http://pubs.acs.

org/journal/acsodf Article

Optimization of a Thermal Cracking Reactor Using Genetic

Algorithm and Water Cycle Algorithm
Peyman Roudgar Saffari, Hesamoddin Salarian, Ali lohrasbi, Gholamreza Salehi,*
and Mohammad Hasan Khoshgoftar Manesh

Cite This: ACS Omega 2022, 7, 12493−12508 Read Online

ACCESS
See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.

Metrics & More Article Recommendations

ABSTRACT: With the global production of 150 million tons in

2016, ethylene is one of the most significant building blocks in
Downloaded via 176.232.224.204 on November 17, 2023 at 00:45:45 (UTC).

today’s chemical industry. Most ethylene is now produced in

cracking furnaces by thermal cracking of fossil feedstocks with
steam. This process consumes around 8% of the main energy used
in the petrochemical industry, making it the single most energy-
intensive process in the chemical industry. This paper studies a
tubular thermal cracking reactor fed by propane and the molecular
mechanism of the reaction within the reactor. After developing the
reaction model, the existing issues, such as the reaction, flow,
momentum, and energy, were resolved by applying heat to the
outer tube wall. After solving the entropy generation equations, the
entropy generation ratio of the sources was evaluated. The
temperature of the tube/reactor was tuned following the reference results, and processes were replicated for different states. The
verification of the modeling and simulation results was compared with the industrial case. The Genetic Programming (GP) machine
learning approach was employed to generate objective functions based on key decision variables to reduce the computational time of
the optimization algorithm. For the first time, this study has proposed a systematic approach for optimizing a thermal cracking
reactor based on a combination of Genetic Programming (GP), Water Cycle Algorithm (WCA), and Genetic Algorithm (GA). In
this regard, multiobjective optimization was performed based on the maximization of the products and entropy generation with the
generation of GP objective functions. The key decision variables in this study included inlet gas temperature, inlet gas pressure, air
mass flow rate, and wall temperature. The results showed that the weighted percentage of products after optimization increased to
61.13% and the entropy production rate of the system decreased to 899.80 J/s, displaying an improvement of 0.85 and 16.51%
compared with the base case, respectively, with the multiobjective GA algorithm. In addition, by applying the multiobjective WCA,
the weighted percentage of products increased to 61.81%. The entropy production rate of the system decreased to 882.72 J/s. So, an
improvement of 1.97% in weights of products and an improvement of 18.77% in entropy generation have been achieved compared
with the base case.

1. INTRODUCTION They intended to enhance the steady-state to gain maximum

The thermal cracking of hydrocarbons is the most common form advantage. Gao et al.2 modeled the tubular reactor in a naphtha
of olefin production. The thermal cracking of propane is cracking furnace to optimize it using the HYSYS software
conducted using a radical mechanism that is highly studied and package. Their method also enhanced the operation profitability
researched by Zhou et al.,1 who conducted an entropy considerably. Process assessment was conducted under steady-
generation minimization in a solid reactor based on construction state circumstances. However, the real-world coke formation is
theory; their model consisted of 10 reactions. The molecular time-dependent, affecting the efficiency; they did not consider
model of reactions performs quickly and is broadly applied in ref coke thickness in the heat transfer coefficient.
2 for the simulation of propane thermal cracking. An industrial
cracker was simulated in ref 2 using the presented scheme. They Received: August 18, 2021
developed a molecular model for nine reactions with nine Accepted: March 16, 2022
components. Masoumi et al.3 studied the thermal cracking Published: April 2, 2022
furnace and monitored its dynamic behavior. To this end, they
built a pilot plant controlled by a computer system. The steady-
state model had one dimension, 543 reactions, and 90 elements.
© 2022 The Authors. Published by
American Chemical Society https://doi.org/10.1021/acsomega.1c04345
12493 ACS Omega 2022, 7, 12493−12508
ACS Omega http://pubs.acs.org/journal/acsodf Article

Ghasemi et al.4 used a 3D model and computational fluid Johannessen and Kjelstrup16 minimized entropy generation in
dynamics to study the ethylene furnace. The temperatures of the the reactor using optimal control theory. They considered
tube skin and heat flux profiles were measured by solving the temperature a control variable and used governing equations of
mass, momentum, and energy equations corresponding to the plug flow reaction and SO2 oxidation to show a way to
furnaces in Ansys Fluent. The results indicated that by minimize the entropy generation rate in plug flow reactors by
increasing the fuel rate, outlet propylene yield decreased, while optimal control theory. It is demonstrated that decreasing the
process gas temperature, pressure drop, propane conversion, entropy generation rate up to 25% can be obtained by changing
and ethylene yield were increased along the reactor tube. In the reactor length and controlling the utility. A hypothesis was
addition, increasing reactor feed flow rate increased the desired proposed that a reactor with high energy efficiency had relatively
product yields, despite the reduction in the coil outlet long sections with constant entropy generation rate and driving
temperature of the reactor. forces.
Ghasghaee and Karimzadeh5 attempted to use a dynamic Kingston and Razzitte17 investigated the unique kinds of
model to explain why cracks show an unsteady-state behavior continuously stirred tank reactors (CSTRs) and plug flow
during their start-up in furnaces. Their mathematical model, the reactors (PFRs). They also verified the temperature and
convection model, comprised four sub-models: cylindrical coil, pressure and gave a graphical picture of entropy generation in
combustion chamber, and shell and tube. They captured a two- chemical reactions to constant volume, allowing access to
dimensional zone to predict the factors at a start-up run using different options efficiently. Also, they indicated that by dividing
these four sub-models. Zheng et al.6 calculated the heat flow and a reactor into two smaller ones, with different operating
temperature distribution using CFD 3D modeling in the thermal temperatures, the entropy generation decreased, going as near
cracking reactor and furnace and compared the results using as 48% less in the case of a CSTR and PFR in series, and
spectrometry. Barazandeh et al.7 investigated a liquid furnace achieving 58% with two CSTR.
using a 1-D mathematical model to analyze the furnace Chen et al.18 produced light olefins using CO2 hydrogenation
temperature profile and how it affects the feed-to-product to minimize entropy production. Røsjorde et al.19 attempted to
conversion percentage. minimize entropy production in propane dehydrogenation
Heat transfer and its entropy generation have gained vast using different tower condenser reactor and exchanger
attention in the research community. Entropy induction components. The structural theory to model and minimize
limitations are variable in a closed system concerning the entropy generation has been used. Abdous et al.20 analyzed the
applied law of heat transfer. Closed systems are important since same problem in helically coiled tubes under flow boiling
they are simple, depict the main principles, and investigate the conditions under a constant heat flux. Kurnia et al.21 applied this
process. Nummedal and Kjelstrup8 employed the thermody- analysis to various cross-sections.
namic equipartition of force (EOF) theory to study entropy Vandewalle et al.22 used computational fluid dynamic (CFD)
generation in a heat exchanger. They showed that the optimal fouling in steam cracking reactors and developed a 3D algorithm
operation of one heat exchanger was the operation that was to simulate coke formation in steam cracking reactors. They
known for long from experience, namely the one given by examined coke layer growth to analyze a reactor’s run length and
specific counter-current flows. The result was aligned with the determined that the ribbed reactor design had the most
EOF principle. extended run length.
Sauar et al.9 and Kjelstrup and Island10 also used the EOF Mu and Gu23 studied the optimal modeling of thermal
principle for a chemical reactor. The EOF theory holds that the cracking furnaces based on the enriched Kumar model by
entropy generated by a system is minimum when the considering free-radical reactions. An improved search engine
thermodynamic forces are constant in terms of time or space. algorithm (IPR) was developed to compute the importance of
This theory applies to processes in which independent events do substances in the Kumar model for efficient model selection.
not communicate. However, the theory of EOF fails to justify the The proposed model indicated that the new model achieved a
comparison between the real minimum entropy generated in a mean relative error (MRE) of less than 0.1% compared to 5% in
chemical reactor and the law of mass conservation issues in the the Kumar model. The proposed model could be applied to
distillation tower.11 modeling extensive feedstocks with high accuracy.
Numerous research studies have addressed the entropy Zhou et al.24 optimized the ethane thermal cracking furnace
generation problem using the governing equations,12−14 via the coupling of the reaction network. The residence time
Nummedal and Kjelstrup8 examined the generation of entropy related to the minimum by-product was determined to be 0.4 s,
and the process of heat transfer using EOF and EOEP methods. and its by-product was 4.3% less than that at the initial residence
Sauar et al.9 and Kjelstrup et al.10 applied EOF theory in time (0.3 s).
chemical reactors as an appropriate approach for analyzing Gold yttrium oxide nanorods for oxidative catalytic cracking
them. Wilhelmsen et al. investigated the energy-efficient reactor of n-propane to light olefins have been studied by Narasimharao
design simplified with entropy analysis.15 The mentioned and Alshehri.25 Gold nanoparticles loading has affected the
studies also model a set of guidelines to achieve reactor design activity of oxidative catalytic cracking. In addition, simple Au-O-
with energy efficiency, which can be applied once the best Y species and Lewis acid sites are provided for the activity.
available heat transfer coefficients have been reached. The Ebrahimian et al.26 proposed an innovative reactor concept
optimal design was related to the heat transfer coefficient relative for thermal coupling of naphtha reformation with propane
size across the wall of the tubular reactor and heat transfer ammoxidation. The production of aromatics, hydrogen, and
coefficients in heat exchangers. acrylonitrile have been considered. The evaluation of the
The main drawbacks are the nonlinear relations of the flow thermally coupled reactors with the conventional reactors has
and force concerned with the chemical reactions. Wilhelmsen et been performedeliminating traditional furnaces in naphtha
al.16 studied hydrogen producer steam reformers extensively. reforming.
12494 https://doi.org/10.1021/acsomega.1c04345
ACS Omega 2022, 7, 12493−12508
ACS Omega http://pubs.acs.org/journal/acsodf Article

Figure 1. Hydrocarbon thermal cracking process.34

A comprehensive investigation on hydrogen production reduced chemistry and post-processing over the CFD fields,
through propane steam reforming inside a reactor has been proving that realistic predictions can be made in this setting
performed by Barnoon et al.27 The results demonstrate that the using plant data at the stack.
hydrogen generation yield can change from 77.5 to 92.2%. The For an ethylene-cracking furnace with many faults and
reaction rate can be varied by the velocity and temperatures of exceptional operation conditions, a two-level completely
the hot gases. Also, for T = 900 K, full propane used is obtained energy-efficient quantitative diagnosis method was developed
at the reformer outlet. by Meng et al.33 The goal of energy efficiency diagnosis was to
Simulation and dynamic optimization to find optimum determine the severity and root causes of various types of
operational parameters of the industrial tubular reactor for inefficiency. More accurate findings are obtained when a
propane cracking have been investigated by Berreni and Wang.28 mechanism and a data-driven multiple benchmark criteria are
This work was performed in gPROMS. The steady-state and coupled.
dynamic optimizations were respectively performed. The results Saffari et al.34 assessed the entropy generation in a thermal
show that dynamic optimization can improve the net profit by cracking reactor. The results indicated for two states that the
10.6% rather than the base case. In addition, the computation three factors were in the order of chemical reactions, heat
demand by dynamic optimization was much wider and higher transfer, and pressure loss in terms of their contribution to
than the steady-state optimization. entropy generation. The variation of feed temperature did not
A coupled CFD simulation method for investigating the affect the entropy generated by the chemical reactions.
pyrolysis process in industrial naphtha thermal cracking furnaces However, the entropy generation in the reactor with optimum
has been performed by Rezaeimanesh et al.29 A typical k−ε wall temperature was the lowest. Also, entropy was higher at the
turbulence model is combined with the molecular kinetic inlet of the reactor when the wall had an optimal temperature.
response for cracking, a thorough combustion model, and Previous studies rely on basic principles of flow, chemical
radiative properties in this full CFD model. The simulation reactions, and fundamental energy equations, which do not
results corresponded well with industrial data acquired from a involve minimization of entropy generation and its adequation
mega olefin plant of a petrochemical complex in terms of for the process. A review of previous research shows no
temperature, product yield, and especially propylene-to-ethyl- investigation of optimizing thermal cracking reactors by
ene ratio (P/E). The difference between P/E resulting from addressing maximum productivity and minimum entropy
industrial data was less than 2%. generation. In addition, no systematic approach has been
From the standpoint of intelligent manufacturing, knowledge adopted for the multiobjective optimization of thermal cracking
expression, numerical modeling, and optimization application of reactors. The innovation of this study is that it has, for the first
ethylene thermal cracking have been investigated by Zhang et time, addressed the maximization of productivity, and
al.30 This study presented a summary of advances and minimization of entropy generation have been considered
contributions to PSE-assisted thermal cracking production; it simultaneously. In this regard, the second law of thermody-
introduces the frameworks, methodologies, and algorithms namics and the entropy production rate were employed.
suggested over the last 10 years; and it explores the benefits, Considering entropy generation sources, the basic factors, and
limitations, and industrial applications. mitigating its generation, enhance the process that may lead to a
Yuan et al.31 performed advanced exergy analysis to assess the better economization. Also, a systematic approach was adopted
energy-saving potential of industrial ethylene cracking furnaces. based on the genetic programming (GP) technique for the
Inside the furnace, a simulation is run that considers the multiobjective optimization of thermal cracking reactors based
interactions of three different components. The highest exergy on key parameters based on minimizing entropy generation and
degradation occurs during the combustion process in the maximizing products. The GP technique reduced the run time
radiation section. High avoidable exergy destruction in tube and obtained high convergence to two objectives optimization.
reactors and combustion processes have been found.
Monitoring of combustion in an industrial cracking furnace
using a combination of CFD and visual techniques has been 2. THERMAL CRACKING PROCESS
investigated by Rebordinos et al.32 A new methodology based on Lighter product yield when the hydrocarbon feed flows at a high
CFD calculations of OH− and CH− radicals via decreased speed within a thermal cracking reactor at 800−900 °C with
chemical kinetics and industrial-scale experimental validation dilution steam. The reactor outlets are mainly ethylene and
using flame spectroscopic measurements and UV CCD cameras propylene, as well as olefin, butadiene, and aromatics. The outlet
have been developed to monitor burner efficiency. Second, NOx reactor products rapidly cooled down to 300 °C to prevent
emissions have been simulated using the standard approach of unwanted reactions. High-pressure steam was generated using
12495 https://doi.org/10.1021/acsomega.1c04345
ACS Omega 2022, 7, 12493−12508
ACS Omega http://pubs.acs.org/journal/acsodf Article

Table 1. Study Reactor’s Specifications28

parameters/variable value
length of the coil in the radiant section 95 [m]
length of the straight portion of the coil 8.85 [m]
length of the bends 0.554 [m]
the radius of the bends 0.178 [m]
tube internal diameter 0.108 [m]
wall thickness 0.008 [m]
total feed per coil 0.7635 [kg/s]
steam dilution rate 0.4 [kg steam/kg propane]
Figure 2. Schematic illustration of a thermal cracking reactor.34

πdt2 ijj yz πd 2
inlet pressure 3 [bar]

= jjj∑ αijri zzzz t

inlet temperature 873.15 [K]

j i z 4
dFj

k {
= Rj
dz 4 (1)
the resulting energy. Then water and oil were injected to reduce where ri represents the speed of the ith reaction. The following
the temperature of the products to virtually room temperature equation captures ri:
and split up the heavy elements. The pressure of the outlets was
|αij|
then increased to 35 bar. They were sent for fractionation to split ri = K i ∏ c j (2)
up the primary products.35
Figure 1 is a representation of the process. Thermal cracking The molar flow/pressure rate should be considered to apply the
reactors are influenced by various performance factors, such as speed law. Assuming all the gases as ideal, the general law of gas
feed composition and feed quantity, reactor pressure and indicates
temperature, the amount of the dilution steam, and reactor
residence time. The coke precipitate remains on the reactor Pj yj Pt
cj = =
walls and impairs the furnace’s performance. Monitoring and RT RT (3)
controlling the optimum factors and parameters enhances the
Fj
operation and improves the furnace performance. Replacing yj (the molar ratio) with its equivalent, yj = F , leads to
The reactor resides on the thermal cracking furnace that t

provides the required heat for the reactions. The inside FjPt
cj =
temperature of the reactor is an essential parameter in thermal FT RT (4)
cracking reactions due to their endothermic characteristic. Heat
flux increasing in the outgoing reactor surface to the highest 3.2. Energy Balance Equation. Equation 5 captures the
possible allowed level may expand the amounts of undesirable energy balance along the reactor28

1 ijjj yz
∑ ri(−ΔHi)zzzzz
products such as coke. Therefore, the reactor temperature
jjJq (z)πdt +
∑j FjCpj j
should be regulated to increase the desired products instead of dT πdt2
k {
undesired products. The temperature was increased due to the =
dz 4 i (5)
resident materials in the outlet of the reactor. Nevertheless, the
flow should get consistent heat during its process inside the
ÄÅ ÉÑ
3.3. Momentum Balance. Equation 6 captures the pressure
ÅÅ 2f ξ ÑÑÑ 2
Å
reactor axle. Overheating leads to hot spots and generates local
= αÅÅÅ + ÑÑρ u + αρg u
ÅÅÇ dt πrb ÑÑÑÖ g
coke.28 −dPt du
The position of the reactors in thermal cracking furnaces is dz dz
different. They may be established either in parallel or vertical to (6)
the floor regarding the furnace type. Normally, variable-diameter The first term on the right-hand side is friction-induced pressure
reactors are used as an option for liquid feeds. However, fixed- loss. The second term captures the pressure loss induced by
diameter reactors are used for gassy feeds. The number of momentum change. Regarding the continuity law and the
reactor paths inside the furnace may vary based on the available assumption of the gas ideality, the following equations are
technical knowledge. derived28
2.1. Reactor Specifications. The employed reactor is the
same as in ref 28 fed by pure propane. Table 1 illustrates the G = pu = const (7)
required details.
M mPt
ρ=
3. MATHEMATICAL MODEL OF THE THERMAL RT (8)
CRACKING REACTOR
G RT
Berreni and Wang prepared an accurate mathematical model of u= .
M m Pt

ÅÄ ÑÉ
an industrial thermal cracking reactor. Accordingly, the (9)

GR ÅÅÅ d(1 / M m) 1 dT ÑÑÑ GRT dP

ÅÅT ÑÑ −
mathematical models were extracted from ref 28. For simplicity,

Pt ÅÅÅÇ M m dz ÑÑÑÖ M mPt2 dz

a simple scheme is presented in Figure 2. In addition, other du
relevant equations were extracted from ref 36. = + . .
dz dz
3.1. Mass Balance Equation. The following equation
(10)
captures the mass balance for the jth component of the reaction
mix along the dz length of the reactor Substituting the equations gives
12496 https://doi.org/10.1021/acsomega.1c04345
ACS Omega 2022, 7, 12493−12508
ACS Omega http://pubs.acs.org/journal/acsodf Article

Table 2. Reactions Involved in the Thermal Cracking of Propane37

no reaction A(1/S, m3/kmol·S or kgS/mol·m) E (kJ/mol) ΔH (kJ/mol, @298 K)
1 C3H8 ↔ C2H4 + H4 4.692 × 1010 211.7 82.66
2 C3H8 ↔ C3H6 + H2 5.888 × 1010 214.6 124.68
3 C3H8 + C2H4 → C2H6 + C3H6 2.536 × 1013 241.7 −11.64
4 2C3H6 → 3C2H4 1.514 × 1011 233.5 117.5
5 2C3H6 → 0.5C6 + 3CH4 1.423 × 109 190.4 −14.48
6 C3H6 ↔ C2H2 + CH4 3.794 × 1011 248.5 96.4
7 C3H6 + C2H6 → C4H8 + CH4 5.553 × 1014 251.1 −11.24
8 C2H6 ↔ C2H4 + H2 4.652 × 1013 272.8 136.32
9 C2H4 + C2H2 → C4H6 1.026 × 1012 172.6 −133.42
10 C4H8 → C6 6.92 × 107 143.6 83.42

d(1 / M m) 1 1 dT For a small element dz, we have ΔSu = −πDJq(z)/Ta(z)dz. The

dPt
=
dz
+ (
M m T dz
+ Fr ) following illustrates the final entropy generation equation by
dz 1
−
Pt
entropy balance16

jij ds zyz = s − s − πd
M mPt αG2RT (11)

j z
L Jq (z)

k dt {irr
The friction function for the direct section of the tube is defined
as28 out in t ∫0 Ta(z)
dz
(22)

i ds
Lj J (z ) y
jij ds zyz = jj − πd q zzzdz
2f
j z jj z
Ta zz
= 0.092Re−0.2/ dt
j dz
Fr =
k dt {irr k {
dt (12) ∫0 t
The bends equations are defined as (23)

Fr = 0.092Re−0.2/ dt + ζ / πR b (13) ds Jq (z)

i y
ζ = jjj0.7 + × 0.35zzzξ
σ= − π dt
dz Ta (24)

k {
Λ
90° (14) Using
dt (FT / Ω)RT
ξ = 0.051 + 0.19 v=
Rb P (25)

i1y
(15)

Δjjj zzz =
k {
The following equation may be replaced in the pressure 1 1
equation28 −

d ijjj ∑j Fj yzzz ∑j dz
T T ( z ) Ta(z) (26)

j z=
dF

dz jj GΩ zz
j 39
the local entropy generation becomes
i 1 dP zy πdt i −ΔGi yz
d(1 / M m)
i1y
k { σ = πdt Jq (z)Δjjj zzz + Ωvjjj− zz + ∑ rijjjj zzz
=
dz GΩ 2

kT { k T dz { k T {
(16)
3.4. Calculation of Convection Heat Transfer in 4 i
Reactor Outlets. The following equations illustrate the (27)
calculation of convection heat transfer28 Equation 27 shows three facts responsible for entropy in a
q = hA(Tw − Tb)av (17) chemical reactor: the heat transfer to the reactor wall, chemical
reactions, and pressure loss. The terms of the above equation
R result from multiplying thermodynamic flux and its relevant
∫0 l 2πdrUCpT force. The first term includes the multiplication of heat flux Jq(z)
Tb = R
∫0 l 2π drUCp and its thermodynamic force Δ(1/T), the second term includes
(18) the multiplication of velocity flux and its thermodynamic force
q = mCp(Tbo − Tbi) (19) (− T1 ddPz ), and the last term includes the multiplication of the
velocity flux of the reaction ri and its thermodynamic force
Nu = 0.023ReD0.8Pr n (20) −ΔGi

where n = 0.4 outlet heating, Tw > Tb, L/D > 60, 6000 < ReD <
( ).
T

107. Integrating σ over the reactor length captures the amount of

ij ds yz
3.5. Entropy Generation in the Thermal Cracking entropy generation as in the following equation

jj zz =
k dt {irr
Reactor. The reactor is supposed to function in a stable or L
steady state. Therefore the entropy balance of the system ∫0 σ dz
(28)
includes three additional terms. Heat transfer induces two
additional entropy factors to the system. So, the following 3.6. Kinetic Model Selected for the Research. The
equation captures the amount of entropy generation16
ij ds yz
molecular mechanism used in this work has been proposed by

