A novel approach to understanding and modelling performance evolution of catalysts during their initial operation under reaction conditions-Case study of vanadium phosphorus oxides for n-butane selective oxidation

Article abstract of DOI:10.1016/j.jcat.2012.11.027

In heterogeneous catalysis, catalyst precursor activation and subsequent initial operation ('conditioning') under reaction conditions are often highly dynamic steps which are important in delivering effective catalyst performance over time scales of years. In many cases however, the phenomena occurring in both steps are often poorly understood. A novel approach to assess the conditioning step, comprising advanced catalyst testing methods in micro-reactors, catalyst characterisation and detailed kinetic and activity modelling techniques is demonstrated for a model system: the selective oxidation of n-butane to maleic anhydride (MA) using a pre-activated vanadium phosphorus oxide (VPO) catalyst. A transient kinetic model for the conditioning step is presented which describes the decline of population of active sites for three reaction pathways on the catalyst surface over time. Two reaction pathways, n-butane - > MA and n-butane - > CO_x decay similarly with time whilst a third, MA - > CO_x declines very sharply. The approach used provides important mechanistic information on catalyst reaction kinetics whilst also providing understanding the impact of reaction conditions on the catalyst during conditioning.

Full text of DOI:10.1016/j.jcat.2012.11.027

Journal of Catalysis 299 (2013) 249–260
Contents lists available at SciVerse ScienceDirect
Journal of Catalysis
journal homepage: www.elsevier.com/locate/jcat
A novel approach to understanding and modelling performance evolution
of catalysts during their initial operation under reaction conditions – Case study
of vanadium phosphorus oxides for n-butane selective oxidation
b
a
a
a
S.K. Wilkinson a,b, , M.J.H. Simmons , E.H. Stitt , X. Baucherel , M.J. Watson
⇑
a Johnson Matthey Technology Centre, P.O. Box 1, Belasis Avenue, Billingham TS23 1LB, UK
b School of Chemical Engineering, University of Birmingham, Birmingham B15 2TT, UK
a r t i c l e i n f o
a b s t r a c t
Article history:
In heterogeneous catalysis, catalyst precursor activation and subsequent initial operation (‘conditioning’)
under reaction conditions are often highly dynamic steps which are important in delivering effective cat-
alyst performance over time scales of years. In many cases however, the phenomena occurring in both
steps are often poorly understood. A novel approach to assess the conditioning step, comprising advanced
catalyst testing methods in micro-reactors, catalyst characterisation and detailed kinetic and activity
modelling techniques is demonstrated for a model system: the selective oxidation of n-butane to maleic
anhydride (MA) using a pre-activated vanadium phosphorus oxide (VPO) catalyst. A transient kinetic
model for the conditioning step is presented which describes the decline of population of active sites
for three reaction pathways on the catalyst surface over time. Two reaction pathways, n-butane - ->
MA and n-butane –> COx decay similarly with time whilst a third, MA - -> COx declines very sharply.
The approach used provides important mechanistic information on catalyst reaction kinetics whilst also
providing understanding the impact of reaction conditions on the catalyst during conditioning.
Ó 2012 Elsevier Inc. All rights reserved.
Received 11 September 2012
Revised 16 November 2012
Accepted 18 November 2012
Available online 28 January 2013
Keywords:
Catalyst
VPO
Conditioning
Micro-reactor
Kinetic modelling
Activity
1. Introduction
are often studied. Less explored are the activation and conditioning
steps, which are often highly dynamic in nature. With the condi-
Solid catalysts are structured products which are invaluable in
assisting a wide range of chemical processes on an industrial scale.
Critical performance features of a ‘good’ catalyst often comprise
high activity, high selectivity towards desired reaction products
and stability during long-term operation. The pursuit of under-
standing and subsequently controlling the development of such
features for different catalyst systems requires knowledge of a
number of stages of catalyst production, which chiefly includes,
but is not limited to:
tioning step in particular, there is very little mechanistic under-
standing of why the catalyst performance changes. Hence, there
is considerable scope to develop techniques for a range of systems
to de-convolute and describe phenomena occurring during both
steps. The conditioning period may also provide valuable informa-
tion around the fundamental reaction kinetics of catalyst systems,
such as an understanding of which active sites are being generated
and lost on the catalyst surface over time.
In this paper, a novel approach will be used which combines ad-
vanced reactor testing methods to track catalyst evolution (chiefly,
parallel difference testing [1]) with kinetic/activity modelling and
supporting characterisation methods during the conditioning per-
iod. Using a model system, evolution of active sites will be demon-
strated and linked through to fundamental physical and chemical
changes occurring to the catalyst itself. Investigation into the effect
of reaction conditions during conditioning will aid in developing
mechanistic models for catalyst evolution.
ꢀ Catalyst precursor preparation.
ꢀ Activation of the precursor.
ꢀ Initial, unsteady state, ‘conditioning’ of the catalyst under reac-
tion conditions. This can also be referred to as ‘equilibration’.
For the last step, catalytic performance can change significantly
over time and may or may not tend towards a quasi-steady-state
(QSS) performance. For many reaction systems, precursor prepara-
tion and performance of the catalyst under QSS reaction conditions
Vanadium phosphorus oxides (VPOs), which carry out the selec-
tive oxidation of n-butane to maleic anhydride (MA), are a model
example of catalysts whose final QSS performance is strongly
dependent on prior preparation, activation and conditioning meth-
ods [2] and have been selected for investigation in this work.
Whilst the effect of the preparation method of the VPO precursor,
⇑
Corresponding author at: Johnson Matthey Technology Centre, P.O. Box 1,
Belasis Avenue, Billingham TS23 1LB, UK.
E-mail address: sam.wilkinson@matthey.com (S.K. Wilkinson).
0021-9517/$ - see front matter Ó 2012 Elsevier Inc. All rights reserved.
http://dx.doi.org/10.1016/j.jcat.2012.11.027

