The mechanism of reduction of iron oxide by hydrogen

Article abstract of DOI:10.1016/S0040-6031(02)00478-1

Precipitated iron oxide samples were characterized using temperature-programmed reduction. H₂ was used as the reduction agents. The two-stage reduction was observed: Fe₂O₃ was reduced to Fe₃O₄ and the

Full text of DOI:10.1016/S0040-6031(02)00478-1

Thermochimica Acta 400 (2003) 61–67
The mechanism of reduction of iron oxide by hydrogen
a
a,∗
b
Hsin-Yu Lin , Yu-Wen Chen , Chiuping Li
a
Department of Chemical Engineering, National Central University, Chung-Li 32054, Taiwan
b
Division of Fine Chemicals and Solvents, Chinese Petroleum Cooperation, Chia-Yi 60036, Taiwan
Received 21 June 2002; received in revised form 9 September 2002; accepted 10 September 2002
Abstract
Precipitated iron oxide samples were characterized using temperature-programmed reduction. H2 was used as the reduction
agents. The two-stage reduction was observed: Fe2O3 was reduced to Fe3O4 and then reduced to metallic Fe. The activation
−1
energy for the two reduction steps of iron oxide are 89.13 and 70.412 (kJ mol ), respectively. The simulation by reduction
models of the TPR patterns presents well fitting of unimolecular model for Fe2O3 → Fe3O4 reduction and two-dimensional
nucleation according to Avarmi–Erofeev model for Fe3O4 → Fe.
©
2002 Elsevier Science B.V. All rights reserved.
Keywords: Temperature-programmed reduction; Iron oxide; Kinetic mechanism
1
. Introduction
Unmuth et al. [12] studied the reduction using TPR
technique. They prepared the catalysts by loading
metal on silica gel, and calcined in air at 200 C. They
◦
Temperature-programmed reduction (TPR) method
has been widely used in the characterizations of solid
materials [1–10]. Kissinger [11] developed a method
to demonstrate the effect of varying order of reac-
tion from differential thermal analysis (DTA) patterns.
Wimmers et al. [5] presented a convenient method ex-
tended from Kissinger’s approach for the calculation
of TPR patterns by using kinetic expressions.
Many works have been done on the reducibility
of bulk iron oxides by TPR. Brown et al. [8] stud-
ied the reducibility of bulk iron oxide materials and
showed that the reducibility of Fe2O3, as measured by
TPR, differed markedly from that of Fe3O4. The first
step in the reduction of Fe2O3 is reduction to Fe3O4,
which was confirmed by the TPR profiles, with the
low-temperature peak in the profile of Fe3O4 reduc-
tion corresponding to this reaction.
found that the reduction profile for 5 wt.% Fe/SiO2
consisted of two peaks, at 307 and 447 C, which
correspond to the following process:
◦
Fe2O3 → Fe3O4 → Fe
TPR data for ␣-Fe2O3 and Au/␣-Fe2O3 systems
was reported by Munteanu et al. [13], they also
found that the TPR profiled for ␣-Fe2O3 consisted
◦
of two peaks, at 280 and 427 C. It must be empha-
sized that the literature data diverge to a large extent
[6–13], since different oxides exist (Fe2O3, Fe3O4
and FeO) and, moreover, these can contain impurities
or dopants. Furthermore, there are large differences in
the literature with respect to, for instance, the selection
of reduction temperature, the particle/crystallite size,
and the reducing agent. The present study gives TPR
results of different heating rate, and shows the good
agreement of the TPR simulation results compared
∗
Corresponding author. Fax: +886-3-425-2296.
E-mail address: ywchen@cc.ncu.edu.tw (Y.-W. Chen).
0
040-6031/03/$ – see front matter © 2002 Elsevier Science B.V. All rights reserved.
PII: S0040-6031(02)00478-1

