You are on page 1of 22

chemical engineering research and design 9 1 ( 2 0 1 3 ) 296–317

Contents lists available at SciVerse ScienceDirect

Chemical Engineering Research and Design

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

A numerical study of pellet model consistency with respect


to molar and mass average velocities, pressure gradients
and porosity models for methanol synthesis process:
Effects of flux models on reactor performance

K.R. Rout ∗ , M. Hillestad, H.A. Jakobsen


Department of Chemical Engineering, Norwegian University of Science and Technology, NTNU, N-7491 Trondheim, Norway

a b s t r a c t

The objective of this work is to compare mass- and mole based diffusion flux models, convection, fluid velocity and
pore structure models for methanol synthesis process. Steady-state models have been derived and solved using least-
squares spectral method (LSM) to describe the evolution of species composition, pressure, velocity, total concentration
and diffusion fluxes in porous pellets for methanol synthesis. Mass diffusion fluxes are described according to the
rigorous Maxwell Stefan model, dusty gas model and the more simple Wilke model. These fluxes are defined with
respect to molar- and mass averaged velocities. The different effects of choosing the random- and parallel pore models
have been investigated. The effects of Knudsen diffusion have been investigated. The result varies significantly in
the dusty gas model. The effectiveness factors have been calculated for the methanol synthesis process for both
mass- and mole based pellet models. The values of effectiveness factors for both mass- and mole based pellet
models do not vary so much. The effect of Wilke-, Maxwell–Stefan- and dusty gas mass diffusion fluxes on the
reactor performance have been studied. Steady-state heterogeneous fixed bed reactor model is derived where the
intra-particle mass diffusion fluxes in the voids of the pellet are described by Wilke-, Maxwell–Stefan- and dusty gas
models. Furthermore, the total computational efficiency of the heterogeneous fixed bed reactor model is calculated
with several closures for the intra-particle mass diffusion fluxes. The model evaluations revealed that:

- The mass- and mole based pellet models are not completely consistent. However, the small deviation (less than 2%) between mass- and
mole based pellet models is due to the model equations are not fully consistent. If one pellet model is to be chosen for the methanol
synthesis process, the optimal diffusion flux model is the Maxwell–Stefan model.
- The parallel pore model is deviating from the random pore model for the methanol synthesis process. The results of both the parallel-
and random pore models have been compared with experimental data available in the literature. It is found that the result of the
parallel pore model is well agreement with experimental data.
- A small but significant differences in the mole fraction profiles of methanol along the reactor axis where the diffusion fluxes are
described according to the Wilke-, Maxwell–Stefan- and dusty gas models.
- The consistent models are about 20 and 60% more computationally expensive than the simplified and not always consistent Wilke
model. It is recommended to use the consistent models instead of Wilke approach.

© 2012 The Institution of Chemical Engineers. Published by Elsevier B.V. All rights reserved.

Keywords: Multicomponent diffusion; Methanol synthesis; Wilke model; Maxwell–Stefan model; Dusty gas model;
Heterogeneous reactor model


Corresponding author. Tel.: +47 73594146; fax: +47 73594080.
E-mail addresses: kumar.rout@chemeng.ntnu.no, ranjan8109@gmail.com (K.R. Rout).
Received 8 December 2011; Received in revised form 15 July 2012; Accepted 8 September 2012
0263-8762/$ – see front matter © 2012 The Institution of Chemical Engineers. Published by Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.cherd.2012.09.003
chemical engineering research and design 9 1 ( 2 0 1 3 ) 296–317 297

1. Introduction and energy fluxes are not measurable quantities. Therefore,


for the calculation of multicomponent problems containing
Methanol has been widely used as a basic feed-stock in many convective transport phenomena the mass averaged velocity
chemical processes. It is used as a fuel additive and it is the based description of the diffusion has en great advantage. If
starting chemical for producing formaldehyde and other sol- this description has been chosen, then all the conservation
vents. Moreover, it is a key reactant for synthesis of methyl equations contain only mass average velocities, while in the
tertiary butyl ether (MTBE), which was a favored gasoline addi- molar description of the diffusion the equations of motion
tive for reducing air pollution, but problems developed with its and energy contain mass average velocities. So in the molar
use because of water pollution. With the increasing concern based model we have converted the molar averaged velocity
over the environment, the demand for methanol will increase to mass averaged velocity to use in the equations of motion
significantly in the near future as a result of increasing pres- and energy (Bird et al., 2002). Hence, in the molar description
sure to convert methane to liquid fuels and to produce olefins of the diffusion model we will have two velocities, one is mass
from methanol. Methanol is also easily transported and is con- averaged velocity and the other one is mole averaged velocity.
sidered to be a future renewable energy (Mullins and Baron, The two velocities are equal only when all the components of
1997). In addition, it might be used potentially as a cleaner and the system have same molar mass.
more reliable fuel than the petroleum-based fuel for the future Until today, the diffusion fluxes have been described by
(Boutacoff, 1989; Gray and Alson, 1989). Methanol is also used mole averaged velocity for methanol synthesis. Lommerts
in direct methanol fuel cell in consumer electronics (Yusup et al. (2000) and Graaf et al. (1990) have developed the multi-
et al., 2010). component diffusion mole based pellet model with respect to
Methanol synthesis is a heterogeneous catalyzed process mole averaged velocity. However, in their steady-state model
where CO and H2 react over Cu/ZnO/Al2 O3 catalyst at low they assumed temperature is uniform throughout the spher-
temperatures and high pressure to produce methanol. The ical particle. In our simulations, we have checked whether
following reactions can be assumed to occur in methanol syn- the temperature- and pressure vary throughout catalyst par-
thesis (Graaf et al., 1990; Vanden Bussche and Froment, 1996). ticle or not for both mass- and mole based pellet models.
Recently, Solsvik and Jakobsen (2011) developed a dynamic
CO + 2H2 = CH3 OH H298 = −90.55 kJ/mol (I) mass- and mole based mathematical model to investigate the
phenomena of mass and heat transport in porous catalyst
CO2 + 3H2 = CH3 OH + H2 O H298 = −49.43 kJ/mol (II) pellet for methanol synthesis. It was found that the mass-
and mole based models differ nearly 4% for the methanol
CO2 + H2 = CO + H2 O H298 = 41.12 kJ/mol (III) synthesis process by Solsvik and Jakobsen (2011), but the
reason for this in-consistency between the mass- and mole
The life cycle of the commercial catalyst Cu/ZnO/Al2 O3 based models has not been verified yet. Methanol synthe-
with 50–70% of CuO in weight is in the order of two years. sis is one of the important processes. It has been studied
Hence, the catalyst deactivation phenomena can reason- extensively by many researchers and a variety of numerical
ably neglected for the steady-state simulation proposed here methods applied to solve the pellet models. Orthogonal col-
(Manenti et al., 2011). location (Finlayson, 1972), in particular has been found to be
Internal mass transfer limitations is significant for an adequate method that is now routinely used in diffusion-
methanol synthesis using industrial Cu/ZnO/Al2 O3 cata- reaction problems. But in this paper we have applied the
lyst (Lommerts et al., 2000; Seyfert and Luft, 1985). When least-squares spectral method (LSM) to solve the diffusion-
modelling an industrial reactor, internal mass transport limi- reaction problem in a porous pellet. Rout et al. (2011) have
tations should be taken into account. There are several models developed mass based multicomponent diffusion models for
reported in the literature for the multicomponent diffusion the sorption-enhanced steam methane reforming process.
fluxes. In the Wilke model (Krishna and Wesselingh, 1996; They have solved the mass based mathematical models with
Jakobsen, 2008), the binary diffusion coefficient is replaced least-squares spectral element method (LS-SEM). However, the
by an effective multicomponent diffusion coefficient. This details regarding the implementation of the LSM method for
model represents a simplified version of the more rigor- the multicomponent diffusion models has not been reported
ous Maxwell–Stefan model (Krishna and Wesselingh, 1996; yet. In this paper, the implementation of LSM method on
Jakobsen, 2008; Maxwell, 1866; Stefan, 1871). The dusty gas diffusion-reaction problem has been described. The least-
model (Jackson, 1997; Mason and Malinauskas, 1983) denotes squares spectral method (LSM) is a well-established numerical
an extension of the Maxwell–Stefan model in which Knudsen method for solving a wide range of mathematical problems
diffusion is considered. (Jiang, 1998; Bochev, 2001; Proot and Gerritsma, 2002; De
The conventional modelling considers the mass averaged Maerschalck, 2003; Pontaza and Reddy, 2004). The basic idea
mixture velocity as a primitive variable instead of the num- in the LSM is to minimize the integral of the square of the
ber or molar averaged mixture velocity. The mass averaged residual over the computational domain.
velocity is natural and most convenient basis for the laws of The models conventionally applied to packed bed reactors
conservation of mixture mass, momentum and energy. It is are normally classified into two major categories: pseudo-
true that total number of moles of the molecules in a mix- homogeneous- and heterogeneous model. The pseudo-
ture is not necessarily conserved, where as the mixture mass homogeneous fixed bed reactor model does not account
must be conserved. Separate equations of motion and energy explicitly for the presence of catalyst, in contrast to the het-
for each species can be derived by continuum arguments. erogeneous models, which led to separate balance equations
However, for mixtures containing a large number of different for the fluid phase species mass and, the fluid inside the cat-
species the resulting sets of separate species mole, momen- alyst pores. The heat transport phenomena are considered
tum and energy equations are neither feasible nor needed for for the solid material and the fluid in the bulk and in the
solving transport problems. Besides, the species momentum pores. However, the pseudo-homogeneous models require an
298 chemical engineering research and design 9 1 ( 2 0 1 3 ) 296–317

