You are on page 1of 14

Applied Catalysis A: General 186 (1999) 27–40

Fischer–Tropsch technology — from active site to commercial process


J.J.C. Geerlings a,∗ , J.H. Wilson a , G.J. Kramer a , H.P.C.E. Kuipers a , A. Hoek a , H.M. Huisman b
a Shell International Oil Products B.V., Shell Research and Technology Centre, P.O. Box 38000, 1030 BN Amsterdam, The Netherlands
bShell International Chemicals B.V., Shell Research and Technology Centre, P.O. Box 38000, 1030 BN Amsterdam, The Netherlands

Abstract
An overview of current research at Shell in the area of Fischer–Tropsch technology is given covering the full range from a
detailed understanding at the atomic level right up to the commercial process. Studies of model catalysts under Fischer–Tropsch
reaction conditions are reviewed, which have yielded new information on the nature of the active site. A recently developed
kinetic network is described which couples these model catalyst studies to the macroscopic world of ‘real catalysts’. Finally,
the advantages and disadvantages of various suitable reactor types will be discussed, with the emphasis placed on a comparison
between multitubular fixed bed and slurry type reactors. ©1999 Elsevier Science B.V. All rights reserved.
Keywords: Fischer–Tropsch reaction; Synthesis gas

1. Introduction the Fischer–Tropsch step should aim to utilise synthe-


sis gas as efficiently as possible. Selectivity consider-
Fischer–Tropsch technology forms the heart of ations are thus extremely important in the design of
many natural gas conversion processes which have the Fischer–Tropsch section of a gas conversion plant.
been developed by various companies in recent years For a plant producing only middle distillates, the C5+
[1]. In these processes, natural gas is first of all par- selectivity (i.e. the fraction of pentanes and heavier
tially oxidised to yield a mixture of carbon monoxide hydrocarbons in the total product) should be as high
and hydrogen, generally referred to as synthesis gas. as possible. If olefins or waxes are co-produced, then
The Fischer–Tropsch reaction forms the second step, their selectivities should be optimised simultaneously.
in which synthesis gas is converted into a mixture Most reviews of Fischer–Tropsch technology sep-
of mainly long straight chain paraffins with water as arate catalyst development and reactor technology.
a by-product. In the final step these hydrocarbons However, one of the main messages of this paper is
undergo further hydrogenation or hydrocracking to that to achieve an optimum in performance for the
yield clean middle distillates as products. complete process, the catalyst and the reactor should
It is well known that the economic viability of gas be optimised in a combined fashion. The first part
conversion is determined by capital costs and average of this paper is devoted to reviewing our fundamen-
product price. In this respect it should be realised that tal studies of model catalysts under Fischer–Tropsch
the manufacture of synthesis gas is by far the most reaction conditions. State-of-the-art surface science
capital intensive part of a gas conversion plant. Hence, techniques have been employed to yield information
on the relationship between the surface atomic struc-
∗ Corresponding author. ture of model catalysts and their performance in the

0926-860X/99/$ – see front matter ©1999 Elsevier Science B.V. All rights reserved.
PII: S 0 9 2 6 - 8 6 0 X ( 9 9 ) 0 0 1 6 2 - 3
28 J.J.C. Geerlings et al. / Applied Catalysis A: General 186 (1999) 27–40

