragkounis, E.M. Papoutsis-Kiachagias, K.C. Giannakoglou, The

continuous adjoint method for shape optimization in Conjugate Heat Transfer problems
with turbulent incompressible flows, Applied Thermal Engineering, vol. 140, pp. 351-
362, (2018).

[5] S.-M. Lee, K.-Y. Kim, S.-W. Kim, Multi-objective optimization of a double-faced
type printed circuit heat exchanger, Applied Thermal Engineering, vol.60, pp. 44-50,
(2013).

[6] A. Biswas, K. K. Mishra, S. Tiwari, A. K. Mishra, Physics-Inspired Optimization

Algorithms: A Survey, Journal of Optimization (2013), 16 pps.

[7] I. Fister Jr., X.-S. Yang, I. Fister, J. Brest, D. Fister, A Brief Review of Nature-
Inspired Algorithms for Optimization, Elektrotehniski Vestnik 80, 3, (2013) 1-7.

[8] S. Salcedo-Sanz, Modern meta-heuristics based on nonlinear physics processes: a

review of models and design procedures, Physics Reports, 655 (2016) 1-70.

[9] F. Glover, Future paths for integer programming and links to artificial intelligence,
Computers & Operations Research 13 (1986) 533–549.

[10] R. Storn, K. V. Price, Differential evolution – A simple and efficient heuristic for
global optimization over continuous spaces, Journal of Global Optimization 11 (1997)
341-359.

[11] J.H. Holland, Genetic algorithms, Scientific American. 267 (1992) 66–72.

[12] R.C. Eberhart, J. Kennedy, A new optimizer using particle swarm theory, in:
Proceedings of the Sixth International Symposium on Micro Machine and Human
Science (1995), Nagoya, Japan, 39–43.

[13] M. Asadi, Y. Song, B. Sunden, G. Xie, Economic optimization design of shell-and-

tube heat exchangers by cuckoo-search-algorithm, Applied Thermal. Engineering. 73
(2014) 1032-1040.
[14] R. V. Rao, V. K. Patel, Thermodynamic optimization of cross-flow plate-fin heat
exchanger using particle swarm optimization, International Journal of Heat Transfer 49
(2010) 1712-1721.

[15] G. N. Xie, B. Sunden, Q. W. Wang, Optimization of compact heat exchangers by a

genetic algorithm, Applied Thermal Engineering 28 (2008) 895-906.

[16] M. Mishra, P. K. Das, S. Sarangi, Second law based optimisation of crossflow

plate-fin heat exchanger design using genetic algorithms, Applied Thermal Engineering
29 (2009) 2983-2989.

[17] P. Wildi-Tremblay, L. Gosselin, Minimizing shell-and-tube heat exchanger cost

with genetic algorithms and considering maintenance, International Journal of Energy
Research 41 (2007) 867-885.

[18] B. Allen, L. Gosselin, Optimal geometry and flow arrangement for minimizing the
cost of shell-and-tube condensers, International Journal of Energy Research 32, n. 10,
(2008) 958-969.

[19] P. D. Chauduri, U. Diwekar, J. Logsdon, An automated approach for the optimal

design of heat exchangers, Industrial & Engineering Chemistry Research 36 (1997)
3685-3693.

[20] R. Selbas, O. Kizilkan, M. Reppich, A new design approach for shell-and-tube heat
exchangers using genetic algorithms from economic point of view, Chemical
Engineering and Processing 45 (2006) 268-275.

[21] A. C. Caputo, P. M. Pelagagge, P. Salini, Heat exchanger design based on

economic optimization, Applied Thermal Engineering 27 (2007) 1151-1159.

[22] M. Fesanghary, E. Damangir, I. Soleimani, Design optimization of shell and tube

heat exchangers using global sensivity analysis and harmony search algorithm, Applied
Thermal Engineering, 29(5-6) (2009) 1026-1031.

