Energy and AI 13 (2023) 100264

Contents lists available at ScienceDirect

Energy and AI
journal homepage: www.elsevier.com/locate/egyai

Coaxial multi-criteria optimization of a methane steam reforming reactor for

effective hydrogen production and thermal management
Marcin Pajak a ,∗, Grzegorz Brus a , Shinji Kimijima b , Janusz S. Szmyd a
a
AGH University of Science and Technology, Krakow, Poland
b
Shibaura Institute of Technology, Tokyo, Japan

GRAPHICAL ABSTRACT

HIGHLIGHTS

• Introduction of a novel approach to the macro-patterning concept.

• Sensitivity analysis conducted for the evolutionary algorithm parameters.
• Enhancement of thermal conditions via a modification of the catalyst insert.
• Increase in hydrogen productivity.

ARTICLE INFO ABSTRACT

Keywords: The advancement in environmental awareness is the recent driving factor of the energy industry development.
Hydrogen The market sentiments dictate the commercialization of unconventional energy sources. Thus, generation
Genetic algorithm via hydrogen conversion gains popularity. The presented research regards the enhancement of the steam
Steam reforming
reforming reaction, used for the production of hydrogen via the conversion of hydrocarbons. The reforming
Design optimization
process characterizes by a strong endothermic nature. The rapid course of the reaction leads to the creation
of temperature gradients of a considerable magnitude. The presented research strives to alleviate the negative
consequences of the reaction character. An original strategy by the name of macro-patterning is suggested
as a remedy. The presented research proposes an updated concept, predicting the introduction of coaxial
segments to the catalytic insert. The segments may consist of catalytic material or metallic foam applied
for local suppression of the reaction. The morphology of specific segments may be altered independently, to
allow for additional control of the reforming reaction. The objective of the research is to define the optimal

∗ Corresponding author.
E-mail address: mpajak@agh.edu.pl (M. Pajak).

https://doi.org/10.1016/j.egyai.2023.100264
Received 24 December 2022; Received in revised form 4 April 2023; Accepted 9 April 2023
Available online 8 May 2023
2666-5468/© 2023 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-
nc-nd/4.0/).
M. Pajak et al. Energy and AI 13 (2023) 100264

segment composition. The optimization process is based on an in-house procedure implementing a genetic
algorithm. The acquired results appear to validate the macro-pattering concept. A significant unification of
the temperature field is obtained, with a simultaneous increase in hydrogen productivity.

Nomenclature 𝑈𝑥 gas phase average local velocity in the 𝑥 direction

(m s−1 )
𝐴 Arrhenius constant (mol s−1 g−1 Pa−(𝛼+𝛽) )
𝑢 velocity (m s−1 )
𝑎𝑗 coefficient in the discretized equation (the unit
u velocity vector (m s−1 )
depends on the case)
𝑉 volume of the reactor (m3 )
𝑏 constant term in the discretization equation (the unit
𝑊 𝑚𝑗 molecular mass (g mol−1 )
depends on the case)
𝑤𝑖 function’s weight (–)
𝐶mult multiplication coefficient used for fitness scaling (–)
𝑤̇ cat catalyst weight density (g m−3 )
𝐶𝑝 specific heat at constant pressure (J kg−1 K−1 )
𝛥𝑥 grid × dimension (m)
𝑐ine inertial coefficient (–)
𝑥𝑐𝑟 methane conversion rate (–)
𝐷𝑗 mass diffusivity of the species 𝑗 (m2 s−1 )
𝑌𝑗 mass fraction of the species 𝑗 (–)
𝑑 dimensionless foam ligament radius (–)
𝑦𝑐𝑟 carbon monoxide conversion rate (–)
𝑑p average pore diameter (mm)
𝐸a activation energy (J mol−1 ) Greek letters
𝑒 dimensionless cubic node length (–)
𝛼 order of the reaction with respect to methane (–)
𝑓 fitness value (–)
𝛽 order of the reaction with respect to water (–)
𝑓 𝑟𝑎𝑐 mole fraction (–)
𝛤 diffusive term (the unit depends on the case)
𝛥𝐺 change of standard Gibbs free energy (J mol−1 )
𝜀 porosity (–)
𝑔s shape function derived for the metallic foam
𝜁 hydrogen productivity (–)
structure (–)
𝜄 ratio of the catalyst amount in a specific reactor to
𝛥𝐻 enthalpy change (J mol−1 )
the amount of catalyst in the reference reactor (–)
𝐾p permeability (m2 )
𝜆 thermal conductivity (W m−1 K−1 )
𝐾 equilibrium constant (–)
+∕− 𝜇 dynamic viscosity (Pa s)
𝑘WGS rate constant of the forward/backward water-gas-
𝜌 density (g m−3 )
shift reaction (mol s−1 m−3 Pa−2 )
𝜌0 density of the gas mixture (kg m−3 )
𝑘MSR rate constant of the methane/steam reforming
reaction (mol s−1 m−3 Pa−2 ) 𝜌̄ pseudo density (the unit depends on the case)
𝐿 maximal × dimension (m) 𝜏 tortuosity (–)
𝑙 node-to-node length (m) 𝜙𝑗 dependent variable (the unit depends on the case)
𝑛 molar flow rate (mol−1 ) 𝛹𝑟 convective term in the 𝑟 direction (the unit depends
on the case)
𝑃 pressure (Pa)
𝛹𝑥 convective term in the 𝑥 direction (the unit depends
𝑝𝑗 partial pressure of the species 𝑗 (Pa)
on the case)
𝑄 heat flux (W m−3 )
𝑄s heat source/sink (W m−3 ) Subscripts
𝑅 reactor radius (m)
A, B, C, D unit cell subsections
𝑅𝑖 thermal resistances of the porous media cell subsec-
avg average value
tions (m2 K W−1 )
CH4 methane or based on the methane conversion rate
𝑅̄ universal gas constant (J mol−1 K−1 )
E node to the right of the central node
𝑅MSR steam reforming reaction rate (mol s−1 m−3 )
e interface to the right of the central node
𝑅WGS water–gas shift reaction rate (mol s−1 m−3 )
eff effective value
𝑟m arithmetic mean between 𝑟n and 𝑟s (m)
in inlet
𝛥𝑟 grid 𝑟 dimension (m)
𝑗 chemical species, grid element’s location, specimen
𝑆𝑗 mass source/sink of the species 𝑗 (kg s−1 m−3 )
loc local average, both over the gas and solid phase
𝑆 source term (the unit depends on the case)
MSR steam reforming reaction
𝑆𝐶 constant in the linear source term
max maximal value
𝑆𝑃 dependent variable coefficient in the linear source
min minimal value
term (the unit depends on the case)
mix gas mixture
𝑆𝐶 steam-to-carbon ratio (–)
N node above the central node
𝑇 temperature (K)
n interface above the central node
𝛥𝑇 difference between maximal and minimal tempera-
norm normalized value
tures inside the reactor (K)
out outlet
𝑈𝑟 gas phase average local velocity in the 𝑟 direction (m
s−1 ) P central node of the grid

2
M. Pajak et al. Energy and AI 13 (2023) 100264

the temperature distribution may not only improve the conditions but
S node below the central node also achieve easier control of the process [36]. The presented research
s interface below the central node aims to alleviate the negative consequences of the strong endothermic
𝑇 temperature based character of the process, via the introduction of radial division of
W node to the left of the central node the catalytic insert. A similar approach is successfully investigated
WGS water-gas-shift reaction by Cherif et al. who introduced a radially-structured catalytic insert,
w interface to the left of the central node with alternately placed platinum and nickel-based catalysts for an
ATR reforming reactor [37]. The concept originates from the approach
Chemical species proposed by Settar et al. [38]. The research predicted an introduction of
CH4 methane macro-patterned active surfaces with an introduction of metallic foam
CO carbon monoxide matrices, focusing on providing advantageous thermal conditions for
the reaction [39,40]. The presented research extends the concept to fill
CO2 carbon dioxide
the whole reactor’s volume with a catalytic composite of nickel and
H2 hydrogen
yttria-stabilized zirconia (Ni/YSZ), to maximize the reaction region in
H2 O steam the reactor. Further, the reforming unit is divided into segments in the
radial direction, instead of the longitudinal division [41]. Non-catalytic
metallic foam is used as a substitute for parts of the catalyst, to adjust
the intensity of the reaction proceeding, leading to the unification of
1. Introduction the thermal field inside the reactor. To define the optimal alignment
of the catalyst, an evolutionary algorithm is coupled with an in-house
Hydrogen technologies are one of the promising directions of the reforming simulation [42]. The presented analysis includes:
clean energy sector development [1]. The research on hydrogen is
conducted to provide a reliable alternative to the currently dominant • Investigation of the macro-patterning concept applicability with
fossil fuel energy sources [2,3]. Hydrogen might be used as an en- the introduction of the catalytic insert radial division.
ergy carrier for internal combustion or fuel cells, both resulting in • Introduction of two separate principles for the configuration of
steam being the main product [4,5]. However, with the application the segments.
of hydrogen technology, some crucial issues arise. The first issue re- • Comprehensive sensitivity analysis to define the finest performing
gards hydrogen acquisition, as it does not occur on Earth in its pure set of the evolutionary algorithm parameters.
form. The second matter is hydrogen storage, with currently no ef- • Analysis of the results robustness, via measuring the hydrogen
fective measures of long-term storage [6,7]. The two most common productivity of specimens defined by the specific algorithms.
processes for on-the-spot production of hydrogen are water electrolysis
and the reforming of hydrocarbons [8,9]. Water electrolysis is a pro- 2. Mathematical model
cess predicting the breaking of chemical bonds between oxygen and
hydrogen in particles of water. The current state of the electrolysis The analyzed process incorporates a plug-flow reactor, maintaining
development is far from meeting economical requirements [10–12]. axial symmetry throughout its whole volume. The reactor consists of
The only reasonable use of water electrolysis is to deplete surplus a straight, cylindrical pipe, allowing for an accessible replacement of
energy generated from renewable sources during low market demand the catalyst insert when required. The introduced macro-patterning
and short-term storage of hydrogen for further use during increased strategy predicts modification of the catalyst insert’s morphology and
demand for energy [13]. The second measure for hydrogen production its partial substitution with a non-catalytic material. Substitution of
is the reforming reaction [14,15]. The reforming process is a catalytic the catalyst material serves to enhance the heat transfer inside the
reaction used for the conversion of hydrocarbons for the production of reactor [41]. Metallic foams are chosen for application to the in-
hydrogen [16,17]. The reforming process can be successfully applied dicated issues [43]. To provide a relevant measure for altering the
to the conversion of biofuels, allowing the reforming process to be temperature distribution inside the reactor, the catalyst insert is divided
considered a renewable hydrogen source [18,19]. Furthermore, the into segments. Each segment may consist of a catalyst or a non-
process can be successfully applied as a measure of carbohydrate-based catalytic material, with their morphological parameters individually
waste gases or plastic recycling, establishing it as a prominent for assigned. The insert’s segmentation is carried out according to two
hydrogen generation [20,21]. The reforming technology brings a series separate strategies. The first strategy predicts division in the radial
of issues regarding the thermal conditions occurring inside the reactor. direction with maintaining a constant width of inlet surfaces of each
The strong endothermic nature of the process results in the occurrence cylinder (Fig. 1(b)). The second strategy predicts the application of
of thermal stresses and may lead to a shortening of the reactor’s segments with an identical area of the inlet surfaces for each of the
lifespan [22]. The presented research aims to reduce the drawbacks of coaxial cylinders (Fig. 1(c)). The design is prepared to contain exclu-
the process, by enhancement of the thermal conditions. The majority sively simple geometrical shapes, allowing to maintain an insignificant
of researchers focused on the parametric study and optimization of computational complexity of the domain [44].
the reaction conditions, resulting in improvements only to a certain
extent [23,24]. Further development of the process is pursued by the 2.1. Chemical reactions model
introduction of new materials and design concepts, including new
catalyst structures [25,26], the introduction of new kinds of catalyst The process is assumed to be dominated by three reactions, as
supports [27], or by rethinking the design of the reactor itself [28– reported in literature [36,45,46]. The reactions are steam reforming of
31]. An opportunity for intensification of the reaction is described in methane (MSR) (Eq. (1)), dry reforming of methane (DRY) (Eq. (2)),
a work by Palma et al. [32], who introduced a structured catalyst and water–gas shift reaction (WGS) (Eq. (3)) [47]. The stoichiometric
for the enhancement of the axial and radial temperature distribution. equations for the reactions are presented below:
The proper handling of heat in the reforming process is confirmed to CH4 + H2 O → 3H2 + CO,
enhance the overall process conduction [33,34]. Furthermore, a rapid kJ
temperature decay at the upstream region of the reactor results in 𝛥𝐻MSR = 206.1 , (1)
mol
thermal stresses forming in the reactor. Thus, leading to its uneven
degradation and reduction of the unit’s lifetime [35]. A unification of CH4 + CO2 ⇌ 2H2 + 2CO,

