MATHEMATICAL MODELING OF VULCANIZATION CHARACTERISTICS

O. H. YEOH*
FREUDENBERG-NOK SEALING TECHNOLOGIES, PLYMOUTH, MI 48170

RUBBER CHEMISTRY AND TECHNOLOGY, Vol. 85, No. 3, pp. 482–492 (2012)

ABSTRACT
It is often useful to describe vulcanization reaction kinetics in terms of a mathematical model so that curing behavior
may be predicted for different heat histories. Commonly used models for describing vulcanization kinetics are reviewed and
discussed with emphasis on the mathematical relationships between them. The Deng-Isayev, Sestak-Berggren, and Kamal
models are regarded as members of a hierarchy of increasing complexity. Simpler members are preferred for robustness.
Tests of their validity against experimental data are recommended to guide the choice of models. [doi:10.5254/rct.12.87982]

CONTENTS
I. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 482
II. Basic Reaction Kinetics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 483
III. Deng-Isayev Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 484
IV. Sestak-Berggren Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 488
V. Kamal Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 490
VI. Other Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 491
VII. Discussion and Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 491
VIII. References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 492

I. INTRODUCTION
On the factory floor, rubber products are commonly molded with complex heat histories.1 For
example, in injection molding, a rubber compound may be kept in the barrel at a moderate
temperature for a significant period of time before it is injected quickly into the hot mold. During the
injection process itself, the rubber heats up by heat transfer from the mold as well as by work done in
forcing the rubber through long runners and small gates. This means that the temperature is transient
and the heat history experienced at different locations within the part will be different.
Vulcanization reactions often initiate while the rubber is still in a nonisothermal state. If
vulcanization proceeds too quickly, the resulting increase in rubber viscosity inhibits flow, causing
molding defects (scorch). If vulcanization reactions are too slow, there will be loss of productivity.
Balancing the heating rate and vulcanization kinetics is important to successful molding. Therefore,
this is a priority during material and process development.
Computer simulation programs are now commonly used to calculate the heat history
experienced during molding.2 Provided the vulcanization characteristics of the rubber compound
have been measured and expressed in the form of a suitable mathematical model, the computer
program can estimate the state of cure at different times and locations in the part. This information is
useful during material, process, and product development.
Clearly, accurate modeling of the vulcanization process is important to the success of
computer simulation. Experimental techniques used for determining vulcanization characteris-
tics will not be discussed here. Instead, it is assumed that satisfactory measurements have already
been made. The focus here is on the mathematical model of vulcanization characteristics itself.
Various mathematical models have been used. The purpose of this article is to review some of
these mathematical models and provide some guidance on their empirical use in computer
simulations.

*Corresponding author. Ph: 734-354-5410; email: OonHock.Yeoh@fnst.com

482
 MATHEMATICAL MODELING OF VULCANIZATION CHARACTERISTICS 483

II. BASIC REACTION KINETICS

In general, chemical reactions may be modeled by a number of suitable rate equations of the
form
da
¼ kf ðaÞ ð1Þ
dt
where a is the dimensionless fraction of reaction (conversion), such that 0 < a < 1; t is the reaction
time; and k is the rate constant at a particular temperature.
The rate constant, k, is usually assumed to have an Arrhenius-type temperature dependence of
the form
 
E
k ¼ k0 exp  ð2Þ
RