i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 4 8 ( 2 0 2 3 ) 3 5 8 2 1 e3 5 8 3 7

Available online at www.sciencedirect.com

ScienceDirect

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

One-dimensional kinetic model with heat transfer

and axial dispersion of molten-metal bubble
column reactors for hydrogen production via
methane pyrolysis

Bang Thanh Le a, Son Ich Ngo a, Young-Il Lim a,*, Uen-Do Lee b
a
Center of Sustainable Process Engineering (CoSPE), Department of Chemical Engineering, Hankyong National
University, 327 Jungangro, Anseong, Gyeonggi-do, 17579, Republic of Korea
b
Energy System R&D Group, Korea Institute of Industrial Technology (KITECH), Cheonan, 331-825, Republic of
Korea

highlights graphical abstract

One-dimensional molten metal-

based bubble column reactor
(MMBCR) model was developed.

MMBCR model for CH4 pyrolysis
Dax

was coupled with heat transfer

rcat
T

and axial dispersion. XM

Dax

Heat transfer and axial dispersion rnon
slightly affect bench-scale
MMBCRs CH4 conversion.

Axial dispersion lowered the CH4
conversion of an industrial-scale
MMBCR by 17%.

The MMBCR model is useful to
design the reactors and optimize
operating conditions.

article info abstract

Article history: A one-dimensional (1D) molten-metal-based bubble-column reactor (MMBCR) model

Received 12 April 2023 coupled with heat transfer and axial dispersion was developed for non-oxidative CH4 py-
Received in revised form rolysis to produce low-carbon H2. The MMBCR model included mass, momentum, and
2 June 2023 energy balances, accounting for catalytic and non-catalytic reaction kinetics, and hydro-
Accepted 3 June 2023 dynamic parameters such as the gas holdup (aG), gas velocity (uG), bubble size (db), and
Available online 18 June 2023 specific interfacial surface area of bubbles (as). The 1D MMBCR model was compared with
other 1D reactor models and bench-scale experimental data for CH4 conversion (XM). The
Keywords: MMBCR model agreed well with the experimental data. When heat transfer and axial
Hydrogen production dispersion were considered for an industrial-scale MMBCR to produce 10,000 Nm3/h, the

* Corresponding author.
E-mail addresses: [email protected] (B.T. Le), [email protected] (S.I. Ngo), [email protected] (Y.-I. Lim), uendol@
kitech.re.kr (U.-D. Lee).
https://doi.org/10.1016/j.ijhydene.2023.06.031
0360-3199/© 2023 Hydrogen Energy Publications LLC. Published by Elsevier Ltd. All rights reserved.
35822 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 4 8 ( 2 0 2 3 ) 3 5 8 2 1 e3 5 8 3 7

Methane pyrolysis reactor length increased to 2.4 times higher than that estimated by the model without axial
Molten metal dispersion to meet 80% XM. The MMBCR model has the potential to design industrial-scale
Bubble column MMBCRs and optimize operating conditions.
Heat transfer © 2023 Hydrogen Energy Publications LLC. Published by Elsevier Ltd. All rights reserved.
Axial dispersion

The use of molten metal (MM) as a liquid catalyst for CH4

Introduction pyrolysis reduces the activation energy and promotes CH4
decomposition at low temperatures [8]. Unlike solid catalysts
Hydrogen is the lightest element and the most abundant gas deactivated by coke, MM is free of catalyst deactivation owing
in the universe [1]. However, obtaining H2 is energy- and to the floating of carbon particles on the surface [6]. In MM-
carbon-intensive [1,2] because H2 is produced from fossil based H2 production from CH4, a bubble column is often
fuels, water splitting, and biomass. In 2020, the global used for effective heat and mass transfer with a high
hydrogen demand was 90 million tons (Mt), with approxi- gaseliquid interfacial area [4,9,10]. The low density, volatility,
mately 80% of it produced using fossil fuels [3]. Hydrogen is and solubility of solid carbon in liquid media are advanta-
typically produced using technologies such as steam methane geous for the separation of carbon particles from MM [11].
reforming (SMR), oil/naphtha reforming, coal gasification, In molten metal-based bubble column reactors (MMBCRs),
methane (CH4) pyrolysis, and water electrolysis [3e5]. SMR is gas enters the bottom of the reactor via a gas distributor to
the main commercial technology used, and it requires 41 kJ/ form gaseous bubbles. Methane pyrolysis occurs inside the
mol-H2 while producing 6.6 kg-CO2/kg-H2 [5]. SMR is techno- bubbles during non-catalytic reactions and on the surface of
logically mature and cost-effective, but decarbonization of the bubbles during catalytic reactions [10]. The solid carbon
carbon-intensive SMR processes is required for the transition produced rises to the MM surface, forming a thin layer that
to low-carbon H2 production [6]. can be continuously removed [5,11]. Upham et al. (2017) car-
The main methodologies for low-carbon H2 production ried out non-oxidative methane pyrolysis in a bubble column
can be divided into three categories: (i) fossil-fuel-based H2 reactor with molten Ni27Bi73 alloys, achieving up to 95%
production coupled with CO2 capture and storage (CCS), (ii) methane conversion at 1065  C [12]. When using molten
non-oxidative decomposition of hydrocarbons to H2 and Ni27Bi73 alloys and an additional layer of sodium bromide,
solid carbon, and (iii) water splitting with non-carbon en- Rahimi et al. (2019) were able to achieve a methane conversion
ergy sources such as nuclear and solar energy [4,7]. The H2 of 37.5% at 1000  C [13]. Pe
rez et al. (2021) observed a CH4
production cost of SMR with CCS increases by 30%e50% conversion of 91% at 1119  C in a U-shaped vessel filled with
compared to that without CCS [6]. Although water elec- molten Ga. Noh et al. (2022) reported a three-stage bubble-
trolysis using renewable energy such as wind and solar column reactor containing molten NieBi alloy, zirconia, and
power is promising for CO2-free H2 production, it is not molten NaBr, enhancing the carbon conversion by 5% at 985  C
economically competitive compared to conventional H2 as compared to the case without the zirconia bed. These ex-
production pathways [5,6] because of the high heat of re- periments were performed using bench-scale MMBCRs to
action (285 kJ/mol-H2), which is seven times greater than investigate the effects of liquid metals on CH4 conversion. To
that of SMR. scale up from bench-scale experiments to industrial-scale
Alternatively, non-oxidative CH4 decomposition (also processes, reactor modeling is often used to design reactors
known as pyrolysis) produces H2 and solid carbon without CO2 under given operating conditions, such as temperature (T),
[4e6]. pressure (P), and feed flow rate (F) [14]. Moreover, reactor
kJ modeling can provide a deeper understanding of the physical
CH4 / C þ 2H2 ; DH0R ¼ 37 (1) and chemical mechanisms involved in MMBCRs [15].
mol  H2
Computational fluid dynamics (CFD) has been used to un-
This reaction is endothermic with a standard reaction
derstand physical phenomena in two-dimensional (2D) and
enthalpy (DH0R ) of 37 kJ/mol-H2, which is slightly lower than three-dimensional (3D) geometries and to scale up, optimize,
that of SMR. Owing to the strong CeH bonds and lack of po- and design chemical and biological processes [16]. Eulerian
larity, the chemical inertness of CH4 molecules is considered a [17] and volume of fluid (VOF) [4] CFD models have been used
significant challenge in the methane pyrolysis process. An for bubble column reactors (BCRs) to investigate the gas
additional heat must be supplied to break the CeH bond, holdup, bubble size, and interfacial surface area, which are
which could lead to CO2 emission. crucial for designing BCRs. Xu et al. (2015) used a 3D VOF-CFD
Non-catalytic CH4 decomposition requires high activation model to track the interface between argon and liquid metal
energy and high-temperature operation (>1200  C) [5]. Solid assuming an initial bubble size [18]. Kudinov et al. (2021)
catalysts with active metals are used for the pyrolysis of hy- investigated non-catalytic methane pyrolysis in a molten tin
drocarbons (CHx) in fixed or fluidized beds. Most solid cata- BCR using a 2D VOF-CFD model [19]. The 2D/3D CFD simula-
lysts are deactivated because the solid carbon byproduct tion is appropriate for identifying the hydrodynamics of
(coke) blocks the active sites [5]. equipment following geometrical and operational
 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 4 8 ( 2 0 2 3 ) 3 5 8 2 1 e3 5 8 3 7 35823

modifications [16] but may be computationally prohibitive.

One-dimensional (1D) reactor models help design reactors 1D PFR model in MMBCRs
and determine a suitable range of operating conditions [14]
within a reasonable computational time. BCRs have the advantages of a simple geometry, excellent
An 1D plug-flow reactor (PFR) model coupled with chemical heat and mass transfer performance, and low operating costs
reactions and axial dispersion was developed for bench-scale [23,24]. However, multiphase flow behaviors are complex and
BCRs for the hydrocracking of vacuum residues [14]. The 1D coupled with physical properties and chemical reactions [17].
PFR models of the MMBCRs for hydrogen production from CH4 Fig. 1 shows a BCR filled with MM for H2 production from CH4.
are listed in Table 1. Keipi et al. (2017) [20] developed an CH4 gas was injected into the bottom of the reactor through a
isothermal 1D PFR model to obtain non-catalytic kinetic pa- gas distributor, forming bubbles that rose through the MM
rameters for the thermal decomposition of CH4. Catalan and along the reactor height (h). In MMBCRs, a non-catalytic re-
Rezaei (2020) [9] proposed a 1D PFR model with mass and action occurs inside the bubbles and a catalytic reaction on
momentum balances of the gas phase in MMBCRs, dis- the bubble surface. Solid carbon (C) floating on the surface of
regarding the heat transfer between the gas and liquid phases the MM was continuously removed (Fig. 1a). There are two
and axial dispersion. Catalan and Rezaei (2022) [10] enhanced main heat transfers, as shown in Fig. 1b: heat transfer be-
the drift-flux PFR model [9] by adding catalytic reactions at the tween the gas and liquid phases and wall-to-bed heat transfer.
gaseliquid interface and presented optimized designs for an The liquid phase (MM) supplies the heat required for the
industrial-scale MMBCR for 10,000 Nm3/h H2 production. Here, endothermic CH4 decomposition into the gas phase (CH4). The
the reactor size was underestimated for a given H2 production external heat is transferred from the wall to the bed. As shown
rate and a given CH4 feed flow rate because the recycling of the in Fig. 1c, bubbles rise strongly near the center of the reactor
unreacted feed was not considered. Moreover, the feed gas because of stationary wall, and the down flow near the wall
should enter the MMBCR at less than 800  C to avoid prema- causes back-mixing of fluids [17], which is expressed as gas
ture CH4 decomposition and carbon deposit in the injection axial dispersion. A radial cross-exchange of fluid (dashed-ar-
area [9,21]. Park et al. (2022) [15] developed a non-isothermal rows) is superimposed on rising bubbles (dotted-arrows) in
1D PFR model in molten salt-based BCRs for H2 production Fig. 1c [25]. The axial dispersion offers a remedy for the ideal
from CH4, in which continuous liquid and discrete bubble PFR model in BCRs [14,17].
phases were considered by incorporating catalytic reactions In this study, a 1D PFR model coupled with heat transfer
and heat transfer between the two phases without axial and axial dispersion was presented under the following as-
dispersion. As the back-mixing of fluids is significant in BCRs sumptions: (i) the bubble column reaches a steady state in the
[22] and reduces the reaction driving force because of lower gas and liquid phases [15], (ii) concentration, pressure, and
reactant concentration [14], it is desirable to consider the axial temperature change along the reactor height (h), (iii) ideal gas
dispersion caused by back-mixing in 1D PFR models, espe- equation of state and Arrhenius equation of chemical reaction
cially for industrial-scale reactors. kinetics are used [26], (iv) gas properties are functions of
This study aimed to develop a 1D PFR model coupled with temperature, (v) the non-catalytic reaction occurs inside the
chemical reactions, heat transfer, and axial dispersion in bubble volume [10], (vi) the catalytic decomposition of CH4 is
MMBCRs for H2 production from CH4. A 1D MMBCR model proportional to the interfacial area (as) between the gas and
based on Catalan and Rezaei (2022) [10] was enhanced by liquid phases [10], (vii) although the gas distributor influenced
considering the heat transfer and axial dispersion. The novel significantly the CH4 conversion [27], the effect of the gas
model was compared and validated using a non-isothermal distributor on the gas holdup and bubble size is ignored [10]
multiphase 1D PFR model [15] and experimental data [12]. because of lack of information on the correlation between the
The enhanced model was applied to the design of an gas distributor and bubble size in MMBCR, (viii) mass transfer
industrial-scale reactor by recycling the unreacted feed. The of solid carbon from the bubble surface to the liquid phase is
1D MMBCR model is useful for understanding the chemical ignored, (ix) the wall temperature is fixed [26], (x) the gas phase
and physical mechanisms of MMBCRs under different oper- inside bubbles is well mixed and homogeneous, and (xi) the
ating conditions and for determining the diameter and height liquid phase is stagnant (uL ¼ 0) because the net liquid velocity
of MMBCRs for a given operating condition at the initial design in a cross-sectional area of the MMBCR is close to zero due to
stage. no liquid injection.

Table 1 e Plug flow reactor (PFR) models in previous publications.

Authors Remarks
Keipi et al. (2017) [20] An isothermal model with non-catalytic reactions for CH4 decomposition
Catalan and Rezaei (2020) [9] An isothermal 1D model with hydrodynamics and non-catalytic reaction
(no heat transfer and no axial dispersion)
Catalan and Rezaei (2022) [10] An isothermal 1D model with catalytic and non-catalytic reactions (no heat
transfer and no axial dispersion)
Park et al. (2022) [15] A non-isothermal 1D model with mass, momentum, and energy balances
(no axial dispersion)
35824 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 4 8 ( 2 0 2 3 ) 3 5 8 2 1 e3 5 8 3 7

Fig. 1 e Chemical reactions, heat transfer, and axial dispersion in an MMBCR.

Governing equation of 1D MMBCR model the range of 0.6  ub  0.94 m/s used in this study for an
industrial-scale reactor (D ¼ 0.1 m). The Dax (solid line in black)
The governing equation of the 1D MMBCR model includes the for a bench-scale MMBCR (D ¼ 0.03 m) [12] lies in a lower range
species mass balances in the gas phase, momentum balance than in Mangartz and Pilhofer's experiment, and varies from
for static pressure, and energy balances in the gas and liquid 2  103 to 2.5  103 m2/s, which is significantly lower than
phases. The mass balances of CH4 and H2 contain convection, Dax of the industrial-scale (0.22e1.25 m2/s). The dispersion
axial dispersion, and reaction terms as follows [28]: coefficients are mainly caused by convective turbulent flows

 and back-mixing rather than by molecular diffusion [22,29].
drG uG xi d dxi
¼ rG aG Dax  ni aG rA Mi (2) Fig. 2b illustrates Dax for the bench-scale MMBCR
dh dh dh
(D ¼ 0.03 m, H ¼ 1.15 m, and uG ¼ 5 mm/s) and the industrial-
where uG (m/s) is the superficial gas velocity, and xi is the mass scale MMBCR (D ¼ 0.1 m, H ¼ 2.11 m, and uG ¼ 134 mm/s) at
fraction of species such as CH4, H2, solid carbon, and Ar as the 1050  C. Dax increased from the bottom to the top of the
inert gas. rA (mol/m3/s) is the reaction rate per unit volume of reactor, because uG increased as the reaction progressed (see
gas for the catalytic and non-catalytic reactions of CH4 Eq. (1)). The Dax (approximately 2.5  103 m2/s) is relatively
decomposition [10]. aG (m3-gas/m3-reactor) is the gas holdup, small for the bench-scale reactor, while Dax (z0.5 m2/s) is high
rG (kg/m3) is the gas density, ni is the stoichiometric factor of for the industrial reactor because of a large diameter and high
the methane decomposition reaction (nCH4 ¼  1, nC ¼ 1, nH2 ¼ velocity. Therefore, axial dispersion is essential in the design
2, and nAr ¼ 0), and Mi (kg/mol) is the molecular weight of the of MMBCRs for industrial-scale reactors.
species. The axial dispersion of solid carbon was neglected in The axial dispersion (or back-mixing) may be weak near
the gas phase, and solid carbon was excluded from the the bottom of the bubble column because of the strong rising
calculation of the gas holdup. flow from the gas distributor. It is reasonable to assume that
Dax (m2/s) is the gas phase axial dispersion coefficient, Dax is close to zero at h ¼ 0, because experimentally measured
which is composed of the molecular and hydrodynamic concentrations showed a smooth change at the entrance [22],
diffusion coefficients [29], depending on the bubble rising ve- while concentrations calculated from dispersed PFR models
locity (ub ¼ uG =aG ) and the diameter (D, m) of the reactor with a constant Dax jump discontinuously to a certain value at
[22,25,30]: the reactor inlet [29]. However, because the discontinuous Dax
with respect to h from 0 to Eq. (3) causes numerical divergence
Dax ¼ f ðuG ; aG ; DÞ ¼ 50ðub Þ3 ðDÞ1:5 (3) when using an ordinary differential equation (ODE) solver, a
The empirical correlation was obtained for bubble columns smoothing function is necessary from the bottom (Dax /0) to a
up to 3.2 m in diameter [25]. The axial dispersion was applied certain height to approach Dax of Eq. (3). The sigmoid function
only to the gas phase because the reaction occurs both on the was used in this study.
surface of bubbles and inside bubbles, as indicated in the

2D*ax h
model assumptions (v) and (vi). In Fig. 2a, Dax with respect to ub Dax ¼ 50ðub Þ3 ðDÞ1:5  h
1 (4)
1þe H:4th H
in Eq. (3) was compared to the experimental data obtained by
Mangartz and Pilhofer (1981) [30]. Dax (red dashed line in where D*ax ð ¼ 50ðub;in Þ3 ðDÞ1:5 Þ is the axial dispersion coefficient
Fig. 2a) shows good agreement with the experimental data in at h ¼ 0. The threshold factor (4th ) was set as 0.001 to minimize
 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 4 8 ( 2 0 2 3 ) 3 5 8 2 1 e3 5 8 3 7 35825

Fig. 2 e Gas-phase dispersion coefficients (Dax) in the bench- and industrial-scale MMBCRs.

the effect of the boundary condition at h ¼ 0 on the ODE where CCH4 and CH2 (mol/m3ebubble) are the molar concen-
solution. trations calculated from the ideal gas equation of state, mass
The reaction rate (rA, mol/m3/s) includes the non-catalytic flux (rG uG xi , kg/m2/s), and methane conversion (XM). The re-
(rnon, mol/m3-bubble) and catalytic (rcat, mol/m3-bubble) reac- action kinetic parameters were obtained from the literatures
tion rates [10]: [10,12]: n ¼ 1.0809, knon;0 ¼ 1.4676  1010 m3(n1)/moln1/s,
Eanon ¼ 284,948 J/mol, kcat;0 ¼ 78,800 m/s, Eacat ¼ 208,000 J/mol. The
rA ¼ rnon þ rcat (5)
concentration-based equilibrium constant (KC ) is defined as:

 !
0
Ea C2H2 P
rnon ¼ knon;0 exp  non CnCH4 1  (6) KC ¼ K (8)
RT CCH4 KC RT

where K is the dimensionless equilibrium constant calculated

 !
Eacat C2H2 from the Gibbs free energies and P0 is the standard pressure
rcat ¼ kcat;0 exp  CCH4 1  $as (7)
RT CCH4 KC (1  105 Pa) [10]. as (m2/m3) is the specific surface area, which is
35826 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 4 8 ( 2 0 2 3 ) 3 5 8 2 1 e3 5 8 3 7

a hydrodynamic parameter. The non-catalytic reaction occurs external electric furnace, was set to be the same as the oper-
inside the bubbles, whereas the catalytic reaction is assumed ating temperature (¼1040  C in the bench-scale MMBCR) [12].
to be proportional to the interfacial area (as) between the The overall heat transfer coefficient between the inner wall
bubbles and molten media [10]. In Eq. (7), the units of the and molten metal (UW) was set to Uw ¼ 1  105 (W/m2/K)
catalytic reaction rate (rcat) were converted from mol/m2- [32,33], which was relatively high due to high thermal con-
bubble to mol/m3-bubble, applying as. ductivity of liquid metal. The liquid temperature (TL) may
The mass balance proposed by Catalan and Rezaei (2022) approximate TW because of high Uw. The Nusselt number of
was based on the methane conversion (XM), whereas the 1D the orifice tube (NNu;orf ) was set to 3.66 for a fully-developed
MMBCR model is based on the mass flux (rG uG xi ). In this study, laminar flow [15] in a tube of the single-tube bench-scale
XM was defined as a function of the methane mole flow rate reactor [12]. For industrial-scale multi-tubular reactors, the
(nCH4 ) at the inlet and outlet surfaces. last term on the right side of Eq. (12) was removed because no

 immersed tube existed [10].
ðnCH4 Þout
XM ð%Þ ¼ 100  1  (9) The 1D MMBCR model is expressed as a set of ODEs coupled
ðnCH4 Þin
with physical properties (r, m, s, kL, and Cp), design parameters
The steady-state momentum balance was simplified as a (D and H), hydrodynamic parameters (aG, uG, db, and as), and
pressure drop along the height [10]: model parameters (Dax, Uh, and rA). Furthermore, boundary
conditions (BCs) at h ¼ 0 and H are required to obtain solutions
dP to the ODEs. BCs are often defined by operating conditions
¼  ½rL ð1  aG Þ þ rG aG g (10)
dh such as temperature (T), pressure (P), and feed flow rate (F).
where rL (kg/m3) is the liquid density and g (9.8 m/s2) is the
gravity. The static pressure drop originates from the masses of Physical properties and boundary conditions
the gas and liquid phases.
The steady-state energy balance of the gas phase includes Physical properties such as density (r), viscosity (m), surface
heat convection, heat transfer between the two phases, and tension (s), thermal conductivity (k), and heat capacity (cp) are
heat of reaction (DHR ) terms [26,28]. pre-described to solve the 1D steady-steady governing equa-
  tion. The equations for the physical properties are presented
d uG rG Cp;G TG
¼  Uh as ðTG  TL Þ  aG DHR rA (11) in Tables S1eS4 in the Supplementary Material.
dh
Ten BCs at h ¼ 0 and h ¼ H are required to obtain the so-
where cp;G (J/kg/K) is the heat capacity of the gas phase, TG and lutions of the six ODEs in Eqs. (2), (10), (11) and (13). The carbon
TL are the gas and liquid temperatures, respectively, and as mass fraction was obtained algebraically by the overall mass
(m2/m3) is the specific surface area. DHR (J/mol) is the heat of balance. Danckwerts BC [29] is applied as follows:
reaction at a gas temperature (TG) [26]. The heat diffusion in
the gas phase was ignored because of its low heat conduc- ðrG uG xi Þjh¼0 ¼ ðrG uG xi Þin


tivity. Uh (W/m2/K) is the overall heat transfer coefficient be- aG;in Dax;in drG uG xi 
þ  ði ¼ CH4 ; H2 ; and ArÞ (14)
tween the gas and liquid phases, which is obtained from the u G;in dh h¼0

Nusselt number (NNu;G ) of the gas phase [31].

Pjh¼0 ¼ Pin (15)

0:7
Uh db NPe
NNu;G ¼ ¼ 1:1 (12)  

kG 1 þ mG =mL 1  aG;in kL;in dTL 
TL jh¼0 ¼ TL;in þ (16)
where kG (W/m/K) is the thermal conductivity of the gas phase uG;in dh h¼0
and db (m) is the bubble diameter. The Peclet number (NPe ¼
rG Cp;G db vG
) is defined for rising bubbles and mG =mL is the viscosity TG jh¼0 ¼ TG;in (17)
aG kG

ratio between the gas and liquid phases. Eq. (12) used for heat

dxi
transfer in MMBCRs by Park et al. (2022) [15] was valid for j ¼ 0 ði ¼ CH4 ; H2 ; and ArÞ (18)
dh h¼H
50  NPe =ð1 þmG =mL Þ  900 [31].
The liquid-phase energy balance includes heat convection,

dTL
heat diffusion by conduction, heat transfer between the gas j ¼0 (19)
dh h¼H
and liquid phases, heat supply through the wall to the liquid
metal, and heat loss to the submerged center tube orifice: For a bench-scale reactor [12], the operating conditions are:
 
 ðrG uG xCH4 Þin ¼ 1:2212  104 kg/m2/s, ðrG uG xAr Þin ¼ 7:6035  105
d uL rL Cp;L TL d dTL 4Uw kg/m2/s, ðrG uG xH2 Þin ¼ 0, Pin ¼ 2  105 Pa, TL;in ¼ 1040 oC, and
¼ aL kL þ Uh as ðTG  TL Þ þ ðTW  TL Þ
dh dh dh D TG;in ¼ 800 oC. For an industrial-scale reactor with 891 tubes
4NNu;orf kG ðTG  TL Þ [10], ðrG uG xCH4 Þin ¼ 0:1606 kg/m2/s, ðrG uG xAr Þin ¼ ðrG uG xH2 Þin ¼ 0,
þ
D2 Pin ¼ 17:84  105 Pa, TL;in ¼ 1050 oC, and TG;in ¼ 800 oC. The inlet
(13)
gas temperature (TG,in) was set to 800  C to avoid premature
where cp;L (J/kg/K) is the heat capacity of the liquid phase. The CH4 decomposition [9,21], as aforementioned. The wall tem-
liquid velocity (uL) was assumed to be zero because no liquid peratures (TW) were 1040 and 1050  C for the bench- and
was injected. kL (W/m/K) is the effective heat conductivity of industrial-scale reactors, respectively, which were taken from
the MM [15]. The wall temperature (TW), controlled by an references [10,12].
 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 4 8 ( 2 0 2 3 ) 3 5 8 2 1 e3 5 8 3 7 35827

Hydrodynamic parameters Here, nCH4 ; nH2 ; and nAr (mol/s) are the mole flow rates of the
methane, hydrogen, and argon gases, respectively. The dis-
The four hydrodynamics parameters such as aG, uG, db, and as tribution parameter (Co) depends on column geometry and
are required to solve the governing equations. The drift ve- flow regime.
locity (VG) of the gas phase is the velocity of the gas phase with rffiffiffiffiffi
rG
respect to the volume center of the mixture in the two-phase Co ¼ 1:2  0:2 (23)
rL
flow, considering the wall shear stress [34]. In MMBCRs, the
gas holdup (aG) was calculated from a dimensionless drift for churn flow in a round column [35].
0:25
velocity (VGþ ¼ VG =ðsgDr=r2L Þ ) and a dimensionless superfi- The gas density (rG ) is calculated by the ideal gas equation
0:25 of state:
cial gas velocity (uþ
G ¼ uG =ðsgDr=rL Þ
2
) [9].
PMG
uþ rG ¼ (24)
aG ¼ G
(20) RT
uþ
G C þ
0 þ VG
where MG (kg/mol) is the mean molar weight of the mixture.
The superficial gas velocity (uG) was obtained from the
volume flow rate of gas (QG), which was derived from the ideal CCH4 MCH4 þ CH2 MH2 þ CAr MAr
MG ¼ (25)
gas law: CCH4 þ CH2 þ CAr

4QG where Ci ¼ QnGi ; mol =m3 is the molar concentration of the
uG ¼ (21)
pD2 species i (CH4, H2, and Ar).
  The dimensionless drift velocity (VGþ ) was determined as a
nCH4 þ nH2 þ nAr RT
QG ¼ (22) function of the range of the viscosity number (Nm ) and the
P

Inputs Model libraries

IC and BC calculations

Hydrodynamic parameters
Dax

rA

kL
Outputs
Sub-models

Fig. 3 e Flow diagram for the calculation of 1D MMBCR model with heat transfer and axial dispersion.
35828 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 4 8 ( 2 0 2 3 ) 3 5 8 2 1 e3 5 8 3 7

Isao and Mamoru (1987) presented a dimensionless drift

þ
velocity (VG;1 ) of bubbles with liquid recirculation in a vessel
[36]:

0:157
þ r
VG;1 ¼ 0:92 L for Nm  2:2  103 and D*H  30 (28)
rG

When Nm  2:2  103 :

0:157
þ 0:809 rL
VG;1 ¼ 0:0019 D*H Nm0:562 for D*H  30 (29)
rG

0:157
þ r
VG;1 ¼ 0:92 L N0:562
m for D*H  30 (30)
rG
Fig. 4 e Comparison of methane conversions (XM) along the The dimensionless drift velocity (VG;2 þ
) was empirically
bench-scale reactor height (h) at 1040  C obtained from measured for the flow regime such as bubbly flow, churn-
experiment [12], reference models [10,15], and present turbulent flow, and slug flow for uþ
G  0:5 [34]:
models with and without heat transfer and axial
þ
pffiffiffi
dispersion. VG;2 ¼ 2ð1  aG Þ1:75 for bubbly flow (31)

þ
pffiffiffi
VG;2 ¼ 2 for churn  turbulent flow (32)

dimensionless hydraulic equivalent diameter of the column sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

(D*H ) [9,36]: þ gDH ðrL  rG Þ
VG;2 ¼ 0:35 for slug flow (33)
rL
mL
Nm ¼  pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 0:5 (26)
rL sL sL =ðgðrL  rG Þ Finally, the dimensionless drift velocity (VGþ ) was obtained
þ þ
by combining ðVG;1 Þ and ðVG;2 Þ using a smoothing function for
D the entire of uþ
G [35]:
D*H ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffi (27)
sL    
gðrL rG Þ VGþ ¼ VG;2
þ
exp A uþ þ þ
G þ VG;1 1  exp A uG with A ¼  1:39 (34)

Fig. 5 e Axial profiles of gas (TG) and liquid (TL) temperatures for a bench-scale reactor at TW ¼ 1040  C.
 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 4 8 ( 2 0 2 3 ) 3 5 8 2 1 e3 5 8 3 7 35829

The bubble diameter (db) in MMBCRs was predicted by diameter (d32) obtained from a range of bubble diameters [37].
Akita and Yoshida [37] based on the following dimensionless The equation including physical properties of the gas and
numbers [10]: liquid phases, gas velocity (uG), and column diameter (D) ig-
nores the effect of the gas distributor on db, which was sig-
db ¼ 26ðNBo Þ0:50 ðNGa Þ0:12 ðNFr Þ0:12 D (35) nificant in MMBCR [27].
2 3 The specific surface area (as) of a spherical bubble is defined
where NBo (¼ gDsLrL ) is the Bond number, NGa (¼ gD
nL
) is the Galilei
uG
as:
number, and NFr (¼ ðgDÞ0:5
) is the Froude number.
rL ðkg =m3 Þ; nL ðm2 =sÞ; and sL ðN =mÞ are the density, kinematic 6
as ¼ (36)
db
viscosity, and surface tension of the liquid phase, respectively.
Eq. (35) was experimentally validated for the ranges: 800  NBo
 48,500, 6.25  106  NGa  1.79  1012, and 8.0  104  NFr  where the units (m2-bubble surface area/m3-bubble volume)
1.35  101 [10]. The db in Eq. (35) was the Sauter mean are the surface area per bubble volume.

Fig. 6 e Axial profiles of hydrodynamic parameters: pressure (P), gas holdup (aG), superficial gas velocity (uG), bubble
diameter (db), specific surface area (as), and bubble rising velocity (ub) in a bench-scale reactor at 1040  C.
35830 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 4 8 ( 2 0 2 3 ) 3 5 8 2 1 e3 5 8 3 7

Solution procedure of 1D MMBCR model and a lower XM in the rear reactor than our 1D MMBCR model.
The main reason for this difference is that the bubble size (db)
Fig. 3 shows the calculation procedure for the 1D MMBCR and interfacial area (as), which are crucial parameters in XM
model. The design parameters (D and H) and operating con- proposed by Park et al. (2022) [15] are different from those of
ditions (Tin, Pin, and Fin) are provided as inputs. The initial our MMBCR model.
condition (IC) and BC were calculated for the mass and heat When the energy equation for the gas and liquid temper-
fluxes, followed by the calculation of the hydrodynamic pa- atures (T) was added to our MMBCR model, XM was slightly
rameters (aG ; uG ; db ; and as Þ. The governing equations, lowered because of the low inlet gas temperature (Tin-
including the mass, momentum, and energy balances, were ¼ 800  C). However, XM of our model with heat transfer
then solved with the IC, BC, and hydrodynamic parameters, exhibited almost the same results as our model without heat
which resulted in the axial profiles of XM ; P; aG ; uG ; db ; as ; rA ; TG ; transfer because the gas temperature (TG) quickly approached
and TL . The physical properties of the species, thermodynamic the reaction temperature (TL). Our MMBCR model with the
data, axial dispersion, reaction rate, and thermal conductivity reactions (represented by rA in Fig. 4), heat transfer
models were incorporated into the governing equations.
For a given reactor dimension (D and H) and operating
conditions (T, P, F, and x), to obtain the axial profiles of
XM ; P; aG ; uG ; db ; as ; rA ; TG ; and TL , the mass, momentum, and
energy balance equations were numerically solved using a
stiff ODE solver (ode15s) for initialization and a boundary
value problem ODE solver (bvp4c) for the final solution using
MATLAB 2022b (MathWorks, USA). The hydrodynamic pa-
rameters (aG, uG, db, and as) were computed along the height
(h).

Results and discussion

The 1D MMBCR model coupled with heat transfer and axial

dispersion was first validated against experimental data [12]
and then compared with other previous 1D PFR models [10,15].
The hydrodynamic parameters, temperature, and reaction
rates along the height (h) were shown and analyzed. Finally,
an industrial-scale MMBCR was designed using a 1D MMBCR
model to produce 10,000 Nm3/h of H2 (or 7 kt/y).

Validation of 1D MMBCR model against experimental data

Upham et al. [12] performed methane pyrolysis experiments

in a bench-scale MMBCR filled with a molten Ni0.23Bi0.73 alloy.
The inner diameter (D) of this column was 0.03 m, and the
depth (H) of the melt above the injection point varied up to Fig. 7 e Axial profiles of non-catalytic reaction rate (rnon)
1.15 m. The feed gas consisted of an 80/20 mol% mixture of and catalytic reaction rate (rcat) versus reactor height (h) for
methane/argon at a standard feed flow rate of 10 cm3/min a bench-scale reactor at 1040  C.
(STP). The experimental methane conversions (XM) [12] were
compared with those of our 1D MMBCR model: the 1D PFR
model without heat transfer and axial dispersion [10] and the
1D PFR model without axial dispersion [15].
Fig. 4 compares the methane conversions (XM) along the
reactor height (h) obtained from experiments at 1040  C [12],
1D drift-flux PFR model without heat transfer and axial
dispersion [10], 1D two-phase PFR model with heat transfer
[15], and our MMBCR models with and without heat transfer
and axial dispersion. All the PFR models shown in Fig. 4 follow
the experimental data at the bench-scale well. The 1D PFR
model by Catalan and Rezaei (2022) [10] showed a slightly
lower XM than our 1D MMBCR model with catalytic and non-
catalytic reactions (rA). This may be attributed to the fact
that the physical property models were different for the two
PFR models. The 1D PFR model with heat transfer developed Fig. 8 e Axial profiles of CH4 conversion (XCH4) along the
by Park et al. (2022) [15] showed a higher XM in the front reactor reactor height (h) for an industrial-scale reactor.
 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 4 8 ( 2 0 2 3 ) 3 5 8 2 1 e3 5 8 3 7 35831

(represented by T in Fig. 4), and axial dispersion (represented heat transfer competed with the endothermic heat of reac-
by Dax in Fig. 4) slightly decreased XM, although it was not tion. In Park et al. (2022) [15], the feed gas was injected at
significant in the bench-scale reactor because of the small 1036.5  C (inset of Fig. 5) and TG increased gradually from 1036
diameter and low volume flowrate. to 1040  C. NPe =ð1 þmG =mL Þ of the bench-scale MMBCR ranged
Fig. 5 shows the axial profiles of the gas and liquid tem- from 0.5 to 12.8 (see Eq. (12)), which was out of the valid range
peratures along the bench-scale reactor height (h). The wall from 50 to 900 [31]. However, Eq. (12) was used because the
temperature (TW) was maintained at 1040  C. The liquid tem- heat transfer model for high temperature MMBCRs was
perature (TL) was almost the same as TW because the heat scarce.
losses to the gas phase and orifice were insignificant (Eq. (12)).
The gas temperature (TG) rose quickly to TL at the entrance Axial profiles of hydrodynamic and reaction parameters
then gradually approached from 1036 to 1038  C. As shown in
Eq. (11), TG increases rapidly at the bottom owing to heat Fig. 6 shows the axial profiles of pressure (P), gas holdup (aG),
transfer with a large temperature difference between the gas superficial gas velocity (uG), bubble diameter (db), specific
and liquid phases. Subsequently, TG increased slowly as the surface area (as), and bubble rising velocity (ub) along the

Fig. 9 e Axial profiles of hydrodynamic parameters: pressure (P), gas holdup (aG), superficial gas velocity (uG), bubble
diameter (db), specific surface area (as), and bubble rising velocity (ub) in an industrial-scale reactor at 1050  C and H ¼ 2.11 m.
35832 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 4 8 ( 2 0 2 3 ) 3 5 8 2 1 e3 5 8 3 7

volume of the present study (approximately 33 cm2) was close

to the experimental value of Upham et al. (2017) (z30 cm2)
[12]. Thus, there was room for improvement in the bubble size
model.
Fig. 6e shows the axial profile of the as for the bench-scale
reactor. Parkinson et al. (2021) reported that the experimental
specific surface area was between 600 and 1200 (m2-bubble
surface area/m3-bubble volume) for CH4 flow rates of 5e40
(mL/min) [11]. The as in this study ranged from 762 to 886 (m2-
bubble surface area/m3-bubble volume) for the bench-scale
reactor. The ub increases slightly from 0.20 to 0.21 m/s along
the reactor height with increasing uG and decreasing db, as
shown in Fig. 6f. Although the bubble size profile of Park et al.
(2022) [15] was opposite to that of this study, ub of Park et al.
(2022) ranged from 0.20 to 0.22 m/s along the reactor height for
the same bench-scale reactor.
Fig. 7 shows the axial profiles of the non-catalytic (rnon in
Fig. 7a) and catalytic (rcat in Fig. 7b) reaction rates along reactor
Fig. 10 e Axial temperature profiles obtained from the
height for the bench-scale reactor at 1040  C of the three
MMBCR models with reaction and heat transfer (rA þ T) and
MMBCR models with rA , rA þ T, and rA þ T þ Dax , respectively.
reaction, heat transfer, and axial dispersion (rA þ Tþ Dax ).
The value of rcat was approximately four times higher than
that of rnon because of the low activation energy of rcat . The
MMBCR model considering only rA exhibited higher reaction
rates than the other models because of the isothermal
column height (h) in the bench-scale MMBCR at 1040  C. The assumption (TG ¼ TL ¼ TW ). The reaction rates obtained from
1D MMBCR models with and without heat transfer and axial the MMBCR model with rA þ T and rA þ T þ Dax were almost
dispersion provided practically the same results for all the identical because of the low TG at the entrance (Fig. 5). Unlike
hydrodynamic parameters. Nevertheless, the MMBCR model the isothermal reactor (MMBCR with rA ), the peaks of rnon and
(rA þ T þ Dax ) with reaction, heat transfer, and axial dispersion rcat appeared at the bottom of the reactor (approximately
showed slightly lower aG and as than the MMBCR model (rA) 0.015 m), as TG approached quickly TL and the reactant con-
with only reaction. Consequently, XM of the MMBCR model centration (CCH4) decreased (Eqs. (5)e(7)).
with rA þ T þ Dax was smaller than that with only rA (see also
Fig. 4). Design of industrial-scale MMBCR
The pressure decreased almost linearly from the bottom to
the top of the column, with a pressure drop of approximately An industrial-scale MMBCR can consist of multiple tubes filled
1.1 bar (Fig. 6a). Abrupt increase of aG, uG, as, and ub as well as with MM and externally heated by burners or electric furnaces
abrupt decrease of db at the bottom of the reactor was caused to produce large amounts of H2 [10]. A single tube with 0.1 m
by a rapid gas temperature (TG) rising from the gas inlet inner diameter and 2.11 m height [10], which was a part of an
temperature (TG,in ¼ 800  C) to the liquid temperature industrial multiple tubular reactor with 891 tubes, was taken
(TL ¼ 1040  C), as shown in Fig. 5. The aG and uG increased into account, targeting a CH4 conversion of 80%. The total
along the height because the gas volume increased as the re- production rate was 7 kt/y H2 (10,000 Nm3/h) with 8000 oper-
action progressed (Fig. 6b and c). At the bottom (h ¼ 0), aG ating hours per year. The tube was filled with Ni0.27Bi0.73 and
started at 0.236% and then increased quickly to 0.286%. aG heated to 1050  C. Pure CH4 was fed at the bottom of the tube
increased up to 0.935% at the top of the reactor. The uG rose at T ¼ 800  C, P ¼ 17.8 bar, and F ¼ 0.0858 mol/s (4.94 kg/h CH4).
from 4.9  104 (m/s) to 2.0  103 (m/s). The total feed flow rate (F) includes the raw feed flow (3.95 kg/
In Fig. 6d, the db obtained from an empirical equation with h) and recycle flow (0.99 kg/h) under the assumption of 100%
dimensionless numbers [37] decreases from 8.1 to 6.8 mm recovery of unreacted CH4 (20% of the total feed flow rate).
uG Thus, the raw feed flow rate in this study was 20% less than
because the NFr (¼ ðgDÞ 0:5 ) increases with the increase in uG ,
that (4.94 kg/h) of Catalan and Rezaei (2022) [10]. In Supple-
while NBo and NGa are almost constant (Eq. (35)), resulting in an mentary Material, Tables S5 and S6 show the stream tables of
increase of as (Eq. (36)). The decreasing tendency of db was the single-tube MMBCR at XM ¼ 80% without and with recy-
consistent with that reported by Catalan et al. (2022) [10]. In cling of unreacted CH4, respectively. Totally-raw feed material
contrast, Park et al. (2022) [15] reported that db increased from (4.942 kg/h CH4) enters the MMBCR without recycling, whereas
5.5 mm to 8.1 mm along the column height, because db was raw feed (3.954 kg/h) and recycle (0.9884 kg/h) flows enter the
based on the initial bubble size and a proportional relation- MMBCR with recycling.
ship with uG assuming the bubbles as a single-bubble size. Fig. 8 shows the methane conversion (XM) along the height
Mohaghegian and Elbing (2018) demonstrated that the mean (h) of the bubble column at the industrial scale at a reaction
bubble size could increase or decrease depending on hydro- temperature of 1050  C for the four MMBCR models coupled
dynamics of BCRs [38]. However, the total bubble interfacial with rA , rA þ T, and rA þ T þ Dax , respectively. The MMBCR
RH
surface area (As ¼ ðpD2 =4Þ 0 as aG dh) within the reactor models were applied to predict XM for a single tube. According
 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 4 8 ( 2 0 2 3 ) 3 5 8 2 1 e3 5 8 3 7 35833