3
M. Pajak et al. Energy and AI 13 (2023) 100264

Fig. 1. The investigated macro-patterinig designs: (a) conventional reactor with homogeneous and continuous catalytic insert, (b) Strategy I - catalytic insert divided in the radial
direction (equal width of inlet surface), (c) Strategy II - catalytic insert divided in the radial direction (equal area of inlet surface).

kJ
𝛥𝐻DRY = 247 , (2) assumption. Following the statements, CO, CO2 , H2 and H2 O have to
mol
satisfy the equilibrium equation, expressed by the following formula:
( )
CO + H2 O ⇌ H2 + CO2 , 𝑘+
WGS
𝑝CO2 𝑝H2 0
𝛥𝐺WGS
kJ 𝐾WGS = − = = exp − . (7)
𝛥𝐻WGS = −41.15 . (3) 𝑘WGS 𝑝CO 𝑝H2 O 𝑅𝑇
mol
The reformer is supplied with a mixture of H2 , CO2 , and CH4 . The The WGS reaction rate can be further expressed using Eq. (8):
composition of the feedstock is determined by the steam-to-carbon ratio 𝑅WGS = 𝑘+ 𝑝 𝑝 + 𝑘− 𝑝 𝑝 . (8)
WGS CO H2 O WGS H2 CO2
(𝑆𝐶) parameter, defining the ratio of steam to methane provided at the
reactor’s inlet. The process conditions are remarkably influenced by the The value of 𝑅WGS can be acquired through the analysis of the re-
forming process stoichiometry and balancing the chemical species [63,
𝑆𝐶 value, as the reactions’ rates depend directly on the composition
64]. Balancing the reaction’s stoichiometry allows for the calculation
of the inlet gases [48]. A proper setting of the 𝑆𝐶 is crucial for the
of the partial pressures included in Eq. (7), leading to the relation
prevention of the carbon deposition phenomenon [49]. Adverse process
describing the WGS reactions’ rate (Eq. (9)) [53].
conditions can result in carbon particles precipitating on the catalyst
surface [50]. A proper setting of the ratios and the process’ temperature 𝑛outlet
CH
𝑛inlet
CH
⋅ 𝑥𝑐𝑟
4 4
𝑅WGS = = 𝑦𝑐𝑟. (9)
are proven to have the most significant influence on the alleviation of 𝑉 𝑉
the poisoning hazard [51]. The enthalpy changes 𝛥𝐻 are taken from Following, the methane conversion rate 𝑥𝑐𝑟 and carbon monoxide
literature [41,52]. Knowledge of their rates is essential to allow the conversion rate 𝑦𝑐𝑟 can be specified with:
inclusion of the reactions into the model. According to the research
𝑛inlet
CH
− 𝑅MSR 𝑉
conducted by Brus et al. [53], the effective rate of MSR and DRY 𝑥𝑐𝑟 = 1 − 4
, (10)
reactions can be expressed with a common equation: 𝑛inlet
CH
( ) ( )𝛽 √ 4
𝐸 𝐾WGS + 3𝑥𝑐𝑟 − 𝜒 − 𝜔
𝑅ef f = 𝑤̇ cat 𝐴MSR exp − a 𝑝𝛼CH 𝑝H2 O + 𝑝CO2 . (4) 𝑦𝑐𝑟 = , (11)
𝑅𝑇 4 2(𝐾WGS − 1)
The individual reaction rates for the MSR and DRY reactions can be where:
distinguished as follows:
𝜒 = (𝐾WGS 𝑆𝐶 + 3𝑥𝑐𝑟)2 , (12)
𝑝H2 O
𝑅MSR = 𝑅ef f , (5)
𝑝CO2 + 𝑝H2 O 𝜔 = 4𝐾WGS 𝑥𝑐𝑟(𝐾WGS − 1)(𝑆𝐶 − 𝑥𝑐𝑟). (13)
𝑝CO2
𝑅DRY = 𝑅ef f . (6) Coupling Eq. (10) with Eq. (9) and applying simple mathematical
𝑝CO2 + 𝑝H2 O transformations results in the formulation of a final expression for the
The reforming reaction is reported to occur rather slowly [54]. WGS reaction’s rate:
However, the WGS reaction has a more unpredictable nature [55]. 𝑅WGS = 𝑅MSR 𝑦𝑐𝑟. (14)
Thus, the preparation of a formula, returning proper values regardless
of the process conditions, is not feasible. According to Ahmed and The mass consumption and production rates of the reforming pro-
Föger, the WGS reaction can be assumed to maintain equilibrium under cess reactions (Eqs. (1)–(3)), are summarized in the Table 1. The values
specific conditions [56]. The equilibrium assumption has been success- are further applied to the mass transfer equation (Eq. (19)), as a part
of its source terms. The heat generation rates by the described reac-
fully applied in other numerical analyses [57–59]. The assumption also
tions are calculated depending on the rates of the reactions (Eqs. (5),
meets a good agreement between the calculations and the experimental
(6), (14)) and the values of enthalpy change 𝛥𝐻 [46,53]. The heat
data [60–62]. Taking into account the provided literature review, the
generation rates are given below:
WGS reaction is assumed to be in the equilibrium state in the presented
analysis. The process conditions are designed to satisfy the equilibrium 𝑄MSR = −𝛥𝐻MSR 𝑅MSR , (15)

4
M. Pajak et al. Energy and AI 13 (2023) 100264

Table 1
Mass sources/sinks.
Species Mass generation MSR Mass generation WGS Mass generation DRY Summarized generation
H2 3𝑅MSR 𝑀H2 𝑅WGS 𝑀H2 2𝑅DRY 𝑀H2 3𝑅MSR 𝑀H2 + 𝑅WGS 𝑀H2 + 2𝑅DRY 𝑀H2
CO 𝑅MSR 𝑀CO −𝑅WGS 𝑀CO 2𝑅DRY 𝑀CO 𝑅MSR 𝑀CO − 𝑅WGS 𝑀CO + 2𝑅DRY 𝑀CO
CO2 0 𝑅WGS 𝑀CO2 −𝑅DRY 𝑀CO2 𝑅WGS 𝑀CO2 − 2𝑅DRY 𝑀H2
CH4 −𝑅MSR 𝑀CH4 0 −𝑅DRY 𝑀H2 −𝑅MSR 𝑀CH4 − 𝑅DRY 𝑀H2
H2 O −𝑅MSR 𝑀H2 𝑂 −𝑅WGS 𝑀H2 𝑂 0 −𝑅MSR 𝑀H2 𝑂 −𝑅WGS 𝑀H2 𝑂

( ) [ ( ) ]
𝑄DRY = −𝛥𝐻DRY 𝑅DRY , (16) 𝜌0 𝜕𝑈𝑟 𝜕𝑈 𝜕𝑃 𝜇 𝜕 2 𝑈𝑟 1 𝜕 𝜕𝑈 𝑈
+ 𝑈𝑟 𝑟 = −
𝑈𝑥 + + 𝑟 𝑟 − 𝑟
𝜀20 𝜕𝑥 𝜕𝑟 𝜕𝑟 𝜀 𝜕𝑥 2 𝑟 𝜕𝑟 𝜕𝑟 𝑟 2