Fischer–Tropsch reaction. These studies represent molecules from the gas phase can seriously influence
part of the scientific basis on which the development the atomic structure of metal surfaces, causing a range
of our Fischer–Tropsch catalysts is founded. Next, we of adsorbate-induced restructuring effects [5–7]. Most
switch from this microscopic picture to a macroscopic catalytic reactions take place at elevated temperatures
kinetic description of the Fischer–Tropsch reaction. and pressures. It is therefore, questionable whether the
This kinetic network bridges the gap between the results obtained in a UHV experiment can be usefully
world of model catalysts and the complex supported related to phenomena occurring under catalytic reac-
catalysts used in practice. We briefly review exist- tion conditions. This problem can be partially circum-
ing kinetic models of the Fischer–Tropsch reaction, vented by combining a UHV system with a high pres-
which are compared to our own approach. In the last sure reaction cell. In this approach, the model cata-
part of this paper, we deal with reactor technology. lyst is given a similar treatment to that encountered in
The advantages and disadvantages of various suitable practice and can subsequently be transferred back to
reactor types are discussed, with particular attention UHV for detailed surface analysis, without the need
paid to both slurry and fixed bed technologies. to expose the surface to ambient conditions. To make
the direct link between structure/composition and cat-
alytic performance, in-situ methods are obviously pre-
2. Basic research on the Fischer–Tropsch synthesis ferred. Unfortunately, there are only a few in-situ anal-
ysis techniques available which can be applied to char-
Heterogeneous catalysts for industrial applications acterise catalyst surfaces under high pressure and high
often consist of a porous inorganic material which is temperature conditions. One of the few in-situ tech-
used to support small metal particles. By dispersing niques available, which is especially useful in model
the metal in this way, a large total area of metal sur- catalyst studies, is polarisation modulation reflection
face is available at which catalytic reactions can oc- absorption infrared spectroscopy (PM-RAIRS) [8].
cur. For high catalytic conversions, a high metal par- We will describe below how cobalt surfaces be-
ticle dispersion is desirable, particularly for reactions have under Fischer–Tropsch reaction conditions. We
showing a low ‘turnover number’ (defined as the num- will give a summary of the main results, the details of
ber of reactant molecules converted per surface metal which can be found in the original publications [8–10].
site per second). It is also possible to influence the The experiments were performed in two experimental
state of the catalyst surface by the addition of separate set-ups, both of which combine a high-pressure reac-
promoter species. There are a large number of dif- tion cell with a UHV system equipped with a num-
ferent ‘recipes’ used for the preparation of heteroge- ber of techniques, including X-ray photoelectron spec-
neous catalysts. The inherent compositional and struc- troscopy (XPS) and facilities for sample preparation.
tural inhomogeneity of these supported systems makes One of the UHV systems is additionally equipped with
the problem of identifying the underlying reasons for a scanning tunneling microscope (STM).
their catalytic performance extremely difficult. Such A Co(0001) surface has been studied with in-situ
knowledge is essential for placing the development PM-RAIRS and ex-situ STM. The experiments clearly
and optimisation of catalysts for industrial processes show the dynamical structural nature of a model cat-
on a scientific, as distinct from empirical, basis [2]. alyst surface under practical CO hydrogenation reac-
In order to reduce the level of complexity, a com- tion conditions. It appears that the number of surface
mon approach is to study model catalysts such as defects is linked to the activity in the Fischer–Tropsch
single-crystal metal surfaces [3,4]. With the contin- reaction.
uing improvement of modern surface science tech-
niques, the preparation and characterisation of clean
and well-defined metal surfaces under ultra-high vac- 2.1. The restructuring of Co(0001) under
uum (UHV) conditions has now become a routine task. Fischer–Tropsch reaction conditions
Much effort over the past two decades has been de-
voted to the effect of exposing metal surfaces to differ- Fig. 1(a) shows a STM image of the atomically flat
ent molecules. It is now known that the adsorption of terraces of a clean Co(0001) surface separated by steps
J.J.C. Geerlings et al. / Applied Catalysis A: General 186 (1999) 27–40 29
30 J.J.C. Geerlings et al. / Applied Catalysis A: General 186 (1999) 27–40

Fig. 2. STM images of the restructured Co(0001) surface after 1 h exposure to high-pressure CO hydrogenation conditions showing (a)
two distinct levels separated by monoatomic steps, and covered with small island features which exhibit a remarkably narrow lateral size
distribution with a mean diameter of 1.75 nm, and (b) straight step edges with parallel island-free zones clearly visible on the upper step
edge. The image in (a) is a plan view representation of the data shown in Fig. 1(b). Taken from Ref. [9].
J.J.C. Geerlings et al. / Applied Catalysis A: General 186 (1999) 27–40 31

of varying height. After exposoure to synthesis gas


(H2 /CO = 2) at a reaction temperature of 523 K and
a pressure of 4 bar the surface structure has changed
considerably. Fig. 1(b) shows the surface at a higher
magnification after 1 h of reaction. The surface now
appears covered with a high density of similarly sized
particles. The average particle diameter is 1.75 nm.
The same image is shown in plan view in Fig. 2(a).
The upper level is just one atomic step higher than
the lower level while, for a given layer, the particles
are of almost identical height. We have interpreted the
particles observed in STM as small islands of cobalt
atoms occupying the threefold hollow sites of the un-
derlying close packed terrace. From the mean island
diameter and the nearest neighbour distance for bulk
cobalt (0.25 nm), the fraction of all cobalt atoms in the
uppermost atomic layer occupying edge sites is calcu- Fig. 3. Sequence of PM-RAIR spectra taken under pure CO on
an annealed (solid line) and a defective Co(0001) surface (broken
lated to be around 50%.
line).
The STM images show that a surface restructuring
process has taken place as a result of the high pressure
and high temperature treatment. Similar island struc- with CO adsorbed at well-ordered terrace sites. In ad-
tures have not been seen on the clean surface follow- dition to this peak the defective surface shows a pro-
ing annealing. A close examination of the structure nounced shoulder at higher frequencies, which has
near steps gives an indication of the mechanism of been attributed to CO adsorbed on defect sites [8,10].
surface restructuring. First, we observe straight steps This qualitative picture remains the same when the
with a low kink density compared to the clean surface. CO pressure is increased to 100 mbar at room temper-
Second, parallel to the upper edges of these straight ature. For the defective surface most of the intensity
steps, we find a clear empty zone where cobalt islands now appears in the highest frequency peak, which has
do not occur. This can be clearly seen in Fig. 2(b), shifted to 2080 cm−1 . The main observation of this
for example. Both observations are consistent with an series of measurements is that this defect peak also
etch-regrowth restructuring mechanism, which does starts to develop in the spectrum taken from the an-
not necessarily require a net loss of cobalt atoms via nealed surface, after the temperature is increased at
the gas phase. In summary, the STM results provide constant CO pressure to 493 K. This observation is
direct evidence that the surface of Co(0001) under- in qualitative agreement with the STM results, which
goes a massive restructuring under Fischer–Tropsch revealed extensive roughening of the surface after a
reaction conditions. As a result of this restructuring similar treatment in synthesis gas.
process, a large fraction of the surface cobalt atoms is Fig. 4 shows another sequence of PM-RAIR spec-
occupying edge sites. tra taken from a sputtered Co(0001) surface. At
To get a better insight into the surface restructur- 298 K under 100 mbar of pure CO, the defect peak
ing process as observed by STM, the adsorption be- at 2080 cm−1 dominates the spectrum. No significant
haviour of CO on Co(0001) was further studied with change in the PM-RAIR spectrum is observed when
PM-RAIRS. Fig. 3 shows a number of PM-RAIR spec- an additional 200 mbar of hydrogen is added at 298 K.
tra as obtained after adsorption of CO on an annealed The rest of the sequence in Fig. 4 shows the effect
Co(0001) surface and a defective Co(0001) surface, of raising the temperature stepwise to 490 K under
respectively. The defective surface has been created synthesis gas (H2 /CO = 2). The sharp defect peak at
via argon bombardment without subsequent anneal- 2080 cm−1 shows a large decrease in intensity with in-
ing. The annealed surface shows only a one symmet- creasing temperature, accompanied by a small shift to
ric signal after 10 Langmuir CO, which is associated lower frequencies. At 490 K, the peak has totally dis-
32 J.J.C. Geerlings et al. / Applied Catalysis A: General 186 (1999) 27–40