effectiveness factor to account for the diffusion resistance past (Graaf et al., 1988; Hartig and Keil, 1993; Skrzypek et al.,
across the external film and inside the pores of the pellet. 1995; Lovik et al., 1998; Lommerts et al., 2000; Jakobsen et al.,
The diffusion resistance decreases the effective reaction rate, 2001; Rezaie et al., 2005; Shahrokhi and Baghmisheh, 2005;
hence the efficiency factors normally have values 0 ≤  ≤ 1. Manenti et al., 2011). However, computational efficiency of
Difference between diffusion flux models on the level of fixed the heterogeneous fixed bed reactor model was not reported
bed reactor may be minor, but can be of interest for a detailed for methanol synthesis where the intra-particle mass diffu-
calculation of a single pellet (Salmi and Warna, 1991). sion fluxes in the voids of the pellet are described by Wilke-,
The internal effectiveness factor , with respect to the Maxwell–Stefan- and dusty gas models. Fixed bed Lurgi reac-
external surface conditions of the pellet, is defined by (Fogler, tor (Lurgi Gmbh, 2009) is selected for this study in line with
1999; Froment and Bischoff, 1990): current trend for shell and tube reactor boilers to be used in
large scale plants.
actual overall rate of reaction The objective of the this paper is to check consistency of the
= (1)
rate of reaction with surface conditions pellet models with respect to mass- and mole averaged veloci-
ties for methanol synthesis process. The various types of pore
The effectiveness factors have been calculated for the both
structure models have been compared with experimental data
mass- and mole based pellet models for the methanol synthe-
found in literature to propose the best suitable pore structure
sis and compared with the literature data as a verification of
model for methanol synthesis process. Several clousers for
the model.
intra-particle mass diffusion fluxes; Wilke, Maxwell–Stefan-
The pellet models require empirical input describing the
and dusty gas have been compared on the pellet level and the
packing, porosity or sub-structure of the pellet through the
impacts of the different flux closures on the reactor level are
parameter ε. In the parallel-pore model, the linking of the
investigated.
porosity complexity to the intra-particle diffusion is done
via the so-called tortuosity factor . Thus, for a sample of a
porous material an effective diffusivity is defined (Aris, 1969;
2. Mathematical model formulation
Satterfield, 1980):
In this study, we have used general model for the pellet in
 which the reactions take place on active sites within the
De = D (2)
 porous body which represents an assembly of individual
grains or channels. A mean pore diameter is assumed and the
Considering the parallel pore model, it is assumed that the ratio between the porosity and tortuosity is used to character-
pellet has several perfectly cylindrical pores with a smooth ize the fixed structure of the pellet. Possible pellet structural
surface (Prachayawarakorn and Mann, 2007). changes are not considered. Steady-state models, contain-
In the random pore model, the pellet pore structure has ing several closure models for the multicomponent diffusion
been broken up into macro- and micro structures (Froment fluxes and including convection times have been investigated.
and Bischoff, 1990). The transport in the pellet is assumed to In addition, the model equations hold the pressure drop term
occur by a combination of diffusion fluxes through the macro- within the pellet is presented by Darcy law relation.
and micro regions and by a series of contributions involving Table 1 gives the mass- and mole based governing pellet
both regions. By considering a pellet containing only macro equations. To obtain a closed set of equations the constitutive
pores, Eq. (2) reduces to: laws are represented in Tables 2 and 3. Table 4 gives the con-
stitutive laws to obtain mass averaged velocity from the mole
De = 2 D (3) averaged velocity which is used in the temperature- and Darcy
law equation in the mole based pellet model.
which resembles the parallel pore model provided that  = The mass diffusion fluxes is in this study are described by
1/ε. Methanol production rate calculated from Maxwell–Stefan Maxwell–Stefan-, dusty gas- and Wilke models. Tables 5 and 6
model has been compared with the experimental data found hold the closures of the mass diffusion fluxes based on molar-
in the literature by Seyfert and Luft (1985) to conclude which and mass averaged velocities. Further, mass and heat transfer
pore structure model is giving the better results. However, it coefficient values are computed from correlation with dimen-
must be noted that Seyfert and Luft (1985) measured the rate of sionless numbers (Wakao and Funazkri, 1978; Rusten et al.,
the methanol production using varying catalyst sizes, ranging 2007). Tables 7 and 8 gives the boundary conditions which
from 1 to 10 mm. The kinetic expressions of Graaf et al. (1988, have been applied in the model equations. A system of non-
1990) used in the present study were derived from experi- dimensionalized model equations is used in the simulations.
ments at relatively low pressures, while Seyfert and Luft (1985) Appendix A outlines the non-dimensionalized equations for
reported experimental data up to 140/,bar. Hence, the kinetic mass- and mole based pellet models respectively.
expressions of the of the Graaf et al. (1988, 1990) require a Table 9 gives the steady-state one-dimensional hetero-
simple correction factor ˛ while simulated with the experi- geneous reactor model with axial dispersion. The axial
mental conditions of Seyfert and Luft (1985) and Lommerts dispersion coefficient in Eq. (78) is given by Edwards and
et al. (2000). Richardson (1968):
In the chemical reactor engineering approach we desire to
eliminate the mechanism that is not essential for the reactor 0.5uz dp
DLj = 0.73Dmj + (4)
performance from the equations to reduce the computational 1 + 9.49Dmj /(uz dp )
cost. An appropriate engineering packed bed reactor model
is thus tailored for its main purpose. It is as simple as pos- The molecular diffusivities Dmj of component j in the gas mix-
sible, but still include a sufficient representation of essential ture are calculated from the Wilke equation (Bird et al., 2002)
mechanisms involved. Some interesting works on fixed bed and the molecular binary diffusivities are calculated from rela-
reactors have been reported by several researchers in the tions by Poling et al. (2001). KD and KV are constants for the
chemical engineering research and design 9 1 ( 2 0 1 3 ) 296–317 299

Table 1 – Mass- and mole based pellet model equations.


Mass based equations:
1 ∂ 2

n

Continuity eq.: (r ur g ) = Swj (40)


r2 ∂r
j=1

∂wj 1 ∂ 2

n

Species mass balance eq.: ur g =− (r Jj,r ) + Swj − wj Swj (41)


∂r r2 ∂r
j=1

n
∂T 1 ∂
Heat balance eq.: g wj Cpj ur = − 2 (r2 Qr ) + ST (42)
∂r r ∂r
j=1

Mole based equations:


1 ∂ 2 

n

Continuity eq.: (r ur cg ) = sxj (43)


r2 ∂r
j=1

∂xj 1 ∂ 2

n

Species mole balance eq.: ur cg =− (r Jj,r ) + sxj − xj sxj (44)


∂r r2 ∂r
j=1

n
∂T 1 ∂
Heat balance eq.: cg xj Cpj ur = − 2 (r2 Qr ) + ST (45)
∂r r ∂r
j=1

Table 2 – Constitutive laws.


g RT
Ideal gas law: P = cg RT = (46)
M

n

n

Definitions (Table 17.8-1 (Bird et al., 2002)) (Lommerts et al., 2000): wj = xj = 1 (47)
j=1 j=1
 n
 n

Jj,r = Jj,r = 0 (48)


j=1 j=1
ˇ0
The Darcy’s law (Krishna and Wesselingh, 1996; Jakobsen, 2008): ur = − ∇P (49)

 d20
Poiseuille flow relationship for permeability (Krishna and Wesselingh, 1996): ˇ0 = (50)
 32
∂T
Heat flux: Qr = −
(51)
∂r

Table 3 – Source density.



3

Source density of species mass balance: Swj = (1 − )cat Mj Rk jk (52)


k=1

3

Source density of species mole balance: sxj = (1 − )cat Rk jk (53)


k=1
3

Source density of temperature eq.: ST = (1 − )cat (−Hrk )Rk (54)


k=1

viscous and kinetic pressure drop in Eq. (82) and are given as for the methanol synthesis on Cu/ZnO/Al2 O3 catalysts which
(Ergun, 1952): have been widely used. Their rate equations were also used in
our simulations.
2
150 (1 − b )
KD = (5)
d2p 3b 3/2 1/2
k1 KCO [fCO fH − fCH3 OH /(fH /KP1 )]
R1 = 2
1/2
2
1/2
× 10−3
1.75(1 − b ) (1 + KCO fCO + KCO2 fCO2 )[fH + (KH2 O /KH fH2 O ))]
KV = (6) 2
dp 3b (7)

Table 10 gives the boundary conditions which have been


applied in the reactor model equations. Non-dimensionalized Table 4 – Mass averaged velocity from mole averaged
model equations are used in the reactor simulations which velocity (Bird et al., 2002).
have been reported by Sporleder et al. (2010). 
n

ur = ur + wj (uj − ur ) (55)


j=1
2.1. Chemical model
where uj is the velocity of species with respect to fixed
co-ordinates. uj can be calculated from the equation:
Graaf et al. (1990) investigated a large number of detailed Jj = cj (uj − ur ) (56)
mechanisms and derived a set of intrinsic rate equations
300 chemical engineering research and design 9 1 ( 2 0 1 3 ) 296–317

Table 5 – Diffusion flux closures with respect to mass averaged velocity.