𝜇 𝜌0 𝑐ine √ 2
𝑄WGS = −𝛥𝐻WGS 𝑅WGS . (17) − 𝑈𝑟 − √ 𝑈𝑟 𝑈𝑥 + 𝑈𝑟2 . (22)
𝐾p 𝐾p
2.2. Heat and mass transfer model The permeability 𝐾p of the specific segment is calculated using (23),
basing on the information about its porosity 𝜀 [71]:
The fundamental transport equations are incorporated into the
𝜀(1 − (1 − 𝜀)1∕3 )
mathematical model. Considering the computational domain defined 𝐾p = 𝑑2, (23)
36((1 − 𝜀)1∕3 − (1 − 𝜀)) p
for the needs of the analysis, the equations are implemented for two
dimensions. Therefore, the model’s equations are solved along the axis where 𝑑p stands for an average pore diameter. The inertial coefficient
and radius of the reactor’s geometry. The volume-averaging method 𝑐ine was calculated using [72]:
was chosen for the derivation of the governing equations implemented √ √
𝜀 (1 − 𝜀) 1 −1
in the model used for this analysis. The process parameters are locally 𝑐ine = 0.0095𝑔s−0.8 (1.18 ) , (24)
3(𝜏 − 1) 3𝜋 𝑔s
averaged for each representative volume and included in the mathe-
matical formulas [65]. Values of continuity (Eq. (18)), mass transfer where tortuosity 𝜏 and shape function 𝑔s are expressed with following
(Eq. (19)), momentum (Eqs. (21) and (22)) and energy equations equations [71,72]:
(Eq. (27)) characterize the transport phenomena occurring during the 𝜀
𝜏= , (25)
reforming process. The equations are derived for the laminar flow. The 1 − (1 − 𝜀)1∕3
analyzed fluids are considered to be Newtonian and incompressible.
Thus, the continuity equation takes the following form [66,67]: 1−𝜀
𝑔s = 1 − exp(− ). (26)
( ) ( ) 0.04
𝜕 𝜌0 𝑈 𝑥 1 𝜕 𝑟𝜌0 𝑈𝑟
+ = 0, (18) The energy conservation equation (Eq. (27)) describes the process of
𝜕𝑥 𝑟 𝜕𝑟 heat transfer during the reaction. The equation includes local thermal
The species conservation is calculated using molar fractions of conditions, the materials’ parameters, and the heat sources calculated
species taking part in the reaction (Eq. (19)). The formulated equation using Eqs. (15)–(17).
is derived from Fick’s law of diffusion [36]. The mass sources and sinks ( ) ( ) ( )
𝑆𝑗 depend on the MSR, DRY, and WGS rates and molar masses of the 𝜕𝑇 𝜕𝑇 𝜕 𝜕𝑇 1 𝜕 𝜕𝑇
𝜌0 𝐶𝑝 𝑈𝑥 loc + 𝑈𝑟 loc = 𝜆ef f loc + 𝑟𝜆ef f loc + 𝑄s .
species taking part in the reaction [22,68]. The exact equations defining 𝜕𝑥 𝜕𝑟 𝜕𝑥 𝜕𝑥 𝑟 𝜕𝑟 𝜕𝑟
the values of 𝑆𝑗 are described in Table 1. (27)
( ) ( )
𝜕𝑌𝑗 𝜕𝑌𝑗 𝜕 𝜕𝑌𝑗
𝜌0 𝑈 𝑥 + 𝑈𝑟 = 𝜌0 𝐷𝑗,ef f Due to the application of metallic foam, a proper relation describing
𝜕𝑥 𝜕𝑟 𝜕𝑥 𝜕𝑥 the 𝜆ef f of the material is essential [72]. The model chosen for calcu-
( )
1 𝜕 𝜕𝑌 𝑗
+ 𝑟𝜌0 𝐷𝑗,ef f + 𝑆𝑗 . (19) lating the value of the 𝜆ef f was proposed by Boomsma and Poulikakos
𝑟 𝜕𝑟 𝜕𝑟 and further corrected by Dai et al. [73,74]. The outcome relation allows
The effective mass diffusivity of species 𝐷𝑗ef f was calculated using for the derivation of equations describing the thermal conductivity of
the equation explained below (Eq. (20)) [69]: metallic foams. According to the literature review, an adequate model
√ including the morphology of metallic foams is prepared [75].
𝐷𝑗,ef f = (1 − 1 − 𝜀)𝐷𝑗 . (20) √
2𝑙
The diffusion of substance 𝑗 in the gas mixture 𝐷𝑗 is computed using 𝜆ef f = , (28)
2(𝑅A + 𝑅B + 𝑅C + 𝑅D )
Fuller’s method and Blanc’s law. The gases’ properties are taken from
the literature [70]. The values assumed for the flow model describe where 𝑅A - 𝑅D stand for the thermal resistances of the porous media
the local phase average of the gas control volume. The momentum cell subsections [41,74].
conservation depends directly on the insert’s morphology. The materi-
als composing the insert are considered porous. Therefore, parameters 3. Numerical model
describing the material structure have to be included in the equations.
The parameters are porosity 𝜀, permeability 𝐾𝑝 , and inertial coefficient
To solve systems of the defined equations a discretization of the
𝑐ine [42]. A separate momentum equation is formulated for each of the
previously prepared continuous computational domain is necessary.
computational domain dimensions (Eqs. (21) and (22)).
( ) [ ( )] The numerical model is prepared using the Finite Volume Method [76].
𝜌0 𝜕𝑈 𝜕𝑈 𝜕𝑃 𝜇 𝜕 2 𝑈𝑥 1 𝜕 𝜕𝑈 A uniform computational grid with evenly spaced nodes is formed,
𝑈𝑥 𝑥 + 𝑈𝑟 𝑥 = − + + 𝑟 𝑥
𝜀20 𝜕𝑥 𝜕𝑟 𝜕𝑥 𝜀 𝜕𝑥2 𝑟 𝜕𝑟 𝜕𝑟 to assure a minimal complexity of the calculations, [44]. The grid’s
𝜌0 𝑐ine √ dimensions are established at 150 elements in the longitudinal and 25
𝜇
− 𝑈𝑥 − √ 𝑈𝑥 𝑈𝑥2 + 𝑈𝑟2 , (21)
𝐾p 𝐾p elements in the radial directions, resulting in square-shaped elements.

5
M. Pajak et al. Energy and AI 13 (2023) 100264

Fig. 2. The coaxial segments configurations: (a) equal width of rings, (b) equal inlet and outlet surface.

Table 2 Table 3
Coefficients and terms for substitution in Eq. (29). Rings’ dimensions.
Eq. 𝜙 𝛹𝑥 𝛹𝑟 𝛤 𝑆 Radius Equal width (cm) Equal surface (cm)
√
(19) 𝑌𝑗 𝜌0 𝑈𝑥 𝜌0 𝑈𝑟 𝜌0 𝐷𝑗,ef f 𝑆𝑗 r1 1 5
𝜌0 𝜌0 𝜌0 𝑐ine √ 2 √
𝜇 𝜕𝑃 𝜇 r2 2 10
(21) 𝑈𝑥 𝑈𝑥 𝑈𝑟 − − 𝑈 − √ 𝑈𝑥 𝑈𝑥 + 𝑈𝑟2
𝜀20 𝜀20 𝜀0 𝜕𝑥 𝐾p 𝑥 𝐾p √
r3 3 15
𝜌0 𝜌0 𝜇 𝜕𝑃 𝜇 𝜌0 𝑐ine √ 2 𝜇𝑈𝑟 √
(22) 𝑈𝑟 𝑈𝑟 𝑈𝑟 − − 𝑈 − √ 𝑈𝑟 𝑈𝑥 + 𝑈𝑟2 − r4 4 2 5
𝜀20 𝜀20 𝜀0 𝜕𝑟 𝐾p 𝑟 𝐾p 𝜀0 𝑟2
r5 5 5
(27) 𝑇 𝜌 0 𝐶p 𝑈 𝑥 𝜌0 𝐶p 𝑈𝑟 𝜆ef f 𝑄s

The values resulting from solving the Eqs. (30) are presented in
3.1. Transport equations Table 3.
The computational domain is indicated with the red dashed line
Integration of the fundamental transport equations of the reforming
(Fig. 3). Segments are parametrized with porosity, average pore size,
process over the created control volumes is conducted [77]. The partial
and catalytic properties. The analyzed model is steady and the flow of
differential equations covering the species conservation (Eq. (19)), the
gases is assumed to be laminar and occurring in one direction. The flow
momentum conservation equation (Eqs. (21)–(22)), and the energy
characteristics are assumed accordingly to flow through porous media
conservation equation (Eq. (27)) may be described in a generalized
form [77], as follows: described by NIeld et al. [78]. Fluids taking part in the process are
( ) ( ) assumed to be Newtonian. The presented numerical analysis is a quasi-
𝜕𝜙 𝜕𝜙 𝜕 𝜕𝜙 1 𝜕 𝜕𝜙 three-dimensional case. Due to the axial symmetry of the reactor’s tube,
𝛹𝑥 + 𝛹𝑟 = 𝛤 + 𝑟𝛤 + 𝑆. (29)
𝜕𝑥 𝜕𝑟 𝜕𝑥 𝜕𝑥 𝑟 𝜕𝑟 𝜕𝑟 a computational domain can be constrained to a two-dimensional cross-
The Eq. (29) is prepared to allow substitution of any dependant section of the reformer. Implementation of proper boundary conditions
variable 𝜙. The coefficients on the left side of the Eq. (29), 𝛹𝑥 and allows assuming the process occurs in the same manner throughout the
𝛹𝑟 describe the convective terms in 𝑥 and 𝑟 directions and may be reactor’s whole volume [77].
substituted accordingly to the use case. The diffusive terms, represented
by 𝛤 , are placed on the right-hand side of the equation, along with 3.3. Boundary conditions and materials
the source terms 𝑆. The corresponding terms and coefficients are
summarized in Table 2. The source terms 𝑆 values regard the catalytic To complete the preparation of the numerical analysis, setting the
segments only. The reaction is assumed to be put to a halt for the non- properties of the materials and the boundary conditions is vital. Proper
catalytic region of the insert. Therefore, no heat and mass production setting of 𝑆𝐶 value is essential for this analysis, due to the threat
nor consumption is predicted for the metallic foam segments.
of carbon deposit formation [79]. If the ratios are chosen incorrectly,
the carbon deposition phenomenon may occur, poisoning the reactor’s
3.2. Coaxial segments configuration
catalyst [80]. Choosing an appropriate steam-to-carbon ratio allows
The reactor has been divided into five concentric segments, consti- for avoiding the described hazard. The value of SC for this analysis
tuting four hollow cylinders and a core. The segments may take one of was set to be equal to 2.0 [-], preventing this occurrence [47,51].
two configurations. The first one predicts equal width of the inlet and The catalyst insert’s segments are assumed to consist of one of two
outlet surfaces, while the second configuration predicts an equal area materials. The first material is a catalytic composite of nickel and yttria-
of inlet and outlet surfaces. The two possible setups are summarized in stabilized zirconia (Ni/YSZ). The second material, serving the purpose
Fig. 2. of heat transport medium, is chosen to be steel foam. The metallic
The exact dimensions of the particular rings for Fig. 2(b) are calcu- foam segments are introduced for the unification of the temperature
lated using the system of Eqs. (30), as follows: distribution inside the reactor. Suppression of the reforming reaction,
combined with a considerable heat exchange surface for foams, is
⎧ 2 2 2
⎪2𝜋𝑟1 = 5 𝜋𝑅 expected to let the gases mixture reheat effectively [81]. The value of
⎪ ( 2 2
) 2 𝜆solid is set at 22 W m−1 K−1 for the catalytic material [82], while the
⎪2𝜋 𝑟2 − 𝑟1 = 𝜋𝑅2
⎪ ( 2 )
5 value of 𝜆solid for the steel foam is established at 30 W m−1 K−1 [83].
2 2
⎨2𝜋 𝑟3 − 𝑟2 = 5
𝜋𝑅2 (30) The dimensionless cubic node length 𝑒 is set to be equal to 0.0339, as
⎪ ( ) 2 explained by Boomsma and Poulikakos [73]. The temperature of the
⎪2𝜋 𝑟24 − 𝑟23 = 5
𝜋𝑅2
⎪ ( ) fuel flowing inside the unit is considered to reach the temperature of
⎪2𝜋 𝑟2 − 𝑟2 = 2
𝜋𝑅2 the reformer instantly. The symmetry boundary conditions are set at
⎩ 5 4 5

