PACKED BED REACTOR SIMULATION FOR ACETALDEHYDE PRODUCTION
BY ORTHOGONAL COLLOCATION
Oscar Sánchez, Hugo Galindo, Ivan Gil, Gustavo Orozco
Departamento de Ingeniería Química, Universidad Nacional de Colombia
e-mail: ofsanchezm@unal.edu.co
Keywords: Orthogonal Collocation, Fixed Bed Reactor, Acetaldehyde and Partial Ethanol Oxidation
Introduction. Acetaldehyde can be produced in
the industry through petrochemical process or
alcohol-chemical process. Petrochemical process
has relatively low conversions (less than 72%)
and yields (less than 95%) and silver is usually
employed as catalyst besides being a very
expensive metal, is hard to recover it downstream
(1). The alcohol-chemical way is a low price
process and is viable in agricultural countries, now
that they dispose of raw materials like sugar cane
and rice for fermentative process in ethanol
production, that is forward used in vapor phase
partial oxidative process to produce acetaldehyde
(1).
Studies reveal the best operating conditions for
acetaldehyde production from partial ethanol
oxidation as being in the temperature range of
350-420 K at atmospheric pressure, an air/ethanol
ratio between 3 and 22 and over a conventional
Fe/Mo catalyst, under these conditions only
acetaldehyde is detected by gas chromatography
(1, 2). Out of these process conditions (specially
temperature) undesired products are gotten like
acetic acid or carbon dioxide, for this reason the
process viability depends on the reactor
temperature control. The reaction rate for the
ethanol transformation to acetaldehyde has been
studied under the Dante and Temkin
mechanisms, being the Temkin’s one which gives
the best correlation when high conversions are
considered (1).
Filho and Domingues proposed a baffled
multitubular reactor but windowless to avoid
parallel flow regions between the coolant flow and
the reaction tubes, it guaranties the temperature
control (1). These reactor disposals increase its
design and operation, now that each baffled
region should be fed with different coolant
conditions to ensure the reactor temperature
control.
The reactor simulation allows you to evaluate the
axial changes of concentration and temperature
when the process conditions change. The reactor
model numerical solution is a critical aspect due to
mistakes propagation and the inherent instability
of the numerical methods used in the solution of
these kinds of problems (3, 4). The most used
methods in the solution of diffusion-reaction
problems are perturbation method and methods
that used weighted residuals like orthogonal
collocation (OC) and orthogonal collocation on
finite elements (OCFE) that used orthogonal
polynomials like Legendre or Hermite polynomials
(3, 4).
The aim of this work is to get the concentration
and temperature profiles for the gas phase
reaction between ethanol and air in a packed bed
reactor with Fe/Mo catalysts with sphere shape
and the coolant temperature profile by one-
dimensional simulation under pseudo-
homogeneous and heterogeneous non-isothermal
models using the OC and OCFE methods to solve
the model. A parametric sensitivity was made with
the main process variables.
Methods. The pseudo-homogeneous and
heterogeneous non-isothermic models differ each
other in the consideration of the concentration and
temperature interface gradients between the bulk
and catalyst pellet surface (4, 5).
The axial dispersion in the model is evidenced by
the second order of the differential equations for
the mass and energy balance, the problem is
reduced to solve differential equations with
boundary value that can be of Newman, Dirichlet
or Robin type.
Besides the differential balances over the bulk is
necessary the coolant energy balance to evaluate
its axial temperature changes due to the heat
transfer between the reaction bulk and coolant
flow. The coolant physic-chemical properties
belong to the Dow Therm Teresso 77.
The differential mass and energy balance over the
catalyst pellet complete the reactor description,
they involve the effective transport properties.
This is a boundary condition problem too, where
available information is in the middle and the
surface of the pellet. The last one is given by the
reactor model.
The Temkin’s reaction rate mechanism for the
global reaction of the ethanol partial oxidation is:
CH 3 CH 2 OH + 1/2O 2 ⇒ CH 3 CHO + H 2 O
1. CH 3 CH 2 OH + ZO 2 ⎯⎯→
K1
CH 3 CHO + H 2 O + ZO
2. O 2 + Z ⎯⎯→
⎯ ZO 2
K2
3. CH 3 CHO + Z ←⎯
⎯→ ZCH 3 CHO
K3
4. ZCH 3 CHO + H 2 O ⎯⎯→
⎯ CH 3 CH 2 OH + ZO
K4
Which analyzed under the hypothesis that 1, 2 conditions of the mass (6) and energy (7)
and 4 are slow and irreversible and step 3 is in balances for the catalytic pellet.
equilibrium and oxygen is molecularly absorbed 1 d ⎡ 2 dC A ⎤ (− γ AW )ρ B
without dissociation in only one active site, the ⎢r dr ⎥ = (6)
r 2 dr ⎣ ⎦ D eff
reaction rate equation is (1):
d ⎡ 2 dT ⎤ (− ∆HRA )(− γ AW )ρ B r 2
r
⎢ dr ⎥ = − (7)
(γ) =
2K 1K 2PO2PEt
(1) dr ⎣ ⎦ K eff
K1PEt + 2K 2PO2 + K 3K 4PAcPH2O + K 3K1PEtPAc
The mass and energy boundary conditions for the
The application of the conservative laws over the pseudo-homogeneous model are:
reactor generates the mass (2), energy (3) and
momentum (4) bulk balances and the energy dC A
r =0→ =0
balance for the coolant (5). For the momentum dr
balance was used the Ergun´s equation (4). dT
=0
dr
d2C A dC A r = R → C A = C Ao
(D x ) − (υ x ) = ρ B η( − γ AW ) (2)
dx dx T = To
d2T dT 4
Kx 2 − (ρυxCp ) = −ρBη(−γAW)(−∆HRA) + U(T − T' ) (3)
dx dz dT while the mass and energy boundary conditions
for the heterogeneous model are:
dP ρυ 2
= −f x
(4)
dx dp dC A
r =0→ =0
(± ) dT' = Uπd T&N T (T − T' ) dr
dT
dx m m C pm =0
dr
Co - current ( + ) , ∨ , Counter − current (-)
dC A
(5) r = R → −D eff = K g a m (C As −C Ao )
dr
The mass and energy boundary conditions for the dT
− K eff = ha m (Ts − To )
bulk are: dr

⎛ D ⎞ dC A The simultaneous solution of the mass (6) and

x = 0 → C A = C A R + ⎜⎜ x ⎟⎟ energy (7) balances over the catalytic pellet under
⎝ υ x ⎠ dx x =0 the conditions of each model and for every axial
⎛ Kx ⎞ dT bulk concentration and temperature point through
T = TAR + ⎜ ⎟ the reactor allows you to calculate the
⎜ ρυ x C p ⎟ dx
⎝ ⎠ x =0
effectiveness factor (η) that is defined as:
dC A
x =L → =0 1
φ2 ∫0 R1(c(r ), T(r ))r a −1dr
dz η= 1
dT
=0 φ2 ∫0 R1(c(1), T(1))r a −1dr
dx
The OC method was employed to solve the pellet
While the coolant energy balance and the bulk differential balances for the pseudo-homogeneous
momentum balance have initial conditions, the model and the OCFE method was employed to
coolant condition can vary according to its flow solve the pellet differential balances for the
course. heterogeneous model and reactor balance
equations. These methods reduce the problem to
x = 0 → T' = Tm (co - current) a system of non-lineal algebraic equations (that
P = PAR was solved by the Newton-Raphson method);
each algebraic equation is function of the
x = L → T' = Tm (counte - current)
concentrations and temperatures at the
collocation points that are unknown variables of
The mass and energy transfer from the bulk to the the system.
catalytic pellet balances over the catalytic pellet Both methods OC and OCFE require
can or not consider interface gradients between normalization of the differential equations system
them. This consideration modifies the boundary due to the definition range of the Legendre
orthogonal polynomials (OP) between 0 and 1.
These OP were employed in the expansion of the d2τ dτ
assay function. For this reason, it is settled down − Peh = −Φ2βRηf (ζ , τ) + Nu(τ − ψ) (11)
dy2 dy
dimensionless variables to normalize the balance
equations of the reactor and catalytic pellet. dϕ Lf ρυ2x
=− (12)
dy PAR dp
For the pellet
dψ
r T C (−γ AW ) = Bi w (τ − ψ ) (13)
z= ; θ= ; u = A ; f(u, θ) = dy
R To CAo (−γ AW )o
The normalized boundary conditions for the
For the reactor reactor balance equations are:
x T CA
y= ; τ= ; ζ=
L TAR C AR 1 dζ
y = 0 ζ = 1+
Tm P (−γ AW ) Pe m dy
ψ= ; ϕ= ; f(ζ , τ) = 1 dτ
TAR PAR (−γ AW ) AR τ = 1+
Pe h dy
The normalized mass (8) and energy (9) balance ϕ =1
equations of the catalytic pellet generate the Tm
ψ = (co - current coolant flow)
dimensionless groups of Thiele (φ) and Prater (β). T AR
dζ
1 d ⎡ 2 du ⎤ y =1 =0
z = φ2 f (u, θ) (8) dy
z 2 dz ⎢⎣ dz ⎥⎦ dτ
=0
1 d ⎡ 2 dθ ⎤ (9) dy
⎢ z dz ⎥ = − φ β f ( u, θ )
2