jj zz = sout − sin + Δsu

the authors in ref 37. Propane and water steam were fed into the
k dt {irr
reactor; they were converted into 12 components throughout 11
(21) reactions. Water steam reactions with other components were
12497 https://doi.org/10.1021/acsomega.1c04345
ACS Omega 2022, 7, 12493−12508
ACS Omega http://pubs.acs.org/journal/acsodf Article

this study. The following equation captures the velocity specific

to the reaction K, which is a function of temperature37

K = A exp(−E / RT ) (29)

where A is the exponential function coefficient or frequency

factor, E is the activation energy (J/mol or cal/mol), R is the
ideal gas constant, and T shows the absolute temperature (K).
3.7. Assumptions and Problem-Solving Procedures.
The reactor model is addressed in this section. The pressure,
temperature, and weight percent data related to the components
and reactor length are used to assess entropy. The reactor
simulation assumptions are as follows:
• It is assumed that the internal flow of the reactor is plug
flow; because the velocity of the gas is high and the reactor
diameter is negligible compared to its length; thus, the
Figure 3. Pseudocode for the MOGA algorithm.
axial dispersion of the reactor is ignorable. This
assumption is valid based on ref 28.

ignored. Table 2 shows these reactions, the activation energy, • Coke deposition is ignored, and the reactor is assumed to
and the Arrhenius equation factor. Coke formation is ignored in be steady-state.

Figure 4. (a) Schematic description of the stream’s flow to a specific river and (b) schematic of the WCA optimization process.39
12498 https://doi.org/10.1021/acsomega.1c04345
ACS Omega 2022, 7, 12493−12508
ACS Omega http://pubs.acs.org/journal/acsodf Article

Figure 5. Flowchart of the water cycle algorithm.

Table 3. Key Decision Variables and Constraints method. Now that all required factors are known, the integral
generated in the reactor may be calculated.
decision variables unit lower limit upper limit
inlet gas temperature K 700 1000 4. OPTIMIZATION
inlet gas pressure bar 2 6
In this study, two optimization algorithms were used: the
inlet mass flow rate of gases g/s 500 900
multiobjective genetic algorithm (MOGA), a well-known and
wall temperature K 950 1300
efficient method for multiobjective optimization, and the
multiobjective water cycle algorithm (MOWCA), which is an
• The internal space of the reactor is assumed to be filled attractive, recent, and simple concept.
only with the combination of gas and steam. 4.1. MOGA. While GA considered problems with one
• The reactions initiate from the input and terminate in the objective function, MOGAs can overcome this limitation and
reactor’s output. simultaneously consider two or more objectives. In many
industrial problems, objectives under consideration conflict with
• The input/output hydrodynamic and thermal zone effects
each other, and optimizing a particular solution related to a
are not considered.
single-objective can result in unacceptable results dealing with
• The concentrations are captured using the law of ideal gas. the other objectives. A reasonable solution to a multiobjective
The relevant equations, i.e., energy, molar, and momentum problem is to investigate a set of solutions, each of which satisfies
balance, need to be solved simultaneously. the objectives at an acceptable level without being dominated by
After performing 10 chemical reactions, the feed and pure any other solution.38 Equation 30 mathematically presents a
propane resulted in 10 individual components. Table 2 multiobjective optimization problem
demonstrates the results. A differential equation is derivable min(f1 (x), f2 (x), ..., fk (x))
for the components. The molar balance questions are initially
equal to zero, except for propane, whose value is given in Table s . t . x ∈ X1 (30)
2.
The input feed temperature is required to solve the energy where K is the number of objectives and set X is the feasible set
equation, which is a function of the temperature of the outer of decision vectors. The vector-valued objective function can be
surface, and its profile is provided based on the length of the written as eq 31
reference reactor.28 Finally, the initial input pressure for the f : X → Rk , (f1 (x), f2 (x), ..., fk (x))T (31)
pressure loss equation is known. There are 12 differential
equations whose initial conditions are known. They were solved MOGA was used via MATLAB-2020a. The following steps
using the MATLAB software package, using the ode15s function, considered the reproductive procedure of the genetic algorithm:
solving complicated differential equations using the GEAR natural selection, crossover, mutation, fitness computation,
12499 https://doi.org/10.1021/acsomega.1c04345
ACS Omega 2022, 7, 12493−12508
ACS Omega http://pubs.acs.org/journal/acsodf Article

repair, updation of population, and elitism. While GA is related

to one objective function, MOGA can consider more than one
objective function. The upper and lower bounds were
considered. The MOGA was run in MATLAB software. The
pseudocode for the MOGA algorithm used in this study is
shown in Figure 3.
4.2. WCA. The water cycle algorithm (WCA) mimics the
flow of rivers and streams toward the sea based on the water
cycle process. In the case of rain or precipitation, the primary
population of design variables (i.e., the population of streams) is
randomly increased following the raining process. The best
individual variable (i.e., the best stream) with minimum cost
function is the sea. Afterward, some good streams with cost
function values close to the current best record are assumed as
rivers, considering the other remaining streams to have flowed
into rivers and the sea. The optimization algorithm should be
started with a primary population representing a matrix of
streams. For an N-dimensional optimization problem, the
following algorithm can be developed using some equations.

Nsr = number of Rivers + 1 (32)

Nstream = Npop − Nsr (33)

where Npop and N represent the total population and the number
of design variables, respectively.

ij yz
j zz
NSn = roundjjjj stream z
z, n = 1, 2, ..., Nsr
zz
jj ∑Nsr Cost i z
Cost n

k {
× N
i = 1
(34)

where NSn is the number of streams that flow to the specific

rivers and sea. The following equations may obtain the new
position for streams and rivers.
i+1 i i i
X⃗Stream = X⃗Stream + rand × C × (XR⃗ i ver − X⃗Stream ) (35)

i+1 i i i
X⃗Stream = X⃗Stream + rand × C × (X⃗Sea − X⃗Stream ) (36)

i+1 i i i Figure 4a shows a schematic view of a stream flowing toward a

XR⃗ i ver = XR⃗ i ver + rand × C × (X⃗Sea − XR⃗ i ver ) (37) specific river along their connection line. Figure 4b shows the
schematic of the WCA optimization process. Also, Figure 5
where rand is a uniformly distributed random numbers between shows the flowchart of the water cycle algorithm.
0 and 1. The new position of the streams is determined by 4.3. Optimization Problem. The objective functions of the
problem are as follows:
New
X⃗Stream = LB⃗ + rand × (UB⃗ − LB)⃗ (38) • Maximum production based on weight percentages
• Minimization of entropy generation
where LB and UB are lower and upper bounds defined by the
given problem, respectively. The value of dmax adaptively Two-objective optimization leads to a two-dimensional
Pareto frontier. Thus, the closest solution in the population to
decreases as follows the ideal point in the 2D space of the frontier would be
i considered as the optimum solution to the problem.
i+1 i dmax The preliminary community should be generated concerning
dmax = dmax −
Max Iteration (39) some limitations in the input data, i.e., the constraints of the
12500 https://doi.org/10.1021/acsomega.1c04345
ACS Omega 2022, 7, 12493−12508
ACS Omega http://pubs.acs.org/journal/acsodf Article

Figure 6. Algorithm using GP to generate OFs.