6
M. Pajak et al. Energy and AI 13 (2023) 100264

Fig. 3. Boundary conditions applied in the analysis.

the symmetry axis, while the no-slip boundary condition is applied at algorithm analyses subsequent, finite populations of test specimens,
the wall of the reformer. The boundary conditions necessary for the which parameters are inherited from the specimens acquiring the best
solution of the governing equations are summarized below. results in the preceding generation [84]. The results are evaluated by
user-defined fitness functions [85]. The functions are exclusive to a
Thermal boundary conditions certain problem. If more than one function is defined, the functions
have to be combined into a single value, which will be further processed
• inlet temperature 𝑇 = 𝑇in = 900 K at 𝑥 = 0 and 0 ≤ 𝑟 < 𝑅, by the genetic algorithm itself [86]. The remaining part of the genetic
• outlet temperature 𝜕𝑇 ∕𝜕𝑥 = 0 at 𝑥 = 𝐿 and 0 ≤ 𝑟 < 𝑅, algorithm operation is the mutation procedure. The mutation is used
• symmetry boundary condition 𝜕𝑇 ∕𝜕𝑟 = 0 at 0 ≤ 𝑥 < 𝐿 and 𝑟 = 0, as an additional procedure for preventing the local extremum trap
• wall heat flux 𝜕𝑇 ∕𝜕𝑟 = 𝑐𝑜𝑛𝑠𝑡 at 0 ≤ 𝑥 < 𝐿 and 𝑟 = 𝑅. and a measure for the introduction of genes that were not present in
the initial population or are just not possible to acquire on the way
Constant heat flux is assumed on the reactor’s wall, and the wall of the crossover procedure [87]. Coupling of the prepared numerical
itself is the medium for heat exchange between the reactor and the analysis with the algorithm requires proper parametrization of reactor
system boundaries. The wall thickness is equal to 0.001 m and consists parameters [86]. The catalyst insert is set to contain 5 coaxial segments.
of the same material as the metallic foams. The value of 𝜆solid for the Each of the segments is assumed to consist of either a catalytic com-
reactor’s wall is equal to 30 W m−1 K−1 [83]. The boundary conditions posite of nickel with yttria-stabilized zirconia (Ni/YSZ) or a stainless
essential for the calculation of the species conservation equations are steel foam [75]. The insert regions may differ in their morphological
presented below. parameters regardless of the constituting material. The parameters
altered by the algorithm are segment catalytic character, porosity 𝜀 and
Mass transport boundary conditions average pore diameter 𝑑p . The porosity values available for application
in the analysis are constrained to the range between 0.5 and 0.8.
• inlet mole fractions 𝑌𝑗 = 𝑌𝑗,in at 𝑥 = 0 and 0 ≤ 𝑟 < 𝑅, The lower boundary is set at 0.5, as lower porosity values resulted in
• outlet mole fractions 𝜕𝑌𝑗 ∕𝜕𝑥 = 0 at 𝑥 = 𝐿 and 0 ≤ 𝑟 < 𝑅, over-significant pressure drops during the preliminary calculations. The
• symmetry boundary condition 𝜕𝑌𝑗 ∕𝜕𝑟 = 0 at 0 ≤ 𝑥 < 𝐿 and 𝑟 = 0, choice of 0.8 as the upper boundary is motivated by lower accessibility
to the manufacturing methods determined for metallic foams of higher
• no-slip boundary condition 𝑌𝑗 = 0 at 0 ≤ 𝑥 < 𝐿 and 𝑟 = 𝑅.
porosities [88]. The average pore diameters 𝑑p are calculated for each
To complete the numerical model’s setting, the boundary conditions segment, based on the values of the porosity [75]. The upper boundary
for solving the momentum conservation equations are needed. The of the 𝑑p range is restricted by the model conditions. According to the
conditions are applied in the manner listed below. volume-averaging method, the size of a single pore is not allowed to
exceed the size of a single numerical grid element. Therefore, the upper
Momentum boundary conditions boundary is limited to 0.002 m [65]. Furthermore, segment porosity 𝜀
has a direct influence on the catalyst density. The catalyst density 𝑤̇ cat
• inlet velocity 𝑈 = 𝑈in = 0.15 m s−1 at 𝑥 = 0 and 0 ≤ 𝑟 < 𝑅, is calculated using Eq. (31):
• outlet velocity 𝜕𝑈 ∕𝜕𝑥 = 0 at 𝑥 = 𝐿 and 0 ≤ 𝑟 < 𝑅, 𝑤̇ cat = 𝜌cat ⋅ (1 − 𝜀), (31)
• symmetry boundary condition 𝜕𝑈 ∕𝜕𝑟 = 0 at 0 ≤ 𝑥 < 𝐿 and 𝑟 = 0,
• no-slip boundary condition 𝑈 = 0 at 0 ≤ 𝑥 < 𝐿 and 𝑟 = 𝑅. where 𝜌cat is equal to 5.3448 ⋅ 106 g m−3 . The chemical reaction kinetics
model derived for the implemented numerical simulation regards a
All of the boundary conditions are summarized in Fig. 3. The accu- Ni/YSZ composite with a 60:40 Ni to YSZ ratio [75]. Nickel serves
racy of the developed numerical model was tested against experimen- the catalytic purpose, while YSZ serves as a scaffold. Therefore, the
tal results and the numerical error was verified using semi-analytical basic catalyst density for the Ni/YSZ element should be calculated
benchmarking. The falsification procedure included in our previous concerning the ratio of the materials. The solid nickel density equals
research has successfully confirmed the model to be reliable [41,53]. to 8.908 ⋅ 106 g m−3 [89]. The value of 𝜌cat is calculated by applying
the given ratio to the value of the solid nickel density [75]. The
4. Genetic algorithm definition of the relations allows commencing the operation of the GA.
The algorithm starts its operation with the preparation of the initial
The conducted optimization includes the application of an evolu- generation, with the specimens’ parameters fully randomized. First,
tionary algorithm, to define the optimal design of the catalytic insert the segments’ character is randomized. The algorithm is expected to
for the coaxial segments. The genetic algorithm (GA) is selected to generate catalytic and non-catalytic segments in a ratio of 50:50, as the
be the optimization procedure used for the needs of the analysis. The randomization procedure uses a uniform distribution for its seed [90].

7
M. Pajak et al. Energy and AI 13 (2023) 100264

Fig. 4. Summary of the genetic algorithm procedure.

When the specific segment’s character is determined, the algorithm Having computed the set of the 𝛥𝑇loc𝑖,𝑗 for the whole numerical grid,
randomizes its porosity 𝜀 and average pore size 𝑑p values. As the final the algorithm selects the highest value among computed 𝛥𝑇loc𝑖,𝑗 and
initialization step, the GA calculates the catalyst density for each of the saves it as a global 𝛥𝑇 value representing a specific reactor (Eq. (34)).
segments designated to consist of Ni/YSZ composite. Having defined { }
the initial parameters, the reforming simulation code is executed for 𝛥𝑇 = max 𝑇loc1,1 , 𝑇loc1,2 , … , 𝑇loc𝑖,𝑗 , (34)
each of the generated reactors. The overall operation of the algorithm where 𝑖 and 𝑗 indices represent the numerical grid’s dimensions. The
prepared for the needs of the research is presented in Fig. 4. final thermal fitness value 𝑓𝑇 is computed using the following formula:
The conducted investigation regards optimization of the tempera-
ture field distribution inside the reactor, simultaneously maintaining
𝛥𝑇
a considerable methane conversion rate [42]. Therefore, the research 𝑓𝑇 = 1 − , (35)
𝛥𝑇max
is considered a multi-objective optimization, as the fitness analysis is
required to investigate two separate factors. The first factor is tempera- where 𝛥𝑇max stands for the highest temperature difference value re-
ture distribution. The main objective is to reduce the temperature gradi- ported for a single generation. The acquired 𝑓CH4 and 𝑓𝑇 values are
ents occurring throughout the reactor’s volume during the reaction. The further combined into a single fitness value 𝑓 , using the weighted-sum
strong endothermic character of the process has a consequence in large method [92]:
temperature decreases in the regions in which the fuel concentration is ∑
𝑁
significant. The constant transition of the temperature field leads to the 𝑓= 𝑤𝑖 𝑓𝑖 . (36)
establishment of thermal stresses, leading to premature degradation of 𝑖=1