250
S.K. Wilkinson et al. / Journal of Catalysis 299 (2013) 249–260
vanadyl phosphohemihydrate (VOHPO4ꢁ0.5H2O) on QSS perfor-
mance has been extensively explored and understood [3–5], the
same cannot be said for the activation and conditioning stages. It
is accepted that both stages are critical in the generation and loss
of desired (MA forming) and undesired (carbon oxide (COx) form-
ing) active sites on the catalyst [4,6]; however, there is a strong dis-
connect between previous literature efforts on the topic.
The existing confusion around the phenomena taking place dur-
ing VPO activation and conditioning is dependent upon two key
issues:
lyst was activated in situ, the findings are in agreement with [11].
Meanwhile, in [15], TPD and TPR served as useful analytical tech-
niques to measure the number of monolayers of oxygen removed
from the VPO catalyst. Monolayers of oxygen were progressively
removed with time due to surface and bulk morphology changes.
Some insights may be afforded via forced unsteady-state exper-
iments using a temporal analysis of products (TAPs) reactor sys-
tem. Schuurman and Gleaves [16,17] applied both steady-flow
and transient-response experiments on a fully conditioned VPO
catalyst. They found that the apparent activation energy for n-bu-
tane oxidation could be reduced following exposure of the catalyst
to oxygen pulses over a short period of time. They proposed a two-
site model comprising high and low activation energy sites associ-
ated with V4+ and V5+ VPO phases, respectively, with the number of
both types of site being a strong function of the overall oxidation
state of the catalyst.
Transient analysis of VPO catalysts has also been carried out
using a tapered element oscillating microbalance (TEOM) [18,19].
The TEOM assessed rates of reduction and oxidation of a VPO cat-
alyst under a variety of conditions by observing changes in catalyst
mass. This technique produced detailed kinetic models; however,
the reduction and oxidation steps of the catalyst were considered
separately rather than under n-butane/air conditions.
Firstly, the fundamental catalytic reaction mechanisms taking
place under n-butane/air conditions are still not fully understood,
particularly regarding the precise roles of the catalyst surface and
bulk phases in the MA and COx producing reactions. Whilst there is
substantial evidence that the presence of a vanadyl pyrophosphate
((VO)2P2O7)) phase (in particular the (100) face) is an important
feature to an active and selective VPO catalyst [6,7], conflicting evi-
dence exists around the role of V5+ phases (such as
a-VOPO4) in the
reaction mechanisms [5,8–10]. Ultimately, the ‘end-product’ bulk
and surface requirements for an active and selective VPO catalyst
at QSS performance are not fully defined, and hence, the actual
purposes of the activation and conditioning processes are some-
what unclear.
The second issue arises due to the vast number of possible strat-
egies that can be employed for these activation and conditioning
processes, resulting in contradictory results and conclusions in lit-
erature. In general, the available strategies fall into two categories
[11]:
Further transient kinetic investigations include decoupling oxi-
dation and reduction steps of VOPO4 in a TAP reactor system [20]
and kinetic investigations under separate reduction and oxidation
steps [21].
It is clear from the aforementioned studies that a lack of a com-
mon and systematic methodology has prevented full understand-
ing of the evolution of VPO catalysts during activation and
conditioning. This relates to the analysis and catalyst testing tech-
niques used but also the large differences in catalyst preparation,
activation and conditioning methods between different research
works. A critical missing element is a kinetic and activity modelling
study which could be utilised to explain the generation and loss of
active sites on the catalyst under reaction conditions during this
period of operation.
Hence, selective oxidation of n-butane using a VPO catalyst has
been identified as a model candidate system to develop experi-
mental and modelling techniques to understand catalyst evolution
during activation and conditioning. The conditioning step of an ex
situ activated VPO catalyst will be examined in detail in an integral
mode reactor under a range of conditions. The objectives for this
study are therefore as follows:
ꢀ Ex situ activation and conditioning. Activation is performed in a
controlled gas atmosphere (typically air or nitrogen) at temper-
atures up to 600 °C. During this step, the VOHPO4ꢁ0.5H2O is
transformed into a mixture of V4+ and V5+ phases. The activated
VPO catalyst is then exposed to n-butane/air reaction conditions
and undergoes the conditioning step.
ꢀ In situ activation and conditioning. The VPO precursor is exposed
to n-butane/air reaction conditions without prior treatment,
undergoing both processes in tandem.
Firstly, considering the ex situ methodologies, the activations
employed by [12,13] resulted in a VPO catalyst dominated by
poorly crystalline (VO)2P2O7, whilst the air activation in [11] re-
sulted in a catalyst containing V5+ and some remaining precursor
as well as (VO)2P2O7. The final QSS performance developed with re-
spect to bulk composition was largely crystalline (VO)2P2O7 in each
case. Catalyst performances are not comparable due to the differ-
ences in reaction conditions employed. Analysis of the catalyst sur-
face is only attempted in [11], which compares surface evolution
between fresh, non-QSS and QSS VPO catalysts using XPS. Surface
evolution was comparable to that of the bulk catalyst, which fea-
tured a gradual phase transformation to a largely crystalline (VO)2-
P2O7 phase (Vox of +4.0) over 1000 h on stream.
ꢀ Develop and demonstrate a methodology to understand the
performance evolution that catalysts, using VPO as a model sys-
tem, undergo during conditioning, which will include the
following:
–
–
–
Experimental catalyst reaction testing methods.
Kinetic and activity modelling application to data.
Supporting evidence from catalyst characterisation methods.
Lombardo et al. [4] investigated a variety of activation strategies
comprising n-butane in air. It was found that n-butane content and
gas flow rate of the feed had a significant effect on catalyst activity
and selectivity towards MA at QSS performance. The final bulk
composition of the catalysts tested was very similar (mostly crys-
talline (VO)2P2O7 phases) suggesting that there is little correlation
between bulk phase and VPO catalyst performance.
Conclusions in [4] were largely based around QSS performance
observations, whilst other studies [14,15] systematically tracked
VPO evolution over time. In [14], the findings revealed a shift in
bulk phase structure from an initial V4+/V5+ mixture to a more crys-
talline V4+ phase over time on stream. The ratio of V4+/V5+ at the
surface was measured by XPS and increased with time on stream,
as did n-butane conversion and MA selectivity. Although the cata-
2. Materials and methods
2.1. Sample preparation and properties
The VPO precursor was synthesized using an organic prepara-
tion method as described in US patent 4,567,158 [22]. This meth-
odology comprises introducing V2O5 into an alcohol medium to
reduce the average vanadium valence state to less than +5. This
forms a slurry which is then contacted with an alcohol modifying
agent, further reducing some of the vanadium in the mixture to
+4. A pentavalent phosphorus compound is then added to the
mixture to form a phosphorus–vanadium mixed oxide catalyst

S.K. Wilkinson et al. / Journal of Catalysis 299 (2013) 249–260
251
precursor. This is subsequently recovered, dried and formed into
pellets. The P:V ratio of the catalyst was 1.0.
(Wheeler Weisz Modulus), Mears’ criterion for intraparticle heat
transport limitations and radial temperature gradients.
180 < dp < 250 m sieve fraction of VPO powder was chosen for
testing to minimise wall effects in the reactor [25] and prevent
large pressure drops [26].
Transport limitations were checked at the extreme of the test
programme operation, namely the 1st sector of a VPO catalyst
bed operating at 410 °C at t = 0 h on stream. Apparent n-butane
conversion rate was 1.08 mmol kg catꢃ1 sꢃ1. Under these condi-
tions, the radial heat transport gradient was 2.2 K, and both inter-
nal and external mass transfer efficiencies were >99%. These values
satisfy threshold criteria.
A
The pellet batch was subsequently activated using a methodol-
ogy based on US patent 5,137,860 [23]. This comprises heating the
formed catalyst precursor in an atmosphere comprising air, steam
and nitrogen to a temperature not exceeding 300 °C. It is subse-
quently maintained at this temperature for a nominal time. The
temperature is then ramped at a rate between 2 and 12 °C per min-
ute to eliminate water of hydration from the precursor. The tem-
perature is then held at a value in the 350–550 °C range, still
under air and steam to provide a vanadium oxidation state be-
tween +4.0 and +4.5. The gas atmosphere is subsequently switched
to a non-oxidising nitrogen–steam atmosphere, at the same tem-
perature for a defined time to complete the activation process.
l
Each reactor was packed, firstly with silicon carbide
(180 < dp < 250 lm), then VPO catalyst (up to 0.66 g), then the
remainder with silicon carbide. This ensured that pressure drops,
although negligible, were constant across all reactor tubes. Ther-
mocouples were positioned in the tube before this process, so that
one registers the temperature at the catalyst bed inlet and the
other at the catalyst bed outlet. A West 6400 temperature control-
ler was used to control the temperature of the furnace block to
within 2 °C of the set point value. A Grant Remote Squirrel
(1250 Series) was used to log temperature data from each thermo-
couple periodically. Throughout the test programme, inter- and in-
tra-tube axial temperature differences were within 2 °C of each
other.
Gases (air or helium) with/without n-butane were fed to each
micro-reactor tube using a Brookes PID mass flow controller. Max-
imum v/v% concentration of n-butane within the feed was 2%. Flow
through each reactor tube was up to 30 ml minꢃ1 (standard tem-
perature and pressure, STP) during the test programme.
Reactor outlet gases were analysed by gas chromatography
(Varian CP-3800); a thermal conductivity detector analysed carbon
oxides, nitrogen and oxygen. The carrier gas used was helium. The
setup comprised a Hayesep Q (flow through: 25 ml minꢃ1 at
1.7 bar gauge) with molecular sieve column (flow through:
27 ml minꢃ1 at 2 bar gauge) in series; the carrier gas was helium.
Typical retention times were 0.7 min for CO2, 1.6 min for O2,
2.1 min for N2 and 3.1 min for CO. A flame ionisation detector ana-
lysed n-butane, MA and trace acrylic and acetic acid. The carrier
gases used were air and hydrogen. A capillary column was used
(flow through: 10 ml minꢃ1 at 0.45 bar gauge). Typical retention
times were 1.2 min for n-butane, 5.5 min for MA, and trace acrylic
and acetic acid were eluted in the 2–3 min range.
2.2. Micro-reactor test rig
Selective oxidation of n-butane was carried out in a fixed-bed,
down-flow, steel micro-reactor at atmospheric pressure. Five par-
allel tubes were used with inner diameter of 3 mm and length of
ꢂ20 cm.
Parallel difference testing, a methodology described in detail in
[1], is used to investigate the performance evolution of VPO cata-
lysts operating in integral mode reactors. The simple setup for this
study comprises four tubes in parallel containing 0.165, 0.33, 0.495
and 0.66 g, respectively. Flow rate (30 ml minꢃ1), inlet reactant
concentration and temperature are the same for each tube
throughout the test. Subtraction between the exit concentrations
of each catalyst bed at discrete times on stream allow performance
to be broken down into four sectors of equal length down the bed.
Experimental repeats were carried out as
reproducibility.
a
check for
Parallel difference tests were carried out for up to 100 h on
stream. At certain times on stream, the flow rate through the reac-
tors was reduced, so that n-butane conversion across the 0.66 g
‘full’ bed was restored to its initial level from the start of the test.
Relative performances of each sector of the catalyst bed (via prod-
uct yields) could therefore be compared to their initial values. This
provides information on the influence of axial concentration gradi-
ents on the evolution of catalyst performance. Selectivity evolution
of separate reaction pathways can also be investigated by this test
which can be linked to fundamental changes to the catalyst.
VPO catalysts were also tested in a differential mode setup;
50 mg catalyst was used to ensure n-butane conversions which
were <10% in 30 ml minꢃ1 (standard temperature and pressure)
flow conditions.
Experiments were designed to ensure the presence of plug-flow
conditions and the absence of heat and mass transport limitations
allowing for the measurement of intrinsic rates. Full details of reac-
tor setup and catalyst properties are shown in Table 1. This was
confirmed by use of key calculations set out in literature [24]
including calculation of intraparticle mass transport limitations
The calculated carbon balance was between 99% and 101% for
all experiments. As the amount of acrylic and acetic acid formed
was negligible, the molar carbon balance across the reactor simpli-
fies to:
Moles n-butane ðInÞ ¼ Moles n-butane ðOutÞ
þ Moles MA ðOutÞ þ 0:25
ꢄ Moles COx ðOutÞ
ð1Þ
Reaction testing was carried out in the range of 350–410 °C,
chosen to encompass typical operating temperatures of an indus-
trial VPO catalyst. For each test run, the catalyst beds were heated
to set point temperature under a helium flow. At the set point, this
was switched to 2% n-butane in air for all experiments. After each
experimental run, catalysts were cooled down in a helium environ-
ment, to minimise changes to the VPO catalyst during the shut-
down step. After testing, VPO catalyst samples were retained for
future characterisation studies.
One micro-reactor tube filled with silicon carbide acted as a
blank experiment. 2% n-butane in air was fed under reaction con-
ditions to ensure no homogenous reaction contributed to the over-
all reaction.
Table 1
List of properties of catalyst and experimental setup conditions.
Experimental property
Value
Catalyst mass (g)
0.165–0.66
0.13
610
20
180–250
3
623–683
2/98
30
Catalyst pore volume (cm3 gꢃ1
Catalyst pore size (Å)
)
Catalyst N2 BET surface area (m2 gꢃ1
)
Catalyst particle size (
Reactor diameter (mm)
Temperature range (K)
lm)
Feedstock (% n-butane–air)
Flow rate (ml minꢃ1
)