In agreement with this interpretation ex-situ XPS re-


veals the presence of hydrocarbons on the surface. In
addition to these analytical results, kinetic measure-
ments under similar conditions showed that a sputtered
Co(0001) surface is approximately a factor 5–10 more
active than the annealed surface. Taken in combina-
tion with the STM, PM-RAIRS and XPS results, the
kinetic measurements constitute clear evidence that
defects are active sites for hydrocarbon synthesis in
the Fischer–Tropsch reaction, an observation that is of
obvious importance in the process of developing im-
proved Fischer–Tropsch catalysts.

3. A quantitative kinetic model for FT catalysis


on pure cobalt

The experimental results described above, which


clearly show that cobalt surfaces are dynamic under
Fig. 4. Sequence of PM-RAIR spectra taken on a sputtered
Co(0001) surface. The starting conditions were 100 mbar of CO Fischer–Tropsch reaction conditions, do not preclude
at 298 K; 200 mbar of H2 was then added and the temperature in- a macroscopic kinetic description of their steady-state
creased stepwise to 490 K. It can be seen that the absorption signal catalytic behaviour. Over the past 5 years, we have
due to CO attached to defect sites is removed in an irreversible actively sought to find an unambiguous description of
process.
Fischer–Tropsch kinetics, based on well defined model
experiments [11,12]. To this end, we have developed
appeared. The peak at about 2020–2060 cm−1 shows a differential reactor in which flat model catalysts can
an increase in intensity with increasing temperature be tested. We are now in a position to measure product
and is similarly accompanied by a shift to lower fre- distributions up to C20 from a 20 cm2 catalyst surface
quencies. Upon lowering the temperature to 298 K the at CO conversion levels less than 1% and at pressures
2080 cm−1 absorption signal does not appear again, a between 1 and 20 bar [13]. Such measurements, some
clear indication of the irreversibility of the process. of which will be presented in this section, are ideal
The behaviour of the defect peak at 2080 cm−1 is a for the development and testing of kinetic hypothe-
central result of the PM-RAIRS studies. The results of ses and conceptual models. Only by using experimen-
Figs. 3 and 4 clearly show that defects can be created tal data from such dedicated model experiments, one
by annealing a Co(0001) surface in a high-pressure can overcome the intrinsic difficulty that complicates
CO environment or, in higher concentrations, by di- any description of the FT process, namely the sub-
rectly roughening the surface through argon sputter- tle interplay between chemical kinetics and the phys-
ing. As can be seen in Fig. 4, heating in synthesis gas ical (transport) properties of product molecules rang-
instead of pure CO, does not lead to the creation of ing from methane to ‘polyethylene’.
a corresponding defect CO signal in RAIRS. Given In this section, we present the outline of a kinetic
the STM observation that surface defects are created model for the FT synthesis which has been fully quan-
during annealing of Co(0001) in synthesis gas, one tified based on kinetic measurements of cobalt foils.
is led to the conclusion that these defects sites are
apparently unable to adsorb CO to a measurable ex- 3.1. Elements of the kinetic model
tent under synthesis gas at 490 K. The interpretation
of these results is that these defect sites are occupied The general consensus is that the Fischer–Tropsch
by hydrocarbons under Fischer–Tropsch reaction con- reaction is a combination of the following elemen-
ditions and obviously not in a pure CO atmosphere. tary chemical processes. To begin with, chemisorbed
J.J.C. Geerlings et al. / Applied Catalysis A: General 186 (1999) 27–40 33