z 2 dz ⎣ ⎦ T
ψ= m (counter - current coolant flow)
TAR
The normalized mass and energy boundary
conditions for the pseudo-homogeneous model
are:
The simultaneous solution of the mass and
energy balance equations of the catalytic pellet for
du
z = 0 → = 0 every of the studied models allow you to have the
dz temperature in function of the concentration.
dθ
= 0
dz θ j = 1 + β(1 − u j ) (pseudo - homogeneous)
z = 1→ u = 1
Bi m
θ=1 θ j = 1 + β(u N+ 2 − u j ) + β (1 − u N+ 2 ) (heterogeneous)
Bi h
while the normalized mass and energy boundary
conditions for the heterogeneous model are: Then the simultaneous solution of this and the
pellet mass balance equation under the
du
respectively boundary conditions let calculate the
z = 0 → = 0 uj for each collocation point.
dz
The OC method was used to evaluate the mass
dθ
= 0 balance residuals for the pseudo-homogeneous
dz model (16), that is settled from 2 to N+1 (now that
dθ 1 and N+2 are boundary points that are held to
z = 1→ − = Bi h ( θ (1) − 1)
dz their own conditions), in this case was used a fifth
du order Legendre orthogonal polynomial. The OCFE
− = Bi m ( u (1) − 1)
dz was used to evaluate the mass balance residuals
for the heterogeneous model (17), in this case
The normalized mass (10), energy (11) and was used a fifth order Legendre orthogonal
momentum (12) bulk balance equations and the polynomial and the domain was divided into two
coolant energy balance equation (13) in the elements (hk = 0.5).
reactor generate additionally the dimensionless
N+ 2
groups of Peclet for mass and energy and Nusselt
and involve the effectiveness factor. ∑ ℵ ji u i − φ 2 f (u j , θ j ) = 0 where j = 2, 3,..., N + 1 (16)
i =1

d2ζ dζ
2
− Pe m = Φ 2 η f (ζ , τ ) (10)
dy dy
Boundary conditions composition, and were calculated at each
N+ 2 collocation point (data not shown).
z=0 → ∑ ℜ1,iui = 0
i =1

z = 1 → uN + 2 = 1
Results and Discussion. The concentration and
temperature profiles of each model (pseudo-
NP NP
homogeneous and heterogeneous) for the
1 2
∑ ℑ ji u i + ∑ ℑ ji u i − φ 2 f (u j , θ j ) = 0
(17)
catalytic pellet showed mean statistics differences
h k2 i=1 z ih k i =1 lower than 1% between them (data not shown). In
where j = 2, 3,..., NP - 1 the Fig. 1 are shown the dimensionless
Boundary conditions concentration profiles and can be noticed that the
1 NP
difference between the models are at higher
z =0 → ∑ ℑ 1,i u i = 0 temperatures, this is due to the fact that near the
h k i =1
420 K is the higher yield and the transport
1 NP
z =1 → ∑ ℑ NP ,i u i + Bi m ( u NP − 1) = 0
h k i =1
properties (mainly diffusion) are sensible to little
changes of composition at higher temperatures.
Pseudo-homogeneous
The simultaneous solution of the non-lineal
Heterogeneous
algebraic equations gotten from the pseudo- 1
homogeneous or heterogeneous model gives the T = 350K
effectiveness factor. 0,8

N+ 2 0,6
3
η= ∑ ℜN+2,iui (pseudo − homogeneous)
u

φ2 i =1 0,4
NP β = 0,0015
3
η= 2
φ
∑ ANP,iui (heterogeneous) 0,2
i =1
T = 420
0
The OCFE was used to solve the mass (18), 0 0,2 0,4 0,6 0,8 1
z
energy (19), momentum (20) reactor balance
equations and the coolant energy balance (21), Fig. 1 Dimensionless concentration profiles in the
for this case was used a tenth order Legendre catalytic pellet for the pseudo-homogeneous and
polynomial and the domain was divided in four heterogeneous models at different bulk temperatures
elements (hk = 0.25) and a Prater number of 0.0015.