252
S.K. Wilkinson et al. / Journal of Catalysis 299 (2013) 249–260
2.3. Catalyst characterisation
A
Fresh and post-reactor used VPO samples were characterised by
the following ex situ techniques:
X-ray diffraction: Samples were tested using a Siemens D5000
diffractometer using Cu Ka X-rays at wavelength 1.5406 Å.
Raman spectroscopy: Samples were tested using a JY Horiba
Infinity Spectrometer.
X-ray photoelectron spectroscopy: Samples were tested using a
Thermo Escalab 250. The exciting radiation used was mono-
chromatised aluminium Ka radiation, 200 W, 650 lm spot size.
Vox determination via titration: Samples were tested using a
Metrohm 809 Titrando auto-titration device using a methodol-
ogy previously described in literature [27]. Dissolved samples
were first titrated using 0.1 N KMnO4 to fully oxidise the VPO
compounds into a V5+ state. A second titration using 0.1 N
Mohr’s Salt was then carried out to fully reduce the VPO com-
pounds into a V4+ state. The following equation was then used
to determine the bulk Vox of the VPO catalyst:
B
ðvol: KMnO4 addedÞ
Vox ¼ 5 ꢃ
ð2Þ
ðvol: Mohr’s salt addedÞ
2.4. Kinetic modelling
For kinetic modelling purposes, parametric estimation within
the kinetic models was carried out using Athena Visual Work-
benchÓ software [28]. A plug-flow reactor model was used. A
Bayesian estimation routine was used for fitting of model parame-
ters with the objective function being minimised using a forward-
difference method. To minimise cross-correlation between activa-
tion energy (Ea) and pre-exponential factor (Ai) parameters, a re-
parameterised Arrhenius equation was used:
C
ꢀꢀ
ꢁ ꢀ
ꢁꢁ
Eai
613 ꢁ R
613
T
ki ¼ Ai;613 ꢁ exp
ꢁ
1 ꢃ
ð3Þ
where base temperature, Tbase, is 613 K, and Ai,613 is the value of the
rate constant ki at 613 K. Kinetic model fitting was carried out at
nine discrete time on stream points between 0.4 and 11.4 h. At each
time on stream, the data set comprised 28 independent observa-
tions (seven temperatures each with four sector profiles).
3. Results
Fig. 1. Time on stream intrinsic rate evolution for individual sectors of an example
run for (a) n-butane consumption, (b) MA formation, (c) COx formation over a VPO
catalyst. Conditions: 30 ml minꢃ1 flow, 2% v/v n-butane/air reactor inlet concen-
tration and 380 °C. Symbols denote: ( ) 1st sector, ( ) 2nd sector, ( ) 3rd sector,
(s) 4th sector.
3.1. Parallel difference testing
Parallel difference tests were carried out over a temperature
range of 350–410 °C. Fig. 1 shows an example of raw VPO catalytic
performance evolution with time on stream. Values are reported
on an intrinsic rate basis for each sector, calculated as the change
in a reactant or product concentration over a sector divided by
the residence time of the gas within that sector. As each sector con-
tains the same mass, these values are directly representable on a
mass basis also. In this example, carried out at 380 °C, total n-bu-
tane conversion across the entire bed was 74% at the start and
61% after 60 h on stream. It is important to note that CO and CO2
formation was considered as a lumped COx term across the study.
This was due to the fact that CO:CO2 ratio across all sectors, tem-
peratures and times on stream remained roughly steady in the
range of 1.5–1.7.
period. The 2nd–4th sectors show a very different behaviour to the
1st sector with respect to MA formation rates. The 2nd sector
shows a very slow decline, whilst the 3rd/4th sectors actually im-
prove slightly at the start of the test. Meanwhile, for COx formation
rates, a similar decline pattern is seen in each sector of the catalyst
bed although the 1st sector shows the greatest losses. On inspec-
tion, evolution of n-butane consumption appears much the same
as that for MA formation, owing to the high selectivity of the cat-
alyst (>70%) towards MA under these conditions. It is important
to note that these changes are happening on a very short (order
of hours) time scale in context to a VPO catalyst’s working life,
which is in the order of years.
There are some key differences in the approach to QSS perfor-
mance for each sector. A sharp initial decline is seen in the 1st sec-
tor for both MA and COx formation rates over the first 5 h which is
followed by a much slower rate of change over the rest of the test
A comparison of full bed (sectors 1–4 combined) n-butane con-
versions in the 350–410 °C temperature range is shown in Fig. 2.
For each temperature, n-butane conversion drops between the ini-