methyl species are formed by means of CO dissocia-


tion and stepwise addition of hydrogen atoms. These
methyl species can either undergo methanation or al-
ternatively act as initiators for chain growth. Chain
growth occurs via sequential addition of CH2 groups
while the growing alkyl chain remains chemisorbed
to the metal by means of a terminal methylene group.
At any point during this chain growth process ter-
mination may occur to give either an ␣-olefin (via
␤-hydrogen elimination) or an n-paraffin (via primary
␣-hydrogenation). The resulting paraffins are assumed
to be inert: the known occurrence of hydrogenolysis Fig. 5. Graphical illustration of the kinetic network.
reactions can be neglected under normal process con-
ditions. In contrast, ␣-olefins are able to re-enter the thickness δ, mass balance in the liquid phase requires
FT chain growth process by reversing the chain ter- that:
mination step (so-called ‘reinsertion’). Alternatively,
␣-olefins may undergo secondary hydrogenation. The d =
 ξ = = ξ
relevance of making a distinction between primary and δCn,l = −(kd + kd + h2 )Cn,l + kd Cn,g + kd ξn
dt
secondary hydrogenation will become clear shortly. (2)
For completeness we mention that a variety of initiat-
ing species has been proposed for the Fischer–Tropsch d  0  =
δCn.l = −kd Cn,l
0
+ ka Cn,g
0
+ h1 ξn + h2 Cn,l (3)
reaction; see for instance a recent example proposing dt
vinyl species [14]. In this paper we do not describe
experiments, yielding information on the initiation of where C0 denotes paraffin concentrations. Note that
the reaction. This subject will therefore, not be dis- wax can build up in the reactor (with a rate˙␦) since the
cussed any further. only outlet is via the gas phase. During steady state
A complete kinetic model should ultimately be operation, however, the liquid phase concentrations
based on a Langmuir–Hinshelwood description of the will remain constant. This implies that
elementary reaction steps, which includes a descrip-
tion of the overall rate dependence on reactant/product d 
δCn,l = δ̇Cn,l (4)
pressures and temperature 1 . Here, we restrict our- dt
selves to an ‘effective rate’ model. The ingredients of
this model are shown graphically in Fig. 5. Using the Finally, mass balance in the gas phase (for olefins and
notation as shown, mass balances for Cn species in the paraffins alike) gives:
different phases can be set up as follows. They read:  
V d 8
d Cn,g = − ka + Cn,g + kd Cn,l (5)
ξ ξ =
ξn = −(kd + h1 + g)ξn + gξn−1 + ka Cn,l (1) A dt A
dt
where V, A and 8 are the reactor volume, the catalyst
for the chemisorbed phase, where the surface concen-
(or liquid) surface area and the gas flow rate, respec-
tration of a chemisorbed species is denoted by ␰ and
tively. Details of the model and its analysis will be
the liquid (olefin) concentration by C= . For flat model
published elsewhere [16]. It suffices to say at this point
catalysts covered with a homogeneous liquid layer of
that the model can be solved analytically in the form
of recursive relations between the Cn and the Cn + 1
1 In the literature one finds two type of kinetic descriptions. The
species.
first one is of the type discussed here and most references in As described in Ref. [16] the gas–liquid equilibrium
this paper. The second type does contain Langmuir–Hinshelwood
adsorption terms but is in its predictions typically limited to the
can be quantified by a gas–liquid collision rate with
CO conversion. See e.g. the review by Wojciechowski [15]. sticking coefficient, S, given by:
34 J.J.C. Geerlings et al. / Applied Catalysis A: General 186 (1999) 27–40
r Table 1
kB T
ka = S which @ 220◦ C equals : Reactor and process parameters
2π M
Reactor volume V 10 ml
216
ka = √ (m/s) (6) Cobalt surface area A 40 cm2
nc Reaction temperature 493 K
Gas flow rate @ 1 bar 8 1 Nml/min
6◦ in combination with an Arrhenius-type desorption @ 18 bar 8 18 Nml/min
term. By fitting the data to the gas–liquid equilibrium Gas residence time 10 min
of pure C5 –C25 one finds: H2 : CO (syngas) ratio 2
CO conversion @ 1 bar 0.8%
@ 18 bar 0.5%
kd = kd0 exp(−Ed /kB T ) with
 0
kd = 290 (m/s)
@ 220◦ C : (7) reaction time scales. These reaction time scales are
Ed /kB = 230nc (K) quantified in the next section.
Note that the desorption energy is proportional to
the number of carbon atoms (nc ); note also that no 3.2. Quantification of Fischer–Tropsch kinetics at 1
distinction is made between paraffins and olefins. The and 18 bar syngas pressure
simple description of Eqs. (6) and (7) predicts the
exponential decrease in hydrocarbon vapour pressure We consider two experiments done sequentially: the
with carbon number and is accurate within 10%. first at 1 bar and the second at 18 bar synthesis gas
It has been pointed out in the literature [17] that pressure. Apart from the difference in total pressure,
this exponential dependence of the vapour pressure all other conditions were kept the same and are sum-
leads to a critical chain growth parameter, α c , of 0.7 marised in Table 1.
(≈ exp(−230/T)) below which the FT reaction is a The product spectra at both pressures are shown
dry process and above which heavy wax is inevitably in Fig. 6 and show the total yield (i.e. ␣-olefins and
present. Under certain experimental conditions (no- n-paraffins) as a function of carbon number, together
tably low pressure operation) the FT process on a Co with the n-paraffin to ␣-olefin (P/O) ratio. At 18 bar
foil too will run under dry conditions. During dry op- this is a complete description; at 1 bar a large num-
eration, hydrocarbon product molecules evaporate di- ber of isomers is also formed. These are neglected
rectly from the cobalt metal surface rather than from in the present discussion since they do not affect
the surface of a hydrocarbon liquid. This leads to an its outcome. The drawn lines in Fig. 6 are fits to
enhancement in the effective desorption energy Ed /KB . the model, presented above, with fixed rate con-
On the basis of experimental work by Brand et al. [18], stants for C3+ products (see Fig. 5). Note that for
we estimate Ed /KB for evaporation from a metal sur- C1 and C2 we have introduced additional parameters
face to be 680 K per CH2 -group, i.e. more than twice to describe their respective deviations from ’ideal’
as strong. It can be shown [16] that under conditions Anderson–Schulz–Flory behaviour for completeness
of dry operation, the effective desorption energy per purposes only. In the present discussion, we are only
CH2 -group can be anywhere between 230 and 680 K, interested in the fixed rate constants used to predict
depending on the hydrocarbon coverage of what can the C3+ product spectrum. The rate constants giving
then be best described as the hydrocarbon physisorp- the best description of the data as shown in Fig. 6 are
tion layer. summarised in Table 2.
The significance of an explicit quantification of the A total of four kinetic parameters only had to be
gas–liquid equilibrium lies in the fact that, in doing fitted. These are:
so, unnecessary freedom in the remaining parameters • The effective desorption energy per CH2 unit
is eliminated. In the language of surface science: the (Ed /kB ): This is used to obtain a correct slope for
hydrocarbon residence times at the catalyst surface the P/O ratio, which depends on the surface resi-
(or in the liquid) have been fixed in accordance with dence time of 1-olefin products and hence on the
experiment (pure-component vapour pressures) so that desorption energy. For the 1 bar experiments, we
they can be unambiguously compared with the various obtain a value markedly higher than that of a hy-
J.J.C. Geerlings et al. / Applied Catalysis A: General 186 (1999) 27–40 35

