Polymer Testing 32 (2013) 9–14

Contents lists available at SciVerse ScienceDirect

Polymer Testing
journal homepage: www.elsevier.com/locate/polytest

Property modelling

A knowledge based approach for elastomer cure kinetic parameters

estimation
Eric Leroy a, *, Anouar Souid b, Alain Sarda b, Rémi Deterre b
a
LUNAM Université, CNRS, GEPEA, UMR 6144, CRTT, 37, avenue de l’université, 44606 St Nazaire Cedex, France
b
LUNAM Université, IUT de Nantes, CNRS, GEPEA, UMR 6144, OPERP ERT1086, 2, avenue du Professeur Jean Rouxel, BP 539, 44475 Carquefou Cedex, France

a r t i c l e i n f o a b s t r a c t

Article history: Isoconversionnal kinetic analysis was applied for the first time to the isothermal cure
Received 9 July 2012 reaction of an elastomer compound. The apparent activation energy obtained was almost
Accepted 27 August 2012 constant, regardless of the degree of cure. Such a result justifies the use of simple
phenomenological kinetic models with a unique activation energy, and allows develop-
Keywords: ment of a knowledge based approach to kinetic parameters estimation, applied to two
Elastomer
models frequently used for the modelling of elastomer cure reactions: the Isayev and
Rheology
Kamal & Sourour models. Simplified Kamal & Sourour models with a reduced number of
Cure
Kinetics kinetic parameters were obtained. Simulations with such models give a better description
Parameter identification of experimental data than the Isayev model, especially taking into account the simulation
of the rate of reaction.
Ó 2012 Elsevier Ltd. All rights reserved.

1. Introduction reaction can be described by a degree of cure «a» ranging

During the last 50 years, a considerable amount of work
has been published on the modelling of rubber cure
kinetics. Different mechanistic models, more or less
complex, have been proposed [1–3]. Nevertheless, as
reviewed by Ghoreishy [4], for application purposes, such
as numerical simulation of the vulcanization state during
reactive moulding, phenomenological models with
a limited number of parameters are generally preferred,
such as the Isayev [5] and Kamal & Sourour models [6]. This
is also the case when numerical simulation is used for cure
cycle optimisation [7].
Both Isayev and Kamal & Sourour models (presented in
detail below), allow modelling the elastomer cure reaction,
but neglect the reversion phenomenon. As shown on Fig. 1,
the state of a rubber during an isothermal vulcanisation
from 0 (uncured) to 1 (complete reaction). This degree of
cure is generally obtained from dynamic rheological
measurements of elastic modulus G0 : During the complex
vulcanization reaction, the formation of sulphur crosslinks
leads to an increase of G0 , initially equal to G0. Three steps
are generally observed: An induction period (Fig. 1), during
which the modulus increases very slowly (a remains close
to zero), followed by the cure step (Fig. 1), with a sharp
increase of the elastic modulus (and a). During this step,
the modulus increases up to a maximum value Gmax. For
high reaction temperatures, typically above 140  C, the
modulus can decrease due to the reversion phenomenon.
This is due to the fact that some of the sulphur crosslinks
previously formed are unstable. In the present work, the
modelling of this phenomenon will not be addressed.
Consequently, the degree of cure at a reaction time t can
be defined as:

GðtÞ  G0
* Corresponding author. Tel.: þ33 (0) 240172660; fax: þ33 (0) aðtÞ ¼ (1)
240172618. Gmax  G0
E-mail addresses: eric.leroy@univ-nantes.fr (E. Leroy), anouar.souid@
univ-nantes.fr (A. Souid), alain.sarda@univ-nantes.fr (A. Sarda), remi. The Isayev model [5] is based on a discontinuous math-
deterre@univ-nantes.fr (R. Deterre). ematical formalism (Equations (2)–(5)): During the

0142-9418/$ – see front matter Ó 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.polymertesting.2012.08.012
10 E. Leroy et al. / Polymer Testing 32 (2013) 9–14