[23] V. K. Patel, R. V. Rao, Design optimization of shell-and-tube heat exchanger using

particle swarm optimization technique, Applied Thermal Engineering 30 (2010) 1417-
1425.
[24] A.S. Sahin, B. Kilic, U. Kilic, Design and economic optimization of shell and tube
heat exchangers using Artificial Bee Colony (ABC) algorithm, Energy Conversion and
Management 52 (2011) 3356-3362.

[25] A. Hadidi, A. Nazari, Design and economic optimization of shell-and-tube heat

exchangers using biogeography-based (BBO) algorithm, Applied Thermal Engineering
51 (2013) 1263-1272.

[26] A. Hadidi, M. Hadidi, A. Nazari, A new design approach for shell-and-tube heat
exchangers using imperialist competitive algorithm (ICA) from economic point of view,
Energy Conversion and Management 67 (2013) 66-74.

[27] M. Amini, M. Bazargan, Two objective optimization in shell-and-tube heat

exchangers using genetic algorithm, Applied Thermal Engineering 69 (2014) 278-285.

[28] D. K. Mohanty, Application of firefly algorithm for design optimization of a shell

and tube heat exchanger from economic point of view, International Journal of Thermal
Sciences 102 (2016) 228-238.

[29] D. K. Mohanty, Gravitational search algorithm for economic optimization design

of a shell and tube heat exchanger, Applied Thermal Engineering 107 (2016) 184-193.

[30] E. H. Vasconcelos Segundo, A. L. Amoroso, V. C. Mariani, L. S. Coelho,

Economic optimization design for shell-and-tube heat exchangers by a Tsallis
differential evolution, Applied Thermal Engineering 111 (2017) 143-151.

[31] J. J. C. Barros, M. L. Coira, M. P C. López, A. C. Gochi, Sustainability

optimisation of shell and tube heat exchanger, using a new integrated methodology,
Journal of Cleaner Production 200 (2018) 552-567.

[32] M. O. Petinrin, T. Bello-Ochende, A. A. Dare, M. O. Oyewola, Entropy generation

minimization of shell-and-tube heat exchanger in crude oil preheat train using firefly
algorithm, Applied Thermal Engineering 145 (2018) 264-276.

[33] R. V. Rao, V. K. Patel, Thermodynamic optimization of plate-fin heat exchanger

using teaching-learning-based optimization (TLBO) algorithm, International Journal of
Advances in Thermal Sciences and Engineering 2 (2011) 91-96.
[34] M. Yousefi, A. N. Darus, H. Mohammadi, Second law based optimization of plate
fin heat exchanger using Imperialist Competitive Algorithm, International Journal of the
Physical Sciences 6, n.20, (2011) 4749-4759.

[35] M. Yousefi, R. Enayatifar, A. N. Darus, Optimal design of plate-fin heat

exchangers by a hybrid evolutionary algorithm, International Communications in Heat
and Mass Transfer 39 (2012) 258-263.

[36] M. Yousefi, R. Enayatifar, A.N. Darus, A. H. Abdullah, Optimization of plate-fin

heat exchangers by an improved harmony search algorithm, Applied Thermal
Engineering 50 (2013) 877-885.

[37] H. Zarea, F. M. Kashkooli, A. M. Mehryan, M. R. Saffarian, E. N. Beherghani,

Optimal design of plate-fin heat exchangers by a Bees Algorithm, Applied Thermal
Engineering 69 (2014) 267-277.

[38] S. Banooni, H. Zarea, M. Molana, Thermodynamic and Economic Optimization of

Plate Fin Heat Exchangers Using the Bees Algorithm, Heat Transfer – Asian Research
43 (5) (2014) 427-446.

[39] E. H. Vasconcelos Segundo, A. L. Amoroso, V. C. Mariani, L.S. Coelho,

Thermodynamic optimization design for plate-fin heat exchangers by Tsallis JADE,
International Journal of Thermal Sciences 113 (2017) 136-144.

[40] H. Peng, X. Ling, E. Wu, An Improved Particle Swarm Algorithm for Optimal
Design of Plate-Fin Heat Exchangers, Industry & Engineering Chemistry Research 49
(2010) 6144-6149.