Fig. 6. Cn -yield of n-paraffins and a-olefins and the paraffin to olefin (P/O) ratio for a 40 cm2 cobalt foil, at 1 bar and 18 bar total pressures.
Process conditions are given in Table 1.

Table 2
that a de-chemisorbed olefin will re-enter the chain
Kinetic parameters required to fit the experimental data of Fig. 6a
growth process. This produces the characteristic
Quantity Unitb 1 bar 18 bar ‘change in slope’ in the Anderson–Schulz–Flory
Desorption energy Ed /kB K 410 230 (ASF) product distribution, usually referred to as
Liquid layer growth δ̇ d0 /s 0 0.096 ’non-ASF behaviour’. At 1 bar, the yield spectrum
Chain growth probability α∞ – 0.575 0.926 is essentially ASF-like so that only an upper limit
P
TOFCO g · ξn 1/s 0.0146 0.168 ξ
ξ for ka can be given.
De-chemisorption kd /g – 0.76 0.55 ξ
Primary hydrogenation h1 /g – <0.05 0.025 The other parameters, kd and δ̇, are not variables but
ξ
Chemisorption ka d0 /s <100 10 000 fixed by boundary conditions. In the former case, this
Secondary hydrogenation h2 d0 /s 400 1100 is the chain growth probability at high carbon num-
a Only the parameters in bold have to be varied to obtain the fits. bers:
This is the limiting value of the chain growth probability for large ξn g
carbon numbers. α∞ lim = ξ
(8)
b The unit of the liquid-to-surface parameters and of the liquid n→∞ ξn−1 g + h1 + kd (1 − Γ )
layer growth is (m/s). For convenience they have been expressed
in units d0 = 3.25 Å, corresponding to the linear dimensions of a where the -insertion probability, Γ , is given by:
CH2 unit or ‘monolayer’ thickness.
ξ
ka
drocarbon liquid, reflecting ‘dry’ operation under Γ = ξ
(9)
ka + h2 + kd (1 + (ka A/8)) + δ̇
these conditions. Indeed, the model does not pre-
dict any wax build-up in this case. At 18 bar, wax A and 8 are given in Table 1. In the limit of large car-
build-up does occur and the desorption energy is bon numbers, evaporation becomes negligibly small
correspondingly reduced to the liquid value. and 0 reaches a constant value determined by δ̇, which
• The primary hydrogenation rate (h1 ): This deter- itself has a fixed value determined by the condition
mines the P/O level at small carbon numbers. Since that the total liquid density must be identical to that
at 1 bar the P/O line is virtually straight, only an of a F–T liquid (675 kg/m3 ).
upper limit can be given.
• The secondary hydrogenation rate (h2 ): This deter- 3.3. Critical discussion of the kinetic parameters
mines the level of the P/O ratios. Basically, it is ad-
justed so that is equals the effective desorption rate While at first glance the above quantification appar-
for the carbon number at which P/O is unity. ently yields a satisfactory description of the FT pro-
ξ
• The olefin chemisorption rate (ka ). This deter- cess, some severe problems show up on a closer in-
mines the re-insertion probability, i.e. the chance spection of the high pressure data. In particular, the
36 J.J.C. Geerlings et al. / Applied Catalysis A: General 186 (1999) 27–40

sure and the second increases, the anticipated overall


pressure dependence is very weak.
We therefore, conclude that the observation of
non-ASF behaviour in the Fischer–Tropsch product
distribution does not necessarily imply that secondary
growth reactions of ␣-olefins are occurring to a sig-
nificant extent. This is an important conclusion to
reach, as these secondary growth reactions play a
central role in current models of C5+ selectivity in the
Fischer–Tropsch reaction. These models are reviewed
in the next section. In our view, the current analysis
shows that the simple kinetic model outlined above
does not yet contain all the essential elements required
Fig. 7. The effective desorption rate kd for the desorption energies to yield a proper description of the Fischer–Tropsch
(per CH2 unit) as appropriate for 1 bar (410 K) and 18 bar (230 K), reaction over cobalt catalysts.
respectively.

