Professional Documents
Culture Documents
IMPORTANT NOTE: You are advised to consult the publisher's version (publisher's PDF) if you wish to cite from
it. Please check the document version below.
Document Version
Publisher's PDF, also known as Version of record
Publication date:
1999
Copyright
Other than for strictly personal use, it is not permitted to download or to forward/distribute the text or part of it without the consent of the
author(s) and/or copyright holder(s), unless the work is under an open content license (like Creative Commons).
Take-down policy
If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately
and investigate your claim.
Downloaded from the University of Groningen/UMCG research database (Pure): http://www.rug.nl/research/portal. For technical reasons the
number of authors shown on this cover page is limited to 10 maximum.
Proefschrift
door
ISBN: 90-367-1136-3
Abstract
4
abstract
O,h E,h
with k f,O , k f,E , and k iso (units: [mol/kgNi s]) the formation of constants of the
half-hydrogenated intermediate in the hydrogenation of cis-monoene and trans-
monoene and in the isomerization path, respectively. Kiso is the equilibrium constant
for the isomerization, KS/KM is the ratio of adsorption constants of saturated and
monoene components, mc is the catalyst load [kgNi/ml3], CO, CE, and CS are the
concentrations of methyl oleate (cis), methyl elaidate (trans), and methyl stearate,
respectively [mol/ml3], KH is the adsorption constant of hydrogen [MPa-1],
O,h E,h
and PH 2 [MPa] is the hydrogen pressure. The activation energies of KH k f,O , KH k f,E ,
iso
and KH k are 32.2, 28.1, and 47.2 kJ/mol, respectively.
5
abstract
as a function of time. Also, the hydrogenation and the cis–trans isomerization rate
constants observed for monounsaturated esters appeared to be relevant to double
unsaturated esters hydrogenation as well. The experimentally observed high selectivity
SI, defined as the ratio of the hydrogenation rates of double unsaturated FAME over
monounsaturated FAME could be explained quantitatively from strong adsorption of
double- relative to monounsaturated esters. Selectivities of 3.1 ≤ SI ≤ 5.2 were found
for 323 ≤ T ≤ 363 K. The model proposed also proved capable of describing literature
data of others equally well.
6
abstract
intercrystalline pores (<30% of the total pore volume) only. Moreover, De measured
under reaction conditions reflected the influence of the micropores, resulting in a 10-
fold lower value.
The development of a reaction mechanism, starting from basic reaction steps for
monoene and diene hydrogenation and isomerization, is a basis for the investigation of
more complex reaction mechanisms of polyunsaturated edible oils and fats. Also, the
presumed large influence of the intraparticle diffusion coefficient on the reaction
selectivity is now quantitatively supported by a wide series of experiments, both under
reaction and under inert conditions. These data enable a more selective hydrogenation
of edible oils and fats by the optimisation of the process conditions and in the
development of new catalysts.
7
table of contents
ABSTRACT 4
1. INTRODUCTION 12
1.1. Background 13
2. MONOENE KINETICS 18
2.1. Introduction 19
2.2. Theory 21
2.2.1. Hydrogenation Mechanism. 21
2.2.2. Rate equations. 22
2.2.3. Transport phenomena. 28
2.5. Conclusions 43
3. DIENE KINETICS 46
3.1. Introduction 47
3.2. Theory 49
3.5. Conclusions. 65
8
table of contents
4. INTRAPARTICLE DIFFUSION 68
4.1. Introduction 69
4.2. Theory 71
4.2.1. De from the Reaction Data. 71
4.2.2. De from HPLC data. 74
4.4. Results 79
4.4.1. Simulation of H2 or FAME/TAG Diffusion Limitation during Reaction. 79
4.4.2. Evaluation of De from FAME Hydrogenations. 80
4.4.3. Evaluation of De from TAG Hydrogenations. 81
4.4.4. Tracer Pulse Experiments: Test of Column Performance. 85
4.4.5. Intraparticle Diffusion under Nonreaction Conditions. 87
4.4.6. Comparison of De as measured by the different methods. 90
4.5. Conclusions 92
5. ADSORPTION 96
5.1. Introduction 97
5.2. Theory 99
5.2.1. Tracer Pulse Experiments. 99
5.2.2. Batch Adsorption Experiments. 99
5.3. Experimental Section 102
5.3.1. Tracer Pulse Experiments. 102
5.3.2. Batch Experiments. 102
6. CONCLUSIONS 112
9
table of contents
NOTATION 118
De katalysator 131
Verantwoording 137
Levensloop 138
List of publications
Jonker, G. H.; Veldsink, J. W.; Beenackers, A. A. C. M. Intrinsic kinetics of 9-
monoene fatty acid methyl ester hydrogenation over nickel-based catalysts. Ind. Eng.
Chem. Res. 1997, 36, 1567–1579 (Chapter 2).
Bouma, M. J.; Jonker, G. H.; Veldsink, J. W.; Beenackers, A. A. C. M. An
intrinsic kinetic model for the hydrogenation of double unsaturated fatty acid methyl
esters over nickel-based catalysts based on the Horiuti–Polanyi mechanism. Submitted
for publication (Chapter 3).
Jonker, G. H.; Veldsink, J. W.; Beenackers, A. A. C. M. Intraparticle diffusion
limitation in the hydrogenation of monounsaturated oils and their fatty acid methyl
esters. Ind. Eng. Chem. Res. 1998, 37, 4646–4656 (Chapter 4).
Jonker, G. H.; Hoffmann, A. C.; Beenackers, A. A. C. M. Classification
mechanism of the chute, a liquid phase removal of fines in the micron range from a
batch of particles. Powder Techn, 1997, 90, 251-258.
Jonker, G. H.; Veldsink, J. W.; Beenackers, A. A. C. M. Kinetics and mechanism
of oleate hydrogenation over nickel-based catalysts. Lecture, held at ‘Catalysis in
multiphase reactions’; December, 7-9, 1994; Lyon (France).
10
table of contents
Beoordelingscommissie:
Prof. dr. N.-H. Schöön (Chalmers university of technology)
Prof. dr. J. Teuben (RuG)
Prof. ir. J.H. Wesselingh (RuG)
11
1. Introduction
In the labyrinth of the porous catalyst particle, the edible oil molecules diffuse
through channels and adsorb on active sites where
they react with hydrogen to saturated components or
where they are converted to other isomers. The
catalysed process investigated in this thesis: the
hydrogenation of edible fats and oils, has been applied
for over 90 years on a large industrial scale, but the
understanding of the process is far from complete.
This thesis investigates the fundamentals of edible oil
hydrogenation on supported nickel catalysts, by
studying intrinsic surface kinetics, intraparticle
diffusion limitation, and adsorption. Chapter 1 is a
general introduction to the thesis and a definition of
the problem area to be covered by this thesis.
12
introduction
1.1. Background
An edible oil consists of a mixture of triacylglycerides (TAGs). These are triesters
of glycerol and three fatty acids, with typical chain
lengths of 18 carbon atoms and degrees of H O
unsaturation varying between 0 and 6. Typical fatty H C O C R1
acids are stearic (octadecanoic), oleic (cis-9-
octadecenoic), elaidic (trans-9-octadecenoic), and H C O C R2
O
linoleic (cis,cis-9,12-octadecenoic). To improve oil
features (odour, melting point), the hydrogenation is H C O C R3
H O
usually carried out in a three-phase semibatch
reactor over a porous nickel catalyst at 0.1–0.6 MPa Oil molecule; R = alkyd chain
and around 443 K (Coenen, 1976). Besides
saturation, also migration and isomerization of double bonds occur, the latter of which
is undesirable in food applications (Katan et al., 1994; Emken, 1994).
At industrially applied conditions, 373 < T < 473 K and 0.1 < PH2 < 1.0 MPa,
polyunsaturated fatty acids (in this thesis also referred to as polyenes) are
hydrogenated preferentially over the monounsaturated species (monoenes). This
selectivity increases with increasing temperature and decreases with hydrogen
pressure; unfortunately, this also applies to the production of trans isomers. The
kinetics of hydrogenation and isomerization of edible oils are poorly understood, and
general rate equations are not available (Albright, 1985). Because of their relevance to
industrial applications, particularly polyunsaturated oils were investigated aiming at
selectivity improvement, which resulted in apparent rate equations containing many fit
parameters. For the selection of the optimal process conditions, knowledge of the
fundamental process steps is desired (see for a recent review Veldsink et al., 1997).
In the elaborate literature about the hydrogenation of edible oils, the kinetics are
often assumed to be first order in the concentration of unsaturated bonds and based on
the degree of saturation: diene ≡ monoene ≡ saturated (Veldsink et al., 1997). The
application of Langmuir–Hinshelwood–Hougen–Watson (LHHW) kinetics is a more
fundamental approach, which consists of a rate determining reaction step (usually a
surface reaction step), combined with Langmuir adsorption of reactants and products
(Weller, 1992). Hashimoto et al. (1971), Gut et al. (1979), and Grau et al. (1986, 1988)
applied LHHW kinetics to oil hydrogenation, based on apparent kinetics. However, a
thorough investigation of the rate determining steps in edible oil hydrogenation is not
available from open literature.
13
chapter 1
A thorough investigation of the elementary reaction steps should start with the
kinetics of monoene hydrogenation to serve as a key component in the elaborate
reaction scheme of the polyene hydrogenation, because the latter implies a large
amount of various side products (Coenen, 1986). The cis–trans isomerization is
commonly explained from the formation of a half-hydrogenated surface intermediate
by its easy rotation at the surface, whereas migration occurs via the π-allyl
intermediate, the latter being mainly present at hydrogen pressures below industrially
applied values. In the range of 0.1 < PH2 < 1.0 MPa, the formation of the half-
hydrogenated intermediate, which is the key component of the well-known Horiuti–
Polanyi mechanism, forms the basis of the reaction scheme for the mono- and double-
unsaturated oil molecules (Van der Plank and Van Oosten, 1975).
Usually, different adsorption strengths of fatty acids on the catalyst surface are
mentioned as the explanation for the difference in reaction rates of different types of
unsaturated methyl esters. Coenen and Boerma (1968) and Heertje and Boerma (1971)
concluded, that if linoleic acid was present, hardly any of the monoenic acid in the
liquid bulk could adsorb on the catalyst surface. However, adsorption of unsaturated
oil molecules or their fatty acids has been scarcely investigated up to now (Albright,
1985; Veldsink et al., 1997). The Horiuti–Polanyi mechanism enables a
straightforward implementation of the competitive adsorption of diene relative to
monoene.
In the hydrogenation of edible fats and oils over porous nickel catalysts,
intraparticle mass transfer limitations not only reduce catalyst effectivities, but also
may change the product selectivities (Tsuto et al., 1978; Coenen, 1986; Colen et al.,
1988). Information from open literature on the possible role of intraparticle diffusion
limitation in oil hydrogenation is very limited (Veldsink et al., 1997). Preferably, the
intraparticle effective diffusion coefficient, De, should be determined under reaction
conditions, which requires a thorough knowledge of the intrinsic kinetics. Additional
data on De can be obtained from independent experiments in the absence of reaction
(Haynes, 1988).
Recent literature data showed the possibility of investigating the pore structure by
applying various experimental techniques for measuring De (Park and Kim, 1984;
Garc⊆a-Ocoa and Santos, 1994). De should be measured with different experimental
techniques: under inert (non-reaction) and reaction conditions, because the difference
in these values reflects the influence of the pore structure in terms of micro- and
macropores and their connectivity (McGreavy et al., 1994).
14
introduction
15
chapter 1
16
introduction
17
2. Monoene Kinetics∗
∗
Published as “Intrinsic kinetics of 9-monoeneic fatty acid methyl ester hydrogenation over nickel-
based catalysts”; by Jonker, G. H.; Veldsink, J. W.; Beenackers, A. A. C. M. Ind. Eng. Chem. Res.,
1997, 36, 1567–1579.
18
monoene kinetics
2.1. Introduction
This chapter considers the kinetics of hydrogenation of monoenes. These are
simpler to understand than those of polyenes because fewer side products are made,
which enables a more thorough investigation of the elementary reaction steps (e.g.,
Hashimoto et al., 1971; Heertje and Boerma, 1971; Bern et al, 1975b; Cordova and
Harriott, 1975; Susu et al., 1978; Tsuto et al., 1978; Gut et al., 1979; Coenen, 1986).
Also in monoene hydrogenation, undesired products are formed, due to migration and
cis–trans isomerization of double bonds. The cis–trans isomerization was explained
from the formation of a half-hydrogenated surface intermediate (Allen and Kiess,
1955; Hashimoto et al., 1971; Gut et al., 1979; Albright, 1985; Grau et al., 1988) and
its easy rotation at the surface (Dutton et al., 1968; Heertje et al., 1974; Van der Plank
and Van Oosten, 1975), whereas migration occurs via the π-allyl intermediate, the
latter of which is mainly present at low hydrogen pressure (Heertje et al., 1974; Van
der Plank and Van Oosten, 1975). In this study, we focus on the formation of the half-
hydrogenated intermediate only. It is the key component of the well-known Horiuti–
Polanyi mechanism (Horiuti and Polanyi, 1934). The different reactions that can occur
in the hydrogenation of
monoenes are shown in Figure V+VH SH V+VH
2.1. The two possible monoenes kf,O k
H kb,O kb ,E f,E
cis (O, oleic) and trans (E, H
k a,O H ka,E
elaidic) can both adsorb on the kh
O kd,O
O E kd,E E
catalyst (ka and kd); this is V+VH
k a,S
denoted by a bar above their S kd,S S
symbols. The adsorbed species
can react reversibly with a single Figure 2.1. Hydrogenation scheme of monounsaturated
hydrogen atom to form an FAME according to the Horiuti–Polanyi mechanism.
intermediate SH (kf and kb). This
intermediate can revert to the original reactant, it may isomerize or it may react further
with hydrogen to the saturated S (stearic), which will then desorb. The second
hydrogen insertion (kh) is practically irreversible; all other reactions are reversible.
Monoene hydrogenation kinetics are often assumed to be first order in the
concentration of unsaturated bonds (Coenen, 1960; Cousins and Feuge, 1960; Nielsen
et al., 1960; Stefanovic and Albright, 1969; Marangozis et al., 1977; Drozdowski and
Zajac, 1980; Stenberg and Schöön, 1985; Colen et al., 1988), but orders varying
between 0 and 1 have also been observed (Vandenheuvel, 1956; Hashimoto et al.,
1971; Bern et al., 1975b; Susu et al., 1978; Lidefelt et al., 1983). The application of
19
chapter 2
20
monoene kinetics
different sets of rate equations, each with a unique set of rate-determining steps. The
rate equations were subsequently evaluated for a wide range of experimental data
obtained with an industrial nickel-based catalyst (Pricat 9910). Bartlett's test was used
to compare the significance of the set of rate equations (Dumez et al., 1977; Graaf et
al., 1988a,b). All possible combinations were evaluated, including sets of rate
equations which are unlikely from a physical point of view, to test thoroughly our
statistical analysis.
2.2. Theory
2.2.1. Hydrogenation Mechanism. The reactions consist of isomerization
between O and E, and saturation of the double bonds to form S, see Figure 2.1. Table
2.1 shows the rate equations of the reaction steps of the Horiuti–Polanyi mechanism
(HP), and the corresponding equilibrium constants in terms of liquid concentrations
(denoted by C) and surface fractions (∑θi = 1, i = O,E,S,SH,V; θH + θ VH = 1).
A distinction can be made between adsorption steps (A1, B1, C1 and D) and
surface reaction steps (A2, B2, C2), see Table 2.1 (and Figure 2.1). The adsorption of
O and E at the surface (A1 and B1) is followed by (the first) insertion of hydrogen to
form the half-hydrogenated intermediate (A2 and B2). A second hydrogen insertion
(C2) yields a completely saturated product, which gives, after desorption (C1), S
(stearate, methyl octadecanoate). The adsorption of hydrogen may be controlled either
by the rate of adsorption on the surface (associative adsorption (D-a)) or by the
subsequent dissociation in hydrogen atoms (dissociative adsorption (D-d)).
Apart from the second hydrogen insertion, all reaction steps are reversible with
characteristic equilibrium constants. The second hydrogen insertion can be regarded as
an irreversible reaction, because dehydrogenation occurs above 670 K only (Feuge,
1955). The equilibrium constant for the O–E reaction is defined as:
CE KO KI,O
Kiso = =
CO KE KI,E ( 2.1)
The reported values vary between 2 and 4 (Feuge et al., 1951; Litchfield et al.,
1963; Gut et al., 1979; Grau et al., 1986). The equilibrium ratio is independent of
hydrogen pressure (see eq 2.1) and depends weakly on temperature, ∆Hiso = –4 kJ/mol
(Rogers et al., 1977). In our kinetic analysis each reaction step is allowed to either
control the overall conversion rate or to proceed close to equilibrium. The adsorption
of the double bonds of O and E on the supported nickel surface (A1 and B1) is usually
21
chapter 2
k f,E θ SHθ VH
B2 k f,Eθ Eθ H − k b,Eθ SHθ VH K I,E = =
k b,E θ Eθ H
C2 k hθSHθ H
2.2.2. Rate equations. To derive rate expressions, we divided the overall, integral
reaction network (Figure 2.1) into three distinct reactions as shown in Table 2.2 (Graaf
22
monoene kinetics
et al., 1988a,b). From these reactions of Table 2.2, only two are independent,
which restricts the choice of rate determining steps for the hydrogenation reactions.
For each model, the rate equations for the observed reaction rates of O and E are
composed of the contributions of two reaction paths
RO = RO,h + Riso (2.2)
RE = RE,h − Riso (2.3)
The RO,h is the reaction or conversion rate of O in the hydrogenation path of O to
S and RE,h of the path of E to S. Although the isomerization path shares reaction steps
with the hydrogenation paths (see Figure 2.1), the reaction rate of the isomerization
Riso is taken to be independent of the hydrogenation of O and E.
Further basic model assumptions are as follows:
1. The intermediates are in a pseudo steady state, which implies equal reaction
rates of elementary steps along one reaction path (Froment and Hosten, 1981; Graaf
1988; Graaf et al., 1988a,b).
2. The catalyst sites are uniform and homogeneously distributed (Marangozis et
al., 1977; Gut et al., 1979; Grau et al., 1986).
3. Two types of adsorption sites are used to account for independent adsorption of
hydrogen (VH) and FAME (V) (Marangozis et al., 1977; Gut et al., 1979; Grau et al.,
1986; Rodrigo et al., 1992). In the gas phase, with FAME concentrations of around 5%
of hydrogen, hydrogen and FAME reportedly compete for the same active nickel sites
(Lidefelt et al., 1983; Magnusson, 1987a,b), but in the liquid phase the FAME is
present in large excess and, owing to the large differences in molecular sizes, the small
hydrogen most likely adsorbs on sites between the large FAME molecules only
(Marangozis et al., 1977).
23
chapter 2
4. O, E, and S are assumed to compete for the same surface sites, with a negligible
amount of vacant sites (Gut et al., 1979).
5. The adsorption constant of O is considered to be equal to that of E, KO=KE≡KM
(Gut et al., 1979; Grau et al., 1986). KM stands for the adsorption constant of the
monoene.
6. The surface occupation of the half-hydrogenated intermediate is negligible
compared to that of bulk components (Zwicky and Gut, 1978; Graaf et al., 1988a,b). In
deriving LHHW-type rate equations, surface intermediates are usually neglected in the
(competitive) adsorption of bulk components for vacant surface sites (Weller, 1992). It
should be remarked that assumption 6 is only made after the rate equations are
derived, to reduce the number of parameters in the denominator of the Langmuir-type
equations.
7. Hydrogen adsorption is at equilibrium, relative to the hydrogen concentration in
the liquid close to the catalyst surface; see Table 2.1, D-a and D-d for definitions.
The mass balances of the mole fractions of FAME and hydrogen at the surface
thus read:
θ O + θE +θS = 1 (2.4)
θ H + θ VH = 1 (2.5)
A kinetic model consists of two rate-determining steps, one for each independent
reaction. The other steps are considered to be at equilibrium. Because each reaction
path involves four elementary steps, there are 16 possibilities. However, the
hydrogenation reactions O and E share identical reaction steps (C1 and C2), which
decreases the number of combinations from 16 to (2 x 2 + 2 =) 6 combinations or
models. These models are identified as A1B1, A1B2, A2B1, A2B2, C1C1 and C2C2;
the coding refers to the rate-determining reaction steps.