Fig. 11 e Axial profiles of non-catalytic reaction rate (rnon) and catalytic reaction rate (rcat) versus reactor height (h) for an
industrial-scale reactor at 1050  C.

to Catalan's simulation [10], XM reached 80% at a reactor

height of 2.11 m. Our MMBCR models with only reactions (rA )
and reactions/heat transfer (rA þ T) showed almost the same
XM as the Catalan's model (thin solid line). It resulted from the
fact that the gas temperature (TG) reached quickly the liquid
temperature (TL) at the bottom of the reactor owing to high
heat transfer rate (see Eqs. (11) and (12)).
However, unlike the bench-scale reactor, XM (66%) ob-
tained from the MMBCR model with rA þ T þ Dax was signifi-
cantly lower than those (80%) of the three other models at the
exit of the industrial-scale reactor. XM reached immediately
10% at the bottom of the reactor (h ¼ 0). This was due to the
high axial dispersion coefficient of the industrial-scale
reactor, which was approximately 200 times higher than
that of the bench-scale reactor (Fig. 2). When the axial Fig. 12 e Axial profiles of methane conversion (XM) in an
dispersion is high, the product is partially mixed with the industrial-scale reactor considering axial dispersion.
35834 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 4 8 ( 2 0 2 3 ) 3 5 8 2 1 e3 5 8 3 7

reactant in the entire reactor [25], and the reactor lies between reactor using the three MMBCR models (only rA , rA þ T, and
the PFR and continuous stirred-tank reactor (CSTR) [39]. rA þ T þ Dax ). The rcat is approximately six times faster than
Consequently, XM with axial dispersion was higher at the the rnon in the industrial-scale reactor at 1050  C. Similar to the
bottom of the reactor, compared to that without axial bench-scale reactor, a peak at 0.01 m of the reactor height for
dispersion. With ±20% changes in the axial dispersion coeffi- rnon and rcat is observed for the MMBCR models with rA þ T and
cient in Eq. (4), XM varies from 63% to 69% at the exit, which rA þ T þ Dax . The reaction rates (rnon and rcat ) obtained from the
seems to be not significant. MMBCR model with rA þ T þ Dax moved from the PFR toward
Fig. 9 depicts the axial profiles of P, aG, uG, db, as, and ub with CSTR as Dax increased [39], as mentioned earlier. Compared
respect to h in the industrial-scale MMBCR at 1050  C. Similar with the bench-scale reactor, the reaction rates of the
to XM , the MMBCR model with rA þ T þ Dax showed different industrial-scale reactor were 20e25 times higher because of
hydrodynamic parameter profiles from the two other models, the high TG, CCH4 , and as (Eq. (6)e(7)).
except for the pressure profile. The reactor pressure drop was The industrial-scale reactor with H ¼ 2.11 m cannot ach-
approximately 1.6 bar for all the three MMBCR models, as ieve the target XM (80%) because of axial dispersion, as shown
shown in Fig. 9a. The aG obtained from the MMBCR models in Fig. 8. Therefore, the reactor height should be increased to
with only rA and rA þ T varies from 11.4% to 13.4%, which is the increase the reaction time and XM. Fig. 12 shows XM for a
same as Catalan's model [10], while that of the MMBCR model reactor extended until 5.05 m to achieve 80% CH4 conversion.
with rA þ T þ Dax is from 10.6 to 13.3% (Fig. 9b). The range of uG When the reactor height was extended to 5.05 m and 100%
obtained from the MMBCR model with rA þ T þ Dax was unreacted CH4 was recycled to the reactor, the raw feed flow
0.057e0.125 m/s for the industrial-scale reactor (Fig. 9c), which rate decreased from 0.0858 [10] to 0.0686 mol/s at the same H2
was 65e115 times higher than that of the bench-scale reactor production rate of 10,000 Nm3/h. The two profiles of XM for
(Fig. 6c). H ¼ 2.11 and 5.05 m at the same Dax are different in the range
For the industrial-scale reactor, NBo ¼ 2600, of 0  h  2.11 m, because the second-order derivative of xi
NGa ¼ 2.11  1011, and 0.055  NFr  0.125, which lie within the with respect to h in Eq. (2) was solved with the Danckwerts BC.
valid range for the bubble size (db) in Eq. (35). The db values Table 2 summarizes the key differences between the
estimated using Eq. (35) decreased from 3.15 to 2.87 mm using MMBCR model with only rA by Catalan and Rezaei [10] and the
the MMBCR model with rA þ T þ Dax (Fig. 9d), which was present study with heat transfer and axial dispersion
2.4e2.6 times smaller than that of the bench-scale reactor (rA þ T þ Dax ). Our model with rA þ T þ Dax showed differences
(Fig. 6c). According to Eq. (35), db decreased as D and uG in hydrodynamic parameters, such as aG, uG, db, and as up to
increased. A rigorous validation of Eq. (35) was necessary for 15.8%. However, the reactor height estimated from the model
the accurate prediction of db, as aforementioned. As shown in with rA þ T þ Dax was extended by 240% to meet the outlet XM
Fig. 9e and f, as and ub of the industrial-scale reactor are (80%), compared to the model with only rA . It is necessary to
approximately 2.5 and 3.5 times, respectively, higher than consider the heat transfer and axial dispersion in the design of
those of the bench-scale reactor (see also Fig. 6e and f) due to industrial-scale MMBCRs. Specifically, the axial dispersion is
high aG and uG, as well as low db. of significance as the reactor diameter increases (see Eq. (3)).
Fig. 10 shows the axial temperature profiles obtained from
the industrial-scale MMBCR models with rA þ T and rA þ Tþ
Dax . The liquid temperatures (TL) were almost identical to TW. Conclusion
The gas temperature (TG) rose quickly within 0.03 m from the
bottom of the reactor and gradually approached TL, which was Non-oxidative molten metal (MM)-based CH4 pyrolysis is a
the same as that of the bench-scale reactor (Fig. 5). For the promising method for low-carbon hydrogen production, but
industrial-scale MMBCR, the range of NPe =ð1 þmG =mL Þ was from commercialization is not possible because of challenges
22 to 128, which was close to the valid range from 50 to 900 regarding efficient catalyst development and scale-up of the
[31]. process. Considerable attention has been paid to the MM-
Fig. 11 compares the non-catalytic (rnon ) and catalytic (rcat ) based bubble column reactor (MMBCR) for CH4 pyrolysis
reaction rates of CH4 decomposition in an industrial-scale because of the low activation energy of the reaction, low
catalyst deactivation, and effective heat and mass transfer. In
this study, a one-dimensional (1D) MMBCR model coupled
with heat transfer and axial dispersion was developed for the
Table 2 e Comparison of gas holdup (aG), superficial gas conceptual design of reactors under specific operating condi-
velocity (uG), bubble diameter (db), interfacial surface area tions. The model included mass, momentum, and energy
(as), and reactor height (H) at XM ¼ 80% between Catalan
balances in the axial direction of the reactor with the aid of
and Rezaei (2022) [10] and this study.
catalytic and non-catalytic reaction kinetics and empirical
Key parameters aG uG db as H hydrodynamic parameters such as the gas holdup (aG), bubble
(units) (%) (mm/s) (mm) (m2/m3) (m)
size (db), gas velocity (uG), and specific interfacial surface area
Catalan and Rezaei 13.1 134 2.91 2062 2.11 of bubbles (as). For bench-scale CH4 pyrolysis with a molten
(2022) [10] (rA)
alloy (Ni0.27Bi0.73), the 1D steady-state model was compared
This study (rA þ T þ Dax) 14.3 155 2.79 2152 5.05
with other 1D reactor models and experimental data in terms
Difference (%)a þ9.0 þ15.8 5.2 þ4.0 þ240
a
of axial CH4 conversions (XM).
Difference ¼ 100  [(value of Catalan and Rezaei) e (value of this
The MMBCR model with heat transfer and axial dispersion
study)]/(value of Catalan and Rezaei).
showed good agreement with the experimental data in a
 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 4 8 ( 2 0 2 3 ) 3 5 8 2 1 e3 5 8 3 7 35835