E1
k1 ¼ A1 exp  (7)
RT

Fig. 1. The three steps of natural rubber sulphur vulcanization: a) induction
period, b) cure step and c) reversion (neglected in the present work).
induction period, the degree of cure a and the reaction rate
da/dt are equal to zero. This period is described by an
induction time ti: Coran [1] defined ti as the intersection of
the time axis with the tangent at the inflexion point of the
curve a(t) (Fig. 1). For reaction times higher than ti, the
degree of cure a and the reaction rate da/dt are given by
analytic functions (Equations (3) and (5), respectively).
The temperature dependence of the induction time and
the rate constant k are described by Arrhenius equations
(Equations (2) and (4), respectively)


Ei
ti ¼ t0 exp
RT


E2
k2 ¼ A2 exp  (8)
RT
It is worth pointing out that, for application purposes
such as numerical simulation of rubber moulding
processes, the discontinuous mathematical formalism of
the Isayev model can lead to numerical convergence
problems, contrary to the Kamal & Sourour model.
Nevertheless, the Isayev model has the advantage of
having only five parameters (t0, Ei, A, E and exponent n), and
an analytical form, while the Kamal & Sourour model has
six parameters (A1, E1, A2, E2 and exponents m and n) and
has to be solved numerically.
Moreover, the identification of these parameters is
easier in the case of the Isayev model: The usual parameter
estimation technique consists of two steps: First, the two
parameters describing the induction time (Equation (2))
are estimated by linear regression after plotting the natural
logarithm of ti as a function of 1/T.
Then, the remaining parameters (Equation (3)) are
identified by numerical optimisation, for example by
minimizing the ordinary least squares criterion (OLS):
X 2
(2) OLS ¼ aexpðtÞ  amodel ðtÞ (9)

where: aexp is the experimental value and amodel is the

prediction of the model.
n
kðt  ti Þ In contrast, for the Kamal & Sourour model, the
aðtÞ ¼ (3)
1 þ kðt  ti Þn parameter estimation technique generally consists of
a single numerical optimization (by minimizing the above
OLS criterion, for example) in which all the six parameters

 are identified “simultaneously”.
E In both cases, it has to be pointed out that the use of
k ¼ A exp  (4)
RT numerical optimization to simultaneously identify the
activation energy Ex and the frequency factor Ax of an
Arrhenius equation is questionable, since it is well known
1 0 that these parameters are strongly correlated [10].
E
da h i Therefore, there is a risk of identifying pairs of param-
B C
¼ a n ð1  aÞ n nAn exp@ n A
n1 nþ1 1
(5) eters (Ex, Ax) that allow simulating the curves a(t), but that
dt RT
do not have any physical meaning. Of course, such a risk is
more important in the case of the Kamal & Sourour model,
In contrast, the Kamal & Sourour model [6] is based on
in which two pairs (A1, E1 and A2, E2) have to be identified,
a differential equation (Equation (6)) that can be solved
as demonstrated by Scoot & Saaas [11]. This perhaps is
numerically (for example using the method of Runge &
another reason why the Isayev model is used more
Kutta [8]) to simulate both the degree a and the reaction
frequently for vulcanization kinetics modelling.
rate da/dt at any reaction time t, using as an initial condi-
The problem of Arrhenius kinetic parameters estimation
tion a ¼ 0 at t ¼ 0. Note that, as described by Arrillaga et al.
for different kinds of chemical reactions has led to
[9], Equation (6) can also be solved to describe only the cure
considerable research efforts in the field of thermal analysis
step (Fig. 1) by taking as initial condition a ¼ 0 at t ¼ ti; in
[10]. An approach known as the “model free” kinetic
such a case, the induction period should be described by
analysis has been developed. It consists of assuming that
Equation (2)). In both cases, the temperature dependence
the reaction kinetics can be described by the following
of the rate constants k1 and k2 are described by Arrhenius
differential equation:
equations (Equations (7) and (8)), respectively).


da da Ea
¼ ðk1 þ am k2 Þð1  aÞn (6) ¼ Aa exp  f ðaÞ (10)
dt dt RT
 E. Leroy et al. / Polymer Testing 32 (2013) 9–14
where: Aa, the apparent frequency factor; and Ea the
apparent activation energy for the degree of cure a.
In the case of a simple reaction for which Aa and Ea are
constants, the following integration of Equation (10) can be
performed for isothermal reaction conditions:


Ea
exp þ
RT da
dt ¼
Aa f ðaÞ


Ea
Z exp þ Z aj
RTi da
taj i ¼ dt ¼
Aa 0 f ðaÞ
2R 3 rubber G0 . Two measurements were performed for each
aj da
  6 0 f ðaÞ7 Ea
ln taj ;i ¼ ln6 7
4 Aa 5 þ RTi
where: taj ;i is the reaction time at which the degree of cure
reaches a value a ¼ a j at a given reaction temperature T i.