⎛ ⎞−1
⎜ ⎟
⎜ n ⎟

= ⎜Mw
wi ⎟
Wilke model (Krishna and Wesselingh, 1996): jj = −g Dejm ∇wj where Dejm ⎟ (57)
⎜ Mwi Deji ⎟
⎝ i=1 ⎠
i=
/ j
n
(Mw /g ) (wj Ji /Mwi Deji ) − wj ∇ ln(Mw ) − ∇wj
i=1
i=
/ j
Maxwell–Stefan model (Jakobsen, 2008): Jj = n (58)
(Mw /g ) (wi /Mwi Deji )
i=1
i= / j
n
2
Mw (wj Ji /Mwi Deji ) − (vj Mw Dkjk ) − (wj ∇Mw + Mw ∇wj )
i=1
i=
/ j
Dusty gas model (Jakobsen, 2008): Jj = n (59)
2
Mw (wi /Mwi Deji ) + (Mw /Dejk )
i=1
i=
/ j
0.5
where DejK can be expressed as (Geankoplis, 1993): Dejk = 97.0 r(T/Mj ) (60)

Table 6 – Diffusion flux closures with respect to mole averaged velocity.


1 − xj
Wilke model (Krishna and Wesselingh, 1996): Jj = −cg Dejm ∇xj where Dejm = n (61)
(xi /Deji )
i=1
i=
/ j
n
(xj Ji /Deji ) − cg ∇xj
i=1
i=
/ j
Maxwell–Stefan model (Jakobsen, 2008): Jj = n (62)
(xi /Deji )
i=1
i=/ j
n
(xj Ji /Deji ) − (ci u∗ /Dejk ) − cg ∇xj
i=1
i=
/ j
Dusty gas model (Jakobsen, 2008): Jj = n (63)
(xi /Deji ) + (1/Dejk )
i=1
i=
/ j
0.5
where DejK can be expressed as (Geankoplis, 1993): Dejk = 97.0 r(T/Mj ) (64)

3/2 3/2 equations consists of several mathematical tasks like defining


k2 KCO2 [fCO2 fH − fCH3 OH fH2 O /(fH /KP2 )]
R2 = 2
1/2
2
1/2
× 10−3 an optimized form of the system of partial differential equa-
(1 + KCO fCO + KCO2 fCO2 )[fH + (KH2 O /KH fH2 O ))]
2 tions and boundary conditions to be solved, approximate the
(8)
corresponding differential and boundary operators, choose
the relevant discretization spaces, and an optimized solution
k3 KCO2 [fCO2 fH2 − fH2 O fCO /KP3 ] method for the resulting set of discrete equations. Thus the
R3 = 1/2 1/2
× 10−3
(1 + KCO fCO + KCO2 fCO2 )[fH + (KH2 O /KH fH2 O ))] application of the least squares method can result in several
2
(9) substantially different algorithms, even for the same problem.
Despite these differences, the spectral and spectral element
where R1 , R2 , R3 correspond to reactions (I)–(III), respectively. methods share as a common principle, the minimization of a
norm-equivalent functional. This allows us to consolidate the
formulation and analysis of a large class of methods into a
3. The least-squares method
single weighted residual framework. Consider the generalized
formulation of a arbitrary set of partial differential equations
3.1. The least-squares formulation
and boundary conditions:

In general, the design of a least-squares spectral element


Lf = g in ˝ (10)
method for the numerical solution of a given set of problem

Table 7 – Boundary conditions for mass based equations. Table 8 – Boundary conditions for mole based equations.
Boundary conditions Boundary conditions
Jj,r = 0 for r = 0 (due to symmetry) (65) Jj,r = 0 for r = 0; due to symmetry (71)
b
Jj,r + ur g wj = −kj (j,g − wj g ) for r = rp (66) Jj,r + ur cg xj = −kj (cj,g
b
− xj cg ) for r = rp (72)
Qr = 0 for r = 0 (due to symmetry) (67) Qr = 0 for r = 0 (due to symmetry) (73)
g Cpg ur T + Qr = −h(T b − T) for r = rp (68) cg Cpg ur T + Qr = −h(T b − T) for r = rp (74)
g = gb for r = rp (69) cg = cgb for r = rp (75)
∂P ∂P
= 0 for r = 0 (70) = 0 for r = 0 (76)
∂r ∂r
chemical engineering research and design 9 1 ( 2 0 1 3 ) 296–317 301

Table 9 – Reactor equations.



Continuity eq.: (uz g ) = 0 (77)
∂z
∂ ∂(b Jj )
Species mass balance eq.: (g wj uz ) + = av kj g (wsp,j − wj ) (78)
∂z ∂z
∂wj
Diffusive flux Jj : Jj = −g DLj (79)
∂z
∂T ∂
Heat balance eq.: g Cpg uz = − (Qz ) + av h(Tps − T) (80)
∂z ∂z
∂T
Heat flux Qz : Qz = −
z (81)
∂z
∂P
Ergun’s equation: = −KD uz − KV u2z (82)
∂z
PMw
Ideal gas law: g = (83)
RT

Bf = f on = ∂˝ (11) The Lagrange basis polynomials hj , representing a particu-


lar kind of basis functions ϕj , having the property,
in which ˝ denotes the domain and ∂˝ indicates the bound-
aries of the domain. Here we assume that L is a linear ϕj (i ) = 1 if j = i and ϕj (i ) = 0 if j=
/ i
operator which corresponds to the system of equations and B
is the boundary condition operator determining the problem where i are called the collocation points. In our case we
domain. are using Lagrange basis functions ϕj () = hj . In the GL
The least-squares formulation seeks to minimize the resid- method, the polynomials contain points in the interior of
uals of Eqs. (10) and (11) through a norm-equivalent functional. the domain only and no boundary points are considered.
The least-squares norm-equivalent functional may be defined In the GLL method, the polynomials contain points in the
like; domain and at the boundary. In the GRL method, the poly-
nomials contain one boundary and the interior points. In
1 our integration we have used either GL or GLL methods.
J(f ) = (Lf − g2Y(˝) + Bf − f 2Y( ) ) (12)
2 2. The solution function f is expanded in a Lagrange polyno-
mial over either GL or GLL points.
with the norms defined as:
3. A quadrature approximation of the norm integrals using
the specified points defined in the above step.
 • 2Y(˝) = •, •Y(˝) = • • d˝ (13)
˝
One further suppose that the function f (x) can be approxi-
and mated by a truncated series expansion like,

 • 2Y( ) = •, •Y( ) = • • d . (14) 
N
f (x) ≈ f N (x) = f j hj (x) (15)
j=0
Since R˝ = Lf − g and R = Bf − f , the norm-equivalent func-
tional, i.e., Eq. (12) expresses that minimizing J(f ) is equivalent
in which hj is the Lagrange fundamental polynomial and the
to solving Eqs. (10) and (11) in the least-squares sense. A gen-
coefficient is meaningful in the sense that, f j = f (xj ).
eral LSQ spectral solution method thus contains the following
Then, the functional may be written as:
steps;

1. The solution function is projected into a series sum of 1 1 2 2
JN (f N ) = (Lf N − g) d˝ + (Bf N − f ) d . (16)
basis functions. This can be obtained using either of two 2 ˝
2
approaches adopting modal or nodal bases. Using a modal
base the coefficients determine the amplitude of the mode. Introducing solution function expansion of Eq. (15), yields:
Examples of modal bases are the Lagurre, Fourier, Her-
mite functions. In the nodal base the basis functions are
⎛ ⎞2
based on orthogonal Lagrange basis polynomials like the 
N
1
Chebyshev and Legendre polynomials. Common choices JN (f N ) = ⎝L f j hj − g⎠ d˝
2
of basis functions for the Lagrange basis polynomials are ˝ j=0
the Gauss–Legendre (GL), Gauss–Labatto–Legendre (GLL) ⎛ ⎞2
and Gauss–Radau–Labatto (GRL) interpolating polynomials. 
N
1
+ ⎝B f j hj − f ⎠ d (17)
2 j=0
Table 10 – Boundary conditions for the reactor model
equations.
Boundary conditions The norm integral for the domain ˝ is computed using a P-
uz = uin for z = 0 (84) point quadrature rule. If we choose P = N, the collocation and
(wj g uz )(0+ , t) = (gb uin wjin )(0, t) − Jj (0+ , t) for z = 0 (85)
quadrature points are equal. Likewise, the norm integral for
Jj = 0 for z = L (86)
the boundary is computed using a P = N point quadrature
(Qz
)(0+ , t) + (Cpg uz g T)(0+ , t) = (Cpg uin gb Tin )(0, t) for z=0 (87)
Qz = 0 for z = L (88) rule.
P = Pin for z = 0 (89) To minimize the norm-equivalent functional, we have to
differentiate Eq. (17) in 1st order sense with respect to the
302 chemical engineering research and design 9 1 ( 2 0 1 3 ) 296–317

corresponding coefficients i.e., f j = f (xj ), and For the other boundary point (x = x =L = xN ), the point source
yields;
⎛ ⎞2 ⎛ ⎞2

N 
N xmax
∂ 1 ∂ 1
⎝ f j Lhj − g⎠ d˝ + ⎝ f j Bhj − f ⎠ F k = F (hk ) = f , Bhk Y( ) = f (x)Bhk (x)ı(x = xN )dx (27)
∂fk 2 ˝
∂fk 2
j=0 j=0 xmin

× d = 0, (18) F k = f (xN )Bhk (xN ), (28)

where ˝ = [xmin , xmax ] × [ymin , ymax ] × [zmin , zmax ].


xmax
⎛ ⎞ ⎛ ⎞
N 
N [B]kj = B(hj , hk ) = Bhj , Bhk Y( ) = Bhj (x)Bhk (x)ı(x = x0 )dx
⇒ ⎝ f j Lhj − g⎠ Lhk d˝ + ⎝ f j Bhj − f ⎠ xmin

˝ j=0 j=0 = Bhj (x0 )Bhk (x0 ) (29)