[41] L. Zhang, C. Yang, J. Zhou, A distributed parameter model and its application in
optimizing the plate-fin heat exchanger based in the minimum entropy generation,
International Journal of Thermal Sciences 49 (2010) 1427-1436.

[42] S. Ghosh, I. Ghosh, D. K. Pratihar, B. Maiti, P. K. Das, Optimum stacking pattern

for multi-stream plate-fin heat exchanger through a genetic algorithm, International
Journal of thermal Sciences 50 (2011) 214-224.

[43] V. A. Tucker, Gliding flight: speed and acceleration of ideal falcons during diving
and pull out, The Journal of Experimental Biology 201 (1998), 403-414.
[44] V. A. Tucker, Gliding flight: drag and torque of a hawk and a falcon with straight
and turned heads, and a lower value for the parasite drag coefficient, The Journal of
Experimental Biology 203 (2000), 3733-3744.

[45] V. A. Tucker, The deep fovea, sideways vision and spiral flight paths in raptors,
Journal of Experimental Biology 203 (2000) 3745-3754.

[46] V. A. Tucker, A. E. Tucker, K. Akers, J. H. Enderson, Curved flight paths and

sideways vision in peregrine falcons (Falco Peregrinus), The Journal of Experimental
Biology 203 (2000), 3755-3763.

[47] White, C. M., N. J. Clum, T. J. Cade, W. G. Hunt, Peregrine Falcon (Falco

peregrinus), In A. Poole and E. Gill. Eds. The Birds of North America, No. 660, The
Birds of North America Inc. Philadelphia, PA USA (2002).

[48] D. H. Wolpert and W. G. Macready, No Free Lunch Theorems for Optimization,

IEEE Transactions on Evolutionary Computation, vol. 1, no. 1, (1997).

[49] R.K. Sinnott, J.M. Coulson, J.F. Richardson, Chemical Engineering Design, vol. 6,

Butterworth-Heinemann, Boston, USA, 1996.

[50] F. P. Incropera, D. P. DeWitt, Fundamentals of Heat and Mass Transfer, John

Wiley and Sons Inc., Hoboken, USA, 2010.

[51] R. Manglik, A. Bergles, Heat transfer and pressure drop correlations for the
rectangular offset strip fin compact heat exchanger, Experimental Thermal and Fluid
Science 10, n.2 (1995) 171-180.

[52] A. Bejan, Entropy generation units: the method of thermodynamic optimization of

finite-size systems and finite-time process, CRC Press, 1995.

[53] X. S. Yang, A new metaheuristic bat-inspired algorithm. In: Gonzalez JR et al.,

editors. Nature-inspired cooperative strategies for optimization (NICSO 2010).
Springer, SCI 284 (2010), 65–74.

[54] X. S. Yang, Firefly algorithm, stochastic test functions and design optimization,
International Journal of Bio-Inspired Computation 2 (2010), 78–84.

[55] R. K. Sinnott, J. M. Coulson, J. F. Richardson, Chemical Engineering Design, vol.

6, Butterworth-Heinemann, Boston, USA, 1996.
[56] D. Q. Kern, Process Heat Transfer, McGraw-Hill, 1950.

[57] R.K. Shah, D.P. Sekulic, Fundamentals of Heat Exchanger Design, John Wiley &
Sons Inc., Hoboken, USA, 2003.
[58] S. Kakac, H. Liu, Heat Exchanger Selection Rating and Thermal Design, CRC
Press LLC, Florida, 2002.
Highlights

> A new metaheuristic based on the hunting behavior of falcons (FOA) is proposed.
> FOA was applied efficiently in several benchmark functions for single-objective.
> Total cost decreasing 28 and 57.8% for shell-and-tube heat exchanger cases using
FOA.
> EGU units reduced 15.42% for plate-fin heat exchanger case 1 using FOA.
> Effectiveness increasing 10% for plate-fin heat exchanger case 2 using FOA.