The overall rate equations are derived as follows. For model A1B1, the
elementary rate equations of the reaction paths are
24
monoene kinetics
The hydrogenation paths (RO,h and RE,h) and the isomerization path (Riso) are
25
chapter 2
independent paths, each with characteristic rate constants, which are in the case of the
O,h E,h iso
rate constant ka,O (eqs 2.6–2.8) denoted by k a,O , k a,O , and k a,O , respectively. The
equilibrium constants for elementary reactions A2, B2, C1 and C2 (the latter goes to
infinity) can be used to solve the unknown surface fractions from eqs 2.6–2.8. In the
final expressions for the rate equations, the equilibrium constant for the isomerization
reaction is substituted (eq 2.1).
For dissociative equilibrium hydrogen adsorption, the hydrogen fraction at the
surface is
KH PH 2 ( 2.9)
θH =
1 + KH PH 2
and for associative equilibrium adsorption
KH PH 2 ( 2.10)
θH =
1 + KH PH 2
If hydrogen adsorption plays a role in the model rate equations, the applied
hydrogen equilibrium adsorption is denoted by ass (associative) or diss (dissociative),
e.g., A2B2-ass and A2B2-diss. Table 2.3 summarizes the overall rate equations for O
and E for each of the six models.
The model C1C1 predicts an unrealistically high reaction rate, because the
irreversible reaction of the half hydrogenated intermediate to S is now 'close to
equilibrium', implying an infinite reaction rate. Therefore, this model not further
considered.
The independent variables (O, E, and S concentrations and hydrogen pressure) of
the remaining five models are preceded by groups of parameters of elementary steps.
Only the value of the group of parameters can be obtained. Therefore, the groups of
parameters are replaced by lumped parameters, for each model denoted by k1…k6.
Table 2.4 gives the rate equations which were used to fit the experimental results, by
adjusting the lumped parameters. Comparison of Tables 2.3 and Table 2.4 gives the
group of parameters of elementary reaction steps which are represented by the lumped
parameters.
The temperature dependency of the lumped parameters is assumed to obey an
Arrhenius relation:
26
monoene kinetics
−1 −1 −1 −1
A2B1 − ( k1 (1 + k6 K isoθ H ) + k2 )θ H CO + k2 Kisoθ HCE − ( k 3 (1 + k 6 K isoθ H ) + k 2 K isoθ H ) C E + k 2θ H C O
−1 −1 −1 −1
CE + (1 + k6 (1 + K iso )θ H ) CO + k 4 (1 + k6 Kisoθ H )CS C E + (1 + k 6 (1 + K iso )θ H )C O + k 4 (1 + k 6 K iso θ H )CS
Ea,i Tref − T
ki ( T ) = ki ,ref exp − i = 1,...,6 ( 2.11)
RgasTref T
Two literature models can be used to compare model A2B2 with associative
hydrogen adsorption: for the simplified hydrogenation scheme of Figure 2.2,
Hashimoto et al. (1971) and Gut et al. (1979)
(and also Zwicky and Gut, 1978) derived on the
basis of the LHHW approach rate equations E O
similar to model A2B2, with k3 equal to k1.
Hashimoto et al. (1971) assumed dissociative
and Gut et al. (1979) associative hydrogen
adsorption. The equations were tested on
Figure 2.2. Generalized hydrogenation
hydrogenations with TAGs over a nickel scheme of Gut et al. (1979).
catalyst (Hashimoto et al., 1971; Gut et al.,
1979). Grau et al. (1986) derived rate equations on the basis of the Horiuti–Polanyi
mechanism, which were evaluated using monoene hydrogenation over nickel catalysts,
but due to their calculation method, only the cis–trans isomerization equilibrium
constant is available to compare to our data.
27
chapter 2
( − Ri ) mNi ρ c d p2
Φi = i = O, E, H 2 ( 2.12 )
36mc De,i Ci
The criterion for the absence of mass-transfer limitation for unknown kinetics is
Φi < 0.03–0.7 (Westerterp et al., 1987). When the rate equation is known, the
intraparticle concentration profiles can be calculated from:
d 2 dCi ρ c mNi
De,i rp = Ri rp2 i = O,E, H 2 ( 2.13)
drp drp mc
The hydrogen uptake rate equals the sum of the oleate and elaidate reaction rate:
R H2 = − ( R O + RE ) (2.14)
The effectiveness factor for component i follows from the calculated flux to the
particle and the maximum possible reaction rate related to the bulk concentrations Cb
(Westerterp et al., 1987):
28
monoene kinetics
Ji ac
µi = i = O, E, H 2 (2.15)
Ri ( Cb,i )
In absence of diffusion limitations, the reaction rates for a dead end batch reactor
are:
dCb,i
= Ri i = O, E, H 2 ( 2.16)
dt
29
chapter 2
samples were stored in a 16-loop valve. Afterwards, their liquid composition (O, E, S)
was determined by gas–liquid chromatography (Hewlett Packard 5890, supplied with a
Chrompack CP-sil 88 column). The standard deviation of the fractions O, E, and S was
found to be 20% of the measured value at low mole fractions (0.0–0.1), 10% at
moderate fractions (0.1–0.3), and 5% at high fractions (0.3–1.0).
We performed two types of experiments:
a. Constant Hydrogen Pressure. In these experiments, the autoclave was
operated at constant hydrogen pressure by continuously supplying hydrogen from a
storage vessel. During the course of the reaction, up to 15 samples were automatically
taken from the reaction mixture and analyzed afterwards. Hence, the experimental data
consisted of mass fraction distributions of O, E, and S (Figure 2.4) together with the
hydrogen consumption as a function of time. Experiments were carried out for 333 ≤ T
≤ 413 K and 0.02 ≤ PH 2 ≤ 0.50 MPa. At partial hydrogen pressures below 0.2 MPa,
nitrogen was used to balance the total reactor pressure at 0.2 MPa, which was
necessary to collect samples.
1.0
1.5
0.8
Ci/Ctot
0.6 1.0
0.4
0.5
0.2
0.0 0.0
0 2 4 6 8 10 12
t [ks]
Figure 2.4. Liquid-phase composition and hydrogen uptake in a constant hydrogen pressure
experiment, as a function of reaction time for a typical oleate hydrogenation run. The hydrogen
uptake is equal to the amount of produced stearate. Conditions: 413 K, 0.5 MPa, 0.120 kgNi/ml3, l,
O; n, E; s, S.
b. Varying Hydrogen Pressure. Here, hydrogen was consumed from the head
space of the autoclave without additional supply from the storage bottle. During such
30
monoene kinetics
an experiment, the hydrogen pressure dropped from about 0.40 MPa to 0.01 MPa.
(Figure 2.5), while the conversion of FAME remained below 7%. As a result, the
liquid-phase composition remained approximately constant. These experiments were
0.6
PH2
0.4
[MPa]
0.2
0.0
0.0 0.2 0.4 0.6 0.8
t [ks]
Figure 2.5. Reactor pressure during a typical experiment with varying hydrogen pressure.
Conditions: 353 K, 0.177 kgNi/ml3.
31
chapter 2
sunflower, at 353 K and 0.50 MPa, yielding maximally 45% O (medium percentage of
O, MPO). As expected, the MPO and HPO mixtures (Nielsen et al., 1960; Edvarsson
and Irandoust, 1993) gave the same results in the conversion domain, where the
isomerization rate relative to the hydrogenation rate is considered only. Although of
most value for model evaluation, purified unsaturated FAMEs were not used because
these are very hard to obtain in sufficiently large quantities and chemically derived O
and E contain different amounts of minor compounds which influence the
hydrogenation.
The experimental series consist of HPO, HPE, and MPO experiments with both
constant and variable hydrogen pressure. The complete conversion range was analyzed
to derive rate equations which are applicable to a large range of reactant
concentrations. Table 2.6 summarizes the experimental series.
32
monoene kinetics
Intraparticle diffusion limitations for hydrogen and FAME were estimated from
the Weisz–Prater number (eq 2.12) and the initial hydrogenation rates. The hydrogen
concentration in the liquid was calculated using mH 2 = 0.1 mg3/ml3 at 393 K
(extrapolated from Ganguli and Van den Berg, 1978; see also Ganguli, 1975).
Intraparticle diffusion coefficients for the Pricat 9910 were taken as De,H2 = 2 × 10-9
m2/s (Andersson et al., 1974; Ganguli, 1975; Ganguli and Van den Berg, 1978; Tsuto
et al., 1978; Stenberg and Schöön, 1985) and De,O = De,E = 2 × 10-11 m2/s (Andersson
et al., 1974; Tsuto et al., 1978; Colen et al, 1988). Other values are: dp = 8.4 µm and ρc
= 264 kgNi/mc 3. In our experiments, the numerical value of the Weisz–Prater modules
always remained below 0.1. Therefore, we have strong evidence that no intraparticle
diffusion occurred in our experiments.
In each model evaluation, Kiso = 3.5 (T = 393 K). This value was obtained from
independent isomerization experiments of monoene FAME with selenium (Litchfield
et al., 1963). Other reported values of Kiso were determined from kinetic data (Feuge et
al., 1960; Stefanovic et al., 1969; Grau et al., 1986) and may contain ratios of kinetic
rate constants, see section 2.4.3. The temperature dependency is ∆Hiso = –4 kJ/mol
(Rogers et al., 1977).
The five models with associative or dissociative hydrogen adsorption were
subsequently evaluated in both the conversion and the time domain. To reduce the
number of parameters, first an optimization was performed on the ratios of rate
constants in the conversion domain, followed by a second optimization in the time
domain. Model selection was performed both by Bartlett's test and physical
interpretation of the model fits.
2.4.1. Conversion Domain. For parameter optimization in the conversion domain,
the rate equations were rewritten as a function of the double-bond conversion or the
degree of hydrogenation. Advantages of this method are as follows: (1) reduction of
the number of parameters, because the parameters are now obtained as a ratio to the
total hydrogenation rate constant, denoted by k2c (=k2/k1) and k3c (=k3/k1). For
example, the rate equations of model A1B1 in the conversion domain read
33
chapter 2
4
Table 2.8. Total Variances (Sh2 /10 ) for Models A1B2 and A2B1 a
model θH case 1 case 2 case 3
A1B2 ass 3.00 3.67 5.98
diss 1.43 1.55 3.72
A2B1 ass 3.96 3.92 18.0
diss 2.94 2.92 10.2
a
Cases 1, 2, 3: see Table 2.7. θH hydrogen adsorption equilibrium, which can be associative (ass)
or dissociative (diss). Total variance defined in Appendix eq 2.A3
34
monoene kinetics
a
Table 2.10: Bartlett's test for conversion domain
χ2 H=6 H=5 H=4 H=3 H=2
χt2 11.1 9.49 7.81 5.99 3.84
χc2: C,393 369 8.88 6.58 3.81 2.42
χc2: C,all 19.5 8.77 1.86 0.08
a
C,393, C,all, see Table 2.9. χc2: critical χ2, Bartlett's test (eq 2.A2). χt 2: tabulated χ2 (Fisher,
1970).
35
chapter 2
This way, the number of parameters in models A1B2 and A2B1 is reduced from 4
(k2c, k3c, k5 and k6) to 2: k2c/k6 and k3c. Optimization in the conversion domain gave a
first selection of models. Table 2.9 shows the total variances per model for all
optimisations in the conversion domain and their ranking. The total variances serve as
an input for Bartlett's test (Table 2.10). The H denotes the number of collated models,
in succeeding order: H = 6 are the best six of the models, H = 5 the best five, etc.
Bartlett's test (Appendix: eq 2.A2) compares χc 2 to the tabulated value χ2t (Fisher,
1970) for H - 1 degrees of freedom. The underlined values represent the corresponding
best H models, which do not differ significantly from the best model. The selected best
models are A1B2-diss, A2B1-diss, A1B2-ass, A2B1-ass, and A1B1, the latter which
also represents A2B2-ass and A2B2-diss because the rate equations are similar in the
conversion domain.
According to Bartlett's test for the first optimization step in the conversion
domain, at constant temperature, model C2C2 is significantly more inaccurate than the
other models (Table 2.9 and 2.10). Model C2C2 is unrealistic, because, it assumes that
all reactions in the isomerization path are at equilibrium and implies that the
isomerization rate should be instantaneous, which is not found experimentally (see
Figure 2.4). The other models are statistically indistinguishable, and therefore in the
first optimization step, only model C2C2 is removed from the list of models.
0.10
Cim- C ie 0.05
C tot
0.00
-0.05
-0.10
0.00 0.15 0.30 0.45 0.60
e
C i C tot
Figure 2.6. Residuals between the model and experiment for a fit in the conversion domain (whole
temperature range) for model A1B2 with associative hydrogen adsorption equilibrium; ¡, oleate;
¨, elaidate.
36
monoene kinetics
The second step was optimization of activation energies in the conversion domain
(using the whole temperature range). Now A1B2-diss, A1B1 (and therefore also
A2B2-ass and A2B2-diss), A2B1-ass, and A2B1-diss remained statistically
indistinguishable. The χ2 of model A1B2-ass proved to be significantly larger
compared to the other models (Table 2.9), which was confirmed by Bartlett's test
(Table 2.10). Bartlett's test is visualized in Figures 2.6 and 2.7, which show the large
deviations between experimental and predicted mole fractions of model A1B2-ass
(Figure 2.6) compared to A1B2-diss (Figure 2.7). So, model A1B2-ass was also
removed from the list of models.
0.10
Cim- C ie 0.05
C tot
0.00
-0.05
-0.10
0.00 0.15 0.30 0.45 0.60
e
C i C tot
Figure 2.7. Residuals between the model and experiment for a fit in the conversion domain (whole
temperature range) for model A1B2 with dissociative hydrogen adsorption equilibrium: ¡, oleate;
¨, elaidate.
2.4.2. Time domain. In the time domain, the remaining models A1B2-diss, A1B1,
A2B1-ass, A2B1-diss, A2B2-ass, and A2B2-diss were tested by estimation of the k1
and the adsorption parameters using the optimised values of k2c and k3c, found in the
conversion domain.
After an estimation of the value of the lumped rate constant, k1, at the reference
temperature, the three parameters Ea1, k1,ref, and k4 were simultaneously optimized
over the whole temperature range. The parameter k4, a ratio of adsorption constants,
appeared to be independent of T.
37
chapter 2
Bartlett's test (Table 2.12), based on the Sh2 of Table 2.11, restricted the possible
best models to A2B2-ass and A2B1-ass. In model A1B1, the hydrogen function is
absent, which is unrealistic in the time domain, and consequently model A1B1 was
rejected by Bartlett's test (H = 6, χc 2 = 33.4). Also rejected were the two models with a
dissociative hydrogen function: A2B2-diss and A2B1-diss.
1.14 < (1 + KH PH 2 ) < 1.23 for 411 < T < 511 K, 0.22 < PH 2 < 0.34 MPa (2.21)
38
monoene kinetics
dependence on hydrogen pressure for 0.25 < PH 2 < 0.55 MPa and T = 433 K. Only
Cordova and Hariott (1975) calculated for Pd-catalyzed methyl linoleate
hydrogenation an order of 0.5 in hydrogen for the methyl oleate part, but they
remarked that further data were needed to confirm the reaction order.
Associative hydrogen adsorption is, however, rather unexpected with regard to
data from vapor-phase hydrogenations. Lidefelt et al. (1983) hydrogenated methyl
oleate (0.015–0.11 kPa) on supported nickel catalyst at 421–487 K in excess of
hydrogen (0.35–5.6 kPa) and nitrogen (to balance to 0.1 MPa total pressure). The
observed order in hydrogen of 1.5 can be explained by dissociative hydrogen
adsorption in competition with adsorption of oleate at the surface (Lidefelt et al.,
1983). In the same equipment, Magnusson (1987a,b) found from H2/D2 exchange
experiments that hydrogen adsorbs in a Rideal–Eley mechanism, which also assumes
dissociation of the hydrogen molecules. But in the liquid phase, the CO/ CH2 ≈ 1 (gas
phase: CO/ CH2 ≈ 0.01) which apparently changes the adsorption behavior of hydrogen
from dissociative to associative.
A2B2-ass and A2B1-ass are indistinguishable according to Bartlett's test, but from
a physical point of view, model A2B2 is preferred to A2B1, because A2B1 assumes
both an adsorption step (B1) and an insertion of hydrogen into the double bond (A2) as
rate-limiting steps. At the same time, the complementary reaction steps (A1 and B2)
should proceed relatively quickly in the case of model A2B1, which is less likely than
having two identical rate-determining steps such as in model A2B2.
From hydrogenation literature on rate-determining steps, the vast amount on
ethylene hydrogenation should be mentioned. The HP mechanism has been proposed
for these reactions as well (Horiuti and Polanyi, 1934). However, even for this
relatively simple and widely investigated reaction, there still remains uncertainty on
the rate-determining step (e.g., Prairie and Bailey, 1987). This shows how difficult the
determination of rate-determining steps is.
Kinetic data on the migration of double bonds may give more insight in rate-
determining steps. Van der Plank (1972b) studied migration effects of monoene
FAME (liquid phase, 323 ≤ T ≤ 448 K) and proved that if a FAME molecule enters the
pores, in the absence of diffusion limitation, a migration reaction only occurs once
before the FAME leaves the pores. Van der Plank and Van Oosten (1975) concluded
from these data that the adsorption process must be relatively fast compared to the
migration reaction and suggested that the rate-limiting step for the hydrogenation
reaction should be the insertion of hydrogen into the half-hydrogenated intermediate
39
chapter 2
(the second hydrogen insertion, step C2). The migration studies thus showed a
relatively fast adsorption process, which is supported by model A2B2, but from our
analysis step C2 could not to be a rate-determining step.
2.4.3. Consequences of model A2B2. Model A2B2-ass can be used to investigate
the hydrogenation rates of O and E, in relation to the isomerization rate, and their
effect on the observed product distributions during the hydrogenation. Table 2.13
summarizes the parameters for model A2B2-ass. The confidence limits are calculated
by use of the inverse Hessian matrix (Press et al., 1987), which only holds for one
parameter fit (Emig and Hosten, 1974). Therefore, we have also applied the more
generally applicable method of constant χ2-boundaries, as is described by Press et al.
(1987). For our rate equations, both methods give almost the same results, where it
should be remarked that the calculation of constant χ2-boundaries costs many more
model evaluations (Emig and Hosten, 1974).
40
monoene kinetics
RE,eq
CE,eq = Kobs CO,eq ⇒ Kobs = ( 2.22)
RO,eq
However, this equilibrium observed during the hydrogenation is not necessarily
the isomerization equilibrium, as can be illustrated by applying model A2B2:
RE − ( k3 + k2 ) Kobs + k2 Kiso
= = Kobs ( 2.23)
RO −( k1 + k2 Kiso ) + k 2 Kobs
0.30
CE/Ctot
0.15
0.00
0.4 0.6 0.8 1.0
ζ O+E
Figure 2.8. Experimental and simulated elaidate fractions for models A2B2 and A2B1 with
associative hydrogen adsorption equilibrium versus conversion of double bonds for experiments at
0.50 and 0.08 MPa hydrogen pressure (393 K). Parameter values are in Table 2.13. ¡, 0.50
MPa/393 K; ¨, 0.08 MPa/393 K; , simulated curve for A2B2; Λ, simulated curve for A2B1.
Only for k3 = k1, is Kobs = Kiso obtained. From early kinetic data, Kiso ≈ 2.0 (T =
393 K) was determined from an observed equilibrium between O and E during the
reaction (Feuge et al., 1960; Litchfield et al., 1963; Stefanovic et al., 1978). This is
rather low compared to independent measurements (Kiso = 3.6 at T = 393 K); Litchfield
et al., 1963). Litchfield et al. (1963) noted an underestimation of the amount of trans
in earlier literature data, due to inaccurate analysis. Another reason may be that the
41
chapter 2
42
monoene kinetics
strongly influences the calculated value of the activation energy. Also the activation
energy of Susu et al. (1978), 17.7 kJ/mol is strongly influenced by (extremely) high
reaction orders in hydrogen (7.1) for the hydrogenation reaction. The activation energy
of 40 kJ/mol calculated by Nielsen et al. (1960) for first-order kinetics in FAME is
comparable to ours (30 kJ/mol), particularly because the reaction order for model
A2B2-ass becomes first order at higher conversion.
2.5. Conclusions
Kinetic expressions for the hydrogenation reaction rates of methyl oleate and
elaidate were derived on the basis of the Horiuti–Polanyi mechanism. Kinetic
experiments in absence of diffusion limitation were carried out for 333 ≤ T ≤ 443 K,
0.02 ≤ PH 2 ≤ 0.50 MPa and various compositions of oleate, elaidate and stearate.
The experiments were fitted in both the conversion and time domain. The model
with a rate limitation in the first hydrogen insertion in the double bonds of adsorbed
oleate and elaidate proved to be the most likely. Activation energies of 30 kJ/mol for
the hydrogenation reactions of oleate and elaidate and 44 kJ/mol for the isomerization
reaction were found. Further, it is shown that the isomerization constant preferably
must be obtained from independent isomerization experiments rather than from fitting
hydrogenation data in kinetic models.
43
monoene kinetics
1 Ne
1
N e,i N e, d
( Cee,i ,d − Cem,i ,d ) 2
χ =
2
∑N N ∑∑ ( 2. A1)
Ne e= 1 e ,d e ,i i =1 d =1 σ i2,d
Here, Cee,i ,d and Cem,i ,d are the experimentally observed and the calculated model
concentration values for the eth experiment, ith component of the dth data point,
respectively, and σi,d2 is the corresponding experimental variance for the ith
component and dth data point. The χ2 values of all models were compared in order to
select the most likely model. Here we used Bartlett’s test to test whether differences in
accuracy of the various models were statistically significant (Dumez et al., 1977; Graaf
et al., 1988a,b).
For a selected set of H models, the Bartlett test calculates the critical χc 2:
H H
2
ln S av ∑ df h − ∑ dfh ln S h2
h= 1 h =1
χ 2c = H H
( 2. A2 )
1
1+ ( ∑ df h − ( ∑ df h ) −1 )
−1
3(H − 1) h=1 h= 1
with Sh2 the total variance between the experimental series and model h and Sav2
the average total variance of H models (Dumez et al., 1977):
N e N e ,i N e ,d
S h2 = df h−1 ∑ ∑ ∑ (Cee,i ,d − Cem,i ,d ) 2 ( 2. A3)
e= 1 i = 1 d =1
44
chapter 2
H
∑ dfh Sh2
2
S av = h =1
H
( 2. A4 )
∑ dfh
h =1
dfh represents the degrees of freedom for the fit of the hth model to the
experimental series and is calculated by
Ne
df h = ∑ [( N e,d − 1) N e ,i ] − N h,fit ( 2. A5)
e=1
45
3. Diene kinetics ∗
Under industrial conditions, 373 < T < 473 K and 0.1 < PH 2 < 1.0 MPa, poly
unsaturated fatty acids are hydrogenated
p
preferentially over the mono unsaturated species. But
many side products are produced too, which
frustrates the construction of a realistic reaction rate
equation. However, based on the selected rate
determining steps, obtained for the monoene kinetics
(chapter 2), we have developed rate equations for the
diene hydrogenation and cis–trans isomerization
with one new parameter only. This chapter reports
on the development of the monoene-based diene
reaction mechanism and on experiments to verify the
obtained rate equations.
∗
Submitted to Ind. Chem. Eng. Res. as “An Intrinsic kinetic model for the hydrogenation of double
unsaturated fatty acid methyl esters over nickel-based catalysts based on the Horiuty–Polanyi
mechanism”; Bouma, M. J.; Jonker, G. H.; Veldsink, J. W.; Beenackers, A. A. C. M.
46
diene kinetics
3.1. Introduction
A vast amount of studies has been assigned to the kinetics of hydrogenation of
polyunsaturated fats, aiming at selectivity improvement (Veldsink et al., 1997).
However, simultaneous migration and cis-trans isomerization during the
hydrogenation hampered the establishment of intrinsic rate equations. Therefore, early
rate equations for the hydrogenation of the double unsaturated fatty acid Linoleate
(diene), were merely based on the degree of unsaturation: Linoleate→Oleate→Stearate
and presented as first order or power law kinetics.
Applying the Horiuti-Polanyi mechanism Hashimoto et al. (1971), Gut et al
(1979) and Grau et al. (1987a,b) incorporated cis-trans isomerization in the
monounsaturated compounds (monoenes) according to the scheme in Figure 3.1.
Hashimoto et al. (1971) derived the rate
equations as overall first order kinetics,
thus omitting possible saturation of the
D
active catalyst sites with reactants. Gut et 1 2
al. (1979) applied Langmuir–
Hinshelwood type adsorption of C T
unsaturates on the active sites and were 3 4
able to describe the initial zero-order
behavior of the diene. Rate constants as Figure 3.1. Diene hydrogenation scheme used
well as adsorption constants were by Hashimoto et al. (1971), Gut et al. (1979)
and Grau et al (1987). D, diene; C, cis
reported. Limited by the applied monoene; T, trans monoene; S, saturated.
calculation method, Grau et al (1987a,b)
were able to determine lumped rate and adsorption constants only.
However, overall schemes, such as in Figure 3.1, neglect isomerization within the
level of dienes and lump mechanistic steps into overall routes. For instance, if linoleic
acid or its methyl ester (cis,cis) diene is hydrogenated, the direct route to trans
monoene, denoted route 2 in Figure 3.1, has no mechanistic meaning. In terms of the
Horiuti–Polanyi mechanism it combines a hydrogenation and an isomerization step.
Though the rate equations derived from such a scheme may satisfactorily predict
hydrogenation and isomerization rates under the conditions applied, extrapolation to
the hydrogenation of diene cis-trans isomers will fail. While direct hydrogenation of a
diene (L) with increased trans content will yield higher trans monoene in practice, it
has no influence on the ratio of the rates of route 1 and 2 in the model.
47
chapter 3
Usually, different adsorption strengths of fatty acids on the catalyst surface are
mentioned as the explanation for the difference in reaction rates of different types of
unsaturated methyl esters (Eldib and Albright, 1957; Albright, 1963; Coenen and
Boerma, 1968; Heertje and Boerma, 1971; Albright, 1985; Babenkova et al. 1994). For
instance, Coenen and Boerma (1968) observed zero-order behavior with respect to the
unsaturated compound in the hydrogenation of safflower oil and explained it by
complete catalyst surface occupancy of linoleic methyl ester, due to strong adsorption
relative to the monoene. In the hydrogenation of a mixture of linoleic acid and labeled
oleic acid, Coenen and Boerma (1968) and Heertje and Boerma (1971) found that none
of the initially formed elaidate and stearate was labeled. From this, they concluded that
unlabeled elaidic acid and stearic acid were formed directly from linoleic acid, and, if
linoleic acid was present, hardly any of the monoenic acid in the liquid bulk could
adsorb on the catalyst surface.
The aim of this chapter is to extend the rate equations for the hydrogenation of
monoenes to the hydrogenation and isomerization of dienes. Here, it will be tested
whether the same (Horiuti–Polanyi) reaction mechanism that proved successful in
describing the monoene hydrogenation (chapter 2) is applicable to the diene
hydrogenation reactions as well. The selected rate equations were:
− ( k + k 2 K iso ) CO + k 2 CE
RO = mc K H PH 2 ( 31
.)
CO + CE + k 4 CS
− ( k 3 + k 2 ) CE + k 2 Kiso CO
R E = mc KH PH 2 ( 3.2 )
CO + CE + k 4 CS
48
diene kinetics
3.2. Theory
An extension of the monoene hydrogenation mechanism of chapter 2 can be
obtained by assuming that the hydrogenation of double bonds of dienes can be
described by the same rate equations expressed in surface concentrations as proved to
be valid for monoene hydrogenation. In this model the observed greater reactivity of
the dienes stems from the strong adsorption of dienes relative to monoenes only. The
latter is generally accepted from indirect evidence in hydrogenation experiments
(Albright, 1963; Coenen and Boerma, 1971; Susu et al., 1978; Babenkova et al., 1994).
The reaction scheme is depicted in Figure 3.2. If it is assumed that the non-attacked
double bond in the half-hydrogenated diene intermediate exerts no influence on the
reaction steps, the adsorbed diene can be considered as a strongly adsorbed monoene.
CC CC Cσ CT CT σT TT TT
monoene
C C σ T T
S
CC hydrogenated to C TT hydrogenated to T
CC isomerised to CT TT isomerised to CT
S
hydogenation to S
The elementary reaction steps of such a single adsorbed diene are then the same as for
monoenes. Both hydrogenation and isomerization proceed via the half-hydrogenated
state and in each reaction path the insertion of hydrogen in the adsorbed double bond
is the rate-determining step.
Equivalents of the monoene hydrogenation triangle (C-T-S, see also Figure 3.1)
appear twice at the diene level of the mechanism (CC-CT-C and CT-CC-T, Figure
49
chapter 3
3.2). For example, the CT diene, adsorbed with the cis double bond, reacts towards the
Cσ-complex in the same way as C forms the σ-complex. The individual reaction steps
and accompanying rate or equilibrium equations of the diene reaction scheme,
including the monoene reactions are given in Table 3.1.
For the establishment of rate equations, the following assumptions are applied,
following the methodology of chapter 2
(1) Noncompetitive adsorption of hydrogen and the fatty acids.
(2) Fatty acids adsorption, reactions 1-7 in Table 3.1, is at equilibrium.
(3) Hydrogen adsorption, reaction 8 in Table 3.1, is at equilibrium,
(4) The surface fractions of the σ-intermediates as depicted in Figure 3.2 are
negligible, due to their high reactivity. With assumption 1 it then follows:
θ CC + θ CT + θCT + θ TT + θ C + θ T + θ S + θ V ≈ 1 (3.3)
(5) The active sites for fatty acids are occupied completely,
θCC + θ CT + θCT + θTT + θC + θ T + θS >> θV (3.4)
(6) The adsorption constants for cis and trans double bonds in the monoenes and
the dienes are equal
KC = K T = KM (3.5)
K CC = K CT = K CT = K TT = K D (3.6)
(7) The rate constants for the formation of the σ-complex from cis, respectively,
trans double bonds, are independent of the degree of unsaturation. Hence:
k f,CC = k f,CT = k f,C
k b,CC = k b,CT = k b,C
k f,TT = k f,CT = k f,T
k b,TT = k b,CT = k b,T
K I,CC = K I,CT = KI,C
KI,TT = K I,CT = KI,T (3.7)
(8) The location of the double bonds has no influence on reaction rates.
(9) The rate and adsorption constants follow an Arrhenius type of temperature
dependence:
50
diene kinetics
k ka,CT θCT
2 CT+V ka,CT CT ka,CTC CTθV - kd,CTθCT KCT= k =
d,C T d,C T CCTθV
k ka,CT θCT
3 CT+V ka,CT CT ka,CTC CTθV - kd,CTθCT KCT= k =
d,CT d,CT CCTθV
ka,TT ka,TT θT T
4 TT+V k TT ka,TTCTTθV - kd,TTθTT KTT= k =
d,T T d,TT CTTθV
ka,C k θC
5 C+V k C ka,CC CθV - k d,CθC KC= ka,C =
d,C d,C CCθV
ka,T k θT
6 T+V k T ka,TCTθV - kd,TθT KT= ka,T =
d,T d,T CTθV
k k θS
7 S+V ka,S S ka,SC SθV - kd,SθS KS= ka,S =
d,S d,S CSθV
ka,H2 ka,H2 θH2
8 H2+VH k H2 KH2= k = PH2θV
d,H2 d,H2
k k θσT θVH
10 CT+H kf,CT σT+VH kf,CTθCTθH - kb,CTθσT θVH KI,CT= k f,CT =
b,C T b, CT θC TθH
k kf,CT θC σθVH
11 CT+H kf,CT C σ+V H kf,CTθCT θH - kb,CTθCσθVH K I,CT= k =
b,CT b,CT θCTθH
k kf,TT θσΤθVH
12 TT+H kf,TT σT+V H kf,TTθTTθH - kb,TTθσTθVH K I,TT= k = θTTθH
b, TT b,TT
k kf,C θσθVH
13 C+H kf,C σ+V H kf,CθCθH - kb,CθσθVH KI,C= k = θCθH
b,C b,C
k k θσθVH
14 T+H kf,T σ+V H kf,TθTθH - kb,TθσθVH KI,T= kf,T =
b,T b,T θTθH
15 C σ+ H
kh C+V
H
khθCσθH
16 σT+H
kh
T+VH khθσTθH
kh khθσθH
17 σ+H S+V H
51
chapter 3
− E a,i Tref − T
k i ( T ) = k i ,ref exp (38
.)
Rgas Tref T
Furthermore, from the definitions of the adsorption equilibrium constants in Table 3.1
and assumption 6, equal surface occupancies of CT and CT follow:
θ CT = θ CT + θ CT (3.10)
θ CT = θ CT = 1
2 θ CT (3.11)
Using the total adsorbed cis-trans diene CT , eq 3.11 can be substituted in the
equations of Table 3.1.
The surface site balance for hydrogen is
θ VH + θ H = 1 (3.12)
Substitution of θ V from eq 3.12 in the equation for associative hydrogen
H
KH PH 2
θH = ( 313
. )
1 + KH PH 2
52
diene kinetics
assuming the intermediates to be in pseudo steady state. Table 3.2 gives the final rate
equations, expressed in bulk concentrations.
Table 3.2. Diene Model Rate Equations for the Bulk Components a
cp. rate equation R i/mc
CC (
− ( k 1 + k 2 K iso ) K~ D CCC + k 2 K
~ 1C θ θ
D 2 CT ) H V
CT ( ~ ~ ~ 1C θ θ
k 2 K iso K D C CC + k 2 K D CTT − ( k 3 + k 2 + k 1 + k 2 K iso ) K D 2 CT)H V
TT ( 3 2 D TT 2 iso D 2 CT ) H V
~
− (k + k ) K C + k K K ~ 1
C θ θ
C (k1 K~ D CCC + k3 K~D 21 CCT − (k1 + k 2 Kiso ) K~M CC + k2 K~M CT )θ Hθ V
T (k1 K~ D 12 CCT + k 3 K~ D CTT + k 2 Kiso K~M CC − ( k2 + k 3 ) K~M CT )θ Hθ V
S ( k1 K~M CC + k 3 K~M CT )θ H θ V
a
θ V = ( CS + K~M ( CC + CT ) + K D CC CT TT )
~ ( C + C + C ) −1
Marangozis et al. (1977), Gut et al. (1979), and Stenberg and Schöön (1985), the
kinetic rate constants were lumped with the hydrogen adsorption constant. However,
strong deviations from first-order hydrogen dependency have been reported at
pressures of PH = 1.0–11.0 MPa, (Wisniak and Albright, 1961). Since the
2
53
chapter 3
determined explicitly, considering that the linear behavior at PH ≤ 0.6 MPa turns into 2
k2KH
a
k iso k f,hi 1.53 b 47.2 from Chapter 2
2
k 3K H a h
k f,T h
k f,C 3.93 b 28.1 from Chapter 2
2
Kiso iso
k f,C iso
k f,T 3.5 c -4.0d literature
~ K M KS 3.13 e 0 from Chapter 2
KM
~ KD KS
f f
this chapter
KD
f f
KH eq 3.15 this chapter
2
a
valid at PH 2 ≤ 0.6 MPa only, where θ H ≈ K H 2 PH 2 .b units: kg l / kg Ni MPa s . c Litchfield et al.,
1963. d ∆H, Rogers et al., 1977. e dimensionless. f parameters determined in this chapter, Table 3.5.
A Runge–Kutta method with adaptive step size control (Press et al., 1989) was
applied to integrate the rate equations for given parameter values. The parameters were
optimized by the Levenberg–Marquardt method (Press et al., 1989) to obtain the
minimal χ2, defined as
1 1
N e,i N e,d
( Cee,i ,d − Cem,i ,d ) 2
χ =
2
∑N N ∑∑ ( 317
. )
Ne e ,d e ,i i = 1 d =1 σ 2i ,d
Further, the procedure of chapter 2 was applied for avoiding intraparticle mass–
transfer limitation (section 2.2.3). Also, the experiments were performed at such
catalyst loads that the criterion for negligible gas-liquid mass transfer resistance was
satisfied (see section 2.3 of Chapter 2).
54
diene kinetics
calculated from the pressure decrease in the supply system. The fractions CC, CT, TC,
TT, C, T, S, and IPM of 10–17 samples, taken during the reaction, gave the product
distribution with time. A typical reaction profile is shown in Figure 3.8A.
b. Varying Hydrogen Pressure. These experiments were carried out at a constant
temperature (323 ≤ T ≤ 363 K), but now the supply of hydrogen was stopped. The
reactor pressure decreased due to hydrogen consumption from the head-space. A
typical reactor pressure transient is shown in Figure 3.4. When the initial pressure was
55
chapter 3
PH > 1 MPa, dPH / dt was measured during successive periods of 40 seconds, where
2 2
in between the pressure was extra reduced in a controlled way by opening a valve
shortly. This way, rate data over the entire pressure range were obtained while keeping
the liquid conversion low (ζCC < 0.1). A typical reactor pressure profiles of this type of
experiment is shown in Figure 3.4.
Figure 3.4. Reactor pressure transients of typical variable pressure experiments at T = 343 K. Line
a, started at low pressure; line b, started at high pressure; dotted line, tangent, representing reaction
rate.
The catalyst applied was Pricat 9910 (Unichema Emmerich, Germany). Properties
of the catalyst particles can be found in the experimental section of Chapter 2. IPM
and SFME were supplied by Unilever Research (Vlaardingen, The Netherlands) and
SSOG (Milan, Italy). The IPM obtained was further purified with active Raney nickel
at T = 473 K, PH = 0.5 MPa, followed by filtering of the Raney-nickel. SFME was
2
stirred with silica gel and molar sieves for 2 h under vacuum at T = 393 K to remove
water and free fatty acids and filtered subsequently. The compositions after
pretreatment are shown in Table 3.4.
3.4. Results
The gas–liquid volumetric mass-transfer coefficient was measured in
hydrogenation experiments at high (12 kgc /m3 l) catalyst load (Stenberg and Schöön,
56
diene kinetics
1985) yielding kla = 0.8 s-1. By adjusting the catalyst load, the gas–liquid hydrogen
mass transfer resistance was kept below 10 % of the total resistance.
3.4.1. Diene adsorption constant. Similarly to Chapter 2, the rate equations were
first evaluated in the conversion domain instead of the time domain. In the conversion
domain, i.e., the mole fractions of the fatty acids plotted as a function of the double
bond conversion, the effects of cis–trans isomerization and diene-monoene selectivity
on the component curves are pronounced, while nonmechanistic rate effects, such as
catalyst activity variations and poisoning effects, are excluded. Further, the number of
fit parameters is lower in this domain.
The ratio of double bond concentration over fatty acid concentration, ξDB is
calculated from the mole fractions of monoenes and dienes according to
ξ DB = x C + x T + 2( x CC + x CT + x TT ) (3.18)
xi is the mole fraction of i in the bulk liquid phase (Ci/Ctot). To get each
experiment in a uniform conversion scale, the actual conversion is scaled to 100% L,
which is equal to ξDB ≡ 2. The double bond conversion is than defined as
2 − ξ DB
ζ DB ≡ ( 319
. )
2
57
chapter 3
Figure 3.5. Parity plot of modeled versus experimental mole fractions after optimization of
K~Dref and ∆HD to the set of constant pressure experiments in the conversion domain. G, dienes; +,
monoenes; ¨, saturated. Experimental domain: 323 ≤ T ≤ 363 K and 0.4 ≤ PH 2 ≤ 3.5 MPa.
the groups of dienes, monoenes and saturated methyl ester only. The K~D,ref and ∆H D
58
diene kinetics
were optimized for the complete set of constant pressure experiments simultaneously.
The complete data set is shown in Figure 3.5. From Figure 3.5 it follows that the
monoene concentrations are slightly underestimated, which is compensated for by the
dienes and the saturated component. However, these deviations are small, so after
∆H H 2
-53.8 ± 8.1 -50.3 ± 0.5 kJ/mol
a
see Table 3.3, 323 ≤ T ≤ 363 K; 0.3 ≤ PH 2 ≤ 3.5 MPa. Including 95% confidence boundary limits
optimization of only two fit parameters ( K~D,ref and ∆H D ) the model curves of
monoenes and dienes show a remarkable agreement with the experimental data points
for the whole set of data.
~
The initial parameter values were set at KD,ref = 3.13 (equal to monoene) and
∆H D = 1 kJ/mol (a symbolic non-zero value). Different initial values resulted in the
same optimal parameter values. The fitted values, presented in Table 3.5 under 'first
step', result in diene adsorption constants of 9.2 ≤ K~ D ≤ 18.3 at 323 ≤ T ≤ 393 K. Since
~
K D denotes the ratio of diene over saturated adsorption, these values show that
preferent adsorption of the double bonds relative to the adsorption of saturated fatty
acid methyl esters increases with temperature and, because K~M is constant, so does the
preferential adsorption of diene over monoene.
3.4.2. Hydrogen Adsorption Constant. The variable hydrogen pressure
experiments were performed to elucidate the influence of the hydrogen pressure on the
conversion rate. Because of the linear dependency of Ri on θH, KH can directly be
obtained from these experiments (see Table 3.2). In all these experiments, the initial
concentrations of unsaturated compounds were the same and the maximum conversion
of the fastest reactant, i.e., methyl linoleate, was always below 10 % and thus
59
chapter 3
Ri KH2 PH 2
= − ki'θ H = − ki' i = CC,CT,TT, C, T ( 3.22 )
mc 1 + KH 2 PH 2
KH 2 PH 2
RH 2 = 2( RCC + RCT + RTT ) + RC + RT = mc k H' 2 ( 3.23)
1+ K H 2 PH2
bond conversion rate, RDB, and therefore represents the overall hydrogenation rate.
This hydrogenation rate in the liquid phase is calculated from the hydrogen pressure
drop in the reactor via the mass balance:
dPHr 2 Vr ( ε G + mH 2 (1 − ε G ))
φ H2 = RH 2 Vl = 10 6 ( 3.24 )
dt RT
Eq 3.24 accounts for the decrease in dissolved hydrogen with decreasing pressure,
which proved to be significant in the applied pressure ranges. The pseudo first-order
rate constant, k H' , can be calculated from the actual composition and the values of the
2
~
parameters k2, k3, Kiso, K~M from Table 3.3 and KD,ref and ∆H D from Table 3.6. The
remaining model parameters K H , ref and ∆H H can then be fitted such that χ2 of eq 3.17
2 2
is minimal for the complete set of variable pressure experiments (Ne=12). The lumped
value of k 1 K H2 of Chapter 2 (Table 3.3) is not applicable here, because it was
determined in the pressure range 0.02 ≤ PH ≤ 0.50 MPa at T ≥ 353 K where the
2
denominator of eq 3.13 at pressures of PH ≤ 0.5 MPa. Therefore, k1,ref and Ea,1 were
2
fitted simultaneously, which gave a close agreement with the experimental points (see
Figure 3.6).
60
diene kinetics
10
m -1
RH mc 8
2
mol 6
kgNi s
4
0
0 2 4 6 8 10
e -1
RH m
c
mol
2 kgNi s
Figure 3.6. Parity plot of the modeled versus the experimental hydrogen reaction rates after
optimization of KH2ref and ∆HH2 to the variable hydrogen pressure experiments. Experimental
domain: 0 ≤ PH2 ≤ 3.2 MPa and 323 ≤ T ≤ 363 K.
The fitted value for ∆H H , see Table 3.5, is in agreement with literature data
2
obtained from H2/D2 exchange experiments (∆H H = –51.3 kJ/mol; Niklasson et al.,
2
71 kJ/mol on Ni catalysts. At lower occupancies θ H < 0.1, they found the adsorption
enthalpy to rise to ∆H H = –130 kJ/mol. Under hydrogenation reaction conditions,
2
Susu and Ogunye (1981) found ∆H H = –52.3 kJ/mol, while Gut et al. (1979) found a
2
completely different value, ∆H H = +1.3 kJ/mol. The latter can be due to their narrow
2
experimental pressure range (0.22 ≤ PH ≤ 0.34 MPa), which results in minor pressure
2
related variations in the data set only, and, therefore, low reliability of the fitted
hydrogen adsorption parameters. The hydrogen surface occupancies calculated from
Gut et al. (1979), Susu and Ogunye (1981), and the present work are shown in Figure
3.7 together with the pressure ranges applied. The results of Susu and Ogunye (1981)
predict an approximately constant θ H for PH > 1 MPa, which implies a constant
2
61
chapter 3
θH
2
PH [MPa]
2
Figure 3.7. Modeled surface occupancies of hydrogen as a function of pressure. The hydrogen
adsorption parameters were determined from fits of Langmuir-type kinetic models to
hydrogenation experiments. 323 K; − − − - 353 K; A, Gut et al. (1979), determined at 0.22 ≤
PH 2 ≤ 0.34 MPa and 411 ≤ T ≤ 511 K; B, Susu and Ogunye (1981), determined at 0.37 ≤ PH 2 ≤
0.65 MPa and 393 ≤ T ≤ 453 K; C, present work, determined at 0.3 ≤ PH2 ≤ 3.5 MPa and 323 ≤ T ≤
363 K.
variable hydrogen pressure experiments. In Table 3.5 the column denoted “final step”
represents the optimized values, fitted on all experiments simultaneously. The start
values of the parameters for the optimization are taken from (the conversion domain)
constant hydrogen pressure fit and the variable hydrogen pressure fit. The minor
adjustment of the parameters justifies the use of both types of data ranges for a
preliminary optimization of the parameters.
The model curves of the "dienes" (sum of diene isomers), "monoenes" and
saturates are in good agreement with the experimental data, see Figure 3.8. The strong
adsorption of the dienes relative to the monoenes, introduces a selective hydrogenation
of diene over monoene. The deviations of the fit of the model to the total data set in
the time domain, presented in Figure 3.9, are below 10% for the constant pressure
experiments. The remaining deviation is mainly caused by variations in catalyst
activity in the various batch experiments.
62
diene kinetics
Ci 0.3 A PH 1.6 B
2
Ctot [MPa]
1.2
0.2
0.8
0.1
0.4
0.0 0.0
0 4 8 12 16 0 1 2 3 4 5
t [ks] t [ks]
Figure 3.8. Model curves after the final optimization (see Table 3.6). A, A typical constant
pressure experiment, T = 353 K, PH 2 = 2.0 MPa, mc = 7.1 × 10-2 kgNi/ml3; G, dienes; +, monoenes;
o, saturated; B, A typical variable pressure experiment, T = 343 K and mc = 7.4 × 10-2 kgNi/ml3;
circles are experimental points.
In the isomers curves, Figure 3.10, the fits of CC, C, T, and S are remarkably good
0.5
m A B
Ci 0.4 m
Ctot H2
0.3
0.2
0.1
0.0
0.0 0.1 0.2 0.3 0.4 0.5
e
e
Ci PH
Ctot 2
Figure 3.9. Parity plots of the modeled versus the experimental data points after the final
optimization. A, constant pressure experiments; G, dienes; +, monoenes; ¨, saturated; B, variable
pressure experiments.
63
chapter 3
if one considers that none of the fitted parameters had effect on cis–trans
isomerization. It can therefore be concluded that the cis–trans isomerization
parameters, as obtained for monoene hydrogention in Chapter 2, are indeed applicable
to diene hydrogenation as well.
Figure 3.11 compares the selectivity of the hydrogenation of diene relative to
monoene, predicted by the present kinetic model with the available literature data. For
this purpose it is customary to consider the rate equations of the scheme
kD' '
D → M → S as first order (RD = mc k’DCD, RM = mc k’MCM) so that the
kM
k D'
SI = ' ( 3.28)
kM
The rate equations in Table 3.2 reduce to first order kinetics for CS/Ctot → 1,
because then ( CS / Ctot ) >> K~M ( CM / Ctot ) + K
~ ( C / C ) in the denominator of
D D tot
0.3
Ci S
Ctot
0.2 CC
C
0.1
T
CT
0.0
0 TT 8 16
4 12
t [ks]
Figure 3.10. Mole fractions of all components for the experiments in Figure 3.8. ∆, CC (cis-cis);
¯, CT (cis-trans and trans-cis); G, TT (trans-trans); ×, T; ð, C; +, S.
64
diene kinetics
those rate equations. Following the definition of the selectivity, eq 3.28, the linoleate
over oleate hydrogenation selectivity consequently reads
K~D
SI = ~ ( 3.29 )
KM
Since the rate equations in Table 3.2 exhibit equal pressure dependency, eq 3.29 is
independent of the hydrogen pressure. Our selectivities, obtained in the pressure range
0.2 ≤ PH ≤ 3.5 MPa correspond nicely with the data of Albright and Wisniak (1962)
2
which were determined at PH > 1.0 MPa, and Pihl and Schöön (1971), who accounted
2
3.5. Conclusions.
Kinetic rate expressions which were developed for the hydrogenation of double
unsaturated fatty acid methyl esters and are based on an extension of the model for
hydrogenation of mono unsaturated methyl esters of Chapter 2 appear to be capable of
accurately describing the observed diene selectivity and the monoene (cis–trans)
isomers transients in diene hydrogenations at the conditions tested (324 ≤ T ≤ 363 K
and 0.3 ≤ PH 2 ≤ 3.5 MPa), provided the preferential adsorption of dienes relative to
monoenes are taken into account. Due to the larger experimental pressure range in the
hydrogenation experiments compared to Chapter 2, it was necessary to determine the
hydrogen adsorption parameters separately from the kinetic constants.
65
chapter 3
SI
T [K]
Figure 3.11. Selectivity SI as a function of temperature. ¡ 1.0 MPa, + 2.0 MPa, Albright and
Wisniak (1962); ¨ 0.05–0.1 MPa, Pihl and Sch¬¬n (1971); × 0.39 MPa, Ray and Carr (1985); - -
- 0.22–0.34 MPa, Gut et al. (1979); 0.3-3.5 MPa, this work; Λ this work extrapolated.
0.8
Ci
Ctot0.6 B
A
0.4 A B
0.2
A B
0.0
0.2 0.4 0.6 0.8
ζDB
Figure 3.12. Model predictions of cottonseed oil hydrogenation data of A, Pihl and Sch¬¬n
(1971), T = 433 K, PH 2 = 0.11 MPa; B, Eldib and Albright (1957), T = 403 K, PH 2 = 0.41 MPa. l,
dienes; s, monoenes; n, saturated.
66
4. Intraparticle Diffusion∗
In the hydrogenation of edible fats and oils over porous nickel catalysts,
intraparticle mass transfer limitations not only reduce catalyst effectivities, but also
may change the product selectivities. Information from open literature on the possible
p role of intraparticle diffusion limitation in oil
hydrogenation is very limited however. In line with
literature we have measured the intraparticle effective
diffusion coefficient De of edible oils, their fatty acid
methyl esters, and of hydrogen under both inert and
reaction conditions. The intrinsic kinetic rate
equations of Chapter 2 were applied to calculate De
from hydrogenation experiments (reaction conditions).
Comparison with De obtained from tracer pulse
column experiments (inert conditions) have led to new
insights on the pore structure of the catalyst. Besides,
this chapter reports on the experimental and
mathematical aspects of both experimental methods.
∗
This chapter is published as “Intraparticle diffusion limitation in the hydrogenation of
monounsaturated oils and their fatty acid methyl esters”, Jonker, G. H.; Veldsink, J. W.; Beenackers,
A. A. C. M.; Ind. Eng. Chem. Res., 1998, 37, 4646-4656.
68
intraparticle diffusion
4.1. Introduction
Coenen (1986) has summarized the data on diffusion limitation in hydrogenation
of edible oils and reported for a narrow pore catalyst (mean pore diameter <dc> ≈ 4
nm) a 50% decrease in activity relative to a medium- (<dc> ≈ 6 nm) and wide-pore
(<dc> ≈ 8 nm) catalyst. Colen et al. (1988) have observed intraparticle diffusion
limitation in trioleate hydrogenation from which they have calculated intraparticle
TAG diffusion coefficients of De,TAG = 2 × 10-11 m2/s (wide-pore catalysts) and 4 ×
10-12 m2/s (medium-pore catalysts) (T = 373 K) and an absence of hydrogen limitation.
For methyl ester hydrogenation over Pd/C catalysts, Tsuto et al. (1978) could verify
the observed shunt reactions by incorporating intraparticle hydrogen diffusion
limitation, though not very accurately, because of the insensitivity of the curves for the
value of De. We are not aware of any other data, which is surprising in light of the
substantial effects intraparticle diffusion limitation may have on the selectivity of the
reaction.
dm
λ= ( 4.2)
< dc >
The empirical eq 4.1 was tested in the range of 0.1 < λ < 0.5 (Satterfield et al.,
1973). There is a lack of data concerning the change of steric hindrance inside catalyst
pores with varying degrees of unsaturation (Veldsink et al., 1997), although the oil's
viscosity and the bulk coefficient remain unchanged (Veldsink et al., 1997). For
TAGs, Db has been measured by Andersson et al. (1974), but because of the
uncertainties in the values of <dc > and τ, only rough estimates are available for De in
edible oil hydrogenation over porous catalysts (Veldsink et al., 1997).
Preferably, De should be determined under reaction conditions, but this requires a
thorough knowledge of the intrinsic kinetics. Additional data on De can be obtained
from independent experiments in the absence of reaction (Haynes, 1988). In line with
the recommendations in the literature (Park and Kim, 1984; García-Ochoa and Santos,
69
chapter 4
1994), we have measured De both under inert and reaction conditions. However, for
catalyst particles with nonuniform pores, a difference of one order of magnitude or
more can be expected between the measured values of De in both types of experiments
(Park and Kim, 1984; García-Ochoa and Santos, 1994), which is well documented for
reactions in zeolites (Chen et al., 1994).
McGreavy et al. (1994) showed the large influence of the pore structure on the
value of De, in terms of micro- and macropores and their connectivity. Among others,
Park and Kim (1984) and García-Ochoa and Santos (1994) investigated the reaction
and diffusion in bimodal-supported catalysts, which were modelled as pellets and
assembled as crystallites. For this macro- and micropore system, they found De under
nonreaction conditions to vary from 2 to 20 times larger than De measured under
reaction conditions. Although the exact reasons for these differences are not fully
understood, the contributions of the pore structure to the value of De can be expected
to vary with the experimental technique (Park and Kim, 1984; García-Ochoa and
Santos, 1994).
For an independent determination of De, i.e., without using reaction data, the
HPLC (high performance liquid chromatography) method is the most suited technique
(Haynes, 1988). Nevertheless, for small non-zeolite catalyst particles, such tracer pulse
experiments have been rarely performed for the determination of the effective liquid
pore diffusion coefficients (Ma and Lin, 1987; Lin and Ma, 1989; Hejtmánek and
Schneider, 1994). The main reasons are the occurrence of instabilities in column
packing due to the small particle sizes (dp < 10 µm) (Cumberland and Crawford, 1987)
and, with respect to the catalyst strength, intolerable high pressure drops over the
column (Lin and Ma, 1989). Moreover, the accuracy of obtaining De is lower for small
particles, since diffusion and mass-transfer times become comparable to dispersion
times (Hejtmánek and Schneider, 1994). An exception are zeolites where De is low
enough to completely determine trace pulse dispersion (De ≈ 2 × 10-12 m2/s). Therefore,
most HPLC results on diffusivity were obtained for zeolites, while data on small non-
zeolite particles are scarce (Ma and Lin, 1987).
We have studied the intraparticle liquid diffusion coefficients of FAME, TAG,
and hydrogen for small nickel-on-silica catalysts under reaction conditions and by
applying tracer pulse column experiments in the absence of the hydrogenation
reaction. For studying De under reaction conditions, we derived the true intrinsic
hydrogenation kinetics from a wide range of hydrogenation experiments in the absence
of diffusion limitation (see Chapter 2). These rate equations are used here to
70
intraparticle diffusion
investigate the effects of intraparticle hydrogen and FAME or TAG diffusion on the
hydrogenation rate. Also, by applying recent improvements in column-packing
techniques and realizing various improvements in experimental accuracy, we now can
report the first data on effective intraparticle liquid diffusion coefficients of FAME and
TAG in porous non-zeolite nickel-on-silica catalysts from HPLC experiments in the
absence of hydrogenation. The De values obtained from these two methods will be
compared.
4.2. Theory
4.2.1. De from the Reaction Data. The intrinsic kinetic rate equations of the
conversion rate of methyl oleate (methyl ester of oleic acid, O) and methyl elaidate (E)
were taken from Chapter 2. The experimentally determined rate equations are
− ( k1 (T ) + k 2 ( T ) Kiso ) CO + k 2 ( T )C E
RO = mc KH PH 2 ( 4.3)
CO + CE + k 4 ( T )CS
− ( k 3 ( T ) + k 2 ( T )) CE + k 2 (T ) K iso CO
RE = mc K H PH 2 ( 4.4)
CO + CE + k 4 (T ) CS
− Ea,i Tref − T
k i ( T ) = k i ,ref exp ( 4.5)
gas ref
R T T
The parameters are given in Table 4.1 for Pricat 9910 (sample 1992, Unichema,
Emmerich).
Additional batch hydrogenations were performed to obtain De under diffusion-
controlled conditions. The coupling of mass transport and intrinsic kinetic rate
constants is described by the particle mass balance, which reads (Westerterp et al.,
1987)
71
chapter 4
∂ 2 ∂Ci
= RV,i rp i = O, E, H 2
2
De,i rp ( 4.6)
∂rp ∂rp
Here RV,i is the local volumetric reaction rate. Boundary conditions are
∂Cs,i ( t , rp )
=0
∂rp r =0
p
C s,i ( t , R p ) = C b,i ( t )
Table 4.1. Reaction Rate Constants with 95% Confidence Limits (Chapter 2) a.
compound parameter value at Tref Ea [kJ/mol]
k1 4.14 ± 0.14 b 32.2 ± 0.7
k2 1.54 ± 0.15 b 47.2 ± 4.2
k3 3.94 ± 0.32 b 28.1 ± 3.1
k4 0.32 ± 0.04 c
a
Kiso = 3.5 (Litchfield et al., 1963); ∆Hiso = –4 kJ/mol (Rogers et al., 1974); Tref = 393 K.b units:
[kgl/kgNi MPa s].c dimension: none.
dCb,S dC
RH2 = = −∑ b,i i = O, E ( 4.8)
dt i dt
The differential equations (4.6) were solved numerically. The time scale for
concentration changes in the bulk liquid is long relative to that in the catalyst particles
(Colen et al., 1988), which leads to pseudosteady states with a constant Cb within one
72
intraparticle diffusion
time interval (Colen et al., 1988). Time integration of the bulk concentration change
was performed using a fifth-order Runga–Kutta method with an adaptive step size.
The experimental curves were used to obtain a value for the unknown parameter De by
fitting the model curves.
The differential equations were solved with estimated values for the
intraparticle effective diffusion coefficients, De, and using the intrinsic kinetic rate
constants of Table 4.1. The deviations of the resulting model calculations for an
estimated De were subsequently used to calculate a χ2-target function for Ne,d
experimental points of the Ne experiments with Ne,i components:
1 Ne
1
N e ,i N e ,d
( Cee,i ,d − Cem,i ,d ) 2
χ = 2
∑N N ∑∑ ( 4.9 )
Ne e =1 e ,i e,d i =1 e =1 σ i2,d
N e N e ,i N e ,d
∑ ∑ ∑ (Cee,i ,d − Cem,i ,d )2
e =1 i =1 d =1
S2 = Ne
( 4.10)
∑ (( N e,d − 1) Ne ,i ) − N fit
e=1
Ji ac
µi = i = O, E, H 2 ( 4.11)
Ri ( Ci )
µi < 0.95 indicates significant diffusion limitation, whereas µi > 0.95 indicates
neither intraparticle TAG or FAME nor hydrogen limitation.
73
chapter 4
In the laminar flow regime (which always holds for HPLC ) theoretically Bo = 0.5
(Westerterp et al., 1987).
The transport inside the particles is described by a Fickian diffusion equation in
which the pore hindrance is accounted for by the effective diffusion coefficient.
Although for λ → 1 (micropores) surface diffusion of adsorbed species should be
incorporated, the adsorption on the pore wall is negligible in this study, so De can be
applied to both the macro- and the micropore regimes (Lin and Ma, 1989). The
external particle mass transfer was neglected (Lin and Ma, 1989). The validity of this
assumption was justified afterwards by incorporating the external particle mass-
transfer into the mass balances and using Sh (= ksdp/Db) = 4 (Bern et al., 1975) and the
experimentally determined De.
Vr follows for linear adsorption (represented by Ks, after Hejtmánek and
Schneider, 1994) from Φv and MO1:
MO1 is the first moment of the RTD curve (Westerterp et al., 1987)
74
intraparticle diffusion
∞
MO1 = θ mτ = τ ∫ θE(θ )dθ ( 414
. )
0
With eq 4.13, Vr can be calculated from the first moment of the RTD curve (4.14),
which served as a start value in the optimization procedure.
Parameters De and Peax and the final value of Vr were obtained from a best fit of
the time domain solution of Hejtmánek and Schneider (1994) to the experimental
elution curves by using the Levenberg–Marquardt algorithm (Press et al., 1987) and
χ2, defined by eq 4.9. The parameter optimization procedure was adjusted to obtain the
most accurate values, using a wide range of experimental RTDs simultaneously.