× Bhk d = 0. (19)
Likewise, for the boundary point (x = x =L = xN ), the matrix
[B]kj yields
Our aim is to find the value of fj , hence Eq. (19) can be repre-
sented on the form,
xmax

N 
N [B]kj = B(hj , hk ) = Bhj , Bhk Y( ) = Bhj (x)Bhk (x)ı(x = xN )dx
xmin
f j Lhj Lhk d˝ + f j Bhj Bhk d = gLhk d˝
˝ j=0 j=0 ˝ = Bhj (xN )Bhk (xN ) (30)

×+ f Bhk d . (20) Alternatively, the problem can be implemented in the matrix
form;

This statement can be represented in the inner product form


as, xmax 
P

[A]kj = Lhj , Lhk Y(˝) = Lhj (x)Lhk (x)dx ≈ wiq Lhj (xiq )

N 
N
xmin iq =0
f j Lhj , Lhk  + f j Bhj Bhk = g, Lhk  + f Bhk . (21)
j=0 j=0 
P

× Lhk (xiq ) = [L]iq k []iq iq [L]iq j = [L]Tk [L]j (31)


In the matrix form, if [A]kj = Lhj , Lhk , [B]kj = Bhj , Bhk  and iq =0
F k = g, Lhk , F k = f , Bhk , Eq. (21) can be represented on the
form, where the elements in the matrix L is defined as, [L]iq j =
Lhj (xiq ). Here, it is noted that wiq and xiq are the weights and
Af + Bf = F + F (22) points of quadrature (Karniadakis and Sherwin, 1999), and 
is the diagonal matrix containing weights of the quadrature.
The matrix [A]kj from Eq. (21) can then be expressed in the Hence, the matrix A can be written in the compact matrix form
pointwise form like, as,

A = LT L. (32)
xmax
[A]kj = A(hj , hk ) = Lhj , Lhk Y(˝) = Lhj (x)Lhk (x)dx
xmin Similarly, the vector F can be written in the compact matrix
form as,

P

≈ wiq Lhj (xiq )Lhk (xiq ) (23)


F = LT g. (33)
iq =0

and the source vector F k from Eq. (21) yields: 3.2. Implementation of LSM to the mass- and mole
based pellet models, and reactor model
xmax
For the cases of the mass- and mole based pellet models,
F k = F(hk ) = g, Lhk Y(˝) = g(x)Lhk (x)dx
15 collocation points have been used for the whole compu-
xmin
tational domain. Symmetry is assumed in the sphere which

P
results in one independent spatial variable in the radial
≈ wiq g(xiq )Lhk (xiq ) (24)
dimension which is discretized by the method of weighted
iq =0
residuals concepts.
The non-linear mass fluxes expressions (57)–(59), total con-
For the boundary point (x = x =0 = x0 ); the point source yields; tinuity equation (40), the species mass balance equation (41),
xmax the heat balance (42), heat flux equation (51), Darcy law (49),
F k = F (hk ) = f , Bhk Y( ) = f (x)Bhk (x)ı(x = x0 )dx (25) density (46) for the mass based pellet model and non-linear
xmin mole fluxes (61)–(63), total continuity equation (43), the species
mole balance equation (44), the heat balance (45), heat flux
F k = f (x0 )Bhk (x0 ), (26) equation (51), Darcy law (49), total concentration equation (46)
chemical engineering research and design 9 1 ( 2 0 1 3 ) 296–317 303

for the mole based pellet model have been solved as a mixed d
 
where {L1jj } = dr∗
+ r2∗ (•j )
system of first order differential and algebraic equations.  ∗ ∗ d

{L2jj } = u  dr∗ (•j )
In this section, we outline the application of the least
{L3jj } = (1) (•j )
squares method for solving the proposed set of governing
1 d
equations. For this case, the unknown f is defined as a vector {L4jj } = ∗ ∗ (•j )
Mw∗  n wi  dr
containing all the unknown vectors of the problem, namely: i = 1 ∗  e∗ 
g∗  Mwi Dji
Mass based pellet model:
i=/ j
T

 d 
f = J1 w1 J2 w2 J3 w3 J5 w5 J6 w6 J7 w7 Q T g P ur L33 = dr∗
+ r2∗ (•)
 Dref ref Cpg

L34 = u∗ ∗ drd∗
(•)
Mole based pellet model: L43 = (•)
T L44 = drd∗ (•)
f = J1 x1 J2 x2 J3 x3 J4 x4 J6 x6 J7 x7 Q T cg P ur L55 = (•)
Mw∗
L56 = − T∗  M (•)
To handle the non-linearity, the L-operator is linearized to L b
d
in order to implement a Picard iteration solution technique. L66 = dr∗
(•)
Dref
Hence, the system of equations can be expressed like: A f = F L67 =
ˇpb
(•)
in which the prime indicate a linearized form. L75 = (u∗ drd∗ )(•)
The governing equation in the solution process, a system of L77 = ∗ drd∗ (•)
non-dimensionalized model equations has been used. Hence,
the unknown f can be written for pellet models where the The mass fraction of the 4th component was calculated a
diffusion flux has been described by Maxwell-Stefan model posteriori, knowing that the sum of the mass fractions by
as: definition must give unity.
T
f = {Jj∗ wj } Q∗ T∗ ∗ P∗ u∗ ω4 = 1 − (ω1 + ω2 + ω3 + ω5 + ω6 + ω7 )

Mass based pellet model For mole based pellet model:


T
g = {Rj rj } C1 0 0 C2 C3
T
f = {J∗j xj } Q∗ T∗ c∗ P∗ u∗ The velocity in the Darcy law is a mass averaged velocity and
it has been calculated from the mole averaged velocity. Hence,
Mole based pellet model in the case of mole based pellet model, there is source term
n
in the Darcy law equation, i.e., C2 = −(Dref /ˇpb ) j=1 wj (u∗j −
where the star indicates non-dimensionalized variable. u∗ ). The rest of the modelling procedure is thus same as for
For mass based pellet model: the mass based pellet model.
T For reactor model:
g = {Rj rj } C1 0 0 0 C2 In the reactor model, 10 collocation points have been used
for the whole reactor axis and 10 collocation points for pel-
where Rj = rp2 (1/ref Dref )Swj
let radius which are discretized by the method of weighted
⎡ ⎤
⎛ ⎞
⎢ ⎥
⎢⎜ ⎟ ⎥
rj = ⎢⎜(Mw∗  /g∗  ) n (J ∗  /M∗  De∗  )⎟ × w ×

1 ⎥−
⎢⎝ i=1 i wi ji ⎠ j ∗  ∗  n
(w  /Mwi Dji ) ⎥
∗  e∗ 
⎣ (Mw /g )
i=1 i ⎦
i=
/ j
i=
/ j
⎡ ⎤
⎢ ⎥ residuals concepts. The implementation of reactor equations
⎢ 1 wj d ⎥
⎢ ) × × × M ∗ ⎥ in the least-squares method have been reported by Sporleder
⎢ (M∗  /∗  ) n  ∗  e∗  ∗
Mw 
dr ∗ w ⎥
et al. (2010).
⎣ w g (w /Mwi Dji
i=1 i ⎦
i=/ j 3.3. Model convergence
C1 = (rp2 /(ref × Tref × Dref )) × (1/Cpg )ST )
 ∗ ∗ 
C2 = − 2u r∗ + rp2  1
Dref Swj . In order to assess the convergence of the model, we define two
ref
convergence criteria like:
⎡ ⎤
{L1jj } {L2jj } 0 0 0 0 0 
⎢ 3 ⎥ Residual = LfN − g, LfN − g ≤ e− 5 (34)
⎢ {Ljj } {L4jj } 0 0 0 0 0 ⎥
⎢ ⎥ 
⎢ 0 L33 L34 0 ⎥
⎢ 0 0 0 ⎥ εiteration = fN it − fN it−1 , fN it − fN it−1 ≤ e− 10
⎢ ⎥ (35)
⎢ 0 0 L L44 0 0 0 ⎥
L = ⎢
43

⎢ 0 0 0 0 L55 L56 0 ⎥
⎢ ⎥ The Residual denotes a measure for the overall error
⎢ 0 ⎥ obtained for the system of equations discretized by the least
⎢ 0 0 0 0 L66 L67 ⎥
⎢ ⎥ squares method. The iteration overall error denotes a mea-
⎣ 0 0 0 0 L75 0 L77 ⎦
sure of the difference in variable values between the second
last- and the last preceding iterations, for a preliminary L
304 chemical engineering research and design 9 1 ( 2 0 1 3 ) 296–317

−2
10
Table 11 – Simulation parameters in a particle for
methanol synthesis (Graaf et al., 1990; Rezaie et al.,
2005). 10
−4

Mole fractions Mass Base Model

2
Mole Base Model
CH4 0.1026 –

L
−6

|| Norm of Residual ||
10
CH3 OH 0.0049999 –
CO 0.045998 –
−8
CO2 0.093998 – 10
H2 0.65901 –
H2 O 0.00039999 – −10
10
N2 0.092997 –
Pressure 80 × 105 Pa
Temperature 523 K 10
−12

Physical parameters
Density of catalyst 1950 kg cat/m3 −14
10
0 1 2
Void fraction/ tortuosity 0.123 – 10 10 10
Pellet diameter 0.0042 m Order
Mean pore diameter 20 nm
Heat capacity of solids 823 J/kg/,K Fig. 1 – L2 norm of the residual vs. order of the
Velocity 0.4 m/s approximation.

representing a linearized integro-differential operator using


the Picard iteration technique. The iteration procedure pro- 0
ceeds updating the L estimates until both error measures are
less than the specified upper limits.
50
4. Results and discussion