S.K. Wilkinson et al. / Journal of Catalysis 299 (2013) 249–260
253
sion readings. Sector MA and COx yields were calculated using (Eqs.
(4a) and (4b)) and were subsequently normalised. It is important
to note that in these equations, the term ‘reactor inlet feed’ refers
to 2% n-butane/air.
% MA formed over sectorðsÞ
Yield MA ¼
ð4aÞ
% n-butane in reactor inlet feed
% COx formed over sectorðsÞ
Yield COx ¼
ð4bÞ
% n-butane in reactor inlet feed
In sector 1 (Fig. 3a), a loss of yield for both MA and COx forma-
tion is seen at restored initial conversion. Meanwhile, sectors 2–4
(Fig. 3b) evolve quite differently over the first 17 h. Relative yields
with respect to MA formation have increased, whilst for COx forma-
tion, the yields remain roughly unchanged.
ο
Yields of MA and COx also change very differently with temper-
ature in sectors 2–4. Whilst COx yields systematically increase with
temperature, this does not happen to the MA yields which appear
to decline at the highest temperatures. The latter observation sug-
gests that MA may either be re-oxidised to COx or inhibited by its
own formation at high n-butane conversions.
Fig. 2. Evolution of n-butane conversion over the full catalyst bed between initial
reading (t = 0.4 h) ( ) and 17 h on stream ( ) for different reaction temperatures.
(d) represents the adjusted reactor inlet flow rates required at 17 h on stream to
achieve initial n-butane conversion over the full catalyst bed.
The observations in Fig. 3 firstly suggest that the mechanism
behind the evolution of catalyst performance is driven by concen-
tration of reactants (n-butane and/or oxygen). The 1st sector of the
reactor sees the greatest apparent loss in catalyst performance, and
it is in this sector that reactant concentrations are highest.
The problem is, however, clearly more complex than this as
there are differences in performance evolution for the MA and
COx formation routes in sectors 2–4. Furthermore, an Arrhenius
Plot analysis of the 1st sector (Table 2) reveals differences in the
two product formation routes. The apparent activation energy
(Ea) for MA formation is ꢂ15–20 kJ molꢃ1 lower at the initial read-
ing compared to 17 h on stream. This ‘drift’ is a similar observation
to that made by [17] under forced unsteady-state testing. Mean-
while, there are no changes in apparent Ea observed for COx forma-
tion during the conditioning period.
A
B
Table 2
Apparent activation energies for MA and COx product formation in the 1st sector
obtained via Arrhenius Plot (R2 > 0.99 in all cases).
Product
Ea (initial)
Ea (17 h)
Ea (17 h, restored)
(kJ molꢃ1
)
(kJ molꢃ1
)
(kJ molꢃ1
)
MA
COx
49.1
77.3
67.2
80.3
62.6
76.6
Fig. 3. Performance evolution over (a) sector 1, (b) sectors 2–4 combined. ( ) and
(
) denotes initial MA and COx yields respectively, ( ) and (s) MA and COx yields,
respectively, under adjusted inlet flows at 17 h on stream.
tial reading and 17 h on stream. At 17 h, the flow rate through the
reactors was adjusted to restore initial n-butane conversion across
the full bed. In Fig. 2, the reduced flow rates required to achieve
this are very similar at all temperature points, with all values in
the 23–24 ml minꢃ1 range. The level of flow rate reduction re-
quired to restore initial conversion is a good indicator of the extent
of activity loss in a catalyst bed. This suggests that the rate of activ-
ity evolution during conditioning may be temperature
independent.
Fig. 4. Evolution of MA intrinsic selectivity of a VPO catalyst across a full catalyst
bed for different reaction temperatures. ( ) denotes initial MA selectivity, ( ) MA
selectivity under adjusted inlet flows at 17 h on stream.
Fig. 3 makes a comparison of the sector contributions to overall
catalyst bed performance at the initial and restored initial conver-

254
S.K. Wilkinson et al. / Journal of Catalysis 299 (2013) 249–260
Fig. 4 shows the evolution of MA intrinsic selectivity, which is
of sector position. By 100 h on stream, the average bulk Vox of
the catalyst bed is seen to decline to +4.10–4.15. It has previously
been noted in the literature that VPO catalysts tend towards a bulk
Vox of +4.0 with a few weeks on stream, such as in [11]. This is
being approached with this particular catalyst, but the changes
do not correlate with catalyst performance evolution observed
during the first few hours on stream.
seen to systematically improve over the first 17 h at all reaction
temperatures. This shows that the relative proportion of MA and
COx producing sites on the VPO catalyst have changed over time.
Hence, there must be a minimum of two different active sites on
the catalyst which are both:
ꢀ Evolving at different rates.
Bulk Vox measurements were also made after 17 h on stream for
VPO catalysts conditioned at 350 °C and 410 °C. Their Vox values
were +4.25 and +4.30, respectively; very similar within measure-
ment error. This suggests conditioning temperature has little influ-
ence upon transitions within the catalyst bulk in the range 350–
410 °C.
ꢀ Declining in population to different extent over the testing
period.
3.2. Characterisation
XRD analysis of fresh and post-reactor use samples revealed
very little change to the bulk structure during the conditioning
period. All catalysts were found to contain significant amounts of
(VO)2P2O7 via the technique [5], with reflections at 18.5°, 23°,
28.5°, 30° and 33.7°. Graphite peaks were also observed at 26.7°
and ꢂ34.5° in some cases. A shoulder between 28.5° and 30° is ob-
The evolution of the catalyst surface as a function of treatment/
time on stream was studied using XPS. Fig. 6 shows that the peak
binding energies for both the V2p3/2 and O1s signals shift during
the initial first 5 h of operation. Using a calculation method in
[29], evolution of surface Vox with time on stream could be ascer-
tained. For the fresh and 1 h conditioned catalyst, surface Vox was
calculated to be +4.2. For observations at 5, 17 and 100 h, this
dropped to +4.1. Also, there appears to be some shifts in the shape
of the V2p3/2 signal, particularly during the first 5 h. This may sug-
gest that the phases present at the VPO surface are changing to a
certain extent. This may be in line with Raman observations which
served. Reflections in this region could be characteristic of
c-
VOPO4 or I-VOPO4. The shoulder is somewhat steeper in the fresh
a
sample. Degree of crystallinity afforded by the sharpness of ob-
served peaks (full width at half maximum) changes minimally with
time, reactor axial position of the catalyst or reaction temperature.
An example of this is shown in Supplementary information for
time on stream evolution of the 1st sector of the catalyst bed.
Raman spectra for all samples, regardless of time on stream,
reaction temperature or axial position, appear to be similar, with
only small variations. Spectra plotted as a function of time on
stream are shown in Supplementary information. Again, the in-
depth work of Guliants et al. [5] has been used for interpretation.
In all cases, a sharp peak is detected around 920 cmꢃ1, which is
characteristic of (VO)2P2O7. A weakly intense response is also seen
at 1180–1190 cmꢃ1, again characteristic of (VO)2P2O7. A mixture of
peaks is also seen in the 1015–1045 cmꢃ1. The presence of a dou-
blet in the fresh sample at c.a. 1020 and 1035 cmꢃ1 is characteristic
suggest the presence of either c-VOPO4 or aI-VOPO4, the former of
which appears to be more predominant in the fresh sample.
Phosphorus–Vanadium (P:V) ratio was also calculated and was
found to be fairly consistent at ꢂ1.9 regardless of time on stream,
temperature or axial position. The values observed agree with lit-
A
of c-VOPO4. The intensities in this region for the used samples are
less clear and could constitute either c-VOPO4 or aI-VOPO4. A mix-
ture of low intensity responses is seen c.a. 940 and 980 cmꢃ1 at
fresh and all times on stream which could again relate to either
c-VOPO4 or
aI-VOPO4.
The measured average bulk Vox of the VPO catalyst is seen to de-
crease slowly with time on stream. The majority of changes to cat-
alyst performance in all parallel difference tests were seen during
the first 10 h of operation; however, Fig. 5 shows that the bulk
Vox of the catalyst changes very little during this time, regardless
B
Fig. 5. Measured bulk Vox via auto-titration of VPO catalyst before and after testing
at 380 °C, 2% n-butane in air reactor inlet conditions, 30 ml minꢃ1 flow. ( ) denotes
1st sector, ( ) 4th sector. Error bars denote standard deviation over three samples
measured.
Fig. 6. XPS analysis of VPO catalyst from the 1st sector of the catalyst bed, before
and after testing at 380 °C, 2% n-butane in air reactor inlet conditions, 30 ml minꢃ1
flow. (a) Denotes V2p3/2 signal, (b) O1s signal. (
) Fresh VPO, ( ) 1 h,
(
) 5 h, ( ) 17 h, (-- -- --) 100 h.

S.K. Wilkinson et al. / Journal of Catalysis 299 (2013) 249–260
255
within the scope of this study. The extent that this is true in the
>410 °C range would need to be ascertained by further study. This
confirms the initial observations made in Fig. 2. The areal rates for
n-butane conversion measured at 350 and 410 °C were
ꢂ1.5 ꢅ 10ꢃ8 and 5.5 ꢅ 10ꢃ8 mol mꢃ2 sꢃ1. These rates are compara-
ble with differential rates measured previously in literature
[8,19,30].
A
ο
ο
ο
Fig. 7 also shows that when the VPO catalyst undergoes an im-
posed temperature step change, a new QSS performance is imme-
diately observed. In effect, changing reaction temperature on a QSS
operation VPO catalyst does not appear to impose a further tran-
sient period of performance on the catalyst (i.e. further condition-
ing). This suggests that the surface is not inherently modified (e.g.
progressively reduced or oxidised) as a result of changing reaction
temperature over the range of test conditions employed.
When the VPO catalyst is returned to its reference ‘conditioning’
temperature (350 or 410 °C) following a temperature step change,
the catalyst also returns to its original QSS reference performance.
This shows that altering reaction temperature at QSS performance
does not invoke any irreversible effects on the catalyst which
would result in a generation or loss of active sites.
ο
B
ο
ο
ο
ο
4. Kinetic modelling
4.1. Apparent kinetics
The parallel difference test provides an axial concentration pro-
file of each reactant and product along the catalyst bed at each dis-
crete time on stream measurement. Simple power law-type
relationships for a parallel reaction network can be applied to
the data (Eqs. (5a) and (5b)). A concentration dependence on n-bu-
tane has been chosen for the both MA and COx forming reactions,
based on differential mode observations. Furthermore, it is the lim-
ο
C
ο
A
ο
ο
Fig. 7. Differential mode testing of VPO catalyst under steady-state performance
following 48 h conditioning. Inlet feed 2% v/v n-butane in air, ( ) denotes catalyst
conditioned at 350 °C, ( ) at 410 °C. Graph (a) n-butane consumption, (b) MA
formation, (c) COx formation.
erature that the surface of a working VPO catalyst is phosphorus
enriched compared to the bulk [30].
B
3.3. Testing in differential mode
In separate experiments, 50 mg of VPO catalyst was conditioned
at either 350 or 410 °C in a flow of 2% v/v n-butane in air. After 48 h
on stream, VPO performance with respect to n-butane consump-
tion, MA formation and COx formation had reached a QSS perfor-
mance level. Following this, the catalyst was exposed to a series
of temperature ramps. These are shown in graphs A–C in Fig. 7
(noting that t = 0 h in this case follows an initial 48 h of
conditioning).
Inspection of the graphs in Fig. 7 shows that the QSS perfor-
mance developed under temperatures of either 350 or 410 °C is al-
most identical. This suggests that conditioning temperature
employed has little effect on the final catalyst surface developed,
Fig. 8. Evolution in apparent reaction order in n-butane with time on stream for (a)
MA formation, (b) COx formation. ( ) 360 °C, ( ) 380 °C, ( ) 410 °C.

256
S.K. Wilkinson et al. / Journal of Catalysis 299 (2013) 249–260
playing a role in the reaction network during reactor start-up.
The transient kinetic model in [17] has also been selected for com-
parison purposes. Fig. 9 shows a diagram of the triangular network,
and Table 3 shows the kinetic models which were tested based on
this network.
The aim of this section of the study is to isolate a kinetic model,
which can describe the entire data set, as the VPO catalyst is not
fundamentally changing in structure during the conditioning peri-
od. Hence, for a correct model, the fundamental nature of the ac-
tive site should not change with time on stream (reflected in
fitted Ea values), and instead, a decay of active site populations
should be observed (via fitted pre-exponential factor, Ai, values).
The triangular networks in Table 3 are representative of three
levels of model. The model of Escardino et al. [31] uses a set of sim-
ple pseudo-first-order reactions, thus providing a pragmatic base
case for describing the system. The model of Buchanan and Sun-
daresan [32] will assess if a Mars–van Krevelen (MvK) type
description is necessary: an important consideration as VPO cata-
lysts fundamentally function via a redox mechanism [6,37]. Finally,
the model of Sharma and Cresswell [35] considered product inhibi-
tion factors in an integral mode reactor and so is applicable to this
study.
Fig. 9. Schematic of triangular reaction network.
iting reactant in the feed (at 2% v/v compared to 21% O2) and is also
unanimously cited as part of unsteady-state [17,20] and steady-
state [31–36] kinetic models in literature.
n1
rMA ¼ k1½C4H10
ꢆ
ð5aÞ
n2
rCO ¼ k2½C4H10
ꢆ
ð5bÞ
x
Log–log plots of (ri) against [C4H10] were generated to determine
apparent order in n-butane (ni) and apparent rate constant (ki) via
gradient and y-intercept, respectively. R2 values from this fitting
process were >0.9 across the data sets. Fig. 8a shows how ni system-
atically changes with time towards apparent 1st order for the MA
formation reaction at different temperatures. In Fig. 8b, ni for COx
formation changes with both time and temperature.
This result instantly shows that the assumption of a simple par-
allel reaction network is wrong; ergo, reaction orders in a correct
model should not change with time or temperature. More complex
models are therefore needed to describe the selectivity evolution
discussed earlier.
4.2.1. 6-parameter model of Escardino et al. [31]
Examples of results returned using a 6-parameter pseudo-first-
order model are shown in Table 4. At each time on stream point
(illustrated here by t = 0.4 and 11.4 h), the quality of fit (R2) is
excellent. Values of Ea1 and Ea2 stay roughly constant at different
times with acceptable confidence intervals. The value of Ea1 is also
lower than Ea2 which is observed in a number of aforementioned
literature models. This suggests that the rate-determining steps
within the mechanisms for n-butane ? MA and n-butane ? COx
are different. For both n-butane oxidation routes, the parameters
which change with time on stream are the pre-exponential factors.
The main shortcoming of the 6-parameter model is the clear
cross-correlation between the A3,613 and Ea3 parameters in the
MA ? COx route. In most cases, the 95% confidence intervals for
4.2. Triangular networks
Triangular network models are the main focus as there is strong
evidence in the aforementioned results that MA re-oxidation is
Table 3
Selected kinetic models from literature applied in this study.
Study
Network type
Rate expressionsa
Activation energies
(kJ molꢃ1
Reaction conditions and mode of
operation
)
Escardino et al. [31]
Triangular (6 param.)
r1 = k1[C4H10
r2 = k2[C4H10
r3 = k3[C4H2O3]
]
]
ꢀ 400–480 °C
ꢀ <0.95% n-butane in air
ꢀ QSS operation
k1 ½C4 H10
ꢆ
ꢆ
Buchanan and Sundaresan
[32]
Triangular (8 param.)
Triangular (8 param.)
E
E
E
a1 = 125.0
ꢀ 390–430 °C
r1
r2
r3
¼
¼
¼
N
k2 ½C4 H10
N
ꢀ 0.62–1.55% n-butane in air
ꢀ Up to 100% n-butane conversion
ꢀ QSS operation
a2 = 145.0
a3 = 180.0
k3 ½C4 H2O3
ꢆ
N
KBu ½C4 H10
KMA½C4 H2 O3ꢆ
N ¼ 1 þ
ꢆ þ
½O2
ꢆ
½O2 ꢆ
n
k1½C4H10
ꢆ
Sharma and Cresswell [35]
E
a1 = 93.0
ꢀ 300–400 °C
r1
¼
ð1þKMA ½C4H2O3 ꢆÞ
ꢀ Up to 3% n-butane in air
ꢀ Up to 90% conversion
ꢀ QSS operation
n
r2 = k2[C4H10
]
E
E
a2 = 93.0
a3 = 155.0
k3 ½C4 H2 O3
¼
ð1þKMA ½C4H2O3 ꢆÞ
ꢆ
r3
r = kB[C4H10
kB = kB1 + kB2
2
Schuurman and Gleaves
[17]
n-Butane ? products only (4
param.)
]
E
E
aB1 = 50.2
aB2 = 96.3
ꢀ 260–425 °C
ꢀ TAP reactor
ꢀ QSS and unsteady state operation
a
=r1 = n-butane ? MA, r2 = n-butane ? COx, r3 = MA ? COx. kB1 and kB2 refer to high and low energy site reaction pathways, respectively.
Table 4
6-Parameter model of Escardino et al. at 0.4 and 11.4 h.
Time on stream (h)
Ln [A1,613] (sꢃ1
)
Ln [A2,613] (sꢃ1
)
Ln [A3,613] (sꢃ1
)
Ea1 (kJ molꢃ1
)
Ea2 (kJ molꢃ1
)
Ea3 (kJ molꢃ1
)
R2
Obj. function
0.4
11.4
0.4 0.1
0.1 0.1
ꢃ1.1 0.2
ꢃ1.4 0.1
ꢃ2.8 1.2
ꢃ7.4 5.1
71.3 7.0
74.7 7.23
95.8 9.4
93.5 8.1
81.5 64.8
268.5 258.9
0.99
0.99
ꢃ435
ꢃ462