R a da
For a given value of a j, the termln 0 j =Aa does not
f ðaÞ
depend on temperature. Consequently, by plotting lnðtaj ;i Þ
as a function of 1=RTi , a straight line should be obtained,
with a slope equal to Ea. Repeating this operation for
different values of a j, the evolution of Ea during reaction
can be obtained experimentally.
Given the assumptions made for the integration of
Equation (10), the apparent activation energy values
calculated by this method will be reliable only if Ea is
constant during the whole reaction. If Ea is not constant,
but has only small variations, a systematic but small error
will be made if it is calculated using Equation (13). Alter-
natively, if Ea shows strong variations throughout reaction,
the use of Equation (13) will only allow detection of these
strong variations, indicating the complexity of the overall
reaction, but the calculated quantitative values of Ea will
not be reliable. In such a case, more advanced calculations
should be performed in order to estimate accurate values of
Ea, as described by Vyazovkin [12].
Nevertheless, in all cases, the interest of this approach is
that the apparent activation energy obtained, t, depends on
neither the function f(a) nor the frequency factor Aa in
Equation (10). In other words, Ea can be determined inde-
pendently of the other kinetic parameters.
To our knowledge, such a “model free” kinetic analysis
has never been applied to the cure kinetics of a rubber
using rheological data. In the present paper, we propose to
use it to study the cure kinetics of a model natural rubber
compound. We would like to point out that our aim is not to
provide a detailed mechanistic model, but phenomeno-
logical models allowing simulation of experimental data,
and which could be used to predict the state of cure during
a moulding process within the temperature range used for
the study of the kinetics.
2. Experimental
The isothermal cure of a natural rubber compound
(Table 1) was studied at four different temperatures (130,
11
Table 1
Composition of the rubber compound.
Component Parts (by weight)
Rubber gum 100
Fillers (including carbon black) 25
Plasticizer 5
Sulphur based curing system 9
Other additives 2
(11)
140, 150 and 160  C) using a MDR type rheometer
(RPA2000, Alpha Technologies) [13] in dynamic mode
(12)
(1 Hz). The raw data obtained is complex torque S* the real
part of which S0 is related to the storage modulus of the
temperature, showing no significant dispersion. For each
(13) temperature, the S0 (t) curves obtained are used to calculate
the aexp(t) curves by assuming that G0 is proportional to S0 .
Equation (1) can, therefore, be modified as follows:
S0 ðtÞ  SminT
aexpðtÞ ¼ (14)
SmaxT  SminT
where Smin T is the lowest initial value of S0 at temperature
T, Smax T the maximum value. As we neglect the reversion
phenomenon, aexp values are not calculated after the value
Smax T is reached.
For each temperature, the induction time ti was esti-
mated by the Coran method [1] described in the intro-
duction section (Fig. 1). The two parameters describing the
induction time (Equation (2)) were estimated by linear
regression after plotting the natural logarithm of ti as
a function of 1/T.
The “model free” kinetic analysis of the aexp(t) curves at
different temperatures was performed using Equation (13),
as described in the introduction. The values of taj ;i for
different degrees of cure were obtained by linear interpo-
lation. For each temperature, the reaction rate curve (da/
dt)exp was also calculated by numerical derivation of the
aexp(t) curve.
All calculations were made using spreadsheet software
(Microsoft EXCEL). The numerical solving of the Kamal &
Sourour model was done using the 4th order Runge & Kutta
method [8]. For the numerical optimisation of the OLS
criterion (Equation (9)), the solver of the spreadsheet
software, based on a Generalized Reduced Gradient (G.R.G)
method [14] was used.
3. Results and discussion
The experimental torque curves (S0 ) obtained at
different temperatures are presented in Fig. 2. One can
observe that the reversion phenomenon is significant only
for the experiment conducted at 160  C. This shows that the
natural rubber compound used is an adequate model
system for studying vulcanisation kinetics neglecting
reversion. Fig. 3a and b shows the corresponding calculated
aexp(t) and daexp(t)/dt curves, respectively.
The plots ln(taj,i) vs 1/Ti used for the determination of
the apparent activation energy at different values of a are
presented in Fig. 4. One can observe that the linearity
assumed by Equation (13) is verified. Fig. 5 shows the
12 E. Leroy et al. / Polymer Testing 32 (2013) 9–14

Fig. 4. “Model free” kinetic analysis.

Fig. 2. Torque curves (S0 ) obtained for the different reaction temperatures.

In the case of the Isayev model, the comparison of

resulting values of Ea. It is clear that they have no significant
Equations (5) and (10) implies:
dependence on the degree of cure, at least for the range of
a tested (0.1 to 0.9). This indicates that the assumption of E
constant Ea, on which Equation (13) was obtained, is valid Ea ¼ (15)
n
for the studied rubber compound. Consequently, the
Thus, we have the condition E/n ¼ 92,700 J/mol, which
average value of Ea, equal to 92,700 J/mol, can be assumed
reduces the number of independent parameters to four.
to be reliable from a quantitative point of view. The
Following the two step procedure previously described, the
dispersion of experimental values and the standard devi-
parameters describing the induction time were first esti-
ation are 800 J/mol (0.86%) and 530 J/mol (0.57%),
mated by linear regression. Then the parameters n and A
respectively.
have been estimated by numerical optimisation of the OLS
The fact that the apparent activation energy of the
criterion. The values of the parameters estimated and of the
vulcanization kinetics is constant also gives us important
OLS criterion are given in Table 2. The corresponding
information for the estimation of the kinetic parameters of
simulations of a (t) and da(t)/dt curves are shown on Fig. 3a
the models of Isayev and Kamal & Sourour. A knowledge
and b, respectively.
based approach can be developed:
In the case of the Kamal & Sourour model, the theoret-
ical apparent activation energy is given by the following
equation [15]:

k1 E1 þ am k2 E2
Ea ¼ : (16)
k1 þ am k2
Since a constant value of Ea is observed, there are only
two possibilities: Either A1 (or A2) should be equal to zero,
and E2 (or E1) equal to 92,700 J/mol. Or, both E1 and E2
should be equal to 92,700 J/mol.

Fig. 3. Experimental curves and simulations obtained using the model of Fig. 5. Apparent activation energy for different values of degree of cure. The
Isayev: Degree of cure (a) and reaction rate (b). average value is shown by the dotted line.
 E. Leroy et al. / Polymer Testing 32 (2013) 9–14 13

Table 2
Kinetic parameters of the model used for simulations of a (t) and da(t)/dt curves: Model of Isayev (Fig. 3) and simplified Kamal & Sourour models (Fig. 7). In
Model 1 exponents n and m have been used as free parameters, while in Model 2, n and m are fixed. The OLS criterion values are given for each model.

Model of Isayev (Equations (2)–(5)) Simplified model of Kamal & Sourour (Equation (16))

“Model 1” n and m are optimized “Model 2” n and m are fixed

Ei 94,590 J/mol. A1 5.34  105 s1 A1 1.67  102 s1
Ai 8.70  1010 s A2 1.00  1010 s1 A2 1.00  1010 s1
E/n 92,700 J/mol E 92,700 J/mol E 92,700 J/mol
A 7.39  1016 s1 m 1.146 m 1
n 1.796 n 1. 930 n 2
OLS 0.314 OLS 0.244 OLS 0.292