The mass- and mole based steady state model describes the 100
evolution of species mass composition, pressure, concentra-
tion, temperature, gas velocity, mass- and mole diffusion flux,
150
heat flux and convection for methanol synthesis have been
simulated by the least-squares spectral method (LSM). The
pellet model consistency, temperature- and pressure profile 200
across the pellet have calculated after converting the mole
averaged velocity to mass averaged velocity for the mole based
model, which has been used in the temperature and Darcy 250
law equation. The pellet model has been verified with result 0 50 100 150 200 250
found in the literature. Table 11 shows the parameter used for nz = 7879
solving pellet equations. The effect of particle flux closures
Fig. 2 – Shape of problem matrix of the mass based model.
on the reactor performance and the computational efficiency
have been calculated. Table 12 shows the parameter used for
solving reactor equations.
Fig. 1 shows the dependence of the error with expansion
order N for the pellet model. It has been shown that the
0.6
error is reduced with an exponential convergence rate. The CH OH (Simulation)
3
CO(Simulation)
0.5
CO2(Simulation)
Table 12 – Reactor specification for methanol synthesis
Mole fraction

H2O(Simulation)
(Rezaie et al., 2005). 0.4
CO (Simulation)
Mole fractions 2
CH OH (Graaf et al.[4])
CH4 0.1026 – 0.3
3
CH3 OH 0.0049999 – CO (Graaf et al.[4])
H (Graaf et al.[4])
CO 0.045998 – 2
0.2 CO (Graaf et al.[4])
CO2 0.093998 – 2
H2 0.65901 – H O (Graaf et al.[4])
2
H2 O 0.00039999 – 0.1
N2 0.092997 –
Inlet pressure 80 × 105 Pa
0
Tube inlet temperature 497 K 0 0.5 1
Tube outlet temperature 533 K Pellet Radius [−]

Physical parameters Fig. 3 – Comparison between the result of derived model


Tube inner diameter 0.051 m and the result of the model given by Graaf et al. (1990) for
Tube outer diameter 0.066 m
the catalyst particle. T = 547.8 K, P = 100 bar. Bulk gas
Length of tube 7 m
composition: yCO = 0.3125; yCO2 = 0.0625; yH2 = 0.6250.
Inlet superficial velocity 0.4 m/s
chemical engineering research and design 9 1 ( 2 0 1 3 ) 296–317 305

0.08
0.05
Mass base Model
0.045 Mass base Model
0.07 Mole base Model
Mole base Model
0.04
Mole fraction CH3OH 0.035 0.06

Mole fraction CO
0.03

0.025 0.05

0.02
0.04
0.015

0.01
0.03
0.005

0 0.02
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1

Pellet Radius [−] Pellet Radius [−]

(a) CH 3 OH (b) CO
0.1 0.8

Mass base Model 0.75


0.095
Mole base Model

0.09 0.7
Mole fraction CO2

2
Mole fraction H
0.085 0.65

0.08 0.6

0.075 0.55

0.07 0.5

0.065 0.45 Mass base Model


Mole base Model
0.06 0.4
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Pellet Radius [−] Pellet Radius [−]

(c) CO 2 (d) H 2
0.02

0.018 Mass base Model


Mole base Model
0.016

0.014
Mole fraction H2O

0.012

0.01

0.008

0.006

0.004

0.002

0
0 0.2 0.4 0.6 0.8 1
Pellet Radius [−]

(e) H 2 O
Fig. 4 – Comparison of the mass- and mole based model in a catalyst particle for the Wilke model.

norm of the residual, shown in Fig. 1 decreases until reach- treating these terms. However, we have calculated the reac-
ing a point of limiting accuracy, close to numerical precision. tion terms from the previous step values and treated them
The convergence rate is strongly affected by the capability of as source terms in the least-squares formulation. Then, we
the solver, which allows for the attainment of more accurate performed iterations to reduce the error. In our simulations,
results (Dorao and Jakobsen, 2006). we have used two convergence criteria: residual and εiteration .
The shape of the problem matrix for the pellet model has The iteration procedure proceeds updating the estimates
been shown in Fig. 2. When the equations involve an identity until both error measures are less than the specified upper
operator, the choice of the search affects appreciably the fill-in limits.
of the problem matrix, the Lagrange polynomials evaluated at
the GLL quadrature points yields zeros and ones. 4.1. Model validation
It is important to discuss the implementation of source
density term given in Table 3 for the mass- and mole based Graaf et al. (1990) have developed the dusty gas mole diffu-
model equations. The reaction rates used to calculate these sion model assuming uniform temperature across the pellet.
source terms are extremely sensitive to variations of the main In Fig. 3, we have presented a comparison of the current steady
parameters. Hence, it is difficult to define an optimal way of state mole fraction profiles for the methanol production
306 chemical engineering research and design 9 1 ( 2 0 1 3 ) 296–317

0.08
0.05
Mass base Model
0.045 Mass base Model
0.07 Mole base Model
Mole base Model
0.04

0.035 0.06
Mole fraction CH3OH

Mole fraction CO
0.03

0.025 0.05

0.02
0.04
0.015

0.01
0.03
0.005

0 0.02
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Pellet Radius [−] Pellet Radius [−]

(a ) CH 3 OH (b) CO
0.1 0.8

Mass base Model 0.75


0.095
Mole base Model

0.09 0.7
Mole fraction CO2

2
0.65
Mole fraction H
0.085

0.08 0.6

0.075 0.55
Mass base Model
0.07 0.5 Mole base Model

0.065 0.45

0.06 0.4
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Pellet Radius [−] Pellet Radius [−]

(c ) CO 2 (d) H 2
0.02

0.018 Mass base Model


Mole base Model
0.016

0.014
Mole fraction H2O

0.012

0.01

0.008

0.006

0.004

0.002

0
0 0.2 0.4 0.6 0.8 1
Pellet Radius [−]

(e) H 2 O
Fig. 5 – Comparison of the mass- and mole based model in a catalyst particle for the Maxwell–Stefan model.

process with the Graaf et al. (1990) with same operating con- 4.2. Comparison between mass- and mole base
ditions, but our model is a non-isothermal model. It has been diffusion models
seen from Fig. 3 that the agreement between the developed
model results and the model developed by Graaf et al. (1990) Figs. 4–6 compare the species mole fraction results of differ-
is very good. The internal effectiveness factor of methanol ent diffusion models obtained from the mole- and mass based
and water has values 0.86 and 0.63 respectively, which are in pellet models. It has been shown that the mole- and mass
agreement with the results reported by Lommerts et al. (2000). based models with respect to mass- and mole averaged veloc-
Moreover, Fig. 13 shows the influence of diffusion limitations ities do not give completely identical results. The simulation
for the experimental results of Seyfert and Luft (1985), the sim- results deviate less than 2% between mass- and mole based
ulation results of Lommerts et al. (2000) and the simulated pellet models. This study thus supports the literature finding
result of the developed model. Fig. 13 shows that the results by Solsvik and Jakobsen (2011). It should be noted that the
of developed model are well agreement with experimental Wilke formula does not satisfy the constraints that the sum
results for the parallel pore model. of the species composition must be unity and that the sum
chemical engineering research and design 9 1 ( 2 0 1 3 ) 296–317 307

0.08
0.05
Mass base Model
0.045 Mass base Model
0.07 Mole base Model
Mole base Model
0.04
Mole fraction CH3OH 0.035 0.06

Mole fraction CO
0.03

0.025 0.05

0.02
0.04
0.015

0.01
0.03
0.005

0 0.02
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Pellet Radius [−] Pellet Radius [−]

(a) CH 3 OH (b) CO
0.1 0.8

Mass base Model 0.75


0.095
Mole base Model

0.09 0.7
Mole fraction CO2

2
0.65

Mole fraction H
0.085

0.08 0.6

0.075 0.55

0.07 0.5

0.065 0.45 Mass base Model


Mole base Model
0.06 0.4
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Pellet Radius [−] Pellet Radius [−]

(c) CO 2 (d) H 2
0.02

0.018

0.016

0.014
Mole fraction H2O

0.012

0.01

0.008

0.006 Mass base Model


Mole base Model
0.004

0.002

0
0 0.2 0.4 0.6 0.8 1
Pellet Radius [−]

(e) H 2 O
Fig. 6 – Comparison of the mass- and mole based model in a catalyst particle for the dusty gas model.

of the diffusion fluxes must equal zero. The internal effec- 4.3. Reason for the in-consistency of the mass- and
tiveness factors calculated for both the mass- and mole based mole based model equations for methanol synthesis
model equations are same which is shown in Table 13.
From Fig. 1, it has been seen that the norm of the residuals
of both the mass- and mole based model behaves identically
with respect to order of approximation. The norm of the resid-
Table 13 – Internal effectiveness factors for the methanol
ual of both the mass- and mole based model decreases until
synthesis.
reaching a point of limiting accuracy, close to machine pre-
Reaction 
cision. But, we have found some in-consistency in the model
(I) 0.65 equations.
Mass- and mole based model (II) 0.12 4.3.1. In-consistency in mass based model equations
(III) 0.58 Eq. (40) is:
CH3 OH 0.86

Mass based modela H2 O 0.63

a 1 ∂ 2  n
Gas composition in mole basis: CH3 OH, 0.0; CO, 0.15; CO2 , 0.05; H2 , (r ur g ) = Swj
0.8; H2 O, 0.0; P = 100 bar; T = 525 K. r2 ∂r
j=1
308 chemical engineering research and design 9 1 ( 2 0 1 3 ) 296–317

Mole based model Mole averaged Velocity (Mole based model)


Mass based model Mass averaged velocity (Mole based model)
523.8 −6 Mass based model
x 10
2
523.7
0

523.6 −2
Temperature [k]

Velocity [m/sec]
523.5 −4

−6
523.4
−8
523.3
−10
523.2
−12
523.1
0 0.2 0.4 0.6 0.8 1 −14
0 0.2 0.4 0.6 0.8 1
Pellet Radius [−] Pellet Radius [−]

Fig. 7 – Comparison of the temperature profile of the mass Fig. 10 – Mass- and mole averaged velocity for the dusty
and mole based dusty gas model. gas model.