‘change in slope’ around C8 can only be explained 4. Strategies for optimal C5+ selectivity
ξ
by assuming unphysically high values for ka . This
can be understood as follows: a change in slope in In Section 1 of this paper, we emphasised the
an ASF plot must, within the context of the kinetic need for operating at high C5+ selectivities in the
model outlined above, necessarily reflect a change in Fischer–Tropsch step for the overall economics of gas
the re-insertion probability of the primarily produced conversion processes. To optimise the C5+ selectivity,
olefins. As seen from its expression (Eq. (9)), 0 is it is necessary to be in possession of an accurate pro-
chain-length dependent by dint of the chain-length de- cess model which relates selectivity to all the relevant
pendence of the desorption rate, kd . This in turn de- catalyst and process parameters. In the literature, such
pends – for given process conditions – solely on the a model has been published by Iglesia et al. [19–22].
effective desorption energy, which is determined by The model of Iglesia et al. not only has direct impli-
fitting the (experimental) P/O values. Fig. 7 shows the cations for the design of F–T catalysts, but also for
actual values of kd as a function of carbon number for the choice of the F–T reactor. While this model is in
the 1 and 18 bar cases. Clearly, in order for 0 to be many ways similar to the kinetic network outlined in
appreciably different from zero, thereby giving rise to Section 3, it differs in its explanation for the observed
the observed non-ASF behaviour of the 18 bar exper- chain length dependence of secondary reactions. We
ξ
iments, ka must be in excess of 103 d0 /s (c.f. Eq. (9)) briefly discuss its main elements here, the details of
ξ
Thus, the 18 bar value for ka required by our model which can be found elsewhere [19–22].
is more than 100 times larger than its upper limit at The model of Iglesia et al. provides a complete de-
1 bar. This is unexpected, indeed we believe it is un- scription of the Fischer–Tropsch reaction, in which
ξ
physical: quite generally, ka is given by secondary reactions such as the secondary growth of
␣-olefins discussed previously play a central role. One
ξ
ka ∝ (1 − θall )a × θHb (10) implication of this model is that, to obtain optimal
values for the C5+ selectivity, the rate of secondary
where (1 − θ all ) is the fraction of all chain growth hydrogenation reactions should be minimised whilst
sites which are occupied, and θ H is the chemisorbed the rate of re-adsorption of ␣-olefins should be max-
hydrogen coverage. The exponents a and b are con- imised. With the latter idea in mind, it is natural to
stants whose precise values will depend on the molec- suppose that the C5+ selectivity can be increased by
ular pathway of the chemisorption process. Since of restricting the removal of ␣-olefins from the pores of
the two constituent factors – free adsorption sites and the catalyst particles as far as is practically possible.
chemisorbed hydrogen – the first decreases with pres- It is argued that the presence of diffusional restric-
J.J.C. Geerlings et al. / Applied Catalysis A: General 186 (1999) 27–40 37

tions on the transport of ␣-olefins out of the catalyst


particles, can result in an increase in the ␣-olefin con-
centration inside the catalyst pores which leads to a
corresponding increase in the number of secondary
chain growth events. A detailed analysis reveals that
one structural parameter only (χ) actually suffices to
describe the C5+ selectivity of Fischer–Tropsch cata-
lysts. This parameter is given by

R02 εθ
χ= (11)
rp

where R0 and  are the radius and porosity of the cata-


lyst particle, respectively, rp is the average pore radius Fig. 8. Bed residence time effects on olefin and paraffin selectivity
of the catalyst support and θ is the density of active within C4 products for a Co/TiO2 catalyst. Experiments were
metal atoms per unit area. It is assumed that the active performed in a tubular packed bed reactor, catalyst pellet size
phase is distributed homogeneously over the support. 0.17 mm. Reaction conditions: p = 20 bar, H2 /CO = 2.05, 9.5–72%
CO conversion. The selectivity is reported on a carbon basis as
The authors show that the C5+ selectivity exhibits a
the percentage of the converted CO appearing as a given product.
maximum as a function of χ. This maximum is quali- After E. Iglesia, Applied Catalysis A: General 161(1997) p.69.
tatively explained as follows. For smaller values of χ
the residence times of ␣-olefins in the catalyst particles
are lower which results in a corresponding loss in C5+ with increasing bed residence time, without a corre-
selectivity. For larger values of χ, differences in the sponding increase in the n-butane selectivity. The in-
diffusion barriers for the reactants (rather than prod- terpretation given by Iglesia et al. is that 1-butene se-
ucts) lead to a significant depletion of CO inside the lectively disappears as a result of its re-attachment to
catalyst particles. This results in additional methane the catalyst surface and subsequent secondary growth.
production with an associated reduction in the C5+ Closer inspection of the data, however, reveals that
selectivity. the activity of the catalyst is not constant as the bed
The above model gives a satisfactory expla- residence time is varied. In Fig. 9 a measure of the
nation for many of the experimental observa- number of C4 molecules actually produced (the yield)
tions in the literature such as the occurrence of is plotted for two bed residence times. The interpre-
non-Anderson–Schulz–Flory product distributions tation of Fig. 9 is completely opposite to that of Fig.
[19–22]. Nevertheless, we believe this description to 8. 1-Butene is apparently hydrogenated without any
be incomplete. In the previous section we showed that evidence for re-adsorption and further growth.
a kinetic model similar to that of Iglesia et al. could in- In the next section, we show that our own pilot
deed be used to fit non-ASF product distributions, but plant experiments, which are fully representative for
we argued that the parameters required are unphysi- the commercial scale, also cannot be described by Eq.
cal. This conclusion is strictly speaking only valid for (11).
experiments with flat model catalysts, which are far
removed from a practical Fischer–Tropsch process.
However, it is also possible to provide a different 5. Fischer–Tropsch reactor technology
interpretation of some of the original data published
by Iglesia et al.. Fig. 8 [19–22] shows the olefin and The Fischer–Tropsch reaction is a very exothermic
paraffin selectivity within the C4 fraction as a function process and can only be operated in a relatively narrow
of bed residence time for a Co/TiO2 catalyst. The bed temperature range. This imposes stringent demands
residence time has been changed by varying the space on both heat removal and temperature control. Nowa-
velocity at otherwise constant process conditions. The days, three different reactor types have been applied
figure suggests that 1-butene is selectively consumed commercially or are considered suitable for commer-
38 J.J.C. Geerlings et al. / Applied Catalysis A: General 186 (1999) 27–40