decision variables. The values of the constraints are shown in mixture. The temperature is 873.15 K at the starting point of the
Table 3. input. As heat is transferred along the tube, the chemical
4.4. Genetic Programming. Genetic Programming (GP) reactions progress, and the outer temperature reaches the wall
generates prediction functions based on decision variables to temperature. Figure 9 depicts the reactor conversion percent,
reduce the optimization run time. In this regard, based on and Figure 10 depicts the weight percent profile of each product.
simulation results associated with varying decision variables, the
5.2. Generation of Correlation Using GP. Different
best correlation for estimating objective functions have been
proposed by the GP code. Figure 6 shows the algorithm of using simulations were performed to generate the prediction
GP to generate OFs. correlations for objective functions with the computer code in
the range of decision variables. These data are transferred to the
5. RESULTS AND ANALYSIS GP code, and the best prediction correlations for objective
functions have been achieved, as shown in Table 5. When the
5.1. Results for the Reactor in the Reference State. In mean square error (R2) range is between 0.9 and 1, the accuracy
this research, the main objective was to produce ethylene. The
is very high.
main outputs of the propane thermal cracking reactor included
ethylene, methane, propane, propylene, and hydrogen. To verify The verification of the following equations has been
the computer code for modeling and simulation of the thermal compared with the simulation data with high accuracy. These
cracking reactor, the simulation results were compared with an equations reduce the computation time by about 85%. Also,
industrial case study.28 their accuracy is very high and acceptable.
The percent of propane conversion and the weight of the 5.3. Optimization Results. 5.3.1. Multiobjective Opti-
products were derived by comparing with an industrial reactor. mization. Table 6 presents the Pareto Front optimum solutions
Although coke production was neglected, according to Table for objective functions and decision variables based on MOGA.
4, the simulation results match the practical results to a high As shown, the optimum objective functions and corresponding
degree of consistency. The weight percent of C3H6 had the decision variables were determined.
highest inconsistency of 12%. The conversion percent difference To draw the Pareto front and find the best solution that trade-
falls to 0.01%, which was initially the main objective.
off between multiple objectives, it is better first to rewrite the
Figure 7 illustrates the outer wall temperature profile of the
tube, derived from the temperature of the reactor surface in the target functions in the Pareto optimal solutions so that the
industrial sample.28 difference between the dimensions of the functions is
Table 4 presents the results according to the temperature of neutralized in the concept of points distance from the ideal
the outer wall (Figure 8); Figure 8 also depicts the calculation of point. To make the answers dimensionless, we use the following
the governing equations and the temperature profile of the gas relation
12501 https://doi.org/10.1021/acsomega.1c04345
ACS Omega 2022, 7, 12493−12508
ACS Omega http://pubs.acs.org/journal/acsodf Article

Table 4. Comparison of the Products of Simulated and

Industrial Reactors
weight percent (wt %) of the weight percent (wt %) of
component simulated reactor an industrial reactor28
conversion 90.61 90.60
coefficient
CH4 22.89 24.00
C2H4 36.82 34.50
C3H6 12.89 14.70
C3H8 9.38 9.30
H2 1.67 1.20

Figure 9. Conversion percent profile per unit length of the reactor with
industrial wall temperature.

Figure 7. Temperature profile of the outer surface of the reactor with

the industrial wall temperature.

Figure 10. Weight percent of the products along the reactor with
industrial wall temperature.

The optimum values of OFs and decision variables based on

MOGA are determined and shown in Table 7. The weight
percentage of the basic system products is 60.615%, and the
entropy production rate is 1048 J/s. The table above shows that
the weighted percentage of products and the system’s entropy
production rate after optimization reached 61.13% and 899.8 J/
Figure 8. Temperature profile of the gas mixture in the reactor with
s, displaying an improvement of 0.85 and 16.51%, respectively.
industrial wall temperature.
The GP and non-GP results also showed similar improvements.
OFi , k − OFmin, i However, the computation time of the GP declined significantly
OFDL, i , k = (about 85% of the base case).
OFmax, i − OFmin, i (40) With MOWCA, the Pareto front optimum solution and its
decision variables are indicated in Table 8.
OFDL,i,k is the dimensionless objective function for the i-M The Pareto front optimal solution based on MOWCA is
function and the kM response, i is the target function counter, demonstrated in Figure 12.
and k is the response function in that target function. The The weight percentage of the products produced in the basic
presentation of the Pareto front for the dimensionless OFs is system is 60.61%, and the entropy production rate is 1048 J/s.
shown in Figure 11. The 2D normalized Pareto front indicates As shown in Table 9, the weighted percentage of products and
normalized entropy generation rate versus normalized products. the entropy production rate of the system after optimization
As shown in Figure 11, the point to the coordinates (1,0) is reached 61.81% and 882.72 J/s, respectively, which resulted in
the equilibrium point or ideal point. From the Pareto front’s an improvement of 1.97 and 18.77%, respectively. As shown in
optimal response set, any response closer to the ideal point is the Table 9, the optimization with GP and non-GP algorithms
optimal response. The concept of distance is used according to shows similar results.
the following equation to find the closest answer to the ideal 5.3.2. Single-Objective Optimization. In this case, two
point optimization processes are implemented with one of the target
functions. Optimization results to maximize product production
dk = (OF1, k − OF1,ideal )2 + (OF2, k − OF2,ideal )2 (41) lead to 64.36% by weight of system products, equivalent to a
6.17% increase in product production. The single-objective
The minimum distance is calculated as 0.5913 and is related to optimization response aims to maximize product production, as
the selected optimal solution, as shown in Table 5. shown in Table 10.
12502 https://doi.org/10.1021/acsomega.1c04345
ACS Omega 2022, 7, 12493−12508
ACS Omega http://pubs.acs.org/journal/acsodf Article

Table 5. Objective Function Correlation Generated by GP

mean square error
objective function best equation (R2)
products (%) 0.00127 × Pgasin + 7045.399 × log(Tw) + 1.55E−28 × Pgasin × Tw8 − 42402.84 − 6.23 × Tw − 1.38E − 9 × Pgasin × Tw2 0.97
4.42E12 3.78E5 + 371.07 × ma
123.876 + 0.00189717 × Pgasin − Tgasin × Pgasin
+ Tgasin
entropy
generation 2 0.94
Tw Tw
+ 1.498 × − 1.60E17 ×
Tgasin × Pgasin 2 Tgasin × Pgasin3