−4
x 10
Mole based model
Mass based model

Mass Diffusion flux of CH OH [kg/m sec]


80.1 Wilke

2
Maxwell Stefan
80 Dusty gas
2
79.9
3
Pressure[bar]

79.8

79.7
1
79.6

79.5

79.4 0
0 0.2 0.4 0.6 0.8 1
Pellet Radius [−]
0 0.2 0.4 0.6 0.8 1
Pellet Radius [−]
(a) CH 3 OH

Fig. 8 – Comparison of the pressure profile of the mass and Fig. 11 – Mass diffusive flux of methanol in a particle for
mole based dusty gas model. methanol synthesis process.

For the summation of all the components, Eq. (66) becomes:


n
For mass based model, S = 0 as the total mass is con-
j=1 wj 
n

n

n
served. After integration, we can write: Jj,r + ur g wj = − kj (gb − wj g ) for r = rp
j=1 j=1 j=1

ur = 0 ∀r (36)
Wilke model
Maxwell−Stefan model
Mole based model
−8 dusty gas model
x 10
Mass based model 2.5
Convective flux of CH3OH [Kg/m sec]

1.832
2

2
1.83
Total Concentration [Kmol/m3]

1.5
1.828
1
1.826

1.824 0.5

1.822 0

1.82 −0.5
1.818
−1
0 0.2 0.4 0.6 0.8 1
1.816
Pellet Radius [−]
0 0.2 0.4 0.6 0.8 1
Pellet Radius [−] (a)CH3 OH
Fig. 9 – Comparison of the total concentration profile of the Fig. 12 – Convective flux of methanol in a particle for
mass and mole based dusty gas model. methanol synthesis process.
chemical engineering research and design 9 1 ( 2 0 1 3 ) 296–317 309

n n
Simulation by parallel pore model In the above equation, J
j=1 j,r
= 0, wj = 1 and
70 Lommerts et al. (2000) n j=1
Experimental by Seyfert and Luft (1985) k (b − wj g ) =
j=1 j g
/ 0. So, the above equation becomes:
Simulation by random pore model
60
Reaction rate [mmol/s/kg]

1
n
50
ur = − kj (gb − wj g ) for r = rp (37)
g
40 j=1

30
So, Eq. (36) contradicts Eq. (37) at r = rp .
20
4.3.2. In-consistency in mole based model equations
2 4 6 8
Eq. (43) is:
Diameter of pellet [mm]

1 ∂ 2   n
Fig. 13 – Comparison of diffusion limitations in methanol (r ur cg ) = sxj
synthesis for the parallel- and random pore model at r2 ∂r
j=1
T = 518.2 K; the influence of particle size.

0.08
0.05
Random pore model
0.045 Random pore model
0.07 Parallel pore model
Parallel pore model
0.04

0.035 0.06
Mole fraction CH3OH

Mole fraction CO
0.03

0.025 0.05

0.02
0.04
0.015

0.01
0.03
0.005

0 0.02
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Pellet Radius [−] Pellet Radius [−]

(a ) CH 3 OH (b) CO
0.1 0.8

Random pore model 0.75


0.095
Parallel pore model

0.09 0.7
Mole fraction CO2

0.65
Mole fraction H

0.085

0.08 0.6

0.075 0.55

0.07 0.5

0.065 0.45 Random pore model


Parallel pore model
0.06 0.4
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Pellet Radius [−] Pellet Radius [−]

(c ) CO 2 (d) H 2
0.02

0.018

0.016

0.014
Mole fraction H2O

0.012

0.01

0.008

0.006

0.004

0.002 Random pore model


Parallel pore model
0
0 0.2 0.4 0.6 0.8 1
Pellet Radius [−]

(e) H 2 O
Fig. 14 – Comparison of the parallel- and random pore models in a catalyst particle of a mass based dusty gas model.
310 chemical engineering research and design 9 1 ( 2 0 1 3 ) 296–317

0.035 0.035

0.03 0.03

Mole fraction CH3OH

Mole fraction CH3OH


0.025 0.025

0.02 0.02

0.015 0.015
Wilke (2nm) Maxwell (2nm)
Wilke (5nm) Maxwell (5nm)
0.01 Wilke (20nm) 0.01 Maxwell (20nm)
Wilke (40nm) Maxwell (40nm)

0.005 0.005
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Pellet Radius [−] Pellet Radius [−]

(a ) CH 3 OH (b) CH 3 OH

0.07

0.06
Mole fraction CH3OH

0.05

0.04

0.03
Dusty (2nm)
0.02 Dusty (5nm)
Dusty (20nm)
0.01 Dusty (40nm)

0
0 0.2 0.4 0.6 0.8 1
Pellet Radius [−]

(c ) CH 3 OH
Fig. 15 – Effect of Knudsen diffusion for methanol synthesis process.

n
For mole based model, s = / 0 as the total mole is not Fig. 8 shows there is almost no variation of pressure from
j=1 xj
conserved. After integration, we can write: surface to the center of the catalyst pellet for the mass based
pellet model. However, the pressure profile is not uniform
r 
n across the pellet for the case of mole based pellet model. The
1
ur = r2 sxj dr ∀r (38) pressure has been calculated from the Darcy law in both the
cg r2 0 j=1 mass- and mole base models. Moreover, in the mole based
formulation, the velocity in the Darcy law is a mass averaged
For the summation of all the components, Eq. (72) becomes: velocity. In our calculation we have converted the mole aver-
aged velocity to mass averaged velocity and used it in the

n

n

n

Jj,r + ur cg xj = − kj (cgb − xj cg ) for r = rp Darcy law. For the case of mole based model, we have two
j=1 j=1 j=1
type of velocity, one is mass averaged velocity and the other
n n one is mole averaged velocity. Fig. 10 shows the mass- and
In the above equation, J = 0, x
j=1 j
=1 and mole averaged velocity for the mole base dusty gas model.
n j=1 j,r
k (cb
j=1 j g
− xj cg ) =
/ 0. So, the above equation becomes: According to physics both the mass- and mole based pel-
let models should give same result and the mass averaged

1
n velocity in the mole based model should zero across the pel-
ur = − kj (cgb − xj cg ) for r = rp (39) let surface. But the small deviation between the mass- and
cg
j=1 mole based models is due to the inconsistency between the
species boundary condition and total continuity equation.
So, Eq. (38) contradicts Eq. (39) at r = rp . However, this type of particle model is commonly used (Rout
et al., 2011; Solsvik and Jakobsen, 2011). It is noted that the
4.4. Common simplification of the pellet model equation system is solvable; the model is well-posed and not
equations for methanol synthesis over determined. But the boundary conditions applied are not
accurately specified, as the boundary conditions applied to
Fig. 7 shows the temperature profile across the pellet for the species mole balances are not fully consistent with the
mass- and mole based Dusty gas model. From this figure continuity equation condition at the surface. The mass formu-
it has been confirmed that there is very little temperature lation of the model also includes the inaccuracies related to
variation between the mass- and mole based models. How- the specification of the boundary conditions, but the velocity
ever, in the mole based formulation, the velocity in the is then equal to zero in accordance with the continuity equa-
temperature equation is a mass averaged velocity. In our cal- tion. From Fig. 10, it is evident that the mass based velocity
culation we have converted the mole averaged velocity to mass is not completely zero close to the surface is still a conse-
averaged velocity and used it in temperature equation. quence of inaccurate closer laws and contradictions between
chemical engineering research and design 9 1 ( 2 0 1 3 ) 296–317 311

Fig. 16 – Results obtained in the reactor simulation of the methanol synthesis.

the continuity equation and boundary condition used for the structural changes occur during the progress of the reaction.
species mole balances. It is noted that the LSM method both This effect has been studied for the commonly used pore mod-
the boundary condition and the continuity equation are solved els such as parallel- and random pore models. Fig. 14 shows
at the boundary nodes minimizing the sum of the norms of the the comparison between the parallel- and random pore mod-
residuals of the two equations. els for dusty gas model. It seems there is significant deviation
According to Burghardt and Aerts (1988) the pressure between parallel- and the random pore model. However, we
changes in a catalyst pellet under conditions normally have calculated the methanol production rate for both the
encountered in industry are probably so small that can be parallel- and random pore model. The calculated methanol
neglected in the process calculation. Hence, our results agree production rate of the random pore model has been com-
with those reported by Burghardt and Aerts (1988). Fig. 9 shows pared with the experimental data and also the result found by
the total concentration profile for the dusty gas model. As for Lommerts et al. (2000) which is shown in Fig. 13. From Fig. 13
the case of methanol synthesis moles are consuming, the total it has been confirmed that the methanol production rate is
concentration is decreasing from the surface to the center of much lower then the experimental value by Seyfert and Luft
the pellet. (1985) by considering random pore model. Hence, the choice
Figs. 11 and 12 show that close to external surface, the diffu- of random pore model for methanol synthesis is not a good
sion fluxes clearly dominate over the convective fluxes, hence option.
neglecting the convective flux terms in the governing equa-
tions for pellet model is a reasonable model approximation.
4.6. Effect of Knudsen diffusion for methanol
It can be concluded that in the pellet model, we can neglect
synthesis process
convective part, we can assume the pellet model as a uniform
temperature across the pellet and isobaric condition. These
In our calculation we have calculated pressure from the Darcy
assumptions will help us to get faster convergence for hetero-
law. The Darcy’s law applies for laminar flow in porous media;
geneous reactor models.
it is thus not valid in the Knudsen regime. However, in the
4.5. Comparison between parallel- and random pore dusty gas model it is assumed that for viscous flow and in
models the transition regime between viscous flow and Knudsen dif-
fusion, the combined flux is expressed as a linear sum of
In most of the chemical reactions, the complicated pore size the two contributions. It is thus assumed that the Darcy’s
distribution exists in the solid pellet. Moreover, significant law is valid for a Knudsen number below the limit where
312 chemical engineering research and design 9 1 ( 2 0 1 3 ) 296–317

Fig. 17 – Axial distribution of components concentration in the fixed bed reactor for the methanol synthesis.