S.K. Wilkinson et al. / Journal of Catalysis 299 (2013) 249–260
257
profiles and times on stream). This has allowed evolution of VPO
performance to be described by changes in Ai for each of the reac-
tion pathways, with time on stream.
A
The evolution of Ai can now be defined via activity–time mod-
els. Firstly, the principle of separability is invoked [38]:
riðtÞ ¼ f1ðC; TÞ ꢁ f2ðtÞ
ð6Þ
Eq. (6) makes the assumption that the fundamental reaction kinet-
ics (f1) on the catalyst are time-independent and can therefore be
separated from catalyst activity (f2), which describes current activ-
ity relative to a reference activity and is time-dependent. This is a
valid approach as VPO behaviour across the entire data set is de-
scribed by the same kinetic model. Activity evolution of VPO can
be ascribed to changes in the pre-exponential factors of each reac-
tion pathway. Hence, VPO evolution can be modelled using the fol-
lowing, normalised activity measure, ai(t), and will represent the
f2(t) term in Eq. (6):
B
AiðtÞ
aiðtÞ ¼
ð7Þ
AiðinitialÞ
Fig. 11 plots activity evolution for each of the three reaction path-
ways, based on pre-exponential factor values generated from ki-
netic modelling. For each pathway, activity is lost over time;
however, there are clear differences between the n-butane oxida-
tion pathways and the MA oxidation pathway. A large amount of
MA oxidation activity is quickly lost in comparison with the n-bu-
tane oxidation pathways. This finding validates the intrinsic selec-
tivity evolution of the VPO catalyst described in Section 3.1.
Intrinsic selectivity towards MA has improved due to a sharp loss
of MA oxidation sites on the catalyst.
Fig. 10. Fitted parameters against time on stream for 5-parameter model using
pseudo-first-order relationships. (a) Activation energies (b) re-parameterised pre-
exponential factors. Error bars denote 95% confidence intervals.
Both n-butane oxidation pathways have a strikingly similar
evolution pattern. suggesting that both reactions are taking place
at the same active site.
A number of activity–time models exist in literature to describe
evolution of catalyst performance. The most appropriate model for
each reaction pathway was a simple first-order activity–time rela-
tionship with residual activity [39]:
these parameters are large. Similarly, Ea3 is seen to drift signifi-
cantly with time on stream which is not feasible in a correct model.
A logical step to overcome the observed cross-correlation was
to reduce the number of parameters in the model to 5. This was
achieved by assuming that Ea for both n-butane ? COx and
MA ? COx are the same.
aiðtÞ ¼ ai;QSS þ ð1 ꢃ ai;QSSÞ ꢁ expꢃw t
ð8Þ
i
Fitting a five parameter model resulted in only a negligible
reduction in R2 and objective function values and, as Fig. 10 attests,
improves the quality of the parameter estimations. In Fig. 10a, the
estimations for Ea1 and Ea2 remain roughly the same when plotted
against time on stream. Meanwhile, in Fig. 10b, each pre-exponen-
tial factor value decays with time. The 95% confidence interval for
A3,613 is significantly improved.
In Eq. (8), aQSS describes the residual catalyst activity at QSS opera-
tion, and is a rate constant of activity evolution. The initial model
for activity evolution of reaction pathways in the system contained
six parameters (three aQSS and three values, respectively). A
methodology described by Quiney and Schuurman [40] which as-
sesses the statistical significance of parameters within a system
w
w
Parity plots for prediction of n-butane, MA and COx concentra-
tions suggest that a simple pseudo-first-order model predicts
VPO performance during the conditioning period extremely well.
These can be found in Supplementary information. Measured
residuals within the model for n-butane, MA and COx prediction
are small and do not follow a systematic pattern, such as increasing
with temperature or axial reactor position.
The models of Buchanan and Sundaresan [32] and Sharma and
Cresswell [35] were also tested and found to be over-parameter-
ised. In both cases, the models could be reduced down to the afore-
mentioned pseudo-first-order network. The model of Schuurman
and Gleaves [17] resulted in poor quality of fit and wide confidence
intervals for the estimated parameter. Detailed discussion of all
three of these models can be found in Supplementary information.
4.3. Activity–time modelling
Fig. 11. Evolution of normalised pre-exponential factors (‘activity’) with time on
stream for the five parameter kinetic model. Lines denote fitted first-order activity–
time models.
The 5-parameter kinetic model accurately describes catalyst
behaviour for the entire data set (all temperatures, concentration

258
S.K. Wilkinson et al. / Journal of Catalysis 299 (2013) 249–260
Table 5
steady-state kinetic modelling approach. An evolution of active
sites for different reaction pathways can be investigated, and this
can be linked back to fundamental mechanisms. For example, the
almost identical evolution of both n-butane oxidation pathways,
suggesting both reactions occur at the same active surface, would
be simply missed in a steady-state investigation.
First-order activity–time model parameters for each reaction pathway.
Reaction
aQSS (–)
w
(hꢃ1
)
R2
n-Butane ? MA
n-Butane ? COx
MA ? COx
0.75 0.02
0.75 0.02
0.28 0.02
0.41 0.09
0.41 0.09
0.75 0.07
0.99
0.99
0.99
5.2. Critically assess impact of activation and conditioning periods
was used to scrutinise this model. Jacobean norm analysis (see Sup-
plementary information) revealed that the parameters for the n-
w
In this work, the systematic investigation of a pre-activated VPO
catalyst under n-butane/air reaction conditions has allowed cata-
lyst conditioning effects to be understood. During conditioning, it
is seen that a pre-activated VPO catalyst undergoes a subtle surface
rearrangement which impacts on the activity of three reaction
pathways. The loss of catalyst activity is largely temperature and
concentration gradient independent. Meanwhile, the catalyst bulk
remains largely unchanged in structure during the key periods of
performance evolution. This suggests that the catalyst surface
developed during prior activation is critical to the catalyst’s even-
tual QSS performance. The catalyst bulk should not be ignored
however as the phases developed within it during activation, and
their degree of crystallinity may provide indications of the nature
of the surface developed.
butane ? MA and n-butane ? COx pathways had a relatively low
influence on the quality of fit compared to the aQSS parameters. Fur-
thermore, the fitted aQSS values for both the aforementioned path-
ways were found to be the same, within their 95% confidence
intervals. As a result, both
w and aQSS values for both n-butane oxi-
dation pathways were equated, reducing the parameters in the
model to 4. Use of the global F-test, which considers the statistical
significance of parameter reduction (within 95% confidence),
showed this to be an acceptable move. The MA ? COx pathway is
distinct from the n-butane oxidation pathways exhibiting higher
w
and lower aQSS values than the latter. Both parameters in the
MA ? COx pathway are have similar norm of the Jacobean matrix
values, showing that both are significant to the model fit and cannot
be reduced further, unlike the n-butane pathways.
Returning to previous literature investigations of ex situ acti-
vated catalysts [11–13], each study used a different activation
method and ended up with different QSS performance levels. In
each case, however, the catalyst bulk chiefly comprised (VO)2P2O7
at QSS performance after many hours on stream, as was found in
this study. This suggests that (VO)2P2O7 is a key component of
QSS performance but not necessarily a marker of an effective work-
ing catalyst surface. This confirms some of the findings in [4].
Modelling results are shown 0069n Table 5 and graphically rep-
resented in Fig. 11. The fitting process for each reaction yielded a
high quality of fit (R2 > 0.99) and small 95% confidence intervals
for the fitting parameters. From the parameter reduction process,
the active site evolution network is reduced to two pathways,
one for n-butane oxidation and the other for MA oxidation. The val-
ues of
w suggest that activity evolution of the MA ? COx pathway
is the fastest, whilst the n-butane oxidation pathway evolves more
slowly.
A comprehensive kinetic and activity model fit across the data
set shows that the evolution of active sites for each reaction path-
way on the VPO catalyst is insensitive to both temperature and ax-
ial concentration gradients during the conditioning period (i.e. aQSS
5.3. Insight into reaction mechanisms
Evolution of active sites, summarised in Fig. 11, provides new
insight into the nature of the working surface of a VPO catalyst.
Both n-butane ? MA and n-butane ? COx pathways decline at
the same rate and to the same extent suggesting that they are asso-
ciated with the same active site; meanwhile, the MA ? COx path-
way declines faster and more sharply suggesting it is associated
with a different active site. This two-site model differs in basis to
the one proposed by Schuurman and Gleaves [17], which proposed
that n-butane oxidation occurs at either a low Ea (V5+ dominated)
or high Ea (V4+ dominated) site. Interestingly, the apparent Ea drift
which was observed in transient operation in a differential TAP
reactor is echoed in the 1st sector apparent Ea observations in this
work. This work, however, considered the full profile of an integral
mode reactor allowing clearer observations of an MA oxidation
route to be made and fitted to a triangular kinetic model.
and
w do not change as a function of process conditions). Hence, a
VPO catalyst conditioned at 350 °C under 2% n-butane/air reactor
inlet conditions will lose the same proportion of active sites at
the same rate as a VPO conditioned at 410 °C. This supports the
observations made in differential mode operation.
5. Discussion
In the spirit of the introduction, the findings are discussed with
reference to the current barriers in furthering understanding of the
effect of activation and conditioning methods on the structure and
intrinsic performance of solid catalysts, with a focus on VPO for
this model study:
Direct comparison with works which utilise separate oxidation
and reduction steps can also be afforded. In Lorences et al. [21], a
three site model is proposed: two different V5+ sites for n-buta-
ne ? MA and MA ? COx, respectively, and a VC4 site, a carbon me-
tal complex which interacts with gas-phase oxygen to form COx. In
the current work, activity–time models observe a 25% loss in n-bu-
tane oxidation sites and 75% loss in MA oxidation sites between
t = 0 h and QSS operation. In [21], pre-exponentials for the n-buta-
ne ? MA and n-butane ? COx pathways decline by 64% and 62%,
respectively, between transient and QSS observations, whilst
MA ? COx is more significant (85% loss). Whilst the active site ba-
sis of the current work is different, as well as the mode of reactor
operation, common ground can be drawn between the nature of
site evolution for each reaction pathway, even though in [21], the
extent of active site loss is greater.
5.1. Methodology development
Whilst previous work using TAP reactors [17] and TEOM setups
[18,19] has provided good insight into transient behaviour of VPO
catalysts under differential conditions, there was a clear gap in the
understanding of integral reactors displaying transient behaviour.
The parallel difference test has demonstrated itself as an effective
methodology for recording a full performance history of a VPO cat-
alyst during conditioning, providing a wealth of data for kinetic
modelling purposes. Extra investigations, such as restoring initial
n-butane conversion after a set time on stream in a parallel differ-
ence test, provides many insights into the drivers behind the per-
formance evolution of the VPO catalyst.
Investigating transient kinetics of a VPO catalyst during condi-
tioning has shown considerable advantages over the traditional
A number of density functional theory (DFT) works have dis-
cussed the importance of surface morphology and adsorbate con-