the catalyst material [91]. The presented research predicts the search where 𝑁 represents the number of the specimens. The weights 𝑤𝑖 are
for an optimal distribution of catalytic and non-catalytic segments and established at 0.2, 0.4, 0.5, 0.6, and 0.8. The weights are applied to
proper selection of their porosity 𝜀 and average pore size 𝑑p . The first both of the fitness functions to conduct a sensitivity analysis, depending
function 𝑓CH4 regards the methane conversion rate. The fuel conversion on the assumed importance of each.
rate is calculated using the molar fractions of the methane at the inlet
and the outlet of a specific reactor. The 𝑓CH4 value is calculated using 5. Numerical analysis
Eq. (32):
(𝑓 𝑟𝑎𝑐CH4in − 𝑓 𝑟𝑎𝑐CH4out ) The presented investigation regards the optimization of the temper-
𝑓CH4 = , (32) ature distribution inside the reforming reactor. Moderation of the tem-
𝑓 𝑟𝑎𝑐CH4in
perature field is achieved via the introduction of the macro-patterning
The remaining fitness function 𝑓𝑇 evaluates the extent of the uni- concept [41]. The concept predicts the modification of the catalytic
fication of the temperature distribution inside the catalytic insert. insert of a reforming reactor, resulting in a change in the process
Essentially, the thermal fitness objective is to find the 𝛥𝑇 value, being conditions. The insert is divided into segments in the radial direction.
a representative temperature difference for a specific reactor. To assure Each segment may consist of catalytic Ni/YSZ or stainless steel foam.
reliability and accuracy of the mathematical formula, the thermal The algorithm conducts the search for an optimal composition of the
fitness is established to perform the temperature distribution analysis insert, altering the percentage of the catalyst used, segments placement,
for each of the grid elements separately [42]. The algorithm iter- and segments structural parameters. The metallic foam is introduced to
ates through each of the created control volumes and calculates the suppress the reforming reaction. A sensitivity analysis is conducted to
temperature difference between the central node 𝑇P and each of the specify the most suitable multiplication factors for the weighted-sum
neighboring control volumes 𝑇N , 𝑇E , 𝑇S , 𝑇W (Fig. 5). Further, each fitness calculation (Eq. (36)) [93]. The computed fitness is used to rank
of the control volumes has its specific temperature difference 𝛥𝑇loc𝑖,𝑗 the specimens in a single genetic algorithm’s iteration. An analysis of
assigned. The 𝛥𝑇loc𝑖,𝑗 value is chosen to be the highest temperature the genetic algorithm behavior, depending on the weight applied to the
difference among the values calculated for each of the neighboring two fitness functions is conducted, to define the most suitable approach
control volumes. The 𝛥𝑇loc𝑖,𝑗 for each of the cells is based on the to the issue, depending on the applied variant of the macro-patterning
following formula: strategy. The analysis includes a comparison of results for five different
𝛥𝑇loc𝑖,𝑗 = sets of fitness weights, as presented in Table 4.
{ } The analysis has to be preceded by a definition of a reference case. A
max |𝑇P𝑖,𝑗 − 𝑇N𝑖,𝑗 |, |𝑇P𝑖,𝑗 − 𝑇E𝑖,𝑗 |, |𝑇P𝑖,𝑗 − 𝑇S𝑖,𝑗 |, |𝑇P𝑖,𝑗 − 𝑇W𝑖,𝑗 | . (33) comparison of subsequent optimization results with the reference case

8
M. Pajak et al. Energy and AI 13 (2023) 100264

Fig. 5. Temperature field evaluation strategy.

Fig. 6. Temperature field distribution - reference case (𝑇in = 900 K, 𝑢in = 0.15 m s−1 , 𝑆𝐶 = 2.0, 𝑤̇ cat = 2.47 ⋅ 106 ).

Table 4
Summary of the analyzed numerical cases.
Strategy I Strategy II 𝑤CH4 𝑤𝑇
(Fig. 1b)) (Fig. 1c))
5.1 5.6 0.2 0.8
5.2 5.7 0.4 0.6
5.3 5.8 0.5 0.5
5.4 5.9 0.6 0.4
5.5 5.10 0.8 0.2

will serve as a benchmark for defining if any significant improvement in

the reaction conditions has been noted for a specific case. The reference
case defined for the presented analysis is a conventional reforming
reactor, with a catalytic insert being homogeneous and continuous
throughout the whole volume of the reactor. The whole insert consists
of a composite of nickel and yttrium-stabilized-zirconia (Ni/YSZ) with
a ratio of 60:40. The porosity 𝜀 of the reactor’s catalyst is set at 50%,
resulting in the catalyst density equal to 2.47 ⋅ 106 g m−3 . The average Fig. 7. Radius-averaged molar fractions - reference case (𝑇in = 900 K, 𝑢in = 0.15 m
pore diameter 𝑑p is established at 1.5 mm. The boundary conditions s−1 , 𝑆𝐶 = 2.0, 𝑤̇ cat = 2.47 ⋅ 106 ).

applied for the reference case are corresponding to the conditions set
for the optimization procedure (Section 3.3).
The thermal characteristics of the reforming reaction in a conven- (Catalyst insert division strategy II - Fig. 1(c)). To allow a direct
tional reactor can be observed in Fig. 6. The reactor dimensions are comparison between each of the composed algorithms, hydrogen pro-
divided by the reactor’s radius (5 cm), to acquire a dimensionless set of ductivity 𝜁 is introduced. Hydrogen productivity delivers insight into
coordinates. Observable temperature gradients are present throughout the increase of the effectiveness of the reforming reaction for each
the whole volume of the reactor. The most significant temperature drop of the optimized specimens. The 𝜁 parameter is an exact ratio of the
can be noticed at the inlet of the reactor, due to a considerable amount
hydrogen output and the amount of the applied catalyst 𝜄 for a specific
of energy required for the activation of the reforming reaction [41,46].
reactor (Eq. (37)). The amount of the catalyst 𝜄 is calculated as the ratio
The molar fractions of specimens taking part in the process are illus-
trated in Fig. 7. The molar fractions are averaged over the reactor’s of the catalyst used in a specific reactor, to the amount of the catalyst
radius. Thus, the graph describes the change of the radius-averaged applied for the reference case.
molar fractions in the longitudinal direction of the reactor. The values H2out
of the reactor’s length are divided by the reactor’s radius, to acquire 𝜁= (37)
𝜄
dimensionless values.
The coaxial segments may accommodate one of two possible con- A sensitivity analysis for each of them is conducted to comprehen-
figurations. The first variant predicts cylinders with equal width of the sively analyze specific catalytic insert division strategies. The sensitivity
inlet surface (Catalyst insert division strategy I - Fig. 1(b)), while the analysis regards changing the weights 𝑤CH4 and 𝑤𝑇 applied for specific
second introduces cylinders with an equal area of the inlet surfaces fitness functions (Section 4).

9
M. Pajak et al. Energy and AI 13 (2023) 100264

Fig. 8. Strategy I - 𝑤CH4 = 0.2, 𝑤𝑇 = 0.8 30th generation: (a) temperature field distribution, (b) radius-averaged molar fractions, (c) segments distribution (𝑇in = 900 K, 𝑢in = 0.15 m
s−1 , 𝑆𝐶 = 2.0).

Fig. 9. Strategy I - 𝑤CH4 = 0.4, 𝑤𝑇 = 0.6 30th generation: (a) temperature field distribution, (b) radius-averaged molar fractions, (c) segments distribution (𝑇in = 900 K, 𝑢in = 0.15 m
s−1 , 𝑆𝐶 = 2.0).

Fig. 10. Strategy I - 𝑤CH4 = 0.5, 𝑤𝑇 = 0.5 30th generation: (a) temperature field distribution, (b) radius-averaged molar fractions, (c) segments distribution (𝑇in = 900 K, 𝑢in =
0.15 m s−1 , 𝑆𝐶 = 2.0).

5.1. Strategy I - Thermal fitness 80%, methane conversion 20% 5.4. Strategy I - Thermal fitness 40%, methane conversion 60%

Fig. 8 represents the results acquired for the best specimen found by Fig. 11 presents the results acquired for the best specimen found by
the algorithm with fitness weights 𝑤CH4 = 0.2 and 𝑤𝑇 = 0.8. Fig. 9(a)
the algorithm with fitness weights set at 𝑤CH4 = 0.6 and 𝑤𝑇 = 0.4
showcases the temperature distribution, (b) the radius-averaged molar
fractions of chemical specimens along the reactor axis, and (c) the
segments distribution designed by the algorithm.
5.5. Strategy I - Thermal fitness 20%, methane conversion 80%

5.2. Strategy I - Thermal fitness 60%, methane conversion 40%

The performance of the reactor designed by the algorithm with
fitness weights 𝑤CH4 = 0.8 and 𝑤𝑇 = 0.2 is shown in Fig. 12.
The results acquired for the algorithm with fitness weights 𝑤CH4 =
To allow a straightforward comparison between the catalyst inserts
0.4 and 𝑤𝑇 = 0.6 are summarized in Fig. 9.
division strategies, hydrogen productivity 𝜁 is calculated for the op-
timal cases found by each algorithm. The hydrogen productivity is
5.3. Strategy I - Thermal fitness 50%, methane conversion 50% summarized in Table 5. It can be noticed, that the algorithm presented
in Section 5.4 acquired the highest value of hydrogen productivity,
The performance of the solution found by the algorithm with fitness simultaneously reducing the amount of used catalyst by 81%, compared
weights 𝑤CH4 = 0.5 and 𝑤𝑇 = 0.5 is summarized in Fig. 10. to the reference case.

10
M. Pajak et al. Energy and AI 13 (2023) 100264

Fig. 11. Strategy I - 𝑤CH4 = 0.6, 𝑤𝑇 = 0.4 30th generation: (a) temperature field distribution, (b) radius-averaged molar fractions, (c) segments distribution (𝑇in = 900 K, 𝑢in =
0.15 m s−1 , 𝑆𝐶 = 2.0).

Fig. 12. Strategy I - 𝑤CH4 = 0.2, 𝑤𝑇 = 0.8 30th generation: (a) temperature field distribution, (b) radius-averaged molar fractions, (c) segments distribution (𝑇in = 900 K, 𝑢in =
0.15 m s−1 , 𝑆𝐶 = 2.0).