Knudsen diffusion dominates which is about 10 (Geankoplis, the numerical result of dusty gas model. As dusty gas model
1993). By decreasing the mean pore diameter, the Knudsen takes account Knudsen diffusion, by changing the mean pore
number will increase. In our simulation for methanol synthe- diameter it affects the over all process.
sis process, the Knudsen number is less than 1, always outside
the Knudsen regime. 4.7. Results of heterogeneous fixed bed reactor model
The effective Knudsen diffusion coefficients are related to for the methanol synthesis
mean pore diameter of the pellets, where as the bulk dif-
fusion is independent of mean pore diameter and inversely A 7 m long fixed bed Lurgi reactor is simulated with an inlet
proportional to pressure. So by decreasing the size of mean pressure of 80/,bar. The given reactor design is a typical for the
pore diameter, we can increase the Knudsen number, which methanol synthesis (Manenti et al., 2011; Rezaie et al., 2005).
increases the Knudsen diffusion and it should influence the The axial distribution of the fixed bed temperature is
simulation results. shown in Fig. 16(b) for the methanol synthesis processes. As
Fig. 15(a) and (b) shows result of Wilke model and the overall methanol synthesis is exothermic, the tempera-
Maxwell–Stefan model. In both of these models, there is no ture near the reactor inlet increases. The pressure decreases
Knudsen diffusion. So by changing the mean pore diameter, along the reactor axis, as shown in Fig. 16(a). As a consequence
then it does not affect the simulation results. Fig. 15(c) shows the velocity is increased. The velocity profile along the reactor
chemical engineering research and design 9 1 ( 2 0 1 3 ) 296–317 313

Fig. 18 – Mole fraction profiles of different components for the methanol synthesis process. Comparison between the
Wilke-, Maxwell–Stefan- and dusty gas models at pellet level at a distance of 3 m from the reactor entrance.

axis is shown in Fig. 16(c). Moles are consumed in the chemical models at a distance of 3/,m from the reactor entrance. In
reactions I, II and III of the methanol synthesis and, conse- accordance with theory, the rigorous Maxwell–Stefan and the
quently, decrease in the total concentration along the reactor simpler Wilke models are unaffected by the mean pore diame-
axis. In Fig. 16(d) the correct trend is observed for the total ter of the pellet. On the other hand, the dusty gas model show
concentration profile of the methanol synthesis. significant influence of Knudsen diffusion for the mean pore
The mole fraction profiles of different components for the diameters studied, that is, 20 nm. From Fig. 18, a difference of
methanol synthesis along the reactor axis are given in Fig. 17. 4% is observed between the dusty gas model and bulk diffusion
From reactions I, II and III, it has been found that methanol models.
along with water is produced at the expense of CO, CO2 and Dependent on the mechanical properties of the catalyst
H2 . Hence, the mole fraction profiles of methanol and water pellet (e.g. mean pore diameter) and reactor operating con-
(Fig. 19(a) and (e)) are increased along the reactor axis, whereas ditions (e.g. pressure), the Knudsen diffusion effect can be of
the mole fraction profiles of CO, CO2 and H2 (Fig. 19(b)–(d)) paramount importance to include in the pellet model to pre-
are decreased. Fig. 18 holds the mole fraction profile of dif- dict accurate conversion on the level of the reactor. Fig. 19
ferent components of methanol synthesis on the pellet level shows a small but significant differences in the mole frac-
computed with the Wilke-, Maxwell–Stefan- and dusty gas tion profiles of different components of methanol synthesis
314 chemical engineering research and design 9 1 ( 2 0 1 3 ) 296–317

Fig. 19 – Mole fraction profiles of different components for the methanol synthesis process. Comparison between the
Wilke-, Maxwell–Stefan- and dusty gas models at the reactor level.

along the reactor axis where the diffusion fluxes are described
according to the Wilke-, Maxwell–Stefan- and dusty gas mod-
els.
Simulations are performed on an INTEL i7 2640M CPU,
2.80 GHz and 4 GB of RAM. The total computational time of
the heterogeneous fixed bed reactor model is calculated where
the intra-particle mass diffusion fluxes in the voids of the pel-
let are described by Wilke-, Maxwell–Stefan- and dusty gas
models. From Fig. 20, it has been found that the dusty gas
model is most computationally expensive. It is seen that the
consistent models are about 20 and 60% more costly than
the simplified and not always consistent Wilke model. It is
recommended to use the consistent models instead of Wilke
Fig. 20 – Computational time consumed by reactor model
approach.
when incorporated with different diffusion flux models.
chemical engineering research and design 9 1 ( 2 0 1 3 ) 296–317 315

5. Conclusion P pressure (J/m3 )


Q heat conductivity flux (W/m2 )
In this work, the pellet model consistency has been checked R gas constant (J/kmol K)
with respect to mole- and mass averaged velocities for dif- R reaction rate (kmol/kg(cat) s)
ferent multicomponent diffusion models and validated with r radial coordinate (m)
literature data for methanol synthesis. The commonly used rp radius of pellet (m)
parallel pore model has been compared with random pore Swj source term (kg/m3 s)
model for methanol synthesis. The effect of Knudsen diffu- sxj source term (kmol/m3 s)
sion has been investigated for this process. The impact of ST heat source term (J/m3 s)
pellet model closures on the reactor performance has been ur mole averaged superficial velocity (m/s)
studied. The calculational efficiency of the heterogeneous u mass averaged superficial velocity (m/s)
reactor model where intra-particle mass diffusion fluxes are uz superficial space velocity (m/s)
described by several closures has been calculated. The model u mole averaged velocity (m/s)
evaluations revealed that: T temperature (K)
v mass averaged velocity (m/s)
- The mass- and mole based pellet model are not fully con- w species mass fraction
sistent for methanol synthesis as the model equations are x species mole fraction
not consistent. y species mole fraction
- The simulated results for pellet models show that there z reactor axial position (m)
are small differences between the different mass diffu-
sion models. The simulated results reveal the limitations Greek letters
of the Wilke model formulation, the species composition ˛ correction factor
approximations are not accurate. Hence, the more rigorous ˇ0 friction factor (m2 )
Maxwell–Stefan model seem to be better choice.  porosity
- The random pore model deviates less than 25% than paral-  internal effectiveness factor
lel pore model. However, in the methanol synthesis process dynamic viscosity (Pa s)
parallel pore model is commonly used in pellet model equa-
conductivity (W/mK)
tions. It has been concluded by comparing the experimental ˝ overall effectiveness factor
data that the parallel pore model is the most reasonable one  density (kg/m3 )
for methanol synthesis.  tortuosity
- The pellet model results generally support the conventional  stoichiometric coefficient of component j in reaction
model approximations like uniform temperature across the k
pellet as the temperature variation between the surface and
the center of the pellet is less than 1K, and for Kn < 10 con- Subscript
stant pressure within the pellets. Moreover, the magnitude b bulk
of the diffusion fluxes generally dominate over the convec- g gas mixture
tive fluxes. j species j
- Knudsen diffusion is significant on the level of the pel- p pellet
let, a combined bulk and Knudsen diffusion model should r radial direction
be adopted to describe the intra-particle diffusion fluxes ref reference
because Knudsen diffusion influences small but significant w gas mixture
on the reactor conversion. However, the dusty gas model is cat catalyst
computationally most expensive. Hence, the reactor model
incorporated with Maxwell–Stefan mass diffusion flux clo- Superscripts
sure is the optimal reactor model for methanol synthesis for b bulk
the simulated pore diameter. * dimensionless variable

Nomenclature Acknowledgment

The PhD fellowship (Rout, K.R.) financed by the Research


av surface area of pellet (1/m) Council of Norway through the GASSMAKS program is grate-
cg total concentration (kmol/m3 ) fully appreciated.
Cp heat capacity of gas mixture (J/kg K)
Cp heat capacity of gas mixture (J/kmol K) Appendix A. Dimensionless equations
D diffusion coefficient (m2 /s)
DL axial diffusivity (m2 /s) A.1. Mass based model equations
d0 average pore diameter (m)
dp diameter of pellet (m) Dimensionless variables are defined and introduced into the
H heat of reaction (J/kmol) balance equations, mass diffusion flux equations, and the
h heat transfer coefficient (W/m2 K) boundary conditions to give the non-dimensionalized model
J mole diffusion flux (kmol/m2 s) equations.
J mass diffusion flux (kg/m2 s)
k mass transfer coefficient (m/s) r
r∗ = where r∗ ∈ [0 1] (A.1)
M molecular mass, (kg/kmol) rp
316 chemical engineering research and design 9 1 ( 2 0 1 3 ) 296–317


∗ = (A.2) Boutacoff, D., 1989. Methanol a fuel for the future? Elect. Power
ref Res. Inst. J. 14, 24.
Burghardt, A., Aerts, J., 1988. Pressure changes during diffusion

Jj.r with chemical reaction in a porous pellet. Chem. Eng.
Jj,r = (A.3) Process.: Process Intensification 23, 77–87.
(Dref ref /rp )
De Maerschalck, B., 2003. Space-time least squares spectral
element method for unsteady flows-application and
T
T∗ = (A.4) evaluation for linear and non-linear hyperbolic scaler
Tref
equations. Master Thesis Report. Delft University of
Technology, Department of Aerospace Engineering, The
ur
u∗ = (A.5) Netherlands.
uref Dorao, C.A., Jakobsen, H.A., 2006. Application of the least-squares
method for solving population balance problems in Rd+1 .
Dref Chem. Eng. Sci. 61, 5070–5081.
uref = (A.6)
rp Edwards, M.F., Richardson, J.F., 1968. Gas dispersion in packed
beds. Chem. Eng. Sci. 23, 109–123.
Qr rp Ergun, S., 1952. Fluid flow through packed columns. Chem. Eng.
Q∗ = (A.7) Process. 48, 89.