Table 6. Pareto front optimum solutions for Objective

Functions Based on MOGA

OF1 OF2 DV1 DV2 DV3 DV4

products entropy Tgas_in Pgas_in
(%) (J/K s) (K) (kPa) ma (g/s) Tw,in (K)
44.19 699.63 852.71 250.002 750.00 1000.00
44.19 699.63 852.71 250.002 750.00 1000.00
48.41 741.06 853.11 250.003 750.76 1010.39
64.45 986.14 852.79 250.004 751.05 1098.55
54.50 807.46 852.88 250.004 750.30 1028.80
56.42 831.17 853.15 250.004 750.40 1035.97
55.44 818.78 852.99 250.003 750.22 1032.20
46.07 717.76 852.91 250.003 750.53 1004.44
52.80 787.75 852.97 250.003 750.25 1023.12
64.17 971.10 853.49 250.005 750.43 1090.57
60.10 882.90 853.03 250.004 750.45 1053.11
61.61 908.49 852.94 250.004 750.22 1062.63
51.84 777.23 853.31 250.004 750.73 1020.13
57.14 840.46 853.07 250.004 750.43 1038.88
63.39 946.53 853.34 250.004 750.45 1078.57
45.68 713.97 852.83 250.003 750.39 1003.51
63.15 940.32 853.40 250.005 750.39 1075.81 Figure 11. Normalized Pareto front optimum solution obtained with
46.94 726.23 853.01 250.003 750.42 1006.60 MOGA.
61.29 902.59 853.46 250.004 750.32 1060.42
57.84 849.82 853.28 250.003 750.53 1041.89
58.72 862.21 852.95 250.002 750.40 1045.97
63.64 953.19 853.46 250.004 750.35 1081.70 Table 7. Pareto Front Optimum Solutions for Objective
62.87 933.99 853.27 250.002 750.60 1073.00 Functions Based on MOGA
53.02 790.38 852.96 250.004 750.40 1023.85 optimum objective function parameters with GP value without GP value
59.33 871.11 852.94 250.004 750.44 1048.98
products (normalized) 0.83 0.83
64.04 966.05 853.28 250.004 750.43 1087.93
entropy generation (normalized) 0.66 0.66
64.56 1001.43 853.54 250.005 751.93 1108.04
products (%) 61.13 61.13
52.43 783.61 852.97 250.004 750.22 1021.96
entropy generation (J/K s) 899.80 899.80
49.20 748.99 852.76 250.003 750.16 1012.52
optimum decision variable parameters with GP value without GP value
45.08 708.22 853.06 250.003 750.46 1002.08
53.46 795.40 852.93 250.003 750.42 1025.28 Tgas_in (K) 853.42 853.42
61.13 899.80 853.42 250.004 750.36 1059.36 Pgas_in (bar) 25 25
46.98 726.65 852.95 250.003 750.40 1006.71 ma (g/s) 750.36 750.36
44.44 701.94 852.72 250.002 750.02 1000.56 Tw_in (K) 1059.36 1059.36
60.60 891.09 852.76 250.004 750.39 1056.04
50.23 759.82 853.21 250.003 750.57 1015.38
The optimization results aimed at minimizing the entropy
58.08 853.21 853.17 250.004 750.41 1043.00
production rate lead to the production of 687.99 (J/K s) entropy
64.25 974.33 853.52 250.004 750.42 1092.29
in the system, equivalent to a decrease of 52.38% in the
60.58 890.60 853.27 250.004 750.43 1055.92
production of the entropy production rate of the system. The
54.33 805.38 853.09 250.004 750.24 1028.21
solution to the single-objective optimization to minimize the
54.96 813.05 853.28 250.004 750.62 1030.44
entropy production of the system is indicated in Table 11.
55.77 823.00 852.72 250.004 750.20 1033.46
Also, a single optimization using WCA has been performed.
61.95 914.86 853.22 250.004 750.54 1065.11
The optimization results aim to maximize the product, leading
62.68 929.68 852.97 250.004 750.37 1071.16
to the formation of 64.56% by weight of system products,
56.08 826.87 853.29 250.003 750.68 1034.62
equivalent to a 6.51% increase in product production as
64.55 996.99 853.48 250.005 751.88 1105.13
determined in Table 12.
47.57 732.49 852.80 250.003 750.27 1008.21
The optimization results aimed at minimizing the entropy
62.39 923.42 853.46 250.005 750.37 1068.63
production rate lead to the production of 687.50 (J/K s) entropy
12503 https://doi.org/10.1021/acsomega.1c04345
ACS Omega 2022, 7, 12493−12508
ACS Omega http://pubs.acs.org/journal/acsodf Article

Table 8. Pareto Front Optimum Solutions for Objective

Functions Based on MOWCA
OF1 OF2 DV1 DV2 DV3 DV4
products entropy Tgas_in Pgas_in
(%) (J/K s) (K) (kPa) ma (g/s) Tw_in (K)
64.56 962.27 943.15 250.000 750.00 1108.01
44.19 689.91 943.15 250.000 750.00 1000.00
44.19 689.91 895.34 309.530 816.06 1119.54
64.56 962.27 943.15 250.000 750.00 1108.01
63.37 912.68 943.15 250.000 750.00 1078.33
62.08 886.71 943.15 250.000 750.00 1066.12
48.95 733.39 943.15 250.000 753.29 1011.83
45.71 702.98 943.15 250.000 750.00 1003.58
44.78 694.93 943.15 250.000 750.00 1001.36
63.97 928.76 943.15 250.000 750.73 1086.63
48.07 724.02 932.18 384.524 766.49 1100.51
56.53 812.96 943.15 250.422 750.90 1036.40
50.27 749.95 913.24 250.067 750.66 1015.50
57.37 820.33 907.80 285.339 837.65 1050.34
50.96 756.77 943.15 250.000 763.98 1017.51
64.42 946.32 943.15 250.000 750.00 1097.14
46.03 705.98 943.15 250.016 750.30 1004.36 Figure 12. Normalized Pareto front optimum solution obtained with
55.95 806.38 943.15 250.435 750.93 1034.14 MOWCA.
46.73 712.18 926.17 253.355 776.41 1028.38
52.59 770.42 943.15 250.504 751.07 1022.44 Table 9. Pareto Front Optimum Solutions for Objective
49.77 744.57 912.72 250.000 750.20 1014.08 Functions Based on MOWCA
54.50 787.35 943.15 250.000 750.00 1028.82
optimum objective function parameters with GP value without GP value
55.38 799.93 943.15 250.448 750.95 1031.97
58.30 832.55 871.60 272.639 757.16 1116.20 products (normalized) 0.86 0.86
49.51 737.11 901.89 341.795 836.10 1132.17 entropy generation (normalized) 0.71 0.71
55.05 793.36 943.15 250.000 750.00 1030.79 products (%) 61.81 61.81
60.70 864.94 943.15 250.000 751.46 1056.65 entropy generation (J/K s) 882.72 882.72
57.87 825.99 869.25 306.906 819.11 1194.70 optimum decision variable parameters with GP value without GP value
54.02 782.21 881.20 310.435 779.45 1051.96 Tgas,in (K) 943.15 943.15
63.59 921.19 934.42 250.020 750.20 1081.01 Pgas,in (bar) 25 25
58.87 838.76 908.91 424.713 779.49 1054.57 ma (g/s) 751.65 751.65
53.15 777.29 943.15 250.000 760.60 1024.25 Tw,in (K) 1064.08 1064.08
47.77 721.33 943.15 250.000 750.00 1008.73
52.18 764.33 926.52 413.089 782.40 1092.34 Table 10. Results of Maximization of the Product by GA
59.53 850.37 943.15 250.341 750.73 1050.06
optimum objective function parameters with GP value without GP value
47.30 717.09 943.15 250.000 750.00 1007.52
59.23 843.62 873.70 280.983 805.55 1158.11 products (%) 64.36 64.36
61.21 872.22 943.15 250.000 750.00 1059.89 entropy generation (J/K s) 1022.20 1022.20
60.19 859.58 943.15 250.000 756.88 1053.65 optimum decision variable parameters with GP value without GP value
61.81 882.72 943.15 250.000 751.65 1064.08 Tgas,in (K) 908.35 908.35
64.45 951.98 923.33 312.665 824.69 1118.13 Pgas,in (bar) 25.29 25.29
51.61 758.82 943.15 250.000 753.03 1019.44 ma (g/s) 785.71 785.71
61.43 875.80 907.61 353.490 820.16 1009.01 Tw,in (K) 1107.42 1107.42
53.40 779.90 943.15 250.000 760.50 1025.09
44.50 692.64 943.15 250.010 750.15 1000.71 Table 11. Results of Minimization of Entropy Generation by
60.02 854.45 943.15 250.000 750.00 1052.70 GA
47.04 714.72 943.15 250.000 750.00 1006.85
optimum objective function parameters with GP value without GP value
51.85 761.11 859.39 334.392 836.31 1131.28
products (%) 48.92 48.92
entropy generation (J/K s) 688.00 688.00
in the system, equivalent to a decrease of 52.49% in the optimum decision variable parameters with GP value without GP value
production of the entropy production rate of the system, as Tgas_in (K) 845.85 845.85
shown in Table 13. Pgas_in (bar) 35.39 35.39
5.4. Computational Time. The computational time for ma (g/s) 750.27 750.27
MOGA, MOWCA, GP-MOGA, and GP-MOWCA are Tw_in (K) 1000.00 1000.00
indicated in Table 14. In the GP-MOGA and GP-MOWCA,
the time for generation of data sets and GP correlations have The computation time is based on Intel Core i7-8750H with 6
been considered. CPU cores and 16 GB memory. As seen, the computation time is
12504 https://doi.org/10.1021/acsomega.1c04345
ACS Omega 2022, 7, 12493−12508
ACS Omega http://pubs.acs.org/journal/acsodf Article

Table 12. Results of Maximization of the Product by WCA It is suggested that future studies will address exergonic and
exergoeconomic parameters. For studies on the optimization
optimum objective function parameters with GP value without GP value
problem, environmental impacts are critical factors that should
products (%) 64.56 64.56 be further investigated.