Fig. 9. Reinterpretation of the data of Fig. 8 for two values of the Bed residence time: t = 2 s (9.5% CO conversion) and t = 12 s (72% CO
conversion). The quantity plotted is proportional to the number of C4 molecules actually produced (the yield). The difference in catalyst
activity for the two residence times has been taken into account (it can be deduced from Fig. 8 that the catalyst activity at t = 12 s is
roughly a factor 1.26 higher than at t = 2 s).

cial application. These reactors are the gas/solid flu- • easy scale-up, hence no expensive demonstration
idised bed reactor, the multitubular fixed bed reactor unit necessary
and the three phase slurry reactor. • no system needed for separation of catalyst and liq-
The (two phase) gas/solid fluidised bed reactor has uid product
been applied extensively by Sasol [23,24]. Today, two • no problems with catalyst attrition
versions of this reactor type are in operation, namely • larger catalyst loading of the reactor possible
the well-known circulating beds and the more recently Similarly, often quoted advantages of the slurry reactor
developed fixed fluidised bed reactor. The latter reac- are:
tor represents a significant improvement compared to • no intra-particle diffusion limitations due to use of
the circulating fluidised bed type both in terms of cost sufficiently small catalyst particles
and operability [23,24]. Both reactors use iron-based • good isothermal operation due to excellent heat
catalysts, which inevitably leads to a large side pro- transfer, both within the slurry as well as to the
duction of CO2 as a consequence of the water gas cooling system
shift reaction. Such inefficient use of expensive syn- • catalyst can be added and removed during operation,
thesis gas however, makes this reactor technology less leading to a larger availability of the reactor
well-suited for natural gas conversion processes. Of While the advantages of either reactor type stand
course, CO2 can always be separated from the product as such, they do not provide sufficient background to
stream and recycled back to the synthesis gas manu- prefer one reactor over the other. In principle, one
facturing plant, but only at very significant additional should compare the performances of optimised reac-
costs. tor/catalyst combinations. For example, if the lifetime
The multitubular fixed bed and three phase slurry of the catalyst is longer than the time between routine
reactors are both much better suited for application in mechanical inspections of the reactor, as is in fact the
natural gas conversion processes. In the remainder of case for the Shell catalyst, one of the advantages of
this section, we will focus on a comparison between the slurry reactor vanishes. Likewise, arguments such
these two reactor types. It is generally recognised as ‘higher catalyst loading’ or no ‘diffusion limita-
that the main advantages of the multitubular reactor tion’ are only meaningful when they lead to an over-
are: all increase in productivity and an associated reduc-
J.J.C. Geerlings et al. / Applied Catalysis A: General 186 (1999) 27–40 39