75
chapter 4
A Vc,tot Sc,tot
150
2
ml/g m/g 0.12
9910 0.68 281
dSc/drc 9933 0.17 203 dVc/drc
100 Pricat 9910 0.08
[m2/g] [ml/g]
50 0.04
0.020
B Vc,IH
dVc/dr c ml/g
9910 0.31
0.015 9933 0.05
[ml/g]
0.010
Pricat 9910
0.005
Pricat 9933
0.000
0.01 0.1 1
rc [µm]
Figure 4.1. Pore surface and volume of Pricat 9910 (sample 1992) and Pricat 9933 (sample 1992).
(A) Differential specific pore surface (left axis) and pore volume (right axis) plot; data from BET
measurements. (B) Specific pore volume measured by mercury porosimetry. Subscripts tot and IH
are “total” and “intercrystalline” holes, respectively.
76
intraparticle diffusion
Figure 4.2. Electron microscope picture of Pricat 9933, representative for Pricat 9910 too.
410). Eluents were degassed, filtered, and stored under nitrogen. A tracer (1–10 mol
%) was injected in an amount of 0.02% of the column volume. The linearity of the
detection cell was examined by varying the injection sample concentration (Lin and
Ma, 1989). The flow was determined by weighing the eluted liquid.
77
chapter 4
Pricat 9933 (from Unichema, Emmerich) was used, which resembles generally
applied hydrogenation catalysts (nickel-on-silica carrier, e.g., Pricat 9910), but with a
low mean pore size (<dc> = 4 nm, “narrow-pores” according to Coenen (1986)) with
the smallest possible pore-size distribution. However, an electron microscopy
photograph (Figure 4.2) shows that the particles are built from sintered crystallites
giving rise to large, inter crystalline, pores. The pore size distribution shows the larger
pores as a peak around 0.5 µm in Figure 4.1B.
It is very difficult to obtain a uniform column packing with small particles. The
packing is improved by a uniform particle size (Cumberland and Crawford, 1987).
Pricat 9933 was freed from fines with the “sludge” method, yielding a narrow
(volume) particle-size distribution curve (for details, see Jonker et al., 1997). The
technology of Chrompack was used to get columns with a good plug-flow behavior.
Eluent and column features are summarized in Table 4.3, for data on tracers
applied, see Table 4.4. Column degeneration was checked after each ten experiments
by repetition of a “standard” experiment, being 5 mol% hexadecane in either n-octane
(column C1) or MCT oil (column C2), with Φv = 4 × 10-9 m3/s. Both columns
appeared to remain stable.
78
intraparticle diffusion
4.4. Results
4.4.1. Simulation of H2 or FAME/TAG Diffusion Limitation during Reaction.
The qualitative effect of either hydrogen or FAME/TAG intraparticle diffusion
limitations on the reactant and product conversion curves can be investigated by
solving the model (eqs 3–7) with the parameters from Table 4.1 for various values of
De. Figure 4.3 gives intraparticle concentration profiles of O, E and S at t = 0 for
experiment 3 (Table 4.2) for µ H 2 = 0.25, in case intraparticle transport of FAME is
limited by diffusion limitation (De,FAME = 1.25 × 10-13 m2/s, De,H 2 → ∞. The different
Ci 1.0 2.0
Ctot mH PH2 2
0.8 [kPa]
1.5
H2
0.6 S
O
1.0
0.4
E 0.5
0.2
0.0 0.0
center of 0 1 2 3 4 5
particle
r p [µm]
Figure 4.3. Typical initial catalyst concentration profile for experiment 3, Table 4.2, µ H 2 = 0.25.
Fame diffusion limitation (De,FAME = 1.25 × 10-13 m2/s). Solid lines, FAME (O, E, S); dotted line,
H2. Model equations, eqs 4.3 and 4.4; k values, Table 4.1.
79
chapter 4
0.8
Ci/Ctot O
0.6
H2 or
no lim.
0.4
E
0.2
FAME
lim.
0.0
0 20 40 60 80 100
% ζDB
Figure 4.4. Influence of H2 and FAME/TAG limitation in a conversion plot (fractions of O and E
as a function of fraction S) for an initially similar catalyst efficiency of experiment 3, Table 4.2.
Solid lines, model lines without diffusion limitation (‘no lim’); dotted lines, H2 limitation (‘H2 lim’)
( D H 2 = 1.5 × 10-10 m2/s, µ H 2 = 0.25); dashed lines, FAME limitation (‘FAME lim.’; De,FAME or
De,TAG = 1.25 × 10-13 m2/s, µ H 2 = 0.25). Model equations, eqs 4.3 and 4.4; k values, Table 4.1.
the influence of FAME limitation is much lower for an initially E-rich mixture.
4.4.2. Evaluation of De from FAME Hydrogenations. To fit H2 and FAME
diffusion coefficients, the experiments at T = 443 K, 0.02 ≤ PH 2 ≤ 0.50 MPa (Table
4.2, experiments 1–4) were used simultaneously. Other parameters were set at a fixed
value (Table 4.5). The optimization procedure was as follows: (1) determine χ2 at T =
443 K, without applying diffusion limitation; (2) fit of De,H 2 by minimization of χ2 (eq
4.9) with presupposed absence of FAME limitation; (3) fit of De,FAME by minimization
of χ2 (eq 4.9) with a presupposed absence of H2 limitation; (4) use values of 2 and 3 to
start the final optimization of both De,H 2 and De,FAME. The resulting values for the most
optimal De and the variances (eq 4.10) are listed in Table 4.6. Figure 4.5 plots the
residuals.
80
intraparticle diffusion
Both Figure 4.5 and Table 4.6 show that assuming H2 limitation improves the fits
considerably, relative to assuming FAME limitation. The low impact of FAME on the
fit also follows from the calculated ratio of De,FAMECFAME/ De,H 2 C H2 > 104 from the
optimized parameters in the final optimization. The two encircled areas in Fig. 5 give
the residuals of the S and E fraction if FAME diffusion limitation is applied. The
smallest residuals are obtained if H2 limitation is applied (full symbols). The optimum
diffusion coefficient for hydrogen appears to be De,H 2 = 1.6 ± 0.7 × 10-10 m2/s. From
the appropriate mass balances and using the parameters of Table 4.5, it appeared that
the external particle mass-transfer limitation was negligible. This is confirmed by the
Biot number (Bi = ksdp/De): Bi > 102. There is a marked influence of H2 diffusion
limitation.
0.6
m e
Ci -Ci
Ctot 0.3
0.0
-0.3
-0.6
0.10 0.25 0.40 0.55
e
C /Ctot
i
Figure 4.5. Residuals plot of FAME experiments of Table 4.2. Model predictions assuming no
diffusion limitation, open symbols; model extension with intraparticle FAME diffusion limitation,
gray symbols, and of H2, black symbols. l = total of O; n = total of E; s = total of S. Model
equations, eqs 4.3 and 4.4; k values, Table 4.1.
81
chapter 4
no experimental data (Veldsink et al., 1997; Van der Plank et al., 1972b). Therefore,
Table 4.1 is also valid for TAG. (2) All three fatty acids in the triacylglycerol are
equally reactive to hydrogenation and to isomerization (Tumer et al., 1964).
The optimization strategy was the same as for FAME. Table 4.6 shows the results
of fitting both DH 2 as well as De,TAG for experiments at T = 443 K. Here, assuming H2
limitation gives a small improvement (103S2 = 0.012) relative to assuming no diffusion
limitation (103S2 = 0.039), see Table 4.6. Also for TAG hydrogenation, Bi > 102. For
the ratio De,TAGCTAG/ DH 2 CH2 a value of around 0.5 was calculated, which indicates
comparable hydrogen and TAG limitation. However, in section 4.4.1, it was
demonstrated that hydrogen limitation or TAG/FAME limitation each has a marked
influence on the reaction curves. Model curves assuming hydrogen limitation predict a
much higher trans formation than is actually observed in TAG hydrogenation (see
82
intraparticle diffusion
Ci
Ctot 0.12 A B
0.08
0.04
0.00
0 20 40 60 0 15 30 45
% ζDB %ζ
DB
Figure 4.6. Diffusion limitation in TAG hydrogenation experiments. (A) 443 K, 0.50 MPa
(experiment 8, Table 4.2). (B) 443 K, 0.30 MPa (experiment 6, Table 4.2). Solid lines, TAG
limitation (De,TAG = 3.3 × 10-12 m2/s); dashed lines, absence of diffusion limitation or H2 limitation;
dotted lines, both TAG and H2 limitation. l = total of O; n = total of E. Model equations, eqs 4.3
and 4.4; k values, Table 4.1.
0.030
Cim-Cie
Ctot
0.015
0.000
-0.015
-0.030
0.00 0.05 0.10 0.85 0.90 0.95
e
Ci /Ctot
Figure 4.7. Residuals plot of TAG experiments of Table 4.2. Model assuming intraparticle
hydrogen diffusion limitation, open symbols; assuming TAG diffusion limitation, gray symbols;
model allowing for both H2 and TAG diffusion limitation, black symbols. l = total of O; n = total
of E; s = total of S. Model equations, eqs. 4.3 and 4.4; k values, Table 4.1.
0.3
Ci
Ctot
0.2
0.1
0.0
70 80 90 100
%ζ DB
Figure 4.8. Data of synthetic trioleate hydrogenation (experiment 9 of Table 4.2) in the conversion
domain, compared to model prediction from sunflower hydrogenation. Sunflower oil: l = total of
O; n = total of E. Trioleate: ¡ = total of O, ¨ = total of E. Lines, De,TAG = 3.3 × 10-12 m2/s; model
equations, eqs 4.3 and 4.4; k values, Table 4.1.
84
intraparticle diffusion
In section 4.4.1 we have argued that in our reaction mechanism, only TAG or
FAME diffusion limitation can account for a reduced production of trans. Therefore,
the intrinsic kinetic rate parameters of FAME hydrogenation with TAG diffusion
limitation form an unique combination to describe the observed significant reduction
of trans in TAG hydrogenation.
4.4.4. Tracer Pulse Experiments: Test of Column Performance. Typical RTD
curves of experimental runs are depicted in Figure 4.9, both for the RTD curves of the
connecting lines only (inlet, outlet, detection cell) and for the whole system including
the column, the latter curves situated at higher θ. To obtain De with reasonable
accuracy, the volume of inlet and outlet connectings were reduced to only 0.7% Vcl
(column C2). The RTD curves of column + connecting lines were subsequently de-
convoluted (Westerterp et al., 1987) with the appropriate (flow, tracer, temperature)
RTD of the connecting lines. The two experimental series (C1 and C2, see Table 4.3)
differ in eluent, but both columns are equally packed, using the same batch of
particles. From Figure 4.9, it can be concluded that a uniform and regular packing was
obtained.
4
E(θ) 1
4
3 2
3
2
0
0 1 2 3 4
θ
Figure 4.9. Residence time distribution curves for hexadecane and methyl palmitate (1), trioleate
and soybean oil (2), cholesterol (3), and responses from connecting lines only (4) in medium-chain
(carbon length, 10–12 atoms) triacylglyceride, T = 313 K. Column C2 (Table 4.3).
The bedporosity εb was measured indirectly from both the column pressure drop
according to Ergun's relation and from comparing RTD curves of adsorbing and non-
85
chapter 4
adsorbing tracers. For the laminar regime, Ergun's relation reads for spheres (Ergun,
1952):
∆Pcl (1 − ε b ) 2 ηu
= 150 ( 4.15)
L ε 3b 2
d p,S
Figure 4.10 shows that Pcl/ηL for C1 and C2 coincide, though the viscosity of the
eluents of C1 and C2 differed (0.45 and 17 mPa s respectively; the porosities of
Ergun's relation of C1 (εb = 0.26) and C2 (εb = 0.28) proved to be identical).
16
Pcl/Lη
12
109 [1/ms]
8
0
0 1 2 3 4 5
-4
10 u [m/s]
Figure 4.10. Pressure drop per meter per unit eluent viscosity, column C1 (♦) and C2 (l), as a
function of the superficial velocity.
In absence of adsorption, the particle porosity, εp, can be used as a check on εb,
because εp is known from the manufacturer (εp = 0.45). For column 1, RTD curves of
TO and HD were found to be independent of temperature, so TO and HD are not
subject to adsorption on the column packing (Ks ≡ 0). Substituting εp = 0.45 and Ks = 0
in eq 4.13, εb can be calculated from the mean relative residence time, θm (=MO1/τ).
Corrected for the response of the connecting lines, θm(HD) = θm(TO) = 2.25, εb = 0.27,
which is (practically) equal to the value obtained from Ergun's relation. Both from
Ergun's relation and from the mean residence time, εb’s around 0.27 were obtained,
86
intraparticle diffusion
which are not uncommon for micron-size particles with a Gaussian particle-size
distribution (Cumberland and Crawford, 1987).
The parameters, Vr, Peax, and De were simultaneously optimized from all the
RTD curves of the tracers. Peax/L could be obtained using all column peak responses
simultaneously (Lin and Ma, 1989), while De and Vr are unique for each tracer. The
input parameters were L, dcl, Φv , and εb and the experimental, deconvoluted tracer
RTD. The fit procedure was as follows:
(1) The values of Peax, Vr and De were separately fitted for all tracers in order to
obtain starting values. A start value of Vr could also be obtained from eqs 4.13 and
4.14.
(2) All experimental series were simultaneously used to optimize Peax, with the
values of each series of Vr and De, of step 1.
(3) The Vr and the De of the tracers were optimised for the value of Peax of step 2.
(4) Step 2 and 3 were repeated, until the values for Vr and De of the tracers and
Peax did not change anymore.
4.4.5. Intraparticle Diffusion under Nonreaction Conditions. The relation of
Haynes and Sarma (1973) can be used to estimate the influence of the larger pores,
compared to the smaller ones on the total mass transfer resistance. Using the moment
method, Haynes and Sarma (1973) calculated for a bidisperse catalyst the contribution
of the mass transfer in small pores relative to the large pore, denoted by α:
2
(1 − ε p,IH )ε 2p,CR (1 + Ks ) 2 Lp,CR Dp,IH
α= ( 416
. )
( ε p,IH + (1 − ε p,IH )ε p,CR (1 + K s )) 2 Lp,IH Dp,CR
In eq 4.16, Lp,CR and Lp,IH denote characteristic diffusion lengths in the crystallites
and the intercrystalline holes, respectively. For both pore types, IH and CR, Dp is the
pore diffusion coefficient, which is equal to the bulk diffusion coefficient, corrected
for extra pore hindrance. εp, is separated in a crystalline and intercrystalline voidage: εp
= εp,CR + εp,IH. The total particle void volume is defined by the BET isotherm, which is
equal to the volume accessible to nitrogen, Vc,tot = 0.17 mL/g, Figure 4.1A. The
volume of the intercrystalline holes can be estimated from mercury porosimetry for the
range of pores up to rc = 0.5 µm, giving Vc,IH = 0.05 mL/g (Figure 4.1B). Peaks above
rc ≈ 0.5 µm arise from interparticle holes, due to particle settlement in the sample tube.
87
chapter 4
The total particle porosity is εp = 0.45, which can now be divided in εp,CR = 0.31 and
εp,IH = 0.14, for the crystalline pores and the intercrystalline holes, respectively. The
characteristic length for the crystallites as estimated from Figure 4.2 is Lp,CR ≈ 0.5 µm.
The representative length of the inter crystalline holes is equal to the particle diameter,
Lp,IH = 10 µm.With Ks < 0.2 (from the experimental results), 1 + Ks ≈ 1, and Dp,CR ≈
0.1Dp,IH (conservative estimation), α < 0.01 is calculated from eq 4.16. Therefore,
even for a relatively well-defined, small-pore-size range, the influence of the
intracrystalline pores on the effective intraparticle diffusion coefficient is negligible.
Despite a relatively large particle size distribution (see Jonker et al., 1997), the
obtained De can be regarded as representative for the batch of catalyst particles,
because of the approximate Gaussian particle size distribution. Lin and Ma (1990)
calculated that the use of an average particle size in obtaining De from pulse
experiments appears to be a very good approximation, maybe because the slower mass
transfer in the larger particles is compensated for by a faster mass transfer in the
smaller particles (Lin and Ma, 1990).
8
1
E(θ) 2
3,4,5
6
0
1.8 2.1 2.4
θ
Figure 4.11. Residence time distribution curves for hexadecane (1), methyl palmitate (2), trioleate
(3), and soybean oil (4) in MCT oil, T = 313 K. For trioleate, the curve at T = 300 K is also shown
(5). Column C2 (Table 4.3).
With MCT oil as an eluent, typical experimental results and fits of TO and CH are
shown in Figure 4.11 and Figure 4.12, respectively. The model curves agree very well
with the nonadsorbing tracers (MP/HD/TO/SB), and showed some discrepancies in
88
intraparticle diffusion
case of CH, which can be ascribed to nonlinear adsorption (see the appendix of this
chapter). The optimized values of Vr, De and Peax are given in Table 4.7, including
95% confidence interval. The 95% confidence limit of De is obtained by calculations
of χ2 boundaries (after Press et al., 1987). Due to the poor fit, the obtained De values of
6
1
E(θ)
4
2
3
2
0
1.5 2.5 3.5
θ
Figure 4.12. Experimental (solid line) and fitted curve (dashed line) residence time distribution
curves for trioleate (1) and cholesterol: T = 353 K (2) and T = 313 K (3); column C2.
CH are only approximations. The value of Peax = 4.5 × 103 agrees nicely with the
literature value Bo = 0.5 for an ideally packed column (Westerterp et al., 1987).
For n-octane the optimization was insensitive for De, because axial dispersion
appeared to determine >97% of the dispersion effects, a commonly encountered
phenomenon in tracer experiments (Haynes, 1988). Also, Hejtmánek and Schneider
(1994) reported an almost negligible influence of a relatively large De, although 10-
fold larger catalyst particles were applied. Peax could be established at (1.35 ± 0.04) ×
103, which leads to Bo = 0.30 ± 0.01 (eq 4.12).
89
chapter 4
De,TAG
6
[10-12 m2/s]
0
0 4 8 12 16
4
T/η [10 K/Pa s]
Figure 4.13. De,TAG as f(T/η) for experiments 5–8 (Table 4.2), l = De,TAG from hydrogenation data
(Pricat 9910, medium pore, <dc > ≈ 6 nm), ¨ = De,TAG from HPLC data (Pricat 9933, narrow pore,
<dc> ≈ 4 nm), ∆ = De,TAG from Colen et al. (1988) for medium-pore size hydrogenation catalyst.
Solid line, regression of hydrogenation data; dotted line, the same but through the origin.
90
intraparticle diffusion
because εb, τ and λ are temperature independent. Figure 4.13 shows De as a function of
T/η for the hydrogenation data, with η(T) from Wakeham and Magne (1946) and
Eiteman and Goodrum (1994). Indeed, the dependency of De on T/η (Deη/T = 2 × 10-7
(m2Pa)/K) appears to be similar to that of Db as reported by Andersson et al. (1974).
The value of De from the hydrogenation experiments can also be related to the data of
Colen et al. (1988), who performed hydrogenations at T = 373 K with trioleate, both
with wide-pore and medium-pore catalysts. Their value of De,TAG = 4 × 10-12 m2/s at T
= 373 K for a medium-pore (<dc > ≈ 6 nm) catalyst is about 5 times higher, but still of
the same order of magnitude as our value (see Figure 4.13). The observed value of
De,TAG from the HPLC experiments can be compared with De,TAG calculated with
equation 4.1 for wide pores. For large pores, τ = 1.5 (Komiyama and Smith, 1974) and
λ → 0, yielding (eq 4.1): De,IH = 3 × 10-12 m2/s, which is rather close to the
experimentally obtained value of 5 × 10-12 m2/s, see Table 4.8. This is in line with
Colen et al. (1988), who found for wide-pore catalysts, in which diffusion limitation
was almost absent, a value of De of 5 × 10-12 m2/s (extrapolated to T = 313 K). In
contrast, the estimated value of De,CR (for intercrystalline pores) is much smaller:
taking an average value of τ =3.5 (Leyva-Ramos and Geankoplis, 1994), 0.25 < λ <
0.50 (see Table 4.2), the range of De,CR can be estimated as 3 × 10-13 < De,CR < 1 × 10-12
m2/s. As the experimental value of De = 5 × 10-12 m2/s is much larger, this confirms the
dominating contribution of the larger pores.