6
2
H.-Y. Lin et al. / Thermochimica Acta 400 (2003) 61–67
with experiment data. The objective of this investiga-
tion was to gain a more systematic understanding of
the reducibility of iron oxide.
3. Results
3.1. Theory
Consider a reaction:
2
. Experimental
gas + solid → product
The reaction can be described by an equation:
2
.1. Materials
−
[gas]
Precipitated sample was prepared using Fe(NO3)2·
n
rate =
= k(T)[gas]
(1)
9
H2O from Fisher Chemical Co. The iron oxide was
dt
prepared by a precipitation method. An aqueous solu-
tion containing Fe(NO3)3 and a second solution con-
taining aqueous NH3 (∼2.7 M) were maintained in
where [gas] is gas concentration, n the order of reac-
tion and, k the rate constant given by the Arrhenius
equation, where T the Kelvin temperature, R the gas
constant and E is the activation energy.
E/RT
◦
stirred glass vessels at 83 C. The two solutions were
separately conveyed by fluid pumps to a stirred tubu-
k(T) = A e−
(2)
◦
lar reaction vessel that was maintained at 82 ± 1 C.
Precipitation (to form FeOOH/ Fe2O3) occurred as the
two solutions were pumped upward through the vessel,
while an in-line pH electrode was used to monitor the
pH of the reactor effluent. The flow rate of NH3 solu-
tion was normally fixed at 60 ml/min, while that of the
Let α be the fraction reacted of solid reactant, and n
be the reaction order. A simplified mathematical form
of the reaction kinetics can be described as
n
f(α) = (1 − α)
3+
Fe solution was adjusted (typically to 90 ml/min) to
give a precipitation pH value of 6.0 ± 0.2. Collection
of the precipitate was made in ice-cooled vessels and
was continued until one of the two solutions was con-
sumed. The precipitate was then thoroughly washed
And the reaction rate can be written as
dα
=
k(T)f(α)
(3)
dt
Integration (3), yields
−
by vacuum filtration to remove excess NH3 and NO3 ,
ꢀ
α
dα
using 10 l of deionized, distilled water per 100 g (dry
weight) of final catalyst. The washed precipitate was
=
g(α) = k(T)t
0 f(α)
◦
dried in a vacuum oven for 60 h at 50 C, to remove
In TPR process, temperature is also a function of
time, thus:
most of the excess water, and then for an additional
◦
2
4 h at 120 C. The product was dried further in a vac-
◦
dT
uum oven for 16 h at 120 C.
ψ = dt
(4)
2
.2. Temperature-programmed reduction
where ψ is the constant heating rate in the TPR ex-
periment.
Temperature-programmed reduction studies were
Thus, g(α) indicates the function related only on
fraction conversion α, and the temperature T. The
concepts are found suitable for obtaining kinetic pro-
cesses under different conditions. However, this form
of f(α) can not describe kinetics of nucleation or dif-
fusion process. Four types of f(α) [5,14] are given in
Table 1 which are some gas–solid reaction models
based on kinetic studies: the Avarmi–Erofeev model
is concerned with the nucleation process from the
statistical probability treatment [15,16]; the unimolec-
ular model is expected to be a first order reaction,
performed using 5% H2/N2. The consumption of H2
was measured by the change in thermal conductivity
of the effluent gas stream, and a dry ice/acetone bath
was used to remove water formed during the hydro-
gen reduction. Catalyst sample weights of 10–15 mg
and reductant flow rates of 12 ml/min were used
for all the experiments. A temperature ramp of 3,
◦
◦
7
and 21 C/min from room temperature to 900 C
was used for all temperature-programmed reduction
experiments.

H.-Y. Lin et al. / Thermochimica Acta 400 (2003) 61–67
63
Table 1
The f(α) and g(α) function of different reduction models
Reduction model
f(α)
g(α)c
2
/3
1/3
1/2
Three-dimensional nucleation according to Avarmi–Erofeev
Two-dimensional nucleation according to Avarmi–Erofeev
Unimolecular decay
(1 − α)(− ln(1 − α))
(− 3 ln(1 − α))
(− 2 ln(1 − α))
− ln(1 − α)
1
/2
(1 − α)(− 2 ln(1 − α))
1 − α
a,b
3/2(1 − α) ((1 − α)−1/3− 1)−1
This model is geometrically defined as shrinking/unreacted core or contraction sphere models, with reaction proceeding topochemically.
1/3
1/3 2
Three-dimensional diffusion according to Jander
(1 − (1 − α)
)
a
b
c
Gas diffusion through the product layer as the rate-determining step.
ꢁ
α
dα
f(α)
g(α) =
.
0
and the three-dimensional diffusion model [17] is
according to the Jander equation which assuming
the reaction is proceeding equally in all faces of the
particles, and the reaction rate is diminishing as a
consequence of increasing thickness of the barrier
layer.
And for computing purposes, P(x) can be ap-
proaches by the following simplification:


−x
2
3
e
674.567 + 57.421x − 6.055x − x 



P(x) =
2 
x
1699.066 + 841.655x + 49.313x
3
4
−
8.02x − x
Combination of Eqs. (2)–(4) leads to:
The integration approaching equation can be used
for 9 ≤ x ≤ 174 [18]. Thus functions of g(α(T)) and
f(α(T)) are given, and combine Eqs. (5) and (6) one
can get the simulation TPR pattern:
dα
dT
1
A
−
E/RT
=
k(T)f(α) = ψ
e
f(α)
(5)
ψ
where
dα
dT
A
= ψ
A
−
E/RT
−E/RT
e f(α(T)
k(T) = Ae−
E/RT
e
f(α) = ψ
Thus TPR patterns dα/dT versus T can be calculated
where A and E were calculated by experimental data.
by integrating Eq. (5):
Since the maximum reaction rate occurs TPR peak,
ꢂ
ꢃ
ꢄꢅ
ꢀ
ꢀ
d
dα
α
T
dα
A
= ψ
= 0
−
E/RT
e
dT = g(α)
dT dT
T=Tmax
f(α)
0
T0
Then,
Combining (4), (5), the integration can be solved
by partial integration method [18,19]:
ꢂ
ꢅ
d
A
ψ
−E/RT
f(α)
e
dT
T=Tmax
g(α) =
AE P(x)
Rψ
(6)
E
A
ψ
A
ψ
−
E/RTmax
−E/RTmax
e
= RT2
e
f(α) +
max
ꢃ
ꢄ
ꢃ
ꢄ
where
df(α)
dα
dα
dT
×
= 0
E
x =
T=Tmax
T
max
RT
From Eq. (5), f(α) = (dα/dT)(ψ/A)(eE/RT)Thus
the equation becomes:
Doyle [19] has tabulated the most commonly found
values of P(x):
ꢃ
ꢄ
dα
dT
ꢀ
∞
−u
−x
e
e
T=Tmax
P(x) =
=
− Ei(−x)
ꢆ
ꢇ
ꢃ
ꢄ
2
u
x
x
E
A
+ ψ
df(α)
dα
−
E/RTmax
×
e
= 0
7)
ꢀ
2
∞
RT
max
T=Tmax
−
u
Ei(−x) = −
e
(
x

6
4
H.-Y. Lin et al. / Thermochimica Acta 400 (2003) 61–67
◦
Since (dα/dT)T=Tmax
ꢀ
∼50 and ∼100 C for the first and the second step of
ꢆ
ꢇ
ꢃ
ꢄ
iron reduction, respectively. The peaks at about 570 K
in all profiles of Fig. 1 are due to the reduction of
Fe2O3 to Fe3O4:
E
A
+ ψ
−
E/RTmax
e
2
RT
max
Thus A can be calculated by:
3Fe2O3 + H2 → 2Fe3O4 + H2O
The peaks whose maxima were located about 770
K were due to the second reduction step from Fe3O4
to the metallic iron:
(11)
−
E
ψ eE/RTmax
A =
×
(9)
2
RT
(df(α)/dα)T=Tmax
max
From Eq. (8) one has:
ꢃ
ꢄ
ꢃ
ꢄ
Fe3O4 + 4H2 → 3Fe + 4H2O
(12)
ψ
−E
E
ln
=
− ln
+ C
(10)
2
According to Eq. (12), the ratio of the first peak area
to second in the TPR profiles should be 1:8, provide
that the reduction has been completed. Computing the
peak area in Fig. 1, an average value of 11% of hy-
drogen was consumed in the first reduction step, in
good agreement with the theoretical prediction. How-
ever, the peak area of the second reduction step is only
T
RTmax
AR
max
where (df(α)/dα)T=T
is assumed as a constant C
since reduction mechanism at the TPR peak did not
change with heating rate.
max
Plotting ln(ψ/T2 ) versus T2 leads to a tem-
max
max
perature-programmed Arrhenius plot, in which the
slope is equal to −E/R.
66%, lower than the theoretical value of 89%. This
indicates that the second reduction step is incomplete.
Fig. 2 is the Arrhenius plot of the reaction which is
plotted by Eq. (10), and the activation energies of both
reduction steps can be calculated form the slop. The
E values of 89.13 and 70.412 (kJ mol ) were deter-
mined for the two steps, respectively.
3
.2. Results
TPR of iron catalysts were carried out using H2/N2
mixtures with a flow rate of 60 ml/min. Fig. 1 contains
a series of TPR profiles for precalcined iron samples,
−
1
obtained at temperature ramping rates of 3, 7, and
Using Eq. (9), activation energy, Tmax and f(α) of
the four-reduction models from Table 1, the A values
of the two-reductions step could be calculated. With
the A values, the TPR pattern of different reduction
models could be calculated by using Eq. (5). Table 2
shows the calculated A values of the two reduction
steps and Fig. 3 shows the calculated TPR pattern.
Compare the figure with the experiment TPR pattern,
it is concluded that the calculated TPR pattern of uni-
molecular model fits best for the first reduction step:
Fe2O3 → Fe3O4, and the two-dimensional nucleation
model according to Avrami–Erofeev describes the sec-
ond reduction step: Fe3O4 → Fe best.
◦
2
1 C/min. The typical hydrogen reduction profiles
present the two-stage reductions which have been re-
ported by other workers [20–22]. An increase in heat-
◦
ing rate from 3 to 20 C produced increases in T, of
Fig. 4 shows TPR patterns of different heating
−
1
−1
rate (a) ψ = 3 (K min ); (b) ψ = 7 (K min ); (c)
−
1
ψ = 21 (K min ) compared with calculated data.
The solid lines are experimental data and dash-dot-dot
lines are the calculated data by unimolecular for
peak 1 and two-dimensional nucleation according to
Avarmi–Erofeev model for peak 2. It is shown that
the calculation patterns fit quite well with the TPR
profiles for all three experiments of different heating
rate.
Fig. 1. TPR patterns of different heating rates (a) ψ = 3 (K min−1);
−1
−1
(
b) ψ = 7 (K min ); (c) ψ = 21 (K min ).