Fig. 6 shows a plot [(da/dt)exp/(da/dt)max] vs aexp. This
kind of master curve is often used in thermal analysis of
chemical reactions to identify the shape of the function f(a)
when the apparent activation energy is constant [10]. One
can observe that the differences between the experimental
curves at different temperatures are relatively small and
that the average shape is typical of an autocatalytic reaction
[10]. Effectively, the prediction of the Isayev model (Equa-
tion (5), with the estimated parameters given in Table 2)
plotted in Fig. 6, gives a qualitative description of the data.
Nevertheless, a qualitatively better description is obtained
by using a simple 2nd order autocatalytic model:
da
fað1  aÞ
2
dt
Taking into account the above discussions, we propose
a simplified version of the Kamal & Sourour model:


da E corresponding kinetic parameters are given in Table 2): In
¼ ðA1 þ am  A2 Þ  exp   ð1  aÞm
dt RT
where: E ¼ 92,700 J/mol, and A1 << A2.
Such a model verifies both Equations (16) and (17) (for
n ¼ 1 and m ¼ 2) and can be solved numerically with an
initial condition a ¼ 0 at t ¼ 0. (Note that if A1 was equal to
zero, it would be impossible to solve the equation without
introducing an additional parameter such as a ¼ ai at t ¼ 0).
The kinetic parameters A1, A2, n and m, for the simula-
tion of experimental data have been estimated by numer-
ical optimization of the OLS criterion, taking the following
knowledge based initial conditions:
Fig. 6. Master curve [(da/dt)exp)/((da/dt)max] vs aexp and predictions of the
model of Isayev and of a simple 2nd order autocatalytic model.
- The initial values of n and m were taken equal to 2 and
1, respectively, according to the observations made on
Fig. 6.
- The initial value of A2 was estimated by considering the
normalization factors used for the plots on Fig. 6.
Effectively, assuming n ¼ 2 and m ¼ 1 and neglecting
A1 << A2, Equation (18) gives:




da E
¼ A2 exp  0:33 ð1  0:33Þ2 (19)
dt maxT RT
For each temperature, a value of A2 can, therefore, be
estimated. The average value obtained was
(17)
(8.5  1)  109 s1.
For A1, various initial values have been tested, ranging
from 1 to 10,000.
Two numerical optimisations have been performed (the
(18) the first optimisation, the four parameters (A1, A2, n and m)
have been identified. In the second optimisation, the
exponents n and m were kept constant (equal to 2 and 1,
respectively). The two simplified Kamal & Sourour models
obtained are referred to as “Model 1” and “Model 2”,
respectively, in Table 2. The Fig. 7a and b shows the cor-
responding simulations of a (t) and da(t)/dt curves,
respectively.
Finally, we obtained three models: the Isayev model and
two simplified Kamal & Sourour models. If we compare
only the simulations of a (t) curves (Figs. 3a and 7a), it
seems that there is qualitatively no significant difference in
terms of performance between the three models.
Nevertheless, the calculated OLS criterion values given
in Table 2 show that, quantitatively, the worst simulation is
given by the Isayev model. The OLS value for “Model 2”,
which requires only three kinetic parameters (E, A1, A2), is
significantly lower. Of course, an even lower value of OLS
criterion is obtained for “Model 1” due to the two addi-
tional free parameters (n and m). The Isayev model
particularly fails to describe the early stage of cure, as
observed by other authors [16].
If we now compare the simulations of the da(t)/dt
curves (Figs. 3b and 7b), it is striking that the simplified
Kamal & Sourour models gives a better description of the
experimental data. This is clearly due to the discontinuous
mathematical formalism of the Isayev model, that intro-
duces a discontinuity in the simulated da(t)/dt curves.
In future studies, the knowledge based kinetic param-
eter estimation approach that we have been able to develop
14 E. Leroy et al. / Polymer Testing 32 (2013) 9–14
Fig. 7. Experimental curves and simulations obtained using the simplified
models of Kamal & Sourour (Model 1 and Model 2): Degree of cure (a) and
reaction rate (b).
from the “model free” kinetic analysis of a model sulphur
cured natural rubber, will be tested on other systems (with
different sulphur contents, other vulcanization agents and/
or types of rubbers). It is important to point out that if
a dependence of Ea on a was to be obtained for these
systems (contrary to the present case), it would be possible
to use Equation (16) for modelling this dependence. Of
course, in such a case, the advanced integral isoconver-
sional methods described by Vyazovkin [12] should be used
in order to obtain reliable values of Ea from experimental
data, as explained in the introduction.
Another perspective is to develop a similar kinetic
parameter estimation approach for models allowing the
simulation of the reversion phenomenon, such as the Han
et al. model [3].
4. Conclusions
The “model free” kinetic analysis approach was applied
for the first time to study the vulcanization reaction of
a natural rubber compound. Such an approach showed that
the apparent activation energy of the vulcanization is
constant, regardless of the degree of cure. This allowed us
to develop a knowledge based approach for the estimation
of the kinetic parameters of two models frequently used to
describe vulcanization kinetics: The Isayev and of Kamal &
Sourour models. Two simplified Kamal & Sourour models
could be identified (with only five and three kinetic
parameters, respectively). Both are shown to give a better
description of experimental data than the Isayev model
(five kinetic parameters), especially taking into account the
simulation of the rate of reaction. The Kamal & Sourour
model has the advantage of allowing a continuous predic-
tion of the degree of cure and rate of reaction during the
induction period and cure step, due to its non-
discontinuous formalism.
Acknowledgments
The authors gratefully thank, for their financial support,
the French Region “Pays de le Loire”.
References
[1] A.Y. Coran, Vulcanization. Part VI. A model and treatment for scorch
delay kinetics, Rubber Chemistry and Technology 37 (1964) 689.
[2] R. Ding, A.I. Leonov, A kinetic model for sulfur accelerated vulcani-
zation of a natural rubber compound, Journal of Applied Polymer
Science 61 (1996) 455.
[3] I.S. Han, C.B. Chung, S.J. Kang, S.J. Kim, H.C. Chung, A kinetic model of
reversion type cure for rubber compounds, Polymer (Korea) 22
(1998) 223.
[4] M.H.R. Ghoreishy, Numerical simulation of the curing process of
rubber articles (Chap. 11), in: W.U. Oster (Ed.), Computational
Materials, Nova Science Publishers, New York, 2009, p. 445.
[5] A.I. Isayev, Injection Moulding of Rubber Compounds in ‘Injection
and Compression Moulding Fundamentals’, Marcel Dekker, New
York, 1987.
[6] M.R. Kamal, S. Sourour, Kinetics and thermal characterization of
thermoset cure, Polymer Engineering and Science 13 (1973) 59.
[7] A. El Labban, P. Mousseau, J.L. Bailleul, R. Deterre, Optimization of
thick rubber part curing cycles, Inverse Problems in Science and
Engineering 18 (3) (2010) 313.
[8] J.C. Butcher, The Numerical Analysis of Ordinary Differential Equa-
tions: Runge-Kutta and General Linear Methods, Wiley-Interscience,
New York, 1987.
[9] A. Arrillaga, A.M. Zaldua, R.M. Atxurra, A.S. Farid, Techniques used
for determining cure kinetics of rubber compounds, European
Polymer Journal 43 (11) (2007) 4783.
[10] S. Vyazovkin, A.K. Burnham, J.M. Criado, L.A. Pérez-Maqueda, C.
Popescu, N. Sbirrazzuoli, ICTAC kinetics committee recommenda-
tions for performing kinetic computations on thermal analysis data,
Thermochimica Acta 520 (1–2) (2011) 1.
[11] E.P. Scott, Z. Saad, Estimation of kinetic parameters associated with
the curing of thermoset resins. Part I: theoretical investigation,
Polymer Engineering & Science 33 (18) (1993) 1157.
[12] S. Vyazovkin, Modification of the integral isoconversional method to
account for variation in the activation energy, Journal of Computa-
tional Chemistry 22 (2) (2001) 178.
[13] J.S. Dick, H. Pawlowski, E. Scheers, Alternate instrumental methods
of measuring scorch and cure characteristics, Polymer Testing 14 (1)
(1995) 45.
[14] L.S. Lasdon, A.D. Waren, A. Jain, M. Ratner, Design and testing of
a generalized reduced gradient code for nonlinear programming,
ACM Transactions on Mathematical Software 4 (1) (1978) 34.
[15] S. Vyazovkin, N. Sbrirrazzuoli, Mechanism and kinetics of epoxy-
amine cure studied by differential scanning calorimetry, Macro-
molecules 29 (1996) 1867.
[16] M. Rafei, M.H.R. Ghoreishi, G. Naderi, Development of an advanced
computer simulation technique for the modeling of rubber curing
process, Computational Materials Science 47 (2) (2009) 539.