Fig. 13. Strategy II - 𝑤CH4 = 0.2, 𝑤𝑇 = 0.8 30th generation: (a) temperature field distribution, (b) radius-averaged molar fractions, (c) segments distribution (𝑇in = 900 K, 𝑢in =
0.15 m s−1 , 𝑆𝐶 = 2.0).

Table 5 5.7. Strategy II - Thermal fitness 60%, methane conversion 40%

Hydrogen productivity for the catalytic insert division strategy I.
𝑤CH4 /𝑤𝑇 H2𝑜𝑢𝑡 𝜄 𝜁
The results of the solution found by the algorithm with fitness
REF 0.581 1.00 0.581 weights 𝑤CH4 = 0.4 and 𝑤𝑇 = 0.6 are summarized in Fig. 14.
0.2/0.8 0.255 0.20 1.278
0.4/0.6 0.268 0.19 1.398
0.5/0.5 0.231 0.20 1.157 5.8. Strategy II - Thermal fitness 50%, methane conversion 50%
0.6/0.4 0.407 0.45 0.900
0.8/0.2 0.218 0.20 1.090 The results acquired for the algorithm with fitness weights 𝑤CH4 =
0.5 and 𝑤𝑇 = 0.5 are presented in Fig. 15.

5.6. Strategy II - Thermal fitness 80%, methane conversion 20% 5.9. Strategy II - Thermal fitness 40%, methane conversion 60%

Fig. 16 shows the performance acquired for the best specimen found
The results acquired after computing thirty subsequent generations by the algorithm with fitness weights set at 𝑤CH4 = 0.6 and 𝑤𝑇 = 0.4
of the algorithm with fitness weights set at 𝑤CH4 = 0.2 and 𝑤𝑇 = 0.8 is
presented in Fig. 13. Fig. 13 (a) showcases the temperature distribution, 5.10. Strategy II - Thermal fitness 20%, methane conversion 80%
(b) the radius-averaged molar fractions of chemical specimens along
the reactor axis, and (c) the segments distribution designed by the The performance of the specimen found for the algorithm with
algorithm. fitness weights 𝑤CH4 = 0.4 and 𝑤𝑇 = 0.6 is summarized in Fig. 17.

11
M. Pajak et al. Energy and AI 13 (2023) 100264

Fig. 14. Strategy II - 𝑤CH4 = 0.4, 𝑤𝑇 = 0.6 30th generation: (a) temperature field distribution, (b) radius-averaged molar fractions, (c) segments distribution (𝑇in = 900 K, 𝑢in =
0.15 m s−1 , 𝑆𝐶 = 2.0).

Fig. 15. Strategy II - 𝑤CH4 = 0.5, 𝑤𝑇 = 0.5 30th generation: (a) temperature field distribution, (b) radius-averaged molar fractions, (c) segments distribution (𝑇in = 900 K, 𝑢in =
0.15 m s−1 , 𝑆𝐶 = 2.0).

Fig. 16. Strategy II - 𝑤CH4 = 0.6, 𝑤𝑇 = 0.4 30th generation: (a) temperature field distribution, (b) radius-averaged molar fractions, (c) segments distribution (𝑇in = 900 K, 𝑢in =
0.15 m s−1 , 𝑆𝐶 = 2.0).

Table 6 configuration’s performance is tested against five sets of fitness function

Hydrogen productivity for the catalytic insert division strategy II. weights. The weight sensitivity analysis is conducted to assure the
𝑤CH4 /𝑤𝑇 H2𝑜𝑢𝑡 𝜄 𝜁 best possible practices during future numerical analyses of the macro-
REF 0.581 1.00 0.581 patterned reforming reactor design optimization. Analyzing the results
0.2/0.8 0.001 0.00 N/A leads to the following conclusions:
0.4/0.6 0.492 0.32 1.527
0.5/0.5 0.175 0.32 0.548 1. Fitness weights set at 0.6 for the methane conversion and 0.4 for
0.6/0.4 0.536 0.53 1.616
thermal fitness tend to deliver better results than the remaining
0.8/0.2 0.200 0.27 0.744
algorithms.
2. Radial segments with equal widths of the inlet surface return
results of a higher quality than radial segments with equal areas
The hydrogen productivity acquired for optimal specimens included of the inlet surfaces.
in the analysis of strategy II is summarized in Table 6. 3. The highest values of hydrogen productivity are acquired for the
reactor with an equal area of inlet surfaces (Strategy II), proving it
6. Conclusions to be the most valid approach to the optimization of the catalytic
insert design.
The conducted numerical investigation regards radial segmentation 4. The introduction of the macro-patterning concept is proven to
of the catalyst insert. Two different configurations of coaxial segments enhance the effectiveness of the reforming reaction. The hydrogen
are proposed. The radial segments may either have equal widths of the productivity is almost tripled for the best solution when compared
inlet surfaces or equal areas of the inlet surfaces. Each possible segment with the reference case.

12
M. Pajak et al. Energy and AI 13 (2023) 100264

Fig. 17. Strategy II - 𝑤CH4 = 0.8, 𝑤𝑇 = 0.2 30th generation: (a) temperature field distribution, (b) radius-averaged molar fractions, (c) segments distribution (𝑇in = 900 K, 𝑢in =
0.15 m s−1 , 𝑆𝐶 = 2.0).

The research indicates the effect of altering the steam-reforming [7] Tarhan C, Ç.il MA. A study on hydrogen, the clean energy of the fu-
thermal conditions on the overall process. The possible improvement ture: Hydrogen storage methods. J Energy Storage 2021;40:102676. http:
//dx.doi.org/10.1016/j.est.2021.102676, URL: https://www.sciencedirect.com/
of the process’s effectiveness on temperature distribution unification
science/article/pii/S2352152X21004151.
is confirmed. The proposed macro-patterning concept proves the pos- [8] Hassan I, Ramadan HS, Saleh MA, Hissel D. Hydrogen storage technologies for
sibility of acquiring even higher methane conversion rates, using the stationary and mobile applications: Review, analysis and perspectives. Renew
same amount of catalyst as in the case of the reference case. However, Sustain Energy Rev 2021;149:111311.
[9] Azizan MT, Aqsha A, Ameen M, Syuhada A, Klaus H, Abidin SZ, et al. Catalytic
the macro-patterned reactor would have to suffer in compactness, as
reforming of oxygenated hydrocarbons for the hydrogen production: an outlook.
an extension of its dimensions is required to acquire the exact level of Biomass Convers Biorefinery 2020;1–24. http://dx.doi.org/10.1007/s13399-020-
conversion. Considering the magnitude of possible economical benefits, 01081-6.
an enlargement to a certain extent is acceptable. [10] Nguyen T, Abdin Z, Holm T, Mérida W. Grid-connected hydrogen production
via large-scale water electrolysis. Energy Convers Manage 2019;200:112108.
http://dx.doi.org/10.1016/j.enconman.2019.112108.
Declaration of competing interest [11] Yukesh Kannah R, Kavitha S, Preethi, Parthiba Karthikeyan O, Kumar G, Dai-
Viet NV, et al. Techno-economic assessment of various hydrogen production
The authors declare that they have no known competing finan- methods – A review. Bioresour Technol 2021;319:124175. http://dx.doi.org/
10.1016/j.biortech.2020.124175, URL: https://www.sciencedirect.com/science/
cial interests or personal relationships that could have appeared to
article/pii/S0960852420314498.
influence the work reported in this paper. [12] Nazir SM, Cloete JH, Cloete S, Amini S. Pathways to low-cost clean hy-
drogen production with gas switching reforming. Int J Hydrogen Energy
Data availability 2021;46(38):20142–58.
[13] Chi J, Yu H. Water electrolysis based on renewable energy for hydrogen produc-
tion. Chin J Catal 2018;39:390–4. http://dx.doi.org/10.1016/S1872-2067(17)
Data will be made available on request 62949-8.
[14] Zhang H, Sun Z, Hu YH. Steam reforming of methane: Current states of catalyst
Acknowledgments design and process upgrading. Renew Sustain Energy Rev 2021;149:111330.
http://dx.doi.org/10.1016/j.rser.2021.111330, URL: https://www.sciencedirect.
com/science/article/pii/S136403212100616X.
The presented work was supported by the Polish National Agency [15] Nkulikiyinka P, Yan Y, Güleç F, Manovic V, Clough PT. Prediction of sorp-
for Academic Exchange (NAWA), within the Strategic Partnerships tion enhanced steam methane reforming products from machine learning
Program, Project No. BPI/PST/2021/1/00023. AVIZO Software used based soft-sensor models. Energy and AI 2020;2:100037. http://dx.doi.org/10.
1016/j.egyai.2020.100037, URL: https://www.sciencedirect.com/science/article/
in this study was purchased thanks to the program ‘‘Excellence Ini-
pii/S2666546820300379.
tiative – Research University’’ for the AGH University of Science and [16] Taherian Z, Khataee A, Han N, Orooji Y. Hydrogen production through methane
Technology. reforming processes using promoted-Ni/mesoporous silica: A review. J Ind
Eng Chem 2022;107:20–30. http://dx.doi.org/10.1016/j.jiec.2021.12.006, URL:
https://www.sciencedirect.com/science/article/pii/S1226086X21006675.
References
[17] Faheem HH, Abbas SZ, Tabish AN, Fan L, Maqbool F. A review on mathe-
matical modelling of direct internal reforming- solid oxide fuel cells. J Power
[1] Qazi UY. Future of hydrogen as an alternative fuel for next-generation industrial Sources 2022;520:230857. http://dx.doi.org/10.1016/j.jpowsour.2021.230857,
applications; challenges and expected opportunities. Energies 2022;15(13). http: URL: https://www.sciencedirect.com/science/article/pii/S0378775321013446.
//dx.doi.org/10.3390/en15134741, URL: https://www.mdpi.com/1996-1073/ [18] Zhao X, Joseph B, Kuhn J, Ozcan S. Biogas reforming to syngas: A review.
15/13/4741. IScience 2020;23:101082.
[2] Zou C, Zhao Q, Zhang G, Xiong B. Energy revolution: From a fossil energy era [19] Gao N, Milandile MH, Quan C, Rundong L. Critical assessment of plasma
to a new energy era. Nat Gas Ind B 2016;3:1–11. http://dx.doi.org/10.1016/j. tar reforming during biomass gasification: A review on advancement in
ngib.2016.02.001. plasma technology. J Hard Mater 2022;421:126764. http://dx.doi.org/10.1016/j.
[3] Liobikienė G, Dagiliūtė R. Do positive aspects of renewable energy contribute jhazmat.2021.126764, URL: https://www.sciencedirect.com/science/article/pii/
to the willingness to pay more for green energy? Energy 2021;231:120817. S0304389421017295.
http://dx.doi.org/10.1016/j.energy.2021.120817. [20] Saad JM, Williams PT. Catalytic dry reforming of waste plastics from differ-
[4] Xu H, Ma J, Tan P, Chen B, Wu Z, Zhang Y, et al. Towards online opti- ent waste treatment plants for production of synthesis gases. Waste Manage
misation of solid oxide fuel cell performance: Combining deep learning with 2016;58:214–20. http://dx.doi.org/10.1016/j.wasman.2016.09.011, URL: https:
multi-physics simulation. Energy and AI 2020;1:100003. http://dx.doi.org/10. //www.sciencedirect.com/science/article/pii/S0956053X16305141.
1016/j.egyai.2020.100003, URL: https://www.sciencedirect.com/science/article/ [21] Zhang B, Chen Y, Zhang B, Peng R, Lu Q, Yan W, et al. Cyclic performance
pii/S2666546820300033. of coke oven gas - steam reforming with assistance of steel slag derivates
[5] Shadidi B, Najafi G, Yusaf T. A review of hydrogen as a fuel in inter- for high purity hydrogen production. Renew Energy 2022;184:592–603. http://
nal combustion engines. Energies 2021;14:6209. http://dx.doi.org/10.3390/ dx.doi.org/10.1016/j.renene.2021.11.123, URL: https://www.sciencedirect.com/
en14196209. science/article/pii/S0960148121017183.
[6] Hassan I, Ramadan HS, Saleh MA, Hissel D. Hydrogen storage technologies for [22] Mozdzierz M, Brus G, Sciazko A, Komatsu Y, Kimijima S, Szmyd JS. An
stationary and mobile applications: Review, analysis and perspectives. Renew attempt to mimimize the temperature gradient along a plug-flow methane/steam
Sustain Energy Rev 2021;149:111311. http://dx.doi.org/10.1016/j.rser.2021. reforming reactor by adopting locally controlled heating zones. J Phys Conf Ser
111311. 2014;530:012040(1)–8).