S.K. Wilkinson et al. / Journal of Catalysis 299 (2013) 249–260
259
figuration on the reactions taking place at the catalyst surface [41–
43]. A key conclusion reached in all these works is that the (100)
face of the (VO)2P2O7 phase is the active-selective surface for the
n-butane ? MA reaction. n-butane ? COx is less well understood
and has been attributed to other faces of (VO)2P2O7 such as
(001), (010), (021) and (010) [41]. As both n-butane oxidation
pathways are declining at the same rate from the kinetic analysis
of the current work, this questions the argument that alternative
faces are responsible for the total oxidation pathways. An
alternative argument may be that selectivity of n-butane ? MA
versus n-butane ? COx may be a function of the configuration in
which the n-butane adsorbs to the (VO)2P2O7 surface. A less
favourable mode of adsorption in this case may lead to C-C
breakage (via attack from electrophilic oxygen species [6]) and
subsequent formation of COx.
Further works which support the single n-butane oxidation site
include that of Herrmann et al. [44], who investigated links be-
tween VPO activity and its p-type semiconducting properties.
The n-butane activation step (C–H cleavage) is hypothesised to oc-
cur at a V5+–O2ꢃ site (in equilibrium with V4+–Oꢃ). Such a coupling
will polarise an n-butane molecule leading to C–H cleavage. How-
ever, the presence of electrophilic oxygen, Oꢃ, can also lead to COx
formation and could occur at the same site. It may be that such the
V5+–O2ꢃ site is stabilised by the (100) face of (VO)2P2O7 [45].
XPS results suggest that there are some significant changes to
the phases present at the surface occurring during the first 5 h of
operation, and that the surface Vox state drops from +4.2 to +4.1
during this time. Whilst tentative, some correlation could be affor-
ded with the decline of the MA ? COx pathway, which loses a
majority of its active sites during this time. Previous works have
proposed that whilst (VO)2P2O7 is the key active phase, nearby
V5+ phases at stacking fault boundaries (surface defects) are critical
to supplying the oxygen needed to transform n-butane ? MA
[46,47]. Similarly, other works have suggested that specific V5+
phases present at the VPO surface before reaction condition expo-
sure are replaced by different V5+ phases upon re-oxidation
[48,49]. The loss in VPO activity observed may be attributed to
healing of surface defects or replacement of more active V5+ phases
with less active ones during re-oxidation steps. An understanding
of the relative quantities and evolution of different V5+ phases in
the catalyst could constitute useful further work, such as carrying
out 31P NMR spectroscopy akin to the work of Abon et al. [14]. This
may show if particular phases are correlated with the MA ? COx
reaction. Such work could eventually be compared to previous ref-
erence phase kinetics studies [8].
ference testing), kinetic modelling and supporting characterisation
has allowed a systematic investigation to be carried out upon the
conditioning period. This methodology has demonstrated, using a
model catalyst system, how phenomena occurring during condi-
tioning can be understood. In a wider sense, the techniques em-
ployed could be applied to the conditioning step of a wider range
of differently activated catalysts. This could shed light on the im-
pact of different activation strategies behind forming different
types of active sites.
This work has demonstrated that a VPO catalyst bed operating
in integral mode under industrially relevant conditions (2% n-bu-
tane/air) undergoes a time-dependent transition during condition-
ing. The changes involved are shown, via kinetic modelling to
involve a sharp loss in active sites attributed to MA ? COx oxida-
tion. Meanwhile, a separate active site, attributed to the oxidation
of n-butane to both MA and COx, is seen to decay in activity much
less severely. The physical and/or chemical changes driving these
active site decays appear to be subtle. XPS suggests a decline in
surface Vox and a change in the nature of the surface phases pres-
ent, in particular during the first 5 h on stream. This may relate
to the sharp decay in MA ? COx sites observed over this time
frame. The result is an intrinsically less active catalyst surface at
QSS performance in comparison with initial performance, but it
is also intrinsically more selective to n-butane oxidation rather
than MA oxidation.
The evolution of VPO performance towards QSS performance
during conditioning is described by a single set of activity–time
models. Within the bounds of the study, this evolution is inde-
pendent of temperature and reactor axial position (concentration
gradient effects). This suggests that the final working surface of
the VPO catalyst at QSS performance is largely dictated by cata-
lyst precursor preparation and activation methods, rather than
the nature of the reaction conditions exposed to it during
conditioning.
Acknowledgments
The authors gratefully acknowledge work carried out by Dr.
Richard A.P. Smith and Malcolm Kett (both of Johnson Matthey).
S.K.W. is supported by an Engineering Doctorate Studentship
sponsored by the Engineering and Physical Sciences Research
Council (EPSRC) and Johnson Matthey in the Industrial Doctoral
Centre in Formulation Engineering, School of Chemical Engineer-
ing, University of Birmingham.
Previous DFT works have discussed the interaction of MA with a
(VO)2P2O7 site [42], finding that orientation of a molecule when
approaching the surface would be critical to its eventual degrada-
tion to COx. The current work, however, has shown that MA ? COx
activity evolves differently to n-butane oxidation pathways and so
cannot be completely associated with the same (VO)2P2O7 active
site. Unfavourably, distributed V5+ ensembles at the VPO surface
during the initial few hours of operation may present an alterna-
tive active site argument for MA ? COx. A previous work [8] sys-
tematically investigated the MA ? COx oxidation behaviour of
Appendix A
Symbol
Description
activity
SI Units
–
a
A
C
Ea
k
n
r
pre-exponential factor sꢃ1 (for 1st order)
concentration
activation energy
rate constant
reaction order
rate of reaction
mol mꢃ3
kJ molꢃ1
sꢃ1 (for 1st order)
–
different VPO reference phases, including (VO)2P2O7 and
All were capable of carrying out this reaction, and the specific
activities of -VOPO4 and b-VOPO4 towards this pathway were
a-VOPO4.
mol mꢃ3 sꢃ1 (for
intrinsic rates unless
noted)
a
higher than that for (VO)2P2O7.
R
universal gas constant J Kꢃ1 molꢃ1
sum of square of
residuals
time on stream
temperature
R2
–
6. Conclusions
t
T
h
K
The impact of activation and conditioning methods on many
catalyst systems is a largely unresolved topic. In this work, a novel
combination of catalyst testing methods (particularly parallel dif-
(continued on next page)