■
entropy generation (J/K s) 1009.20 1009.20
optimum decision variable parameters with GP value without GP value AUTHOR INFORMATION
Tgas_in (K) 915.27 915.27 Corresponding Author
Pgas_in (bar) 35.39 35.39 Gholamreza Salehi − Department of Mechanical Engineering,
ma (g/s) 25 25 Central Tehran Branch, Islamic Azad University, Tehran
Tw_in (K) 1108.01 1108.01 11936-53471, Iran; orcid.org/0000-0002-7866-358X;
Phone: +989122031671; Email: [email protected]
Table 13. Results of Minimization of Entropy Generation by
WCA Authors
Peyman Roudgar Saffari − Department of Mechanical
optimum objective function parameters with GP value without GP value Engineering, Adiban Institute of Higher Education, Garmsar
products (%) 59.66 59.66 35819-69855, Iran; orcid.org/0000-0002-8419-8602
entropy generation (J/K s) 687.50 687.50 Hesamoddin Salarian − Department of Mechanical
optimum decision variables parameters with GP value without GP value Engineering, Nour Branch, Islamic Azad University, Nour
Tgas_in (K) 878.08 878.08 15858-54289, Iran
Pgas_in (bar) 28.03 28.03 Ali lohrasbi − Department of Mechanical Engineering, Islamic
ma (g/s) 799.67 799.67 Azad University of Nowshahr Branch, Nowshahr 46517-
Tw_in (K) 1161.73 1161.73 39948, Iran
Mohammad Hasan Khoshgoftar Manesh − Energy,
Table 14. Comparison of Computation time for MOGA, Environmental and Biological Systems Research Lab
MOWCA, GP-MOGA, and GP-MOWCA (EEBRlab), Division of Thermal Sciences and Energy Systems,
Department of Mechanical Engineering, Faculty of Technology
GP- GP-
methodology MOGA MOWCA MOGA MOWCA & Engineering, University of Qom, Qom 37137-44003, Iran
computation time (s) 63 056.8 61 753 34 270 33 380 Complete contact information is available at:
computation time (h) 17.51 17.15 9.52 9.27 https://pubs.acs.org/10.1021/acsomega.1c04345

Notes
The authors declare no competing financial interest.

■
reduced significantly in GP-MOWCA and GP-MOGA
compared with MOGA and MOWCA.40
NOMENCLATURE
6. CONCLUSIONS j jth component
A thermal cracking reactor was investigated in this study. The dz the length dz
input feed was propane and water steam, and the main output Fj the intensity of the molar flow of the jth component
products were propane, methane, ethylene, propylene, and αij stoichiometric coefficient of the jth component in the ith
hydrogen. The amount of entropy generated by the sources and reaction
the governing equations was investigated. Rj overall velocity of the production and/or consumption of
Single-objective and two-objective optimizations using GA the jth component
and WCA were performed to maximize products and minimize ri the velocity of the ith reaction
entropy generation. Decision variables were the inlet gas Ki the specific velocity of the ith reaction
temperature, inlet gas pressure, the mass flow rate of the inlet cj the molar concentration of the jth component
air, and the wall. GP was employed to generate the best yj molar ratio
correlations to predict objective functions based on key variables ΔHi the heat of the ith reaction
to reduce the run time of optimization runs. Jq(z) heat flux transferred from the outer heat source to the
Based on single-objective optimization, the following results reaction flow in the location z of the reactor
were obtained: Cpj heat capacity of the jth component
α unit conversion factor
• Maximizing product production using GA and WCA led
ρ gas density
to 64.36 and 64.56% by weight of system products
u gas speed
equivalent to a 6.17 and 6.51% increase in the product
Mm gas mixture molecular weight
compared to the base case.
Λ the angle of the bends, which is equal to 180°
• Minimizing the entropy production rate using the GA and
Tw surface temperature
WCA led to a decrease of 52.38 and 52.49% compared to
T∞ fluid temperature
the base case in the entropy generation rate of the system.
h the coefficient of the convection heat transfer
Based on two-objective optimization, the following results Tc the temperature of the tube center
were obtained: Tb bulk temperature
• The weighted percentage of products and the entropy Tbo outlet bulk temperature
production rate improved 0.85% and 16.51% compared Sin entropy of the inlet flow
with the base case using the MOGA algorithm, and 1.97 Sout entropy of the outlet flow
and 18.77% compared to the base case. ΔSu entropy induced by heat transfer
12505 https://doi.org/10.1021/acsomega.1c04345
ACS Omega 2022, 7, 12493−12508
ACS Omega http://pubs.acs.org/journal/acsodf Article

σ locally generated entropy (19) Røsjorde, A.; Kjelstrup, S.; Johannessen, E.; Hansen, R.
coefficient of exponential function or coefficient of Minimizing the Entropy Production in a Chemical Process for
A frequency Dehydrogenation of Propane. Energy 2007, 32, 335−343.
E the activation energy (expressed in J/mol or cal/mol) (20) Abdous, M. A.; Saffari, H.; Avval, H. B.; Khoshzat, M.
Investigation of Entropy Generation in a Helically Coiled Tube in
R ideal gas constant
Flow Boiling Condition under a Constant Heat Flux. Int. J. Refrig. 2015,
T absolute temperature, K 60, 217−233.
(21) Kurnia, J. C.; Sasmito, A. P.; Shamim, T.; Mujumdar, A. S.
Numerical Investigation of Heat Transfer and Entropy Generation of