13
M. Pajak et al. Energy and AI 13 (2023) 100264

[23] Simpson AP, Lutz AE. Exergy analysis of hydrogen production via steam methane [46] Xu J, Froment GF. Methane steam reforming: II. Diffusional limitations and
reforming. Int J Hydrogen Energy 2007;32:4811–20. http://dx.doi.org/10.1016/ reactor simulation. AIChE J 1989;35:97–103.
j.ijhydene.2007.08.025. [47] Brus G, Nowak R, Szmyd JS, Komatsu Y, Kimijima S. An experimental and
[24] Naseri AT, Peppley BA, Pharoah JG. A systematic parametric study on the theoretical approach for the carbon deposition problem during steam reforming
effect of a catalyst coating microstructure on its performance in methane of model biogas. J Theoret Appl Mech 2015;53:273–84. http://dx.doi.org/10.
steam reforming. Int J Hydrogen Energy 2015;40:16086–95. http://dx.doi.org/ 15632/jtam-pl.53.2.273.
10.1016/j.ijhydene.2015.10.043. [48] Usman M, Wan Daud W, Abbas HF. Dry reforming of methane: Influence
[25] Ali S, Al-Marri MJ, Abdelmoneim AG, Kumar A, Khader MM. Catalytic evaluation of process parameters—A review. Renew Sustain Energy Rev 2015;45:710–44.
of nickel nanoparticles in methane steam reforming. Int J Hydrogen Energy http://dx.doi.org/10.1016/j.rser.2015.02.026.
2016;41:22876–85. http://dx.doi.org/10.1016/j.ijhydene.2016.08.200. [49] Koncewicz W, Moździerz M, Brus G. A fast Gaussian process-based method
[26] Kolaczkowski ST, Awdry S, Smith T, Thomas D, Torkuhl L, Kolvenbach R. Poten- to evaluate carbon deposition during hydrocarbons reforming. Int J Hydrogen
tial for metal foams to act as structured catalyst supports in fixed-bed reactors. Energy 2021, In Press. http://dx.doi.org/10.1016/j.ijhydene.2021.07.213.
Catal Today 2016;273:221–33. http://dx.doi.org/10.1016/j.cattod.2016.03.047. [50] Kaczmarczyk R. Thermodynamic analysis of the effect of green hydrogen addition
[27] Yadav AK, Vaidya PD. Renewable hydrogen production by steam reforming of to a fuel mixture on the steam methane reforming process. Energies 2021;14(20).
butanol over multiwalled carbon nanotube-supported catalysts. Int J Hydrogen http://dx.doi.org/10.3390/en14206564.
Energy 2019;44:30014–23. http://dx.doi.org/10.1016/j.ijhydene.2019.09.054. [51] Tomiczek M, Kaczmarczyk R, Mozdzierz M, Brus G. A numerical analysis of
[28] Butcher H, Quenzel CJ, Breziner L, Mettes J, Wilhite BA, Bossard P. Design of heat and mass transfer during the steam reforming process of ethane. Heat Mass
an annular microchannel reactor (AMR) for hydrogen and/or syngas production Transf 2018;54:2305–14.
via methane steam reforming. Int J Hydrogen Energy 2014;39:18046–57. http: [52] Mazhar A, Khoja AH, Azad AK, Mushtaq F, Naqvi SR, Shakir S, et al. Performance
//dx.doi.org/10.1016/j.ijhydene.2014.04.109. analysis of TiO2-modified Co/MgAl2O4 catalyst for dry reforming of methane
[29] Cherif A, Nebbali R. Numerical analysis on autothermal steam methane re- in a fixed bed reactor for syngas (H2, CO) production. Energies 2021;14:3347.
forming: Effects of catalysts arrangement and metal foam insertion. Int J http://dx.doi.org/10.3390/en14113347.
Hydrogen Energy 2019;44(39):22455–66. http://dx.doi.org/10.1016/j.ijhydene. [53] Brus G, Komatsu Y, Kimijima S, Szmyd JS. An analysis of biogas reforming
2018.12.203. process on Ni/YSZ and Ni/SDC catalysts. Int J Thermodyn 2012;15:43–51.
[30] Yun J, Yu S. Transient behavior of 5 kW class shell-and-tube methane steam [54] Brus G, Kimijima S, Szmyd JS. Experimental and numerical analysis of transport
reformer with intermediate temperature heat source. Int J Heat Mass Transfer phenomena in an internal indirect fuel reforming type Solid Oxide Fuel Cells
2019;134:600–9. http://dx.doi.org/10.1016/j.ijheatmasstransfer.2019.01.078. using Ni/SDC as a catalyst. J Phys Conf Ser 2012;395:012159. http://dx.doi.
[31] Meloni E, Martino M. A short review on Ni based catalysts and related org/10.1088/1742-6596/395/1/012159.
engineering issues for methane steam reforming. Catalysts 2020;10:352. http: [55] Nagata S, Momma A, Kato T, Kasuga Y. Numerical analysis of output
//dx.doi.org/10.3390/catal10030352. characteristics of tubular SOFC with internal reformer. J Power Sources
[32] Palma V, Ricca A, Martino M, Meloni E. Innovative structured catalytic systems 2001;101:60–71.
for methane steam reforming intensification. Chem Eng Process: Process Intensif [56] Ahmed K, Föger K. Approach to equilibrium of the water-gas shift reaction on
2017;120:207–15. a Ni/zirconia anode under SOFC conditions. J Power Sources 2001;103:150–3.
[33] Dubinin AM, Shcheklein SE, Tuponogov VG, Ershov MI. Mini CHP based on the [57] Iwai H, Yamamoto Y, Saito M, Yoshida H. Numerical simulation of intermediate-
electrochemical generator and impeded fluidized bed reactor for methane steam temperature direct-internal-reforming planar solid oxide fuel cell. Energy
reforming. Int J Hydrogen Energy 2018;43:13543–9. http://dx.doi.org/10.1016/ 2011;36:2225–34.
j.ijhydene.2018.05.151. [58] Sciazko A, Komatsu Y, Brus G, Kimijima S, Szmyd JS. A novel approach to
[34] Cherif A, Nebbali R, Sen F, Sheffield JW, Doner N, Nasseri L. Modeling and improve the mathematical modelling of the internal reforming process for solid
simulation of steam methane reforming and methane combustion over continuous oxide fuel cells using the orthogonal least squares method. Int J Hydrogen Energy
and segmented catalyst beds in autothermal reactor. Int J Hydrogen Energy 2014;39:16372–89. http://dx.doi.org/10.1016/j.ijhydene.2014.07.130.
2022;47(15):9127–38. http://dx.doi.org/10.1016/j.ijhydene.2021.12.250. [59] Peng X, Jin Q. Molecular simulation of methane steam reforming reaction
[35] Mozdzierz M, Brus G, Sciazko A, Komatsu Y, Kimijima S, Szmyd JS. Towards for hydrogen production. Int J Hydrogen Energy 2022;47(12):7569–85. http:
a thermal optimization of a methane/steam reforming reactor. Flow Turbul //dx.doi.org/10.1016/j.ijhydene.2021.12.105.
Combust 2016;97:171–89. [60] Brus G. Experimental and numerical studies on chemically reacting gas flow
[36] Mozdzierz M, Chalusiak M, Kimijima S, Szmyd JS, Brus G. An afterburner- in the porous structure of a solid oxide fuel cells internal fuel reformer. Int J
powered methane/steam reformer for a solid oxide fuel cells application. Heat Hydrogen Energy 2012;37:17225–34.
Mass Transf 2018;54:2331–41. http://dx.doi.org/10.1007/s00231-018-2331-5. [61] Sciazko A, Komatsu Y, Brus G, Kimijima S, Szmyd JS. An application of
[37] Cherif A, Lee J-S, Nebbali R, Lee C-J. Novel design and multi-objective optimiza- generalized least squares method to an analysis of methane/steam reforming
tion of autothermal steam methane reformer to enhance hydrogen production process on a Ni/YSZ catalyst. ECS Trans 2013;57:2987–96. http://dx.doi.org/10.
and thermal matching. Appl Therm Eng 2022;217:119140. http://dx.doi.org/ 1149/05701.2987ecst.
10.1016/j.applthermaleng.2022.119140, URL: https://www.sciencedirect.com/ [62] Sciazko A, Komatsu Y, Brus G, Kimijima S, Szmyd JS. A novel approach to the
science/article/pii/S1359431122010717. experimental study on methane/steam reforming kinetics using the Orthogonal
[38] Settar A, Nebbali R, Madani B, Abboudi S. Numerical investigation on the wall- Least Squares method. J Power Sources 2014;262:245–54. http://dx.doi.org/10.
coated steam methane reformer improvement: Effects of catalyst layer patterns 1016/j.jpowsour.2014.03.097.
and metal foam insertion. Int J Hydrogen Energy 2017;42:1490–8. [63] Brus G, Miyawaki K, Iwai H, Saito M, Yoshida H. Tortuosity of an SOFC anode
[39] Settar A, Abboudi S, Lebaal N. Effect of inert metal foam matrices on hydrogen estimated from saturation currents and a mass transport model in comparison
production intensification of methane steam reforming process in wall-coated with a real micro-structure. Solid State Ion 2014;265:13–21. http://dx.doi.org/
reformer. Int J Hydrogen Energy 2018;43:12386–97. http://dx.doi.org/10.1016/ 10.1016/j.ssi.2014.07.002.
j.ijhydene.2018.04.215. [64] Wolf M. Thermodynamic assessment of the stability of bulk and nanoparticulate
[40] Settar A, Mansouri Z, Nebbali R. Impact of Ni-based catalyst patterning on cobalt and nickel during dry and steam reforming of methane. RSC Adv
hydrogen production from MSR : External steam reformer modelling ScienceDi- 2021;11(30):18187–97.
rect Impact of Ni-based catalyst patterning on hydrogen production from MSR : [65] Carbonell RG, Whitaker S. Heat and mass transfer in porous media. In: Bear J,
External steam reformer modelling. Int J Hydrogen Energy 2018;44:11346–54. Corapcioglu MY, editors. Fundamentals of transport phenomena in porous media.
http://dx.doi.org/10.1016/j.ijhydene.2018.09.171. Dordrecht: Springer; 1984.
[41] Pajak M, Mozdzierz M, Chalusiak M, Kimijima S, Szmyd JS, Brus G. A numerical [66] Nishino T, Komori H, Iwai H, Suzuki K. Development of a comprehensive
analysis of heat and mass transfer processes in a macro-patterned methane/steam numerical model for analyzing a tubular-type indirect internal reforming SOFC.
reforming reactor. Int J Hydrogen Energy 2018;43:20474–87. http://dx.doi.org/ In: 1st International fuel cell science, engineering and technology conference
10.1016/j.ijhydene.2018.09.058. Rochester, New York, USA:521-528. 2003.
[42] Pajak M, Brus G, Szmyd JS. Genetic algorithm-based strategy for the steam [67] Nishino T, Iwai H, Suzuki K. Comprehensive numerical modeling and analysis
reformer optimization. Int J Hydrogen Energy 2021, In Press. http://dx.doi.org/ of a cell-based indirect internal reforming tubular SOFC. J Fuel Cell Sci Technol
10.1016/j.ijhydene.2021.10.046. 2006;3:33–44.
[43] Boomsma K, Poulikakos D, Zwick F. Metal foams as compact high performance [68] Tan WC, Iwai H, Kishimoto M, Brus G, Szmyd JS, Yoshida H. Numerical
heat exchangers. Mech Mater 2003;35:1161–76. http://dx.doi.org/10.1016/j. analysis on effect of aspect ratio of planar solid oxide fuel cell fueled with
mechmat.2003.02.001. decomposed ammonia. J Power Sources 2018;384:367–78. http://dx.doi.org/10.
[44] Kaw A, Kalu EK, Nguyen D. Numerical methods with applications. Florida: 1016/j.jpowsour.2018.03.011.
University of South Florida; 2011. [69] Suzuki K, Iwai H, Nishino T. Electrochemical and thermo-fluid modeling of a
[45] Komatsu Y, Kimijima S, Szmyd JS. A performance analysis of a solid oxide fuel tubular solid oxide fuel cell with accompaning indirect internal fuel reforming.
cell - micro gas turbine hybrid system using biogas. ECS Trans 2009;25:1061–70. In: Sunden B, Faghri M, editors. Transport phenomena in fuel cells. Southampton:
http://dx.doi.org/10.1149/1.3205631. WIT Press; 2005, p. 83–131.