tion in the costs. A good measure for this, in the case


of the Fischer–Tropsch process, is the combination
of selectivity and volumetric productivity. The volu-
metric productivity determines the total reactor vol-
ume required, which directly relates to the costs of the
Fischer–Tropsch section.
Regarding this point some clear opinions can be
found in the literature. The model of Iglesia et al., for
example, reveals that for large particles, which are re-
quired in fixed bed operation, optimal values of the
parameter χ and thus of C5+ selectivity can only be
obtained for low values of the site density θ (see Eq.
(11)). As a consequence of the low site density, the
volumetric productivity will be low which makes the
multitubular reactor far more expensive than the slurry Fig. 10. Performance of the second generation fixed bed
reactor. A similar conclusion has been drawn recently Fischer–Tropsch catalyst as measured in a pilot plant under fully
representative conditions. Plotted is the liquid (C5+ ) selectivity as
by de Swart [25,26], in a modelling study of both re- a function of catalyst activity for two process conditions.
actor types. The outcome of this study was that the
production of 5000 t middle distillates per day would
require either 10 multitubular reactors (6 m diameter, 6. Summary
20 m high) or 4 slurry reactors (7.5 m diameter, 30 m
high). On this basis it was concluded that the capi- An overview of the research at Shell in the area
tal costs of the slurry reactors would be 60% lower of Fischer–Tropsch technology has been given, cover-
than the multitubular reactors option. The main reason ing the full range from the surface structure of model
identified for this difference between the two reactor catalysts up to the commercial process. Evidence has
types was that the heat transfer rate in the slurry reac- been presented that ’surface defects’ are active sites
tor was approximately a factor of 5 higher than in the for hydrocarbon synthesis. The work on model cata-
multitubular reactor. lysts has been used to quantify a kinetic model, which
At Shell we have a long standing experience in the can be subsequently applied to real catalyst systems.
application of multitubular fixed bed reactors for the Finally, the advantages and disadvantages of the mul-
Fischer–Tropsch synthesis. An illustration of the pos- titubular fixed bed and the slurry reactors have been
sibilities of this reactor type is given in Fig. 10, which discussed. It has been argued that only a comparison
shows the C5+ selectivity as a function of volumetric between optimised reactor/catalyst combinations is a
productivity as measured in a pilot plant under fully meaningful one.
representative conditions. These results show that the
catalyst limitations, which might limit the scale-up
potential of multitubular fixed bed technology for the Acknowledgements
reasons outlined above have been removed. It can for
instance be deduced from Fig. 10 that in the exam- We are grateful to M.M.G. Senden for his helpful
ple worked out by de Swart significantly less than 10 comments.
multitubulars suffice. We are therefore, confident that
our multitubular fixed bed technology is competitive
with current slurry technology. While we are in a po- References
sition to have a very clear view on the potential of
multitubular fixed bed reactors, we are also interested [1] D. Knott, Oil and Gas Journal 95 (1997) 16.
in benchmarking the potentials of other technologies [2] R. Schlögl, Angew. Chem. Int. Ed. Engl. 32 (1993) 381.
such as slurry. This will enable a balanced, long term [3] G. Ertl, Angew. Chem. Int. Ed. Engl. 29 (1990) 1219.
strategy in the gas conversion field to be established. [4] G.A. Somorjai, Acc. Chem. Res. 9 (1976) 248.
40 J.J.C. Geerlings et al. / Applied Catalysis A: General 186 (1999) 27–40

[5] J. Wintterlin, R.J. Behm, in: H.J. Guntheroth, R. [16] G.J. Kramer, J.H. Wilson, submitted for publication to Chem.
Wiesendanger (Eds.), Scanning Tunneling Microscopy I, Eng. Sci.
Springer, Berlin, 1992, and references therein. [17] S.T. Sie, J. Eilers, J.K. Minderhoud, Proc. 9th Int. Congr.
[6] X. Bao, J.V. Barth, G. Lehmpfuhl, R. Schuster, Y. Uchida, Catal., vol. 2, 1988, p. 743.
R. Schlögl, G. Ertl, Surf. Sci. 285 (1993) 14. [18] J.L. Br, M.V. Arena, A.A. Deckert, S.M. George, J. Chem.
[7] F. Besenbacher, C. Klink, E. Laegsgaard, L.P. Nielsen, L. Phys. 92 (1990) 5136.
Ruan, I. Stensgaard, J. Vac. Sci. Technol. A 11 (1993) 1637. [19] E. Iglesia, S.C. Reyes, R.J. Madon, J. Catal. 129 (1991) 238.
[8] G.A. Beitel, A. Laskov, H. Oosterbeek, E.W. Kuipers, J. Phys. [20] E. Iglesia, S.C. Reyes, R.J. Madon, S.L. Soled, Advances
Chem. 100 (1996) 12494. in Catalysis, vol. 39, Academic Press, New York, 1993, p.
[9] J.H. Wilson, C.P.M. de Groot, J. Phys. Chem. 99 (1995) 7860. 239.
[10] G.A. Beitel, C.P.M. de Groot, H. Oosterbeek, J.H. Wilson, J. [21] E. Iglesia, S.C. Reyes , S.L. Soled in: E.R. Becker, C.J. Pereira
Phys. Chem. B 101 (1997) 4035. (Eds.), Computer-Aided Design of Catalysts and Reactors,
[11] E.W. Kuipers, I.H. Vinkenburg, H. Oosterbeek, J. Catal. 152 Marcel Dekker, New York, 1993, p. 199
(1995) 137. [22] E. Iglesia, Appl. Catal. A 161 (1997) 59.
[12] E.W. Kuipers, C. Scheper, J.H. Wilson, I.H. Vinkenburg, H. [23] B. Jager, M.E. Dry, T. Shingles, A.P. Steynberg, Catal. Lett.
Oosterbeek, J. Catal. 158 (1996) 288. 7 (1990) 293.
[13] J.H. Wilson, C.P.M. de Groot, J.E. Holewijn, submitted for [24] B. Jager, R. Espinoza, Catal. Today 23 (1995) 17.
publication. [25] J.W.A. de Swart, R. Krishna, S.T. Sie, Stud. Surf. Sci. Catal.
[14] H.C. Long, P.M. Maitlis, J. Am. Chem. Soc. 115 (1993) 4417. 107 (1997) 213.
[15] B.W. Wojciechowski, Catal. Rev.-Sci. Eng. 30 (1988) 629. [26] J.W.A. de Swart, Thesis, University of Amsterdam, 1996.

You might also like