91
chapter 4
and Santos (1994) found a huge effect of the pore structure on De, whose influence
varies with the experimental technique. De measured under inert conditions can be 2–
20 times larger than De obtained from reaction data. They concluded that tracer
techniques particularly give diffusivities in the macropores, resulting in diffusion
coefficients up to 10 times larger, relative to diffusivities obtained from diffusion
limited reaction data, depending on the pore structure. Although these effects were not
completely understood, the latter, smaller, values are more relevant for describing
diffusion-limited reaction, because the reaction proceeds in the micropores, where
most of the active surface area is found.
4.5. Conclusions
FAME hydrogenations in Pricat 9910 (sample 1992) at T = 443 K, appear to be
controlled by intraparticle diffusion of hydrogen. The intraparticle D e,H 2 could be
calculated as D e,H 2 = (1.6 ± 0.6) × 10-10 m2/s at T = 443 K. As far as we know, this is
the first result on intraparticle hydrogen diffusion in FAME hydrogenation on
supported nickel catalysts. In TAG hydrogenation, both at 373 K and 443 K, TAG
appears to be diffusion limited (De,TAG = (3.3 ± 1.1) × 10-12 m2/s, T = 443 K), rather
than H2 , which is in line with literature.
Intraparticle pore diffusion was measured for paraffins, triacylglycerides and
methyl esters of fatty acids in Pricat 9933 by a so-called HPLC tracer technique.
Because of the small particles and relatively fast intraparticle mass transfer, these
experiments are unique in the literature. The response peaks indicate a uniform
packing of the particles over the column, without bypasses or dead zones, which was
confirmed by consistent bed voidages (εb = 0.27 ± 0.01), and Peclet numbers for axial
dispersion of 0.5, which is the value expected for ideally packed columns. Intraparticle
diffusivities of methyl palmitate and n-hexadecane in MCT oil were measured as 1.4 ×
10-11 < De < 2.7 × 10-11 m2/s in n-octane (95% confidence limit, T = 313 K). For
trioleate and soybean oil 4 × 10-12 < De < 7 × 10-12 m2/s was observed.
The relatively large value of De in HPLC experiments indicate a small hindrance
in the pores, which can be ascribed to a large influence of intercrystalline holes,
relative to the crystalline pores. The observed De,TAG under reaction conditions
appeared to be 10-fold lower than measured with a tracer technique under nonreaction
conditions. This difference could be qualitatively explained from the fact that tracer
techniques measure the diffusivity in the macropores, whereas the chemical reaction
mostly occurs in micropores.
92
intraparticle diffusion
The cholesterol RTD curves show back tailing, which may indicate nonlinear
adsorption (Ruthven, 1984). However, nonlinear adsorption necessarily leads to a
dependency of the elution pattern on the concentration of injected tracer (Ma and Lin,
1990), which is in contrast with our findings. Back tailing can also be caused by an
extra kinetic effect, as nonequilibrium adsorption (Giddings, 1963). A molecule is then
assumed to adsorb on two different sites, with and without the extra kinetic effect
respectively, with a chance calculated with a probability function (Giddings and
Eyring, 1955; Giddings, 1963). The use of the probabilities enables the numerical
addition of an extra kinetic effect on the column dispersion.
Giddings (1963) demonstrated that tailing can have a kinetic background, such as
a nonequilibrium adsorption–desorption processes, which is superimposed on the
dispersion effects. Two kind of sites are assumed in the column packing: apart from
the “normal” sites with equilibrium adsorption, a category of independently operating
sites are added, on which an extra kinetic effect occurs. The extra kinetic effect is
modeled as a first-order forward rate and a first-order backward rate, without further
specification, although it can be seen as nonequilibrium adsorption. The probability of
a molecule adsorbing at the equilibrium sites, P’(y) is equal to:
P ' ( y) = e − k f tm δ ( y ) (4.A1)
with kf the first order (forward) kinetic constant on the nonequilibrium sites (s -1)
and δ(y) the input (Dirac) response pulse. The y is defined as ts/tm, with t m the mean
residence time and ts the (variable) time spent adsorbed on the tail-producing site
which is thus the time measured from the appearance of a normal peak-i.e., a peak
undisturbed by the extra kinetic effect. The probability of the molecule adsorbing on
the non-linear part is described by the Poisson-distribution, which leads to (Giddings
and Eyring, 1955; Giddings, 1963)
( )
0.5
k k
P (y ) = f b e − ( kf + kb y ) t m I 1 4k f k b t m2 y ( 4. A2 )
y
93
chapter 4
with kb (backwards) first-order kinetic rate constant for desorption from the
nonequilibrium sites (s -1). The I1 is the first order Bessel function of imaginary
argument. By definition, P(y) and P'(y) count up to one. Because eqs 4.A1 and 4.A2
represent a probability distribution, the dispersion effects of axial and intraparticle
mass transfer can be treated as separate and serial processes (Giddings, 1963). The
peak dispersion of the column is therefore convoluted with eq 4.A1 and eq 4.A2. The
resulting response curves contain subsequently the influence of an extra,
nonequilibrium effect.
Figure 4.A1 shows the curves P(y) and P’(y), which account for the adsorption on
equilibrium adsorption sites and on nonequilibrium sites, respectively. Both curves are
corrected for dispersion effects and influences of the connecting lines. The summation
of P(y) and P’(y) fairly better describes the experimental RTD than eq 4.6 alone. The
better fit results in a more accurate value of De than the fit of Figure 4.12: De,CH = 0.63
× 10-12 and 2.2 × 1012 m2/s for T = 313 and 353 K respectively. The fit values of kf and
kb are strongly dependent on the start values, which implies that the CH curves contain
too less information to obtain reliable values for kf and kb. As Giddings (1963) already
stated, nonequilibrium adsorption curves may in some cases better explain severe back
tailing of RTD curves, instead of the more commonly applied nonlinear adsorption.
4
E(θ )
3
2
2
1
1
0
2.0 2.5 3.0 3.5
θ
Figure 4.A1. Fit of residence time distribution curve of cholesterol (T = 353 K). Solid line:
experiment; dotted line ‘1’: contribution of equilibrium adsorption; dotted line ‘2’: contribution of
nonequilibrium adsorption; dashed line: summation of dotted line ‘1’ and ‘2’.
94
5. Adsorption
96
adsorption
5.1. Introduction
Data on adsorption of FAMEs on heterogeneous catalysts are scarce; Table 5.1
lists the range of adsorption enthalpies obtained from literature. Lidefelt (1983)
performed gas chromatography experiments in packed columns using saturated and
unsaturated FAMEs and calculated comparable adsorption enthalpies for mono- and
diunsaturated FAME from the mean residence time. The interaction of saturated
FAME on silica, a commonly used support for hydrogenation catalysts, was
investigated by Mills and Hockey (1975). Under hydrogenation conditions, saturated
FAME usually is the solvent. Then the adsorption enthalpy of unsaturated FAMEs is
obtained relative to the adsorption enthalpy of saturated FAME (Table 5.1, Gut et al.,
1979; Chapter 2 and 3). The ambiguity of the limited data of liquid phase adsorption
of fatty acids (Table 5.1) indicate the necessity of further research into adsorption
phenomena (Veldsink et al., 1997).
97
chapter 5
98
adsorption
on pure nickel was performed in batch adsorption experiments, which allowed the use
of Raney nickel in active form. To increase the experimental accuracy, very high
catalyst loads, up to 60 w/w%, were applied in a specially developed autoclave.
5.2. Theory
5.2.1. Tracer Pulse Experiments. The column experiments were performed to
investigate the interaction of the FAME with supported nickel catalyst. In
chromatographic experiments performed with packed columns, any intraparticle
adsorption of tracer is reflected in the average residence time of the tracer pulse
(Ruthven, 1984). The Residence Time Distribution (RTD) curves of the tracers are
normalised, applying the dimensionless residence time θ, which is defined, in line with
Chapter 4 (4.2.2)
tv
θ= (5.1)
L
The adsorption parameters can be directly calculated from the dimensionless mean
residence time θm:
∞
θ m = ∫ θE (θ ) dθ = 1 + (1 − ε b ) ε −b1ε p (1 + K s ) (5.2 )
0
To use eq 5.2, the solute concentration should be sufficiently low to assume linear
adsorption (Cs = KsCl, with Cs solid concentration [mol/mc 3] and Cl liquid
concentration [mol/ml3]. The linearity should be checked using the experimental data
(Ruthven, 1984). Details of the chromatographic method can be found in Chapter 4.
The catalyst used, the silica-supported-nickel catalyst Pricat 9933, was inactivated and
therefore the hydrogenation reaction was absent.
5.2.2. Batch Adsorption Experiments. Raney-nickel could not be applied in
packed columns, so batch adsorption experiments were carried out, in which activated
Raney-nickel was used. The interaction of double bonds of FAME with pure, active,
nickel surface was studied relative to the adsorption of fully saturated FAME. For this
purpose, a mixture of FAMEs, which contains chemically almost equivalent
molecules, apart from their degree of unsaturation, was defined as a pseudo-binary
system of saturated and unsaturated molecules. We have used pseudo-bicomponent
99
chapter 5
nσM( n) n ∆xl,M
0
= (5.3)
m m
For a binary mixture, ∆xl,M follows directly from measuring the mole fraction of
monounsaturated species in a liquid sample of the slurry. It should be mentioned that
the theory only holds for a (pseudo) binary mixture, that is all monounsaturated or all
double unsaturated versus saturated molecules, but it fails for a mixture of all kinds of
unsaturated (mono and double) molecules. The superscript σ(n), denotes the excess of
the monounsaturated FAME adsorbed at the surface relative to the amount adsorbed if
no preferential adsorption of the monounsaturated FAME would occur, the latter
refered to by n in σ(n).
The surface excess of the saturated components, nσS ( n) , follows from Everett
(1986)
σ ( n)
nM = − nSσ ( n) = − n 0 ( xS0 − xSeq ) (5.4)
with x S0 and xSeq initial and equilibrium mole fraction of the saturated
components, respectively. The specific reduced surface excess of the unsaturated
species is then calculated from (Everett, 1986):
with nmax specific catalyst capacity for adsorption [mol/kgc] and with the
adsorption partition coefficient Ω defined by
100
adsorption
xs,M xl,S
Ω= (5.6)
xl,M xs,S
with subscripts s and l denoting surface and liquid fractions, respectively. All
parameters of eq 5.5 can be measured, apart from Ω, but the latter can be fitted by
applying eq 5.5 on the experimental data.
In kinetic rate equations, the Langmuir adsorption coefficient is often applied
instead of Ω. A typical Langmuir expression for FAME hydrogenation is (Gut et al.,
1979; Chapter 2 and 3):
KM
( xl,M / xl,S )
KM xl,M KS Ω( xl,M / xl,S )
xs,M = = = (5.7)
KM xl,M + KS xl,S KM
( xl,M / xl,S ) + 1 Ω( xl,M / xl,S ) + 1
KS
with KM and KS Langmuir adsorption coefficients of the monounsaturated and
saturated species, respectively (note that the subscript s in Ks denotes solid and S in KS
saturated components). The behavior of Ω with concentration becomes similar to Ks,
the adsorption coefficient from tracer pulse experiments, if Ω is determined in the
linear regime which is achieved for Ω(xl,M/xl,S) << 1 and xl,M << xl,S (see eq 5.6). Only
with knowledge of the surfance occupancy in m2/kgc , their numerical values can be
mutually compared.
For hydrogenation catalysts, adsorption experiments under reaction conditions
imply that the concentrations of the saturated and unsaturated species not only change
due to adsorption effects, but also due to hydrogenation of unsaturated bonds. In the
calculation of adsorption parameters, both effects have to be separated, which can be
done as follows. Each mixture contains a fraction of saturated C16 methyl ester (P,
methyl palmitate), which is chemically almost equivalent to the C18 methyl ester (S),
but can be distinguished from it by gas chromatography. Because unsaturated C16
always was below the detection limit (<0.5%), the amount of methyl palmitate is not
affected by hydrogenation and can be used as an internal standard for the amount of
methyl stearate. The latter is formed due to hydrogenation of monounsaturated FAME.
After completion of the hydrogenation, the mole fraction of saturated C18 corrected
for hydrogenation, xS0, is related to the measured mole fractions of methyl palmitate by
101
chapter 5
x 0P
xS0 = xSeq (5.8)
xPeq
102
adsorption
2 3
4
6 5
7
8
Figure 5.1. Novel mixing cell for mixing at high catalyst loads (60 w/w%). 1 = motor, 2 =
thermostatted oil bath, 3 = autoclave, 4 = entrance for tracer, 5 = sample outlet, 6 = gas inlet, 7 =
sealing of upper and lower half, 8 = filters.
103
chapter 5
6
5
E(θ )
4
3,4
2
2 1
0
2.0 2.5 3.0 3.5 4.0
θ
Figure 5.2. Residence time distribution curves for methyl palmitate at T = 313 K (1), 327 K (2),
and 353 K (3,4); Cinj = 0.8 mol% (1,2,3), Cinj = 4 mol% (4). A typical RTD curve of n-hexadecane
is also shown (curve 5, T = 313 K, Cinj = 0.8 mol%)
0
ln(Ks)
-1
104
adsorption
5.4. Results
5.4.1. Adsorption Data from Tracer Pulse Experiments.
Adsorption phenomena of FAME on supported nickel catalysts were observed for
methyl palmitate with n-octane as an eluent. Figure 5.2 shows the RTD curves of
methyl palmitate for various temperatures and initial concentrations. The curves of
methyl palmitate are more symmetrical at elevated temperatures which is due to a
decreasing adsorption strength. n-Hexadecane does not show any adsorption effects,
indicating that the observed adsorption of methyl palmitate originates from adsorption
of the methyl ester group. A typical response curve of n-hexadecane is included in
Figure 5.2; the curves are remain for other temperatures. The injected concentration of
methyl palmitate does not affect the RTD curve indicating that a linear adsorption
constant can be applied. Figure 5.3 shows Ks, obtained with tracer pulse experiments,
for methyl palmitate in n-octane in form of a Van 't Hoff plot. Although it should be
8
1
E(θ) 2
3,4,5
6
0
1.8 2.1 2.4
θ
Figure 5.4. Residence time distribution curves for hexadecane (1), methyl palmitate (2), trioleate
(3), and soy bean oil (4) in medium chain triacylglyceride, T = 313 K. For trioleate, the curve at T =
300 K is also shown (5).
mentioned that the applied temperature range is rather small, the resulting adsorption
enthalpy ∆H(K s) of methyl palmitate on silica supported catalysts of –42 ± 5 kJ/mol
appears to be reasonable. The value is between the adsorption enthalpies of FAME on
105
chapter 5
silica in benzene (∆H(K ad) = –35 kJ/mol) and in carbon tetrachloride (∆H(K s) = –52
kJ/mol), see Table 5.1.
The influence of the solvent is pronounced if n-octane is replaced by a medium
chain triglyceride (MCT) as an eluent. With medium chain triglyceride, the RTD
curves of methyl palmitate and n-hexadecane completely coincide (see Figure 5.4).
Moreover, RTD curves of trioleate are insensitive to the column temperature (Figure
5.4), denoting negligible adsorption in medium chain triglyceride oil. The small
difference between methyl palmitate/hexadecane and trioleate may stem from a
different accessibility to the (crystalline) pores. Though methyl palmitate in n-octane
adsorbs on supported nickel catalysts, it does not show any preferent adsorption with
an oil as a solvent, indicating an interaction of the ester group with the surface.
5.4.2. Preliminary Tests for Batch Adsorption Experiments. For the batch
adsorption experiments, which are far more sensitive to inaccuracies than tracer pulse
experiments (Parfitt and Rochester, 1983), an elaborate sensitivity analysis of the
experimental setup was carried out, as recommended by Everett (1986). The
inaccuracy in the selectivity factor Ω appeared to originate for >99% from the
inaccuracies in the mole fractions obtained from GC chromatography. Optimization of
the injection and temperature programme minimized the standard deviation of the
mole fractions to 0.005.
Parameter sensitivity calculations showed, that a catalyst load of >50% is needed
to attain the experimentally determined nσM(n) or nσD (n) with confidence limits below
25%. The mixing efficiency of the cell was tested for several catalyst loads and mixing
times. Within a few minutes, the catalyst appeared to be homogeneously mixed with
the fluid, which could be observed by pouring the reactor contents on to a glass plate.
Activation of the Raney nickel catalyst, which was preactivated by the
manufacturer, required the removal of surface bonded hydrogen before the liquid
mixture was added. Therefore the cell was degassed by applying vacuum at the
maximum allowable reactor temperature of 353 K. However, at this temperature part
of the hydrogen may remain at the catalyst surface (Fouilloux, 1983; Babenkova et al.,
1994). If so, some hydrogenation still may take place. Only at T > 600 K (Fouilloux,
1983; Babenkova et al., 1994) all hydrogen can be removed. Applying such high
temperatures is not recommended because, apart from the experimental difficulties, the
catalyst surface may be subject to sintering (Fouilloux, 1983; Kinza et al., 1985;
Mikhailenko et al., 1993). A possible advantage of removing only a part of the surface
106
adsorption
bonded hydrogen is that the adsorption of FAMEs can be studied under circumstances
which are as close as possible to reaction conditions.
The effect of hydrogenation on the adsorption measurements was established by
varying the time of degassing and stabilisation. It appeared that after 18 h under 0.5
mbar, the hydrogenation of FAMEs by surface hydrogen was completed for about
90%, see Figure 5.5. At this point, the hydrogenation rate is sufficiently reduced to
keep changes in the liquid concentrations negligible within the sampling time (<2
minutes). Then, 0.6 mol/kgc is hydrogenated (see Figure 5.5), which is about half of
the initially 1.2 mol/kgc (Fouilloux, 1983) of hydrogen. This way, FAME adsorption
on an hydrogen-rich surface is measured. In some experiments, the equilibration time
was varied between 1.5 and 88 h. It appeared that the equilibration time had no
influence on the amount adsorbed. Therefore, we choose the recommended 18 h of
equilibration in all experiments (Everett, 1986).
100
n hyd nhyd (t)
0.6
[mol/kgc] 80 nhyd(t )
%
0.4 60
40
0.2
20
0.0 0
0 20 40 60 80 100
t [h]
Figure 5.5. Hydrogenated unsaturated fatty acid per kg catalyst with 95% confidence interval, as a
function of the stabilistation period at T = 323 K.
107
chapter 5
0.08
σ(n)
n M
m 0.06
[mol/kg ] c
0.04
0.02
0.00
0 25 50 75 100
xM [-]
Figure 5.6. Specific reduced surface excess for monounsaturated FAMEs mixtures (T = 323 K) on
Raney-nickel. Solid line is a fit of the experimental points (with 95% confidence interval),
calculated with eq 5.5 and Ω > 100 and nmax = 0.08 mol/kgc. As an example Ω = 10 is also drawn
(dotted line). For the compositions of the initital mixtures, see Table 5.2, M1 and M2.
108
adsorption
lower value of Ω measured on supported nickel, compared with Ω obtained for pure
nickel, may stem from additional adsorption of the methyl ester group on the catalyst
support.
5.4.4. Batch Adsorption Experiments with Double unsaturated Mixtures. In
the calculation of adsorption of diene (D), relative to monoene (M), the consecutive
reaction, D→M→S, should be incorporated, to distinguish between the hydrogenation
of D and M. The change in concentration of D and M cannot be separated in an
adsorption and a hydrogenation part, without knowledge of the exact values of the
hydrogenation rate of D→M and M→S. However, calculations showed that very small
variations in the hydrogenation parameters (within 3%) largely change the calculated
surface fractions (>50%). Literature data of kinetic parameters of these reactions are
too inaccurate and too catalyst specific to calculate the amount of diene hydrogenated
to monoene. Therefore, because the D(iene) mixture (see Table 5.2) inevitably also
contains monounsaturated components, the concentration changes in the batch
adsorption experiments of monoenes and dienes cannot be accurately separated the
contributions of adsorption and hydrogenation. Consequently, quantitative data on
diene adsorption could not be gathered.
However, the formation of specific reaction products enabled us to perform a
qualitative evaluation of adsorption of dienes and monoenes. Apart from
hydrogenation and adsorption, we
also observed chain shortening and Table 5.3. Chain Shortening of
C18 or Demethylation as f(T,t)a
double bond isomerization. The chain T, K t, h C17/C18
shortening from C18 to C17 to C16 mol%
and so on appeared to take place on a 323 18 1.5
353 6 2.0
low but significant level. Table 5.3
353 27 2.3
summarizes the calculated ratio of 373 19 3.0
a
C17 to C18 for different temperatures values are corrected for initial concentrations
and times. These observations are in
line with literature: Machiels and
Anderson (1979) reported that nickel
catalysts preferentially attack terminal carbon–carbon bonds and successive
demethylation is found for a number of alkanes, ranging from pentane to decane.
Double bond isomerization, both positionally and geometrically, is well known (see
e.g., Van der Plank, 1972). Applying the diene mixture, which mainly contains the
conjugatable methyl linoleate (9,12, cis, cis), we also observed the formation of
109
chapter 5
5.5. Conclusions
For studying adsorption phenomena of saturated and unsaturated fatty acid methyl
esters, a novel type of mixing cell was developed, which could mix slurries with solids
loads as high as 60 w/w%. Also experiments were carried out with tracer pulse
experiments.
110
adsorption
111
6. Conclusions
f
The chapters of this thesis report on various parts
of an investigation of the nickel-catalyzed
hydrogenation of edible oils, starting from the intrinsic
reaction kinetics of the monounsaturated FAME
hydrogenation. The rate equations obtained could be
applied succesfully in the extension to diene kinetics
and in the measurement of intraparticle diffusion
coefficients (D e) from hydrogenation experiments.
Apart from this, independent experiments were
performed for De and adsorption coefficients (K s). This
chapter is a brief general discussion on the methods
applied and on the results.
112
conclusions
6.1. Methodology
In this section, the methodology of the research is discussed, per chapter:
Monoene Kinetics. Kinetic expressions for the
p
hydrogenation reaction rates of methyl oleate and elaidate
were derived on the basis of the Horiuti–Polanyi mechanism,
in which the half-hydrogenated surface intermediate is the
key component. Kinetic experiments in the absence of
diffusion limitation were carried out for 333 ≤ T ≤ 443 K,
0.02 ≤ PH 2 ≤ 0.50 MPa, and various compositions of oleate,
elaidate and stearate, in constant and variable-hydrogen pressure experiments.
The experiments were fitted in both the conversion and the time domain, and the
rate equations derived were statistically evaluated on the ability of describing the
experimental series. For this purpose, Bartlett’s test was used. The reaction scheme
was constructed from adsorption of the reactants, followed by the formation of the
half-hydrogenated intermediate (first hydrogen insertion) and the second hydrogen
insertion to the fully saturated molecule. From this scheme, we derived sets of rate
equations, based on all possible rate limiting steps.
The model with a rate limitation in the first hydrogen insertion in the double
bonds of adsorbed oleate (cis) and elaidate (trans) proved to be the most likely, rather
than adsorption steps or the irreversible second hydrogen insertion step to the fully
saturated stearate. In this way, we have used a relatively simple system to derive and
test a systematical approach in determining the key reaction steps of an otherwise
complex system, which enables a further straightforward modeling of other kinetic
steps or mass transfer limitations.
The selected rate equations also showed that the isomerization constant preferably
must be obtained from independent isomerization experiments rather than from fitting
hydrogenation data in kinetic models, because of the effect of inaccuracies in the
hydrogenation rate constants on the isomerization constant.
p Diene kinetics. The kinetic rate expressions for the
hydrogenation of double unsaturated fatty acid methyl esters
(FAME) were based on an extension of the model for
hydrogenation of monounsaturated FAMEs of chapter 2. The
experimental series contained both constant and variable
hydrogen pressure experiments. The parameters were obtained
by a stepwise evaluation of the rate equations on the
113
chapter 6
experimental results.
Assuming similar rate determining steps in the diene hydrogenation, and similar
adsorption constants for the cis and trans isomers of diene, the monoene model could
be relatively easily extended to dienes. Using apparent kinetics or power law kinetics,
this extension would have needed the introduction of a number of unknown kinetic
parameters (for all cis and trans isomers of dienes), which are difficult to determine
experimentally. With the addition of the preferential adsorption of dienes relative to
monoenes only, the developed rate equations appear to be capable of describing
accurately both the observed diene selectivity and the monoene cis–trans isomers
transients in diene hydrogenations at the conditions tested (324 ≤ T ≤ 363 K and 0.3 ≤
PH 2 ≤ 3.5 MPa).
Due to the larger experimental pressure range applied in the diene hydrogenation
experiments, relative to the monoene experiments, it was necessary to determine the
hydrogen adsorption parameters separately from the kinetic constants.
p Intraparticle Diffusion. The intrinsic rate equations for
monoene hydrogenation were used to derive intraparticle
diffusion coefficients under reaction conditions. FAME
hydrogenations in Pricat 9910 (sample 1992) at T = 443 K,
appear to be controlled by intraparticle diffusion of hydrogen.
As far as we know, this is the first result on intraparticle
hydrogen diffusion in FAME hydrogenation on supported
nickel catalysts. In edible oil (triacylglyceride, TAG) hydrogenation, both at 373 K
and 443 K, TAG appears to be diffusion limited rather than H2, which is in line with
literature. For the diffusion limited conditions, we were able to determine the relevant
intraparticle effective diffusion coefficients for both hydrogen and TAG.
Under inert conditions, intraparticle pore diffusion was measured for paraffins,
triacylglycerides and methyl esters of fatty acids in Pricat 9933 by a so-called HPLC
tracer technique. Because of the small particles and the relatively fast intraparticle
mass transfer, these experiments are unique in the literature. The response peaks
indicate a uniform packing of the particles over the column and an absence of by-
passes in the column packing. Intraparticle diffusivities of methyl palmitate, n-
hexadecane, trioleate, and soybean oil in MCT oil could be measured.
114
conclusions
6.2. Results
A systematical evaluation of all kinetic and transport processes of monoene and
diene hydrogenation and isomerization on supported nickel catalysts has led to the
following insights in the hydrogenation of edible oils and their fatty acid methyl esters.
As discussed in the different chapters, our findings are in line with the elaborate
literature on results as far as available.
In hydrogenation of a mixture of methyl oleate (O) and methyl elaidate (E), the
hydrogenation and isomerization process is not limited by intraparticle diffusion in the
range 333 ≤ T ≤ 428 K and 0.02 ≤ PH 2 ≤ 0.50 MPa (Pricat 9910, sample 1992, dp = 8.4
µm). At these conditions, the intrinsic kinetics of the hydrogenation process can be
described by Langmuir kinetics with competitive adsorption between the
monounsaturated compounds (KM), O and E, versus saturated (KS), S (KM/KS=3.3,
independent of temperature). The selected rate equations (see e.g., eqs 1 and 2 of the
abstract) predict O and E being directly irreversibly hydrogenated to stearate, at almost
an equal conversion rate. Besides, O is favorably isomerized to E, which was detected
115
chapter 6
116
conclusions
117
notation
Notation
ac external particle surface [mc 2/ml3]
Bo Bodenstein number, Bo = vdp/Dax
Bi Biot number, Bi = Shex = ksdp/De
C concentration [mol/ml3]
CH2 (= mH 2 PH 2 /Rgas T) hydrogen bulk-liquid concentration [mol/ml3]
Cinj injected concentration in pulse experiment [mol%]
Cs concentration in solid [mol/mc 3]
D diffusion coefficient [m2/s]
Dax axial dispersion coefficient [m2/s]
<dc > mean pore size [m]
De effective intraparticle diffusion coefficient [(mc 2/s)(ml3/mc 3)]
df degrees of freedom
d diameter [m]
dp,S mean-surface-based spherical particle diameter [m]
Ea activation energy [kJ/mol]
E(θ) normalized column response
∆H enthalpy of adsorption [kJ/mol]
∆H~ adsorption enthalpy, relative to the solvent [kJ/mol]
∆Hiso enthalpy of isomerization reaction [kJ/mol]
H number of models
J mole flux [mol/(mc 2 s)]
k’ (pseudo) first-order rate constant [mol/(kgNi s)] or [m3/kgNi s]
ka adsorption rate constant [ml3 /(kgNi s)]
kb, kd, kf, kh surface rate constant [mol/(kgNi s)]
ki lumped parameter in time domain, [mol/(kgNi s MPa)] or [MPa-1] or [-]
kic lumped parameter in conversion domain, see section 2.4.1
k iso overall isomerization rate constant, see Table 2.3 [mol/(kgNi s)]
kla volumetric gas–liquid mass-transfer coefficient [s-1]
118
notation
119
notation
Greek Symbols
α contribution of macropore mass transfer; eq 4.16
χ2 target function in optimisation procedure (eq 2.A1)
χc 2 critical χ2 of Bartlett's test (eq 2.A2)
χt2(H) tabulated χ2 for H degrees of freedom
εb bed porosity
εG volume fraction of gas phase in reactor
εp particle porosity [m3pore/m3c ]
φ H2 hydrogen reaction rate, defined by eq 3.24 [mol/s]
Φ Weisz–Prater modulus, see eq 2.12
120
notation
Abbreviations
A1, A2, B1, B2, C1, C2: see Table 2.1
ass associative hydrogen adsorption function
B (methyl) behenate (C22:0)
C FAME with 1 double bond in the cis configuration
CT FAME with 2 double bonds, cis and trans configuration
C1, C2 column 1 or 2
CH cholesterol
CP constant-pressure experiment
D diene; all FAME dienes
diss dissociative hydrogen adsorption function
E elaidate (trans-monoene)
Er (methyl) erucate (C22:1)
FAME fatty acid methyl ester
GC gas chromatograph
H,H2 hydrogen
HD n-hexadecane
121
notation
HP Horiuti–Polanyi mechanism
HPE; HPO high percentage of elaidate or oleate mixture
HPLC high-performance liquid chromatography
IPM isopropyl myristate
LHHW Langmuir–Hinshelwood–Hougen–Watson kinetics
MCT medium-chain triacylglyceride
MPO medium percentage of oleate mixture
L (methyl) linoleate (cis,cis-C18:2)
M monoenes; all FAME monoenes
O (methyl) oleate (cis-monoene, C18:1))
P (methyl) palmitate (C16:0)
RTD residence time distribution (curve)
S (methyl) stearate (C18:0); all saturated species
SB soy bean oil
SFME sunflower methyl ester
σ;SH half-hydrogenated intermediate
T FAME with 1 double bond in the trans configuration
TAG triacylglyceride
TO trioleate
V;VH vacancy for FAME or hydrogen
VP variable-pressure experiment
122
notation
DB double bond
e experimental; effective
eq equilibrium
E,h hydrogenation path of E
ES elementary reaction step
f first hydrogen insertion forward
h model number
h;hyd hydrogenation (path)
H,H2 hydrogen
i component
IH intercrystalline holes
iso isomerization (path)
l liquid
m model; molecule
Ni nickel
O,h hydrogenation path of O
p particle
r reactor
ref reference
s solid; second hydrogen insertion
solv solvent
tot total
0 initial
O adsorbed (O)
~
∆H relative (dimensionless), see e.g., Table 3.3
123
literature cited
Literature Cited
Albright, L. F. Mechanism of hydrogenation of triglycerides. J. Am. Oil Chem. Soc. 1963, 40,
16–17, 26, 28–29.
Albright, L. F. Partial hydrogenation of triglycerides-Current status and recommendations
relative to the mechanism and processes. Fette, Seifen, Anstrichm. 1985, 87, 140–146.
Albright, L. F.; Wisniak, J. Selecitivity and isomerization during partial hydrogenation of
cottonseed oil and methyl oleate: Effect of operating variables. J. Am. Oil Chem. Soc.
1962, 39, 14–19.
Allen, R. R.; Kiess A. A. Isomerization during hydrogenation. I. Oleic acid. J. Am. Oil Chem. Soc.
1955, 32, 400–405.
Andersson, K.; Hell, M.; Löwendahl, L.; Schöön, N.-H. Diffusivities of hydrogen and
glyceryl trioleate in cottonseed oil at elevated temperature. J. Am. Oil Chem. Soc. 1974,
51, 171–173.
Babenkova, L. V.; Naidina, I. N.; Kokh, I. G. Investigation of chemisorption of hydrogen and
methyl linoleate on Ni-Ce/kieselgur. Kin. Catal. 1994, 35, 106–110.
Beenackers, A. A. C. M.; van Swaaij, W. P. M. Mass transfer in gas–liquid slurry reactors.
Chem. Eng. Sci. 1993, 48, 3109–3139.
Bern, L.; Hell, M.; Schöön, N.-H. Kinetics of the hydrogenation of rapeseed oil. I. Influence
of transport steps in kinetic study. J. Am. Oil Chem. Soc. 1975a, 52, 182–187.
Bern, L.; Hell, M.; Schöön, N.-H. Kinetics of the hydrogenation of rapeseed oil. II. Rate
equations of chemical reactions. J. Am. Oil Chem. Soc. 1975b, 52, 391–394.
Chen, N. Y.; Degnan Jr., T. F.; Morris Smith, C. Molecular transport and reaction in zeolites.
Design and application of shape selective catalysts. VCH Publishers, USA, 1994.
Coenen, J. W. E. The mechanism of the selective hydrogenation of fatty oils, in J.H. de Boer
(ed.), The mechanism of heterogeneous catalysis; Elsevier: Amsterdam, 1960 pp. 126–
158.
Coenen, J. W. E. Hydrogenation of edible oils. J. Am. Oil Chem. Soc. 1976, 53, 382–389.
Coenen, J. W. E. Catalytic hydrogenation of fatty oils. Ind. Eng. Chem. Fundam. 1986, 25,
43–52.
Coenen, J. W. E.; Boerma, H. Absorption der Reaktionspartner am Katalysator bei der
Fetthydrierung. Fette, Seifen, Anstrichm. 1968, 70, 8–14.
Colen, G. C. M.; Van Duijn, G.; Van Oosten, H. J. Effect of pore diffusion on the
triacylglycerol distribution of partially hydrogenated trioleoylglycerol. App. Catal. 1988,
43, 339–350.
Cordova., W. A ; Harriott, P. Mass transfer resistances in the palladium-catalyzed
hydrogenation of methyl linoleate. Chem. Eng. Sci. 1975, 30, 1201–1206.
Cousins, E. R.; Feuge, R. O. Hydrogenation of methyl oleate in solvents. J. Am. Oil Chem.
Soc. 1960, 37, 435–438.
Cumberland, D. J.; Crawford, R. J., The Packing of Particles. Elsevier Amsterdam, 1987.
Dietrich, E.; Mathieu, C.; Delmas, H.; Jenck, J. Raney-nickel catalyzed hydrogenations: gas–
liquid mass transfer in gas-induced stirred slurry reactors. Chem. Eng. Sci. 1992, 47,
3597–3604.
Drozdowksi, B.; Zajac, M. Kinetics of nickel catalyst poisoning. J. Am. Oil Chem. Soc. 1980,
57, 149–153.
Dumez, F. J.; Hosten, L. H.; Froment, G. F. The use of sequential discrimination in the kinetic
study of 1-butene dehydrogenation. Ind. Eng. Chem., Fundam. 1977, 16, 298–301.
Dutton, H. J. Hydrogenated fats: processing, analysis and biological implications. Chem. Ind.
1982, 2 jan., 9–17.
124
literature cited
Dutton, H.; Scholfield, C.; Selke, E.; Rohwedder, W. K. Double bond migration, geometric
isomerization, and deuteric distribution during heterogeneous catalytic deuteration of
methyl oleate. J. Catal. 1968, 10, 316–327.
Edvarsson, J.; Irandoust, S. Poisoning of nickel-based catalyst in fat hydrogenation: a
literature review. J. Am. Oil Chem. Soc. 1993, 70, 1149–1156.
Eiteman, M. A.; Goodrum, J. W. Density and viscosity of low-molecular weight triglycerides
and their mixtures. J. Am. Oil Chem. Soc. 1994, 71, 1261–1265.
Eldib, I. A.; Albright, L. F. Operating variables in hydrogenating cottonseed oil. Ind. Eng.
Chem. 1957, 49, 825–831.
Emig, G.; Hosten, L. H. On the reliability of parameter estimates in a set of simultaneous
nonlinear differential equations. Chem. Eng. Sci 1974, 29, 475–483.
Emken, E. A. Dispelling misconceptions with stable isotopes. INFORM, 1995, 5, 906-921.
Ergun, S. Fluid flow through packed columns. Chem. Eng. Process 1952, 48, 89–94.
Everett, D. H. Reporting data on adsorption from solution at the solid/solution interface, Pure
and Appl. Chem. 1986, 58, 968–984.
Feuge, R. O. Hydrogenation of glyceride oils. pp. 413–451. Catalysis-III: Hydrogenation and
dehydrogenation. Reinhold Publishing, 1955.
Feuge, R. O.; Pepper Jr., M. B. ; O'Connor, R. T.; Field, E. T. Modification of vegetable oils.
XI. The formation of trans isomers during the hydogenation of methyl oleate and
triolein. J. Am. Oil Chem. Soc. 1951, 28, 420–426.
Fisher, R. A. Statistical methods for research workers, 14th ed., Macmillan: New York, 1970.
Fouilloux, P. The nature of Raney nickel, its adsorbed hydrogen and its catalytic activity for
hydrogenation reactions (review). App. Catal. 1983, 8, 1–42.
Froment, G. F.; Bischoff, K. B. Chemical reactor analysis and design; John Wiley & Sons:
New York, 1979.
Froment, G. F.; Hosten, L. H. Catalysis kinetics: modelling. Catalysis, Science and
technology; vol. 2, edited by Anderson, J. R.; Boudart, M.; Springer-Verlag: Berlin,
1981.
Ganguli, K. L. Measurements of H2/edible oil interfacial area in an agitated hydrogenator
using a Ziegler–Natta catalyst. PhD thesis. Technical University of Delft, 1975.
Ganguli, K. L.; Van den Berg, H. J. Edible oil hydrogenation rates in the presence of a
homogeneous Ziegler–Natta catalyst in a film reactor. Chem. Eng. Sci. 1978, 33, 27–34.
García-Ochoa, F; Santos, A. Effective diffusivity under inert and reaction conditions. Chem.
Eng. Sci. 1994, 49, 3091–3102.
Giddings, J. C. Kinetic origin of tailing in chromatography. Anal. Chem. 1963, 35, 1999–
2002.
Giddings, J. C.; Eyring, H. A molecular dynamic theory of chromatography. J. Phys. Chem.
1955, 59, 416–421
Graaf, G. H. The synthesis of methanol in gas–solid and gas–slurry reactors. PhD thesis,
University of Groningen, 1988.
Graaf, G. H.; Winkelman, J. G. M.; Stamhuis, E. J.; Beenackers, A. A. C. M. Kinetics of the
three phase methanol synthesis. Chem. Eng. Sci. 1988a, 43, 2161–2168.
Graaf, G. H.; Stamhuis., E. J.; Beenackers, A. A. C. M.; Kinetics of low-pressure methanol
synthesis. Chem. Eng. Sci. 1988b, 43, 3185–3195.
Grau, R. J.; Cassano, A. E.; Baltan↔s, M. A. Kinetics of methyl oleate catalytic
hydrogenation with quantitative evaluation of cis–trans isomerization equilibrium. Ind.
Eng. Chem. Process Des. Dev. 1986, 25, 722–728.
Grau, R. J.; Cassano, A. E.; Baltan↔s, M. A. The Cup-and-Cap reactor: a device to eliminate
induction times in mechanically agitated slurry reactors operated with fine catalyst
particles, Ind. Eng. Chem. Res. 1987a., 26, 18–22.
Grau, R. J.; Cassano, A. E.; Baltan↔s, M. A. Solution of a complex reaction network. The
methyl-linoleate catalytic hydrogenation. Chem. Eng. Comm. 1987b, 58, 17–36.
125
literature cited
Gut, G.; Kosinka, J.; Prabucki, A.; Schuerch, A Kinetics of the liquid-phase hydrogenation
and isomerization of sunflower seed oil with nickel catalysts. Chem. Eng. Sci. 1979, 34,
1051–1056.
Hashimoto K.; Muroyama, K.; Nagata, S. Kinetics of the hydrogenation of fatty oils. J. Am.
Oil Chem. Soc. 1971, 48, 291–295.
Haynes, H. W. The experimental evaluation of catalyst effective diffusivity. Catal. Rev.-Sci.
Eng. 1988, 30, 563–627.
Haynes Jr., H. W.; Sarma, P. N. A. Model for the application of gas chromatography to
measurements of diffusion in bidisperse structured catalysts. AIChE J. 1973, 19, 1043–
1046.
Heertje, I; Boerma, H. Selectivity and monoene isomerization in the catalytic hydrogenation
of polyenoic fatty acid methyl esters. J. Catal. 1971, 21, 20–26.