260
S.K. Wilkinson et al. / Journal of Catalysis 299 (2013) 249–260
References
Appendix A (continued)
Symbol
Description
SI Units
[1] J.J. Birtill, Catal. Today 81 (2003) 531.
[2] J.C. Burnett, R.A. Keppel, W.D. Robinson, Catal. Today 1 (1987) 537.
[3] B.K. Hodnett, Catal. Rev. Sci. Eng. 27 (3) (1985) 373.
[4] E.A. Lombardo, C.A. Sánchez, L.M. Cornaglia, Catal. Today 15 (1992) 407.
[5] V.V. Guliants, J.B. Benziger, S. Sundaresan, I.E. Wachs, J.M. Jehng, J.E. Roberts,
Catal. Today 28 (1996) 275.
[6] G. Centi, F. Cavani, F. Trifiro, Selective Oxidation by Heterogeneous Catalysis,
Kluwer Academic/Plenum, 2001.
[7] J. Ziolowski, E. Bordes, P. Courtine, J. Catal. 122 (1990) 126.
[8] Y. Zhang-Lin, M. Forissier, R.P. Sneeden, J.C. Védrine, J.C. Volta, J. Catal. 145
(1994) 256.
[9] Y. Zhang-Lin, M. Forissier, J.C. Védrine, J.C. Volta, J. Catal. 145 (1994) 267.
[10] F. Cavani, D. De Santi, S. Luciani, A. Löfberg, E. Bordes-Richard, C. Cortelli, R.
Leanza, App. Catal. A: Gen. 376 (2010) 66.
[11] S. Albonetti, F. Cavani, F. Trifiro, P. Venturoli, G. Calestani, M. Lopez-Granados,
J.L.G. Fierro, J. Catal. 160 (1996) 52.
[12] G. Bergeret, M. David, J.P. Broyer, J.C. Volta, Catal. Today 1 (1987) 37.
[13] V.V. Guliants, J.B. Benziger, S. Sundaresan, N. Yao, I.E. Wachs, Catal. Lett. 32
(1995) 379.
[14] M. Abon, K.E. Bere, A. Tuel, P. Delichere, J. Catal. 156 (1995) 28.
[15] K.C. Waugh, Y.-H. Taufiq-Yap, Catal. Today 81 (2003) 215.
[16] Y. Schuurman, J.T. Gleaves, Ind. Eng. Chem. Res. 33 (1994) 2935.
[17] Y. Schuurman, J.T. Gleaves, Catal. Today 33 (1997) 25.
[18] D. Wang, H.K. Kung, M.A. Barteau, App. Catal. A: Gen. 201 (2000) 203.
[19] D. Wang, M.A. Barteau, J. Catal. 197 (2001) 17.
Greek letters
w
activity evolution rate sꢃ1
constant
Subscripts
g
i
gas
of component, i
of component, i at QSS
performance
observed
oxidation state
At time on stream
i,QSS
obs
ox
t
Acronym
BET
Brunauer Emmett
Teller
carbon oxides
density functional
theory
infra-red
maleic anhydride
(C4H2O3)
COx
DFT
[20] P.L. Mills, H.T. Randall, J.S. McCracken, Chem. Eng. Sci. 54 (1999) 370.
[21] M.J. Lorences, G.S. Patience, F.V. Diez, J. Coca, App. Catal. A: Gen. 263 (2) (2004)
193.
[22] J.T. Wrobelski, J.W. Edwards, C.R. Graham, R.A. Keppel, H. Raffelson (to
Monsanto Company), U.S. patent 4,567,158, 1986.
IR
MA
[23] J.R. Ebner, W.J. Andrews (to Monsanto Company), U.S. Patent 5,137,860, 1992.
[24] J. Perez-Ramirez, R.J. Berger, G. Mul, F. Kapteijn, J.A. Moulijn, Catal. Today 60
(2000) 93.
[25] C.F. Chu, K.M. Ng, AIChE J. 35 (1989) 148.
[26] A. Ergun, Chem. Eng. Prog. 48 (1952) 89.
[27] G. Busca, F. Cavani, G. Centi, F. Trifiro, J. Catal. 99 (1986) 400.
[28] Athena Visual Workbench 14.2, Stewart & Associates Engineering Software,
Inc.
[29] G.W. Coulston, E.A. Thompson, N. Herron, J. Catal. 163 (1996) 122.
[30] B.K. Hodnett, P. Permanne, B. Delmon, Appl. Catal. 6 (1983) 231.
[31] A. Escardino, C. Sola, F. Ruiz, An. Quim. 69 (1973) 1157.
[32] J.S. Buchanan, S. Sundaresan, Appa. Catal. 19 (1985) 65.
[33] P. Schneider, G. Emig, H. Hofmann, Ind. Eng. Chem. Res. 26 (1987) 2236.
[34] S.K. Bej, M.S. Rao, (3 parts), Ind. Eng. Chem. Res 30 (1991) 1819.
[35] R.K. Sharma, D.L. Cresswell, Ind. Eng. Chem. Res. 30 (1991) 1819.
[36] K.G. Golbig, J. Werther, Chem. Eng. Sci. 52 (1997) 583.
[37] B. Schiott, K.A. Jorgensen, Catal. Today 16 (1993) 79.
[38] O. Levenspiel, Wiley, New York, 1972 (Chapter 15).
[39] G.A. Fuentes, Appl. Catal. 15 (1985) 33.
MvK
NMR
Mars van–Krevelen
mechanism
nuclear magnetic
resonance
quasi-steady state
scanning electron
microscopy
Temporal Analysis of
Products
Tapered Element
Oscillating
Microbalance
vanadium phosphorus
oxide
X-ray photoelectron
spectroscopy
X-ray diffraction
QSS
SEM
TAP
TEOM
VPO
XPS
XRD
[40] A.S. Quiney, Y. Schuurman, Chem. Eng. Sci. 62 (18–20) (2007) 5026.
[41] P. Courtine, E. Bordes, Appl. Catal. A: Gen. 157 (1997) 45.
[42] D.J. Thompson, M.O. Fanning, B.K. Hodnett, J. Mol. Catal. A: Chem. 198 (2003)
125.
[43] D.J. Thompson, I.M. Ciobîca, B.K. Hodnett, R.A. van Santen, M.O. Fanning, Catal.
Today 91–92 (2004) 177.
[44] J.-M. Herrmann, P. Vernoux, K.E. Bere, M. Abon, J. Catal. 167 (1997) 106.
[45] N. Duvauchelle, E. Bordes, Catal. Lett. 57 (1999) 81.
[46] P.T. Nguyen, A.W. Sleight, N. Roberts, W.W. Warren, J. Solid State Chem. 122
(1996) 138.
[47] A. Haras, M. Witko, D.R. Salahub, H.A. Duarte, Surf. Sci. 538 (2003) 160.
[48] E. Bordes, J.W. Johnson, P. Courtine, J. Solid State Chem. 55 (1984) 270.
[49] J.R. Ebner, M.R. Thompson, in: R.K. Grasselli, A.W. Sleight (Eds.), Stud. Surf. Sci.
Catal., vol. 67, 1991, p. 31.
Appendix B. Supplementary data
Supplementary data associated with this article can be found, in
the online version, at http://dx.doi.org/10.1016/j.jcat.2012.11.027.