H.-Y. Lin et al. / Thermochimica Acta 400 (2003) 61–67
65
Fig. 2. Temperature-programmed Arrhenius plots for the two-step reduction. (a) Fe2O3 → Fe3O4, (b) Fe3O4 → Fe.
Table 2
A values of the two-reaction steps calculated by using E = 89.13 (kJ mol ), Tmax = 598.5 K for the first reduction step Fe2O3 → Fe3O4,
−
1
−1
−1
E = 70.41 (kJ mol ), Tmax = 723.2 K for the second reduction step Fe3O4 → Fe, and ψ = 7 K min
−
1
−1
Reduction model
A, s for Fe2O3 → Fe3O4
A, s for Fe3O4 → Fe
5
Three-dimensional nucleation according to Avarmi–Erofeev
Two-dimensional nucleation according to Avarmi–Erofeev
Unimolecular decay
166.3700
105.3500
122.7900
16.6800
1.46 × 10
4
9.22 × 10
5
1.07 × 10
4
Three-dimensional diffusion according to Jander
1.46 × 10
3
.3. Discussion
shoulder appearing on the low temperature side, and
the good fit of the unimolecular model indicates that
the first reduction step is a first order reaction: r =
The TPR simulation patterns methods can be ap-
−
E/RT
plied to determine the mechanism of the reduction re-
action. Wimmers et al [5] have calculated the TPR
pattern of the second peak for Fe3O4 → Fe step, but
the first reduction peak was too low of an accurate
quantitative analysis due to the sample size, so the
Fe2O3 → Fe3O4 step was not studied. In this work
we observed the first peak of Fe2O3 → Fe3O4 with a
Ae
(1 − α) where A and E are listed in Table 3.
The second peak of the Fe3O4 → Fe was fit-
ted with the two-dimensional nucleation model of
Avrami–Erofeev. The general form of Avrami–Erofeev
equation is:
1/n
[−ln(1 − α)]
= k(t − t0)
Table 3
−
1
A values of the two-reaction steps calculated by using E = 89.13 (kJ mol ), Tmax = 598.5 K for the first reduction step Fe2O3 → Fe3O4,
−1
E = 70.41 (kJ mol ), Tmax = 723.2 K for the second reduction step Fe3O4 → Fe
Reduction model (K min 1)
−
A, s for Fe2O3 → Fe3O4
−1
A, s for Fe3O4 → Fe
−1
4
ψ = 3
ψ = 7
ψ = 21
98.62
122.7900
99.14
9 × 10
5
1.07 × 10
5
1.16 × 10