■ REFERENCES
(1) Zhou, S.; Chen, L.; Sun, F. Constructal Optimization for a Solid-
Laminar Flow in Helical Tubes with Various Cross Sections. Appl.
Therm. Eng. 2016, 102, 849−860.
(22) Vandewalle, L. A.; Van Cauwenberge, D. J.; Dedeyne, J. N.; Van
Geem, K. M.; Marin, G. B. Dynamic Simulation of Fouling in Steam
Gas Reactor Based on Triangular Element. Sci. China, Ser. E 2008, 51,
1554−1562. Cracking Reactors Using CFD. Chem. Eng. J. 2017, 329, 77−87.
(2) Gao, G.-Y.; Wang, M.; Pantelides, C. C.; Li, X.-G.; Yeung, H. (23) Mu, P.; Gu, X. Thermal Cracking Furnace Optimal Modeling
Mathematical Modeling and Optimal Operation of Industrial Tubular Based on Enriched Kumar Model by Free-Radical Reactions. Processes.
Reactor for Naphtha Cracking. Comput.-Aided Chem. Eng. 2009, 27, 2020, 8, No. 91.
501−506. (24) Zhou, L.; Li, K.; Hang, P.; Liu, G. Optimization of the Ethane
(3) Masoumi, M.; Shahrokhi, M.; Sadrameli, M.; Towfighi, J. Thermal Cracking Furnace Based on the Integration of Reaction
Modeling and Control of a Naphtha Thermal Cracking Pilot Plant. Network. Clean Technol. Environ. Policy 2021, 23, 879−887.
Ind. Eng. Chem. Res. 2006, 45, 3574−3582. (25) Narasimharao, K.; Alshehri, A. Gold Supported Yttrium Oxide
(4) Ghasemi, H. M.; Gilani, N.; Daryan, J. T. CFD Simulation of Nanorods for Catalytic Oxidative Cracking of N-Propane to Light
Propane Thermal Cracking Furnace and Reactor: A Case Study. Int. J. Olefins. Fuel 2020, 278, No. 118375.
Chem. React. Eng. 2016, 15. DOI: 10.1515/ijcre-2016-0125. (26) Ebrahimian, S.; Iranshahi, D.; Bahmanpour, A. M. A Novel
(5) Ghashghaee, M.; Karimzadeh, R. Dynamic Modeling and Reactor Concept for Thermal Integration of Naphtha Reforming with
Simulation of Steam Cracking Furnaces. Chem. Eng. Technol. 2007, Propane Ammoxidation. Chem. Eng. Process. Process Intensif. 2019, 146,
30, 835−843. No. 107659.
(6) Zheng, S.; Zhang, X.; Qi, C.; Zhou, H. Modeling of Heat Transfer (27) Barnoon, P.; Toghraie, D.; Mehmandoust, B.; Fazilati, M. A.;
and Pyrolysis Reactions in Ethylene Cracking Furnace Based on 3-D Eftekhari, S. A. Comprehensive Study on Hydrogen Production via
Combustion Monitoring. Int. J. Therm. Sci. 2015, 94, 28−36. Propane Steam Reforming inside a Reactor. Energy Rep. 2021, 7, 929−
(7) Barazandeh, K.; Dehghani, O.; Hamidi, M.; Aryafard, E.; 941.
Rahimpour, M. R. Investigation of Coil Outlet Temperature Effect (28) Berreni, M.; Wang, M. Modelling and Dynamic Optimization of
on the Performance of Naphtha Cracking Furnace. Chem. Eng. Res. Des. Thermal Cracking of Propane for Ethylene Manufacturing. Comput.
2015, 94, 307−316. Chem. Eng. 2011, 35, 2876−2885.
(8) Nummedal, L.; Kjelstrup, S. Equipartition of Forces as a Lower (29) Rezaeimanesh, M.; Asghar Ghoreyshi, A.; Peyghambarzadeh, S.
Bound on the Entropy Production in Heat Exchange. Int. J. Heat Mass M.; Hassan Hashemabadi, S. A Coupled CFD Simulation Approach for
Transfer 2001, 44, 2827−2833. Investigating the Pyrolysis Process in Industrial Naphtha Thermal
(9) Sauar, E.; Nummedal, L.; Kjelstrup, S. The Principle of Cracking Furnaces. Chin. J. Chem. Eng. 2021, DOI: 10.1016/
Equipartition of Forces in Chemical Reactor Design: The Ammonia j.cjche.2021.03.028.
Synthesis. Comput. Chem. Eng. 1999, 23, S499−S502. (30) Bi, K.; Zhang, S.; Zhang, C.; Li, H.; Huang, X.; Liu, H.; Qiu, T.
(10) Kjelstrup, S.; Island, T. V. The Driving Force Distribution for Knowledge Expression, Numerical Modeling and Optimization
Minimum Lost Work in a Chemical Reactor Close to and Far from Application of Ethylene Thermal Cracking: From the Perspective of
Equilibrium. 2. Oxidation of SO2. Ind. Eng. Chem. Res. 1999, 38, 3051− Intelligent Manufacturing. Chin. J. Chem. Eng. 2021, 38, 1−17.
3055. (31) Yuan, B.; Zhang, Y.; Du, W.; Wang, M.; Qian, F. Assessment of
(11) de Koeijer, G. M.; Kjelstrup, S.; Salamon, P.; Siragusa, G.; Energy Saving Potential of an Industrial Ethylene Cracking Furnace
Schaller, M.; Hoffmann, K. H. Comparison of Entropy Production Rate Using Advanced Exergy Analysis. Appl. Energy 2019, 254, No. 113583.
Minimization Methods for Binary Diabatic Distillation. Ind. Eng. Chem. (32) Rebordinos, J. G.; Herce, C.; González-Espinosa, A.; Gil, M.;
Res. 2002, 41, 5826−5834. Cortés, C.; Brunet, F.; Ferré, L.; Arias, A. Evaluation of Retrofitting of
(12) de Koeijer, G.; Røsjorde, A.; Kjelstrup, S. Distribution of Heat an Industrial Steam Cracking Furnace by Means of CFD Simulations.
Exchange in Optimum Diabatic Distillation Columns. Energy 2004, 29, Appl. Therm. Eng. 2019, 162, No. 114206.
2425−2440. (33) Meng, D.; Shao, C.; Zhu, L. Two-Level Comprehensive Energy-
(13) Nummedal, L.; Kjelstrup, S.; Costea, M. Minimizing the Entropy Efficiency Quantitative Diagnosis Scheme for Ethylene-Cracking
Production Rate of an Exothermic Reactor with a Constant Heat- Furnace with Multi-Working-Condition of Fault and Exception
Transfer Coefficient: The Ammonia Reaction. Ind. Eng. Chem. Res. Operation. Energy 2022, 239, No. 121835.
2003, 42, 1044−1056. (34) Saffari, P. R.; Salarian, H.; lohrasbi, A.; Salehi, G. Entropy
(14) Røsjorde, A.; Kjelstrup, S. The Second Law Optimal State of a Generation Analysis of a Thermal Cracking Reactor. ACS Omega 2021,
Diabatic Binary Tray Distillation Column. Chem. Eng. Sci. 2005, 60, 6, 6335−6347.
1199−1210. (35) Meyers, R. A.; Meyers, R. A. Handbook of Petrochemicals
(15) Wilhelmsen, Ø.; Johannessen, E.; Kjelstrup, S. Energy Efficient Production Processes; McGraw-Hill, 2005.
Reactor Design Simplified by Second Law Analysis. Int. J. Hydrogen (36) Froment, G. F.; Bischoff, K. B.; De Wilde, J. Chemical Reactor
Energy 2010, 35, 13219−13231. Analysis and Design; Wiley: New York, 1990; Vol. 2.
(16) Johannessen, E.; Kjelstrup, S. Minimum Entropy Production (37) Sundaram, K. M.; Froment, G. F. Modeling of Thermal Cracking
Rate in Plug Flow Reactors: An Optimal Control Problem Solved for KineticsI: Thermal Cracking of Ethane, Propane and Their Mixtures.
SO2 Oxidation. Energy 2004, 29, 2403−2423. Chem. Eng. Sci. 1977, 32, 601−608.
(17) Kingston, D.; Razzitte, A. C. Entropy Production in Chemical (38) Shamshirband, S.; Khoshnevisan, B.; Yousefi, M.; Bolandnazar,
Reactors. J. Non-Equilib. Thermodyn. 2017, 42, 265−275. E.; Anuar, N. B.; Abdul Wahab, A. W.; Khan, S. U. R. A Multi-Objective
(18) Chen, L.; Zhang, L.; Xia, S.; Sun, F. Entropy Generation Evolutionary Algorithm for Energy Management of Agricultural
Minimization for CO 2 Hydrogenation to Light Olefins. Energy 2018, SystemsA Case Study in Iran. Renewable Sustainable Energy Rev.
147, 187−196. 2015, 44, 457−465.

12506 https://doi.org/10.1021/acsomega.1c04345
ACS Omega 2022, 7, 12493−12508
ACS Omega http://pubs.acs.org/journal/acsodf Article

(39) Eskandar, H.; Sadollah, A.; Bahreininejad, A.; Hamdi, M. Water

Cycle Algorithm − A Novel Metaheuristic Optimization Method for
Solving Constrained Engineering Optimization Problems. Comput.
Struct. 2012, 110−111, 151−166.
(40) Ghosh, P. K.; Sadhu, P. K.; Basak, R.; Sanyal, A. Energy Efficient
Design of Three Phase Induction Motor by Water Cycle Algorithm. Ain
Shams Eng. J. 2020, 11, 1139−1147.

12507 https://doi.org/10.1021/acsomega.1c04345
ACS Omega 2022, 7, 12493−12508
ACS Omega http://pubs.acs.org/journal/acsodf Article

12508 https://doi.org/10.1021/acsomega.1c04345
ACS Omega 2022, 7, 12493−12508