bench-scale reactor (0.03 m in diameter). The model with axial Cp heat capacity [J/kg/K]
dispersion lowered XM by 2.6% compared with that without db mean bubble diameter [m]
axial dispersion because the axial dispersion coefficient (Dax) D reactor diameter [m]
was low for the bench-scale reactor. However, for an D*H dimensionless hydraulic equivalent diameter of
industrial-scale reactor (0.1 m in diameter) to produce 10,000 reactor []
Nm3/hr (7 kt/y) H2, XM decreased significantly by 17% because Dax axial dispersion coefficient [m2/s]
Dax was high owing to its large diameter and back-mixing. The D*ax axial dispersion coefficient at h ¼ 0 [m2/s]
reactor height estimated by the 1D model with axial disper- Eacat activation energy of catalytic reaction [J/mol]
sion was 2.4 times higher than that estimated by the 1D model Eanon activation energy of non-catalytic reaction [J/mol]
without axial dispersion to achieve the same XM. g gravitational acceleration [m2/s]
Dax should be rigorously determined for industrial-scale h dimensionless column height []
MMBCRs, as axial dispersion significantly affects XM. The H reactor height [m]
mass transfer of solid carbon from the bubble surface to the kG thermal conductivity of gas phase [W/m/K]
liquid phase may be useful for understanding the reaction kL thermal conductivity of liquid metal [W/m/K]
rates reduced by the solid carbon on the bubble surface. The kcat;0 forward pre-exponential factor of catalytic reaction
hydrodynamic parameters confirmed by the experimental [m/s]
and numerical studies are expected to improve the accuracy knon;0 forward rate constant of non-catalytic reaction
of the 1D MMBCR model. This study elucidated the effects of [m3(n1).mol(1n).s1]
heat transfer and axial dispersion on the MMBCR, which can KC reaction equilibrium constant based on
be used to design a reactor and optimize the operating con- concentrations (mol/m3)
ditions of an industrial-scale reactor. M molar weight [kg/mol]
n order of the forward non-catalytic reaction with
respect to CH4 [-]
Data availability NFr Froude number []
NGa Galilei number []
The MATLAB GUI stand-alone program developed using NBo Bond number []
MATLAB (MathWorks, 2022) has been deposited on a website NNu Nusselt number []
under the filename MMBCR_1D_HKNU2023.zip (http:// NPe Peclet number []
cospe.hknu.ac.kr > ARCHIVES). Nm viscosity number []
P pressure [Pa]
QG volumetric flow rate [m3/s]
Declaration of competing interest rA reaction rate [mol/m3/s]
rcat catalytic reaction rate per unit volume of gas [mol/
The authors declare that they have no known competing m3/s]
financial interests or personal relationships that could have rnon non-catalytic reaction rate per unit volume of gas
appeared to influence the work reported in this paper. [mol/m3/s]
R universal gas constant [J/mol/K]
T temperature [K]
Uh heat transfer coefficient between molten metal and
Acknowledgments
bubble [W/m2/K]
Uw heat transfer coefficient between wall and molten
This work was supported by National Research Foundation of
metal [W/m2/K]
Korea (NRF) grant funded by the Korean Government (MSIT)
uG superficial gas velocity [m/s]
(Grant number: 2022M3J5A1051728) and (Grant number:
uþ dimensionless superficial gas velocity [m/s]
2021R1A2C1011618). G
VG drift velocity [m/s]
VGþ dimensionless drift velocity []
xi mass fraction of species i
Appendix A. Supplementary data
XM methane conversion [%]
Supplementary data to this article can be found online at Greek letters
https://doi.org/10.1016/j.ijhydene.2023.06.031. a gas holdup [%]
r density [kg/m3]
m dynamic viscosity [Pa.s]
Nomenclatures n kinematic viscosity [m2/s]
k viscosity ratio []
as specific surface area per unit of gas volume s factor []
[m2(interfacial area)/m3(gas volume)] 4 surface tension [N/m]
Cj concentration of component j in the gas phase [mol/
m3] Subscripts
Co distribution parameter [] cat catalytic reaction
35836 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 4 8 ( 2 0 2 3 ) 3 5 8 2 1 e3 5 8 3 7

G gas containing molten metals and molten salts. Carbon

L liquid 2019;151:181e91. https://doi.org/10.1016/
non non-catalytic reaction j.carbon.2019.05.041.
[14] Pham HH, Lim SH, Go KS, Nho NS, Kwon EH, Kim KH, Lim Y-I,
orf orifice
Ryu H-J, Park S-Y. Modeling and simulation of a bench-scale
th threshold bubble column reactor for slurry phase hydrocracking of
w wall vacuum residue. Fuel 2021:122481. https://doi.org/10.1016/
j.fuel.2021.122481.
[15] Park S, Kim M, Koo Y, Kang D, Kim Y, Park J, Ryu C.
references Numerical modeling of methane pyrolysis in a bubble
column of molten catalysts for clean hydrogen production.
Int J Hydrogen Energy 2022;48:7385e99. https://doi.org/
[1] Abdin Z, Zafaranloo A, Rafiee A, Me
rida W, Lipin
ski W, 10.1016/j.ijhydene.2022.11.068.
Khalilpour KR. Hydrogen as an energy vector. Renew Sustain [16] Ngo SI, Lim Y-I. Multiscale Eulerian CFD of chemical
Energy Rev 2020;120:109620. https://doi.org/10.1016/ processes: a review. Chem Engineering 2020;4:23. https://
j.rser.2019.109620. doi.org/10.3390/chemengineering4020023.
[2] Keipi T, Tolvanen H, Konttinen J. Economic analysis of [17] Tran BV, Ngo SI, Lim Y-I, Pham HH, Lim S-H, Go K-S, Nho N-
hydrogen production by methane thermal decomposition: S. Estimation of physical properties and hydrodynamics of
comparison to competing technologies. Energy Convers slurry bubble column reactor for catalytic hydrocracking of
Manag 2018;159:264e73. https://doi.org/10.1016/ vacuum residue. Chem Eng J 2021;418:129378. https://doi.org/
j.enconman.2017.12.063. 10.1016/j.cej.2021.129378.
[3] Global IEA. Hydrogen review 2021. Paris: International Energy [18] Xu Y, Ersson M, Jo € nsson P. Numerical simulation of single
Agency; 2021. argon bubble rising in molten metal under a laminar flow.
[4] Ngo SI, Lim Y-I, Kwon HM, Lee U-D. Hydrodynamics of Steel Res Int 2015;86:1289e97. https://doi.org/10.1002/
molten-metal bubble columns in the near-bubbling field srin.201400355.
using volume of fluid computational fluid dynamics. Chem [19] Kudinov IV, Pimenov AA, Kryukov YA, Mikheeva GV. A
Eng J 2023;454:140073. https://doi.org/10.1016/ theoretical and experimental study on hydrodynamics,
j.cej.2022.140073. heat exchange and diffusion during methane pyrolysis in
[5] Msheik M, Rodat S, Abanades S. Methane cracking for a layer of molten tin. Int J Hydrogen Energy
hydrogen production: a review of catalytic and molten media 2021;46:10183e90. https://doi.org/10.1016/
pyrolysis. Energies 2021;14:3107. https://doi.org/10.3390/ j.ijhydene.2020.12.138.
en14113107. [20] Keipi T, Li T, Løvås T, Tolvanen H, Konttinen J. Methane
[6] Parkinson B, Balcombe P, Speirs JF, Hawkes AD, Hellgardt K. thermal decomposition in regenerative heat exchanger
Levelized cost of CO2 mitigation from hydrogen production reactor: experimental and modeling study. Energy
routes. Energy Environ Sci 2019;12:19e40. https://doi.org/ 2017;135:823e32. https://doi.org/10.1016/
10.1039/C8EE02079E. j.energy.2017.06.176.
[7] Muradov N. Low to near-zero CO2 production of hydrogen [21] Parkinson B, Tabatabaei M, Upham DC, Ballinger B, Greig C,
from fossil fuels: status and perspectives. Int J Hydrogen Smart S, McFarland E. Hydrogen production using methane:
Energy 2017;42:14058e88. https://doi.org/10.1016/ techno-economics of decarbonizing fuels and chemicals. Int
j.ijhydene.2017.04.101. J Hydrogen Energy 2018;43:2540e55. https://doi.org/10.1016/
[8] Noh Y-G, Lee YJ, Kim J, Kim YK, Ha J, Kalanur SS, Seo H. j.ijhydene.2017.12.081.
Enhanced efficiency in CO2-free hydrogen production from [22] Deckwer WD, Burckhart R, Zoll G. Mixing and mass transfer
methane in a molten liquid alloy bubble column reactor with in tall bubble columns. Chem Eng Sci 1974;29:2177e88.
zirconia beads. Chem Eng J 2022;428:131095. https://doi.org/ https://doi.org/10.1016/0009-2509(74)80025-4.
10.1016/j.cej.2021.131095. [23] Nguyen LX, Tran BV, Ngo SI, Lim Y-I, Go K-S, Nho N-S.
[9] Catalan LJJ, Rezaei E. Coupled hydrodynamic and kinetic Hydrodynamics of gas-liquid bubble columns under
model of liquid metal bubble reactor for hydrogen bubbling, transient, and jetting flow regimes using volume of
production by noncatalytic thermal decomposition of fluid computational fluid dynamics. Chem Eng Res Des
methane. Int J Hydrogen Energy 2020;45:2486e503. https:// 2022;182(6):616e28. https://doi.org/10.1016/
doi.org/10.1016/j.ijhydene.2019.11.143. j.cherd.2022.04.033.
[10] Catalan LJJ, Rezaei E. Modelling the hydrodynamics and [24] Wang T, Wang J, Jin Y. Slurry reactors for gas-to-liquid
kinetics of methane decomposition in catalytic liquid metal processes: a review. Ind Eng Chem Res 2007;46:5824e47.
bubble reactors for hydrogen production. Int J Hydrogen https://doi.org/10.1021/ie070330t.
Energy 2022;47:7547e68. https://doi.org/10.1016/ [25] Deckwer W-D. Bubble column reactors. New York: Wiley;
j.ijhydene.2021.12.089. 1992.
[11] Parkinson B, Patzschke CF, Nikolis D, Raman S, Hellgardt K. [26] Plevan M, Geißler T, Aba
nades A, Mehravaran K,
Molten salt bubble columns for low-carbon hydrogen from Rathnam RK, Rubbia C, Salmieri D, Stoppel L, Stückrad S,
CH4 pyrolysis: mass transfer and carbon formation Wetzel T. Thermal cracking of methane in a liquid metal
mechanisms. Chem Eng J 2021;417:127407. https://doi.org/ bubble column reactor: experiments and kinetic analysis. Int
10.1016/j.cej.2020.127407. J Hydrogen Energy 2015;40:8020e33. https://doi.org/10.1016/
[12] Upham DC, Agarwal V, Khechfe A, Snodgrass ZR, Gordon MJ, j.ijhydene.2015.04.062.
Metiu H, McFarland EW. Catalytic molten metals for the [27] Kim J, Oh C, Oh H, Lee Y, Seo H, Kim YK. Catalytic methane
direct conversion of methane to hydrogen and separable pyrolysis for simultaneous production of hydrogen and
carbon. Sci 2017;358:917e21. https://doi.org/10.1126/ graphitic carbon using a ceramic sparger in a molten NiSn
science.aao5023. alloy. Carbon 2023;207:1e12. https://doi.org/10.1016/
[13] Rahimi N, Kang D, Gelinas J, Menon A, Gordon MJ, Metiu H, j.carbon.2023.02.053.
McFarland EW. Solid carbon production and recovery from [28] Jakobsen HA. Chemical reactor modeling: multiphase
high temperature methane pyrolysis in bubble columns reactive flows. Berlin: Springer; 2008.
 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 4 8 ( 2 0 2 3 ) 3 5 8 2 1 e3 5 8 3 7 35837

[29] Deckwer W-D, Ma € hlmann EA. Boundary conditions of liquid [35] Hibiki T, Saito Y, Mishima K, Tobita Y, Konishi K,
phase reactors with axial dispersion. Chem Eng J Matsubayashi M. Study on flow characteristics in gas-
1976;11:19e25. https://doi.org/10.1016/0300-9467(76)80003-2. molten metal mixture pool. Nucl Eng Des
[30] Mangartz KH, Pilhofer T. Interpretation of mass transfer 2000;196:233e45. https://doi.org/10.1016/S0029-5493(99)
measurements in bubble columns considering dispersion of 00293-9.
both phases. Chem Eng Sci 1981;36:1069e77. https://doi.org/ [36] Kataoka I, Ishii M. Drift flux model for large diameter pipe
10.1016/0009-2509(81)80093-0. and new correlation for pool void fraction. Int J Heat Mass
[31] Tokunaga H, Iguchi M, Tatemichi H. Heat transfer between Tran 1987;30:1927e39. https://doi.org/10.1016/0017-9310(87)
bubbles and molten Wood's metal. ISIJ Int 1995;35:21e5. 90251-1.
https://doi.org/10.2355/isijinternational.35.21. [37] Akita K, Yoshida F. Bubble size, interfacial area, and liquid-
[32] Arpaci VS. Conduction heat transfer. Massachusetts: phase mass transfer coefficient in bubble columns. Ind Eng
Addison-Wesley Reading; 1966. Chem Process Des Dev 1974;13:84e91. https://doi.org/
[33] Luchesi VM, Coelho RT. Experimental investigations of heat 10.1021/i260049a016.
transfer coefficients of cutting fluids in metal cutting [38] Mohagheghian S, Elbing BR. Characterization of bubble size
processes: analysis of workpiece phenomena in a given case distributions within a bubble column. Fluid 2018;3:13.
study. Proc Inst Mech Eng, Part B 2012;226:1174e84. https:// https://doi.org/10.3390/fluids3010013.
doi.org/10.1177/0954405412442459. [39] Hooshyar N, Hamersma PJ, Mudde RF, van Ommen JR.
[34] Hibiki T, Ishii M. One-dimensional drift-flux model and Intensified operation of slurry bubble columns using
constitutive equations for relative motion between phases in structured gas injection. Can J Chem Eng 2010;88:533e42.
various two-phase flow regimes. Int J Heat Mass Tran https://doi.org/10.1002/cjce.20338.
2003;46:4935e48. https://doi.org/10.1016/s0017-9310(03)00322-3.