1 NP Pe m NP The effectiveness factor was gotten for different

∑ BI J ζ J − ∑ A I J ζ J = Φ 2 η f ( ζ I , τI ) (18) catalytic shapes (slab, cylinder and sphere), Fig 2.
hk2 J=1 hk J=1 shows a normal behavior at different Thiele
1 NP
Peh NP module values, the spherical shape has the higher
hk2
∑ BI JτJ − hk
∑ AI JτJ = −Φ2βRηf (ζI , τI ) + χ effectiveness factor. The comparison of the
J =1 J =1
(19) effectiveness factor for the spherical shape under
⎡ T ⎤ the pseudo-homogeneous and heterogeneous
χ = Nu⎢τI − m ⎥
⎣ TAR ⎦ models evidence that the effectiveness factor is
1 NP Lf ρυ greater for the heterogeneous model in the way
∑ dJ (J − 1)uJ−2 = − P d x
hk J=1
(20) the Thiele module increases, but this difference is
AR p less at low Prater numbers Fig. 3.
1 NP
∑ A I Jψ J = Bi W ( τI − ψ I ) (21) Slab Cylinder Esphere
hk J=1 1,1

con i = 2, 3,., NP − 1 1
Effectiveness Factor

Boundary values 0,9

1 NP
1 1 NP , ∑ A NP , J ζ J = 0 0,8
ζ1 = 1 + ∑ 1, J J
hk Pe m J=1
A ζ hk J=1 0,7

NP NP 0,6
1 1 1
τ1 = 1 +
hk Pe h
∑ A 1, J τ J , hk
∑ A NP , J τ J = 0 0,5
J=1 J =1
0,4
Tm
ϕ u=0 = 1 , ψ = 0,3
TAR 0,01 0,1 1 10
Thiele

All the physic-chemical properties and transport Fig. 2 Effectiveness factor for different pellet geometry
shapes.
properties were correlated to temperature and
 Pseudo-homogeneous Heterogeneous Pseudo-homogeneous Heterogeneous
1,1 1,12

1 1,1
β = 0,0015

Effectiveness Factor
0,9 1,08

0,8 1,06

ϕ
1,04
0,7
β = 0,004
1,02
0,6
1
0,5 (c)
0,98
0,4 0 0,2 0,4 0,6 0,8 1
0,01 0,1 1 10 y
Thiele
Pseudo-homogeneous Heterogeneous
Fig. 3 Effectiveness factor for a spherical pellet under 0,98
the pseudo-homogeneous and heterogeneous models.
0,97
The bulk concentration, temperature and
0,96
momentum dimensionless profiles like the coolant
temperature profile show a mean statistic 0,95
difference lower than 1.5% between the pseudo- ψ
0,94
homogeneous and heterogeneous model. In Fig.
4. can be noticed that the higher difference can be 0,93
found in the concentration and pressure profiles,
this difference increase at higher difference, as is 0,92
(d)
expected due to the behavior of the effectiveness 0,91
factor with the temperature. 0 0,2 0,4 0,6 0,8 1
Pseudo-homogeneous Heterogeneous
y
1,004 Fig. 4 Bulk dimensionless profiles, temperature (a),
1,002 (a)
concentration (b), momentum (c) and coolant
temperature dimensionless profile (d). The operation
1
conditions were: coolant at 350 K and 0.1 Kg s-1 at
0,998 co-current flow and a feed air: ethanol relation of 20 and
0,996 380K.
0,994
τ

0,992
The reaction temperature was control in a better
way by the counter-current disposal, changes in
0,99
the coolant mass flow and inlet temperature have
0,988
a strong effect over the bulk temperature and
0,986 yield. In Fig. 5 is shown the effect of the fluid flow
0,984 disposal over the bulk and coolant temperature
0 0,2 0,4 0,6 0,8 1
y profile, can be noticed that the counter-current
arrangement is the most effective for the heat
removal, in Fig. 6 is shown the effect of the
Pseudo-homogeneous Heterogneous
coolant inlet temperature over the bulk
1,02
temperature and concentration profiles, as is
(b)
1 noted a coolant inlet temperature under or near
the lower optimal process temperature (350K) turn
0,98 down the reaction, the best coolant inlet
0,96
temperature is in the range of 370 and 390K (data
not shown) and in Fig. 7 is shown the effect of
ζ

0,94 coolant mass flow over the temperature and

concentration bulk profiles and the coolant
0,92
temperature profile, is noticed that at higher mass
0,9 flow the bulk temperature and yield are lower at a
coolant inlet temperature 350K, the optimal
0,88 coolant mass flow is in the range of 0.05 and 0.1
0 0,2 0,4
y
0,6 0,8 1
kg s-1(data not shown) this results can vary
according to the physic-chemistry properties of
the Dow Therm or according to the coolant
system.
The calculated data showed a mean deviation in
Coolant 350 Coolant 380
the range from 10 to 13% to the experimental data
1,02
for the different profiles (data not shown), the (b)
1
higher deviation was in the ethanol yield. These
could be due to the lack of adjust of the physic- 0,98
chemical and transport correlations to the real 0,96
reactor behavior and the numerical error gotten by
0,94
elements of very low order (10-4 and 10-6) in the