Tref
Finlayson, B.A., 1972. The Method of Weighted Residuals and
Variational Principles. Mathematics in Science and
P
P∗ = (A.8) Engineering, vol. 87. Academic Press, Inc., New York.
Pb Fogler, H.S., 1999. Elements of Chemical Reaction Engineering,
3rd ed. Prentice Hall, Inc., Upper Saddle River, NJ.
∗ Mw
Mw = (A.9) Froment, G.F., Bischoff, K.B., 1990. Chemical Reactor Analysis and
Mb Design, 2nd ed. John Wiley and Sons, Inc., Hoboken, New
Jersey.
where the dimensionless variables are denoted by a star (*). Geankoplis, C.J., 1993. Transport Processes and Unit Operations,
3rd ed. Prentice-Hall, Inc., New Jersey.
A.2. Mole based model equations Graaf, G.H., Stamhuis, E.J., Beenackers, A.A.C.M., 1988. Kinetics of
the low-pressure methanol synthesis. Chem. Eng. Sci. 43,
Dimensionless variables are defined and introduced into the 3185–3195.
Graaf, G.H., Scholtens, H., Stamhuis, E.J., Beenackers, A.A.C.M.,
balance equations, mole diffusion flux equations, and the
1990. Intra-particle diffusion limitations in low-pressure
boundary conditions to give the non-dimensionalized model methanol synthesis. Chem. Eng. Sci. 45 (4), 773–783.
equations. The following dimensionless variables are defined: Gray Jr., L.C., Alson, J.A., 1989. The case for methanol. Sci. Am. 261
(5), 108.
r
r∗ = where r∗ ∈ [0 1] (A.10) Hartig, F., Keil, F.J., 1993. Large-scale spherical fixed-bed
rp reactors-modeling and optimization. Ind. Eng. Chem. Res. 32
(3), 424–437.
cg Jackson, R., 1997. Transport in porous catalyst. In: Churchill, S.W.
c∗ = (A.11)
cb (Ed.), In: Chemical Engineering Monographs, vol. 4. Elsevier,
Amsterdam.
Jj.r Jakobsen, H.A., Lindborg, H., Handeland, V., 2001. A numerical
J∗j,r = (A.12) study of the interactions between viscous flow, transport and
(Dref cref /rp )
kinetics in fixed bed reactors. Comput. Chem. Eng. 26,
T 333–357.
T∗ = (A.13) Jakobsen, H.A., 2008. Chemical Reactor Modelling: Multiphase
Tref
Reactive Flows, 1st ed. Springer Verlag, Berlin, Heidelberg.
Jiang, K., 1998. The Least-square Finite Element Method: Theory
ur ur
u∗ = or (A.14) and applications in Computational Fluid Dynamics and
uref uref Electromagnetics. Springer, Berlin.
Karniadakis, G., Sherwin, S.J., 1999. Spectral/HP Element Methods
Dref for CFD. Oxford University Press, New York.
uref = (A.15) Krishna, R., Wesselingh, J.A., 1996. The Maxwell–Stefan approach
rp
to mass transfer. Chem. Eng. Sci. 52 (6), 861–911.
Qr rp Lommerts, B.J., Graaf, G.H., Beenackers, A.A.C.M., 2000.
Q∗ = (A.16) Mathematical modeling of internal mass transport limitations

Tref
in methanol synthesis. Chem. Eng. Sci. 55, 5589–5598.
Lovik, I., Hillestad, M., Hertzberg, T., 1998. Long term dynamic
P
P∗ = (A.17) optimization of a catalytic reactor system. Comput. Chem.
Pb Eng. 22, 707–710.
Lurgi Gmbh, 2009. Lurgi MegaMethanol. Available from:
where the dimensionless variables are denoted by a star (∗). www.lurgi.com
Manenti, F., Silvia, C., Restelli, M., 2011. Considerations on the
References steady-state modeling of methanol synthesis fixed bed
reactor. Chem. Eng. Sci. 66, 152–162.
Mason, E.A., Malinauskas, A.P., 1983. Gas Transport in Porous
Aris, R., 1969. Chemical Reactor Analysis. Prentice Hall Engewood
Media: The Dusty Gas Model. Elsevier, New York.
Cliff, New Jersey.
Maxwell, J.C., 1866. On the dynamical theory of gases. Philos.
Bird, R.B., Stewart, W.E., Lightfoot, E.N., 2002. Transport
Trans. R. Soc. London 157, 49–88.
Phenomena, 2nd ed. John Wiley and Sons, Inc., New York.
Mullins, F., Baron, R., 1997. Policies and Measures for Common
Bochev, P., 2001. Finite element methods based on least-squares
Action. Working Paper 9. International GHG emission trading,
and modified variational principles. Technical Report,
OECD, Annex I, Expert Group on UN FCCC.
POSTECH.
chemical engineering research and design 9 1 ( 2 0 1 3 ) 296–317 317

Poling, B.E., Prausnitz, J.M., O’Connel, J.P., 2001. The Properties of Seyfert, W., Luft, G., 1985. Untersuchungen zur Methanolsynthese
Gases and Liquids, 5th ed. McGraw-Hill, New York. im Mitteldruckbereich. Chem. Ing. Tech. 57 (5), 482–483.
Pontaza, J.P., Reddy, J.N., 2004. Spectral-time coupled spectral/HP Shahrokhi, M., Baghmisheh, G.R., 2005. Modeling, simulation and
least squares finite element formulation for the control of a methanol synthesis fixed-bed reactor. Chem. Eng.
incompressible Navier–Stokes equation? J. Comput. Phys. 197 Sci. 60 (15), 4275–4286.
(2), 418–459. Skrzypek, J., Lachowska, M., Grzesik, M., Sloczynski, J., Nowak, P.,
Prachayawarakorn, S., Mann, R., 2007. Effects of pore assembly 1995. Thermodynamics and kinetics of low pressure
architecture on catalyst particle tortuosity and reaction methanol synthesis. Chem. Eng. J. 58, 101–108.
effectiveness. Catal. Today 128, 88–99. Solsvik, J., Jakobsen, H.A., 2011. Modelling of multicomponent
Proot, M.M.J., Gerritsma, M.I., 2002. A least-squares spectral mass diffusion in porous pellets: application to steam
element formulation for stokes problem? J. Sci. Comput. 17 methane reforming and methanol synthesis. Chem. Eng. Sci.
(1-4), 285–296. 66, 1986–2000.
Rezaie, N., Jahanmiri, A., Moghtaderi, B., Rahimpour, M.R., 2005. A Sporleder, F., Dorao, C.A., Jakobsen, H.A., 2010. Simulation of
comparison of homogeneous and heterogeneous dynamic chemical reactors using the least-squares spectral element
models for industrial methanol reactors in the presence of method. Chem. Eng. Sci. 65, 5146–5159.
catalyst deactivation. Chem. Eng. Process. 44, 911–921. Stefan, J., 1871. Uber das Gleichgewicht und die Bewegung
Rout, K.R., Solsvik, J., Nayak, A.K., Jakobsen, H.A., 2011. A insbesondere die Diffusion von Gasgemengen,
numerical study of multicomponent mass diffusion and Sitzungsberichte der Kaiserlichen Akademie der
convection in porous pellets for the SE-SMR and desorption Wissenschaften Wien. 2te Abteilung a 63, 63–124.
processes. Chem. Eng. Sci. 66, 4111–4126. Vanden Bussche, K.M., Froment, G.F., 1996. A steady-state kinetic
Rusten, H.K., Ochoa-Ferdez, E., Lindborg, H., Chen, D., Jakobsen, model for methanol synthesis and the water gas shift
H.A., 2007. Hydrogen production by sorption-enhanced steam reaction on a commercial Cu/ZnO/Al2 O3 Catalyst. J. Catal. 161,
methane reforming using lithium oxides as CO2 -acceptor. Ind. 1–10.
Eng. Chem. Res. 46, 8729–8737. Wakao, N., Funazkri, T., 1978. Effect of fluid dispersion
Salmi, T., Warna, J., 1991. Modelling of catalytic packed bed coefficients on particle-to-fluid mass transfer coeffcients in
reactors-comparison of different diffusion models. Comput. packed beds. Chem. Eng. Sci. 33, 1375–1384.
Chem. Eng. 15, 715–727. Yusup, S., Anh, N.P., Zabiri, H., 2010. A simulation study of an
Satterfield, C.N., 1980. Heterogeneous Catalysis in Practise. industrial methanol reactor based on simplified steady-state
McGraw-Hill, New York. model. Int. J. Res. Rev. Appl. Sci. 5 (3), 213–222.

You might also like