14
M. Pajak et al. Energy and AI 13 (2023) 100264

[70] Poling BE, Prausnitz JM, O’Connell JP. The properties of gases and liquids. New [82] Kawashima T, Hishinuma M. Thermal properties of porous Ni/YSZ composites
York: McGraw-Hill; 2001, http://dx.doi.org/10.1016/0894-1777(88)90021-0. at high temperatures. Mater Trans JIM 1996;37:1518–24.
[71] Yang X, Lu TJ, Kim T. An analytical model for permeability of isotropic [83] Peet MJ, Hasan HS, Bhadeshia HKDH. Prediction of thermal conductivity of
porous media. Phys Lett Sec A: General, Atomic and Solid State Physics steel. Int J Heat Mass Transfer 2011;54:2602–8. http://dx.doi.org/10.1016/j.
2014;378:2308–11. http://dx.doi.org/10.1016/j.physleta.2014.06.002. ijheatmasstransfer.2011.01.025.
[72] Bhattacharya A, Calmidi VV, Mahajan RL. Thermophysical properties of high [84] Agharezaei P, Sahu T, Shock J, O’Brien PG, Ghuman KK. Designing catalysts via
porosity metal foams. Int J Heat Mass Transfer 2002;45:1017–31. http://dx.doi. evolutionary-based optimization techniques. Comput Mater Sci 2023;216:111833.
org/10.1016/S0017-9310(01)00220-4. [85] Tai XY, Ocone R, Christie SD, Xuan J. Multi-objective optimisation with hybrid
[73] Boomsma K, Poulikakos D. On the effective thermal conductivity of a machine learning strategy for complex catalytic processes. Energy and AI
three- dimensionally structured fluid-saturated metal foam. Heat Mass Transf 2022;7:100134. http://dx.doi.org/10.1016/j.egyai.2021.100134.
2001;44:827–36. http://dx.doi.org/10.1016/S0017-9310(00)00123-X. [86] Goldberg DE. Genetic algorithms in search optimization & machine learning.
[74] Dai Z, Nawaz K, Park YG, Bock J, Jacobi AM. Correcting and extending the Boston: Addison-Wesley Longman Publishing Co., Inc.; 1989.
Boomsma-Poulikakos effective thermal conductivity model for three-dimensional, [87] Raja BD, Patel V, Jhala RL. Thermal design and optimization of fin-and-tube
fluid-saturated metal foams. Int Commun Heat Mass Transfer 2010;37:575–80. heat exchanger using heat transfer search algorithm. Therm Sci Eng Prog
http://dx.doi.org/10.1016/j.icheatmasstransfer.2010.01.015. 2017;4:45–57.
[75] Pajak M, Buchaniec S, Kimijima S, Szmyd JS, Brus G. A multiobjective opti- [88] Banhart J. Manufacture, characterisation and application of cellular metals
mization of a catalyst distribution in a methane/steam reforming reactor using a and metal foams. Prog Mater Sci 2001;46:559–632. http://dx.doi.org/10.1016/
genetic algorithm. Int J Hydrogen Energy 2021;46:20183–97. http://dx.doi.org/ S0079-6425(00)00002-5.
10.1016/j.ijhydene.2020.02.228. [89] Everhart J. Engineering properties of nickel and nickel alloys. Springer Science
[76] Moukalled F, Mangani L, Darwish M. The finite volume method in computational & Business Media; 2012.
fluid dynamics: An advanced introduction with OpenFOAM and Matlab. Cham: [90] Jerrum MR, Valiant LG, Vazirani VV. Random generation of combinatorial
Springer International Publishing; 2016, p. 103–35. structures from a uniform distribution. Theoret Comput Sci 1986;43:169–88.
[77] Patankar SV. Numerical heat transfer and fluid flow. Washington: Hemisphere; http://dx.doi.org/10.1016/0304-3975(86)90174-X.
1980. [91] Pajak M, Brus G, Szmyd JS. Catalyst distribution optimization scheme for effec-
[78] Nield DA, Bejan A, et al. Convection in porous media. vol. 3, Springer; 2006. tive green hydrogen production from biogas reforming. Energies 2021;14:5558.
[79] Bartholomew CH. Carbon deposition in steam reforming and methanation. Catal http://dx.doi.org/10.3390/en14175558.
Rev 1982;24(1):67–112. http://dx.doi.org/10.1080/03602458208079650. [92] Zitzler E, Thiele L. Multiobjective evolutionary algorithms: a comparative case
[80] Lakhapatri SL, Abraham MA. Deactivation due to sulfur poisoning and carbon study and the strength Pareto approach. IEEE Trans Evol Comput 1999;3:257–71.
deposition on Rh-Ni/Al2O3 catalyst during steam reforming of sulfur-doped n- http://dx.doi.org/10.1109/4235.797969.
hexadecane. Appl Catal A: General 2009;364:113–21. http://dx.doi.org/10.1016/ [93] Davahli A, Shamsi M, Abaei G, Khosravi A. Empirical analyses of genetic
j.apcata.2009.05.035. algorithm and grey wolf optimiser to improve their efficiency with a new multi-
[81] Zhao CY. Review on thermal transport in high porosity cellular metal foams objective weighted fitness function for feature selection in machine learning
with open cells. Int J Heat Mass Transfer 2012;55:3618–32. http://dx.doi.org/ classification: the roadmap. J Exp Theor Artif Intell 2022;8:1–36.
10.1016/j.ijheatmasstransfer.2012.03.017.

15