Heertje, I.; Koch, G. K.; Wösten, W. J. Mechanism of heterogeneous catalytic cis–trans
isomerization and double-bond migration of octadecenoates. J. Catal. 1974, 32, 337–
342.
Hejtmánek, V.; Schneider, P. Diffusion of large molecules in porous glas. Chem. Eng. Sci.
1994, 49, 2575–2584.
Horiuti, J.; Polanyi, M. Exchange reactions of hydrogen on metallic catalysts. Trans. Faraday
Soc. 1934, 30, 1164–1172.
Johnson, A. D.; Daley, S. P.; Utz, A. L.; Ceyer, S. T. The chemistry of bulk hydrogen:
reaction of hydrogen embedded in nickel with adsorbed methyl. Science 1992, 257, 223-
225.
Jonker, G. H.; Hoffmann, A. C.; Beenackers, A. A. C. M. Classification mechanism of the
chute, a liquid phase removal of fines in the micron range from a batch of particles.
Powder Techn. 1997, 90, 251-258.
Katan, M. B.; Zoch, P. L.; Mensink, R. P. Trans fatty acids and their effects on lipoproteins in
humans. Ann. Rev. Nutr. 1995, 15, 473–493.
Kinza, H.; Paseka, I.; Popova, N. M. Activity and physical properties of Ni/SiO2
hydrogenation catalysts. Coll. Czech. Chem. Commun. 1985, 50, 912–919.
Komiyama, H.; Smith, J. M. Intraparticle mass transport in liquid-filled pores. AIChE J. 1974,
20, 728–734.
Konvalinka, J. A., van Oeffelt, P. H., Scholten, J. J. F. Temperature programmed desorption
of hydrogen from nickel catalysts. Applied Catal 1981, 1, 141–158.
Le Page, J.-F.; Cosyns, J.; Courty, P.; Freund, E.; Franck, J.-P. and others. Applied
heterogeneous catalysis. Design, manufacture and use of solid catalysts. Éditions
Technip, Paris, 1987.
Leyva-Ramos, R.; Geankoplis, C. J. Diffusion in liquid-filled pores of activated carbon. I.
pore volume diffusion. Can. J. Chem. Eng. 1994, 72, 262–271.
Lidefelt, J.-O. Adsorption equilibrium constants of methyl oleate and methyl linoleate in
vapor phase on supported copper and nickel catalysts. J. Am. Oil Chem. Soc. 1983, 60,
593–599.
Lidefelt, J.-O.; Magnusson, J.; Schöön, N.-H. Vapor-phase hydrogenation of methyl oleate in
the presence of a supported nickel catalyst. J. Am. Oil Chem. Soc. 1983, 60, 603–607.
Lin, Y. S.; Ma, Y. H. A comparative chromatographic study of liquid adsorption and
diffusion in microporous and macroporous adsorbents. Ind. Eng. Chem. Res. 1989, 28,
622–630.
Lin, Y. S.; Ma, Y. H. Analysis of liquid chromatography with nonuniform crystallite
particles. AIChE J. 1990, 36, 1569–1576.
Litchfield, C.; Lord, J. E.; Isbell, A. F.; Reiser, R. Cis–trans isomerization of oleic, linoleic
and linolenic acids. J. Am. Oil Chem. Soc. 1963, 40, 553–557.
Ma, Y. H.; Lin, Y. S. Adsorption and diffusion of liquids in Silicate using HPLC. AIChE
Symposium series 1987, 83 (no. 259), 1–10.
Machiels, C. J.; Anderson, R. B. Hydrogenolysis of isopentane on Nickel. J. Catal. 1979, 60,
339–340.
126
literature cited
Magnusson, J. Competition for active sites between hydrogen and methyl esters of fatty acids
in vapor phase on α-Al 2O3-supported copper and nickel catalysts. Ind. Eng. Chem. Res.
1987a. 26, 874–877.
Magnusson, J. H2/D2 exchange as a model reaction for studying hydrogen adsorption on α-
Al2O3-supported copper and nickel catalysts. Ind. Eng. Chem. Res. 1987b, 26, 877–881.
Marangozis, J.; Keramidas, O. B.; Paparisvas, G. Rate and mechanism of hydrogenation of
cottonseed oil in slurry reactors. Ind. Eng. Chem., Process Des. Dev. 1977, 16, 361–369.
McGreavy, C; Draper, L.; Kam, E. K. T. Methodologies for the design of reactors using
structured catalysts: modelling and experimental study of diffusion and reaction in
structured catalysts. Chem. Eng. Sci. 1994, 49, 5413–5425.
Mendioroz, S.; Muñoz, V. Effect of the method of preparation on the activity of nickel-
kieselguhr for vegetable oil hydrogenation. Appl. Cat. 1990, 66, 73–90.
Mikahilenko, S. D.; Khodareva, T. A.; Leongardt, E. V.; Lyashenko, A. I.; Fasman, A. B. The
effect of redox treatment on the structural, adsorptive and catalytic properties of Raney
Nickel. J. Catal. 1990, 141, 688–699.
Mills, A. K.; Hockey, J. A. Selective adsorption of methyl esters of n-fatty acids at the
Silica/Benzene and Silica/Carbon tetrachloride interface. part 1: Adsorption isotherms
and part 2. Heats of adsorption. Trans. Faraday Soc. 1975, 71, 2384–2397.
Münzing, M. Kinetische untersuchungen zur selektiven hydrierung und isomerisierung
pflanzlicher Öle an Kupferchromit. PhD thesis. E.T.H. Zürich, 1986.
Münzing, M.; Kut, O. M.; Gut, G. Kinetik der Fetthärtung und Vergleich verschiedener
Katalysatoren. Fette, Seifen, Anstrichm. 1986, 88, 387–391.
Niklasson, C.; Andersson, B.; Schöön, N.-H. Influence of hydrogen pressure on selectivity in
consecutive hydrogenation reactions. Ind. Eng. Chem. Res. 1987, 26, 1459–1463.
Nielsen, K.; Hansen, H. J. M.; Nielsen, V. R. Selectivity in the hydrogenation of oleic-linoleic
acid oils with commercial nickel catalysts. J. Am. Oil Chem. Soc. 1960, 37, 271–274.
Park, S. H.; Kim, Y. G. The effect of chemical reaction on effective diffusivity within
biporous catalysts, Chem. Eng. Sci, 1984, 39, 523–531 (I) and 533–549 (II).
Parfitt, G. D.; Rochester, C. H. Adsorption of small molecules. From: Adsorption from
solution at the solid/liquid interface.
Pihl, M.; Schöön, N.-H. Rate factors in liquid phase hydrogenation. I. Kinetics of the
hydrogenation of cottonseed oil in the presence of a solid catalyst. Acta Polytech. Scand.
Chem. Ind. Met. Ser. 1971, 100, 4–36.
Prairie, M. R; Bailey, J. E. Experimental and modelling investigations of steady-state and
dynamic characteristics of ethylene hydrogenation on Pt/Al 2O3. Chem. Eng. Sci. 1987,
42, 2085–2102.
Press, W. H.; Flannery, B. P.; Teukolsky, S. A.; Vetterling, W. T. Numerical Recipes,
Cambridge University Press, 1987.
Reid, R. C.; Prausnitz, J. M.; Poling, B. E. The properties of gases and liquids. 4th ed.
McGraw-Hill, New York, 1987.
Ray, J. D. and Carr, B.T. Emperical modeling of soybean oil hydrogenation. J. Am. Oil Chem.
Soc. 1985, 62, 1218–1222.
Rodrigo, M. T.; Daza, L.; Mendioroz, S. Nickel supported on natural silicates. Activity and
selectivity in sunflower seed oil hydrogenation. App. Catal. 1992, 88, 101–114.
Rogers, D. W.; Hoyte, O. P. A.; Ho, R. K. C. Heats of hydrogenation of large molecules. Part
2-six unsaturated and polyunsaturated fatty acids. J. Chem. Soc. Faraday Trans. 1 1977,
74, 46–52.
Ruthven, D. M. Principles of adsorption and adsorption processes. John Wiley&Sons, 1984.
Satterfield, C. N.; Colton, C. K.; Pitcher, Jr., W. H. Restricted diffusion in liquids within fine
pores. AIChE J. 1973, 19, 628–635.
Stenberg, O.; Schöön, N.-H. Aspects of the graphical determination of the volumetric mass-
transfer coefficient (kla) in liquid-phase hydrogenation in a slurry reactor. Chem. Eng.
Sci. 1985, 40, 2311–2319.
127
literature cited
Stefanovic, S.; Albright, L. F. Catalyst studies for hydrogenation of vegetable oils. J. Am. Oil
Chem. Soc. 1969, 46, 139–142.
Stefanovic, S.; Price, R. H.; Albright, L. F. Hydrogenation and isomerization of methyl 9-
octadecenoates. Fette, Seifen, Anstrichmittel 1978, 80, 464–469.
Susu, A. A.; Ogunye, A. F. Nickel-catalyzed hydrogenation of soybean oil: I. Kinetic,
equilibrium and mass transfer determinations. J. Am. Oil Chem. Soc. 1981, 58, 657–661.
Susu, A. A.; Ogunye, A. F.; Onyegbado, C. O. Kinetics and mechanism of nickel-catalysed
palm oil hydrogenation. J. Appl. Chem. Biotechn. 1978, 28, 823–833.
Tumer, H. V.; Feugo, R. O.; Ward, T. L.; Cousins, E.R. Relative reactivity toward
hydrogenation of the oleoyl group in the 2- and 1,3 positions of triglycerides. J. Am. Oil
Chem. Soc. 1964, 6, 413–415.
Tsuto, K.; Harriott, P.; Bischoff, K. B. Particle mass transfer effects and selectivity in the
palladium-catalyzed hydrogenation of methyl linoleate. Ind. Eng. Chem. Fundam. 1978,
17, 199–205.
Van der Plank, P. Isomerization phenomena during hydrogenation of methyl oleate and
methyl elaidate over nickel-silica catalysts. J. Am. Oil Chem. Soc. 1972a, 49, 327–332.
Van der Plank, P. A criterion for detecting mass transport effects on double bond migration
during the hydrogenation of methyl oleate using nickel-on-silica catalysts. J. Catal.
1972b, 26, 42–50.
Van der Plank, P.; Van Oosten, H. J. Study of the mechanism of double-bond isomerization in
methyl 9-octadecenoates. J. Catal. 1975, 38, 223–230.
VandenHeuvel, F. The kinetics of the hydrogenation of pure fatty esters. I. Mono-olefinic
systems: methyl oleate. J. Am. Oil Chem. Soc. 1956, 33, 347–350.
Veldsink, J. W.; Bouma, M. J.; Schöön, N.-H.; Beenackers, A. A. C. M. Heterogeneous
hydrogenation of vegetable oils: a literature review. Cat. Rev.-Sci. Eng. 1997, 39, 253–
318.
Wakeham, H.; Magne, F. Viscosities and densities of hydrogenated cottonseed oils. Ind. Eng.
Chem. 1946, 36, 568–570.
Weller, S. W. Kinetics of heterogeneous catalysed reactions. Catal. Rev.-Sci. Eng. 1992, 34,
227–280.
Westerterp, K. R.; Van Swaaij, W. P. M.; Beenackers, A. A. C. M. Chemical reactor design
and operation. John Wiley & Sons: New York, 1987.
Wisniak, J.; Albright, L. F. Hydrogenating cottonseed oil at relatively high pressure. Ind. Eng.
Chem. 1961, 53, 375–380.
Zwicky, J. J.; Gut, G. Kinetics, poisoning and mass transfer effects in liquid-phase
hydrogenations of phenolic compounds over a palladium catalyst. Chem. Eng. Sci. 1978,
33, 1363–1369.
128
Een Overzicht van het Onderzoek
-Hydrogeneren van eetbare oliën en vetten-
Het Hydrogeneringsproces
De stad met huizen en straten op de omslag vormen een voorstelling van het
chemische proces dat centraal staat in dit proefschrift: hydrogeneren van eetbare oliën
en vetten. Het hydrogeneringsproces wordt al bijna een eeuw industrieël toegepast in
onder andere het maken van margarine uit zonnebloem- en sojaolie. Hydrogeneren is
het chemisch binden van waterstof aan de vloeibare oliën, waardoor ze veranderen in
vet. Ondanks de vele ervaring worden er nog steeds teveel ongewenste, voor de mens
minder gezonde, bijprodukten gevormd. In het proces zijn vele reaktiestappen en
(bij)produkten te onderscheiden die in het, vaak praktisch gerichte, onderzoek aan dit
proces zelden apart worden gemeten. In dit onderzoek is de aanpak om juist eerst de
deelstappen te onderzoeken en zo stap voor stap de kennis over het proces op te
bouwen.
De eetbare olie- of vetmolekulen heten triacylglycerides en bestaan uit drie
onderling verbonden staarten, de vetzuren, zie Figuur 1. Bij een volledig ‘verzadigd’
vet (hoog smeltpunt) zijn alle koolstofmolekulen van de vetzuren geheel bezet
(verzadigd) met waterstof. Deze vetzuren heten stearine(zuur) (S). Echter, in geval van
een olie komen in de vetzuren zogenaamde onverzadigde verbindingen voor, in twee
oliezuur H H H H O H
H
H H H H
C C C C C C C C C O C H stearinezuur
C H H H H H H H H H H H H H
H H H H O H H H H H H H H
H
C C C C C C C C CH
C C
CH O C H C H C H C H C H C H C H C H C
H
H
H
H H H H O H H H H H H H H
H
H H H H
C C
H C C C C C C C C C O C H
H
C
H
H H H H H
H H H H
R1 R2 R1
H
H
C C
C C
H
C C
H
linolzuur C C
H
H H H R2
H
H
C C
isomerisatie
C
H
C C
H
H
H
H H H2 H H
C C
R1 C C R2 R1 C C R2
H H
C
hydrogenatie
Figuur 1. Een oliemolekuul met een geheel verzadigde (stearinezuur), een enkelvoudig (oliezuur)
en een dubbel (linolzuur) onverzadigde vetzuurketen. De hydrogenerings en isomerisatiereaktie.
130
een overzicht van het onderzoek
ruimtelijke strukturen (isomeren), namelijk cis, zoals oleaat uit Figuur 1, en trans
(elaidaat). Beide isomeren verlagen het smeltpunt. Bij het hydrogeneren van de
onverzadigde verbindingen (verzadigen met waterstof, zie Figuur 1) stijgt het
smeltpunt en daarom wordt hydrogeneren ook het ‘harden van olie’ (tot vet) genoemd.
Als voorbeeld het hydrogeneren van zonnebloemolie. Zonnebloemolie bevat veel
onverzadigde verbindingen, onder
andere het linoleaat, L, (nauw
verwant met linolzuur) uit Figuur
gehalte (%)
1 (twee cis verbindingen). In het 90
verloop van de reaktie ontstaan S
allerlei tussenprodukten. Figuur 2 (cis,cis)
toont hoe tijdens de 60 L
hydrogenering de samenstelling O O E
(cis) (trans)
van het reaktiemengsel in het
30 S
reaktievat verandert gedurende de L (verzadigd)
De katalysator
131
een overzicht van het onderzoek
Onderzoek en Resultaten
Hydrogenering van enkelvoudig onverzadigde verbindingen. Het proefschrift
begint met een eenvoudig hydrogeneringsproces: de hydrogenering van een
enkelvoudig onverzadigde verbinding. De reaktie verloopt via een paar elementaire
stappen, zoals adsorptie van de enkelvoudige onverzadigde verbinding en de waterstof
aan het katalysatoroppervlak, en de eigenlijke reaktie van het waterstof met de
onverzadigde verbinding. Als benadering stellen we nu (een bekend gegeven uit de
literatuur) dat in feite maar 1 of 2 snelheidsbepalende stappen de reaktiesnelheid
dikteren. De formules om de hydrogenering van een willekeurig mengsel
onverzadigde verbindingen bij een bepaalde reaktietemperatuur en waterstofdruk te
voorspellen worden hiermee een stuk versimpeld. Voor deze reaktie is die
snelheidsbepalende stap niet bekend. Door allerlei formules (ook wel modellen
132
een overzicht van het onderzoek
133
een overzicht van het onderzoek
134
een overzicht van het onderzoek
3
2
135
een overzicht van het onderzoek
2
4
1 3
Figuur 6. De adsorptie-
opstelling. 1 = reaktor, 2 =
toevoer gassen, 3 =
monstername, 4 = bevestiging
schudmechanisme.
136
een overzicht van het onderzoek
Verantwoording
Dit proefschrift is de weerslag van het wetenschappelijk gedeelte van mijn
promotieonderzoek in de groep van prof. A.A.C.M. Beenackers. Onder de kritische en
gedreven leiding van Ton Beenackers en Jan-Willem Veldsink is het onderzoek
geconvergeerd tot een serie artikelen en uiteindelijk tot dit proefschrift. Ik ben jullie
hiervoor zeer erkentelijk. Verder wil ik ook de (overige) leden van de
beoordelingscommissie, prof. Nils-Herman Schöön, prof. Jan Teuben en prof. Hans
Wesselingh bedanken voor hun kritische beschouwingen en nauwkeurige correcties
van het proefschrift.
Er staat echter heel veel niet vermeld in dit boekje: theoretische beschouwingen
op een achteraf doodlopend spoor, het bouwen, testen en uitvoeren van series
experimenten, rapportage en diskussies ∗. Oftewel de dagelijkse gang van zaken tijdens
het onderzoek, in een groep van begeleiders, afstudeerders en technici. De kern van het
onderzoek is voor mij de samenwerking met 12 eerst-fase studenten geweest: Joost
Demmink, Antoine Wellink, Steven Osse, Iwan van Gils, Erik Roelofs, Bastiaan van
Hasselt, Paul Koekoek, Jeroen van Schreven, Bart Broens, Rein Schurer, Erik
Westerhof en Luuk Smeets. Al bouwend, metend, redenerend en ‘cheffend’ vormden
we jarenlang een enthousiaste en gezellige groep op zaal 18-256. Jullie inbreng was
een grote drijfveer voor mijn onderzoek.
Het reilen en zeilen op het lab en daarbuiten is nauw verbonden geweest met de
collega-aio’s Reinoud Noordman, Tjaart Molenkamp, Machteld van den Burgh,
Marcel Ottens, Gerard van der Laan en Martin Oosterom. Joost Demmink en Martin
Bouma nemen hierin een speciale plaats in. Met Joost begon mijn promotieonderzoek,
Joost was al die tijd mijn kamergenoot, mede-cursist en collega-fokkenist, heeft
hoofdstukken gelezen en is tenslotte paranimf. Martin was niet alleen altijd bereid om
mee te denken of mee te klussen, maar ook in zijn parallel/vervolg onderzoek op dit
gebied was hij een prima compagnon, hetgeen onder andere geresulteerd heeft in een
gezamelijk hoofdstuk (hoofdstuk 3).
Mijn onderzoek is financieel en wetenschappelijk ondersteund door Unilever
Research Vlaardingen, waarvoor ik met name dr. Guus Trommelen, ir. Ad Rozendaal
en dr. ir. Wicher T. Koetsier erkentelijk ben.
∗
“Experience is the name everyone gives to their mistakes.” (Oscar Wilde, 1854–1900)
137
een overzicht van het onderzoek
Jan-Henk Marsman, Oetze Staal, Luuk Balt en Karel van der West dank ik voor
hun technische en praktische inbreng. In een stage zette Frits Lichtenbelt adequaat de
GC-analyse op poten. Technicus Jan Bolhuis heeft veel bijgedragen aan de opbouw
van de opstellingen.
Ik wil Janine Terhorst bedanken voor het mee-ontwerpen van de voorkant.
Tenslotte: de steun van paranimf (en zusje) Elly, mijn ouders en Esther is misschien
wel het belangrijkste geweest.
Levensloop
Gerald Jonker is op 25 april 1967 geboren in Geleen, maar woont nu al een kleine
30 jaar in Groningen, waar hij na het VWO, de opleiding technische scheikunde
voltooide in de groep van prof. A.A.C.M. Beenackers. Aan de RuG heeft hij verder
aktief deelgenomen aan de studentenpolitiek en bijgedragen aan het ontwikkelen van
cursussen. Sinds enkele jaren is hij als zelfstandig ondernemer veel betrokken bij
(culturele) evenementen in de stad Groningen. Met zijn eenmanszaak Gromit
Maatwerk & Technologie neemt hij advieswerk aan voor de chemische technologie,
maar ook de fabricage van décor- en meubelstukken.
138