6
6
H.-Y. Lin et al. / Thermochimica Acta 400 (2003) 61–67
And the differential form of the Avrami–Erofeev
equation is:
dα
/n
nk1 [−ln(1 − α)1−1/n](1 − α)
=
dt
The exponent n = β + λ where β is the number of
steps involved in nucleus formation (frequently β = 1
or 0, the latter corresponding to instantaneous nucle-
ation) and λ is the number of dimensions in which the
nuclei grow (λ = 3 for spheres or hemispheres, 2 for
discs or cylinders and 1 for linear development). The
equation can be used over a great of ∼0.05 < α <
0
.9 for many solid phase reactions [14]. In this work
the selection model is the two-dimensional nucleation
model of Avrami–Erofeev, which is n = 2 for the gen-
eral Avrami–Erofeev equation, where the in-situ for-
mation of Fe3O4 indicated β = 1 and λ = 2, and the
first-order nuclei formation of the interface is followed
by linear growth of the nuclei in two dimensions.
4. Conclusion
Precipitated iron oxide samples were character-
ized using the temperature-programmed reduction
method to investigate the reduction behaviors. The
two-step reduction was observed. Fe O was reduced
Fig. 3. Shows the TPR pattern of ψ = 7 (K min−1) calculated by
2
3
different reduction mode listed in Table 1.
to Fe3O4 and then reduce to metallic Fe. FeO was
not detected as an intermediated. The activation en-
ergies for the two reduction steps of iron oxide were
−
1
8
9.13 and 70.41 (kJ mol ). The simulation by re-
duction models of the TPR patterns presents well
fitting of unimolecular model for Fe2O3 → Fe3O4
reduction and two-dimensional nucleation according
to Avarmi–Erofeev model for Fe3O4 → Fe.
References
[1] J.W. Jenkins, B.D. McNicol, S.D. Robertson, Chem. Technol.
7
(1977) 316.
[
[
2] C. Li, Y.W. Chen, Thermochim. Acta 256 (1995) 457.
3] P. Arnoldy, J.C. M de Jonge, J.A. Moulijn, J. Phys. Chem.
8
9 (1985) 4517.
4] N.W. Hurst, S.J. Gentry, A. Jones, B.D. McNicol, Catal. Rev.
4 (1982) 233.
5] O.J. Wimmers, P. Arnoldy, J.A. Moulijn, J. Phys. Chem. 90
1986) 1331.
6] M. Sgunijawabe, R. Furuichi, T. Ishii, Thermochim. Acta 28
1979) 287.
Fig. 4. TPR patterns of different heating rates compared with cal-
culated data. The solid lines are measured data and dash–dot–dot
lines are the calculated data by unimolecular for peak 1 and
two-dimensional nucleation according to Avarmi–Erofeev model
[
[
[
2
(
−1
−1
for peak 2. (a) ψ = 3 (K min ); (b) ψ = 7 (K min ); (c)
−
1
ψ = 21 (K min ).
(

H.-Y. Lin et al. / Thermochimica Acta 400 (2003) 61–67
67
[
[
[
7] A. Baranski, J.M. Langan, A. Pattek, A. Reizer, Appl. Catal.
(1982) 207.
8] R. Brown, M.E. Copper, S.A. Whan, Appl. Catal. 3 (1982)
77.
9] D.B. Bukur, X. Lang, J.A. Rossin, W.H. Zimmerman, M.P.
Rosynek, E.B. Yeh, C. Li, Ind. Eng. Chem. Res. 28 (1989)
[15] M. Avrami, J. Chem. Phys. 7 (1939) 1103;
3
M. Avrami, J. Chem. Phys. 8 (1940) 212;
M. Avrami, J. Chem. Phys. 9 (1941) 117.
[16] B.V. Erofeev, C.R. Dokl. Acad. Sci URSS 52 (1946)
511.
[17] W. Jander, Z. Anorg. Allg. Chem. 163 (1927) 1;
W. Jander, Angew. Chem. 41 (1928) 79.
[18] J. Sesták, V. Satava, W.W. Wendtland, Thermochim. Acta 7
(1973) 333.
[19] C.D. Doyle, J. Appl. Polym. Sci. 5 (1961) 285.
[20] A. Lycourghiotis, D. Vattis, React. Kinet. Catal. Lett. 18
(1981) 377.
1
1
130.
[
10] D.B. Bukur, X. Lang, J.A. Rossin, W.H. Zimmerman, M.P.
Rosynek, C. Li, Ind. Eng. Chem. Res. 29 (1990) 1588.
11] H.E. Kissinger, Anal. Chem. 29 (1957) 1702.
[
[
12] E.E. Unmuth, L.H. Schwartz, J.B. Butt, J. Catal. 63 (1980)
4
04.
13] G. Munteanu, L. Ilieva, D. Andreeva, Thermochim. Acta 291
1997) 171.
[
[
[21] A.J. Kock, H.M. Fortuin, J.W. Gecis, J. Catal. 96 (1985)
261.
(
14] L. G. Harrison, in:Comprehensive Chemical Kinetics, vol. 2,
C. H. Bamford, C. F. H. Tipper (Eds.), Elsevier, Amsterdam,
[22] I.R. Leith, M.G. Howden, Appl. Catal. 37 (1988) 75.
1
969, p. 57.