ζ
matrixes of the OC and OCFE numerical 0,92

methods. 0,9

0,88
Co-current Counter-current
1,004 0,86

(a) 0,84
0 0,2 0,4 0,6 0,8 1
y
0,999

Fig. 6 Coolant inlet temperature effect over the bulk

0,994
τ

temperature (a) and concentration (b) profiles. The

operation conditions were: coolant at 350 K and 0.08
(—), 0.1 (− −) or 0.25 (− — −) Kg s-1 and a feed air:
0,989
ethanol relation of 20 and 380K.
Bulk Coolant
0,984 1,12 1,04
0 0,2 0,4 0,6 0,8 1 (a)
y 1,1 1,02

1,08 1
Co-current Counter-current
0,99 1,06 0,98
(b)

ψ
τ

0,98 1,04 0,96

0,97 1,02 0,94

0,96 1 0,92
ψ

0,95 0,98 0,9

0 0,2 0,4 0,6 0,8 1
0,94 y
0,93

0,92 1
0 0,2 0,4 0,6 0,8 1
y (b)
0,97
Fig. 5 Dimensionless temperature profiles for the bulk
(a) and for the coolant (b) at co-current and counter- 0,94
current. The operation conditions were: coolant at 350
K and 0.3 Kg s-1 and a feed air: ethanol relation of 20 0,91
ζ

and 380 K.
0,88
Coolant 350K Coolant 380K

1,08 0,85

1,06
(a)
0,82
1,04 0 0,2 0,4 0,6 0,8 1
y
1,02
Fig. 7 Coolant mass flow effect over the bulk and
τ

1 coolant temperature profiles (a) and bulk concentration

profile (b). The operation conditions were: coolant at
0,98 350 K and 0.08 (—), 0.1 (( () or 0.25 (( — () Kg s-1 and
a feed air: ethanol relation of 20 and 380K.
0,96

0,94
0 0,2 0,4 0,6 0,8 1
y
Conclusions. The OC and OCFE numerical ∆HRA Reaction heat
methods allow a fast and easy convergence to the R, r Pellet radius and radial position
solution of the catalytic pellet and reactor models. T Temperature
The OCFE has the advantage that the domain u, ζ Dimensionless concentration
division can help to avoid convergence problems z, y Dimensionless radial and axial
when you have to use OC with a high order position
orthogonal polynomial (tenth order or higher) θ , τ, ψ Dimensionless temperature
because the matrixes gotten have elements of f(u, θ) Dimensionless kinetic
very low order (10-8 and lower) that cause ρB Pellet density
instability in the program, based in the continuity φ, Φ Thiele module
criteria we used OCFE for the reactor modeling.
β Prater number
The models showed the presence of hot spots in
hk Domain division
the ethanol catalytic conversion to acetaldehyde
TAR Coolant inlet temperature
and thermal instability like run-away in the reactor.
PAR Initial pressure
The counter-current operation showed to be the ℑ, ℜ,ℵ Matrix notation for OC
most efficient system for the temperature control.
The critical parameters to be control in the reactor A, B Matrix notation for OCFE
are mainly the inlet temperature and the mass
flow of the coolant. Bibliography.
We recommend to make more experimental runs
1. Filho, M. R. and Domingues, A., (1992). A
to verify the prediction power of the program and Multitubular Reactor for Obtention of Acetaldehide
to extend the numerical method to a two- by oxidation of Ethyl Alcohol. Chem. Eng. Sci., vol
dimensional simulation to verify if there is a higher (47): 2571 - 2576.
accuracy. And develop optimization studies to get 2. Dente, M., Poppi, R. and Pasquon, I., (1964).
the better acetaldehyde production conditions. Cinetica dell´ossidazione del metanolo a
formaldeide con catalizzatore a base di ossidi di Fe
Notation. (I.S.U) e Mo. La Chimica e lÌndustria., vol (46) 1226 -1236.
3. Finlayson, B. A. (1980). Non Linear Analysis in
Bim, Bih or w Biot of mass and heat Chemical Engineering. McGraw Hill United States
of America.
Pem, Peh Peclet of mass and heat 4. Finlayson, B. A. (1971). Packed Bed Reactor
CA Ethanol concentration Analysis by Ortogonal Collocation. Chem. Eng. Sci.
Deff, Keff Effective Diffusivity and thermal , vol (26): 1081 - 1091.
conductivity
(-γAW) Kinetic expression
Kg , h Mass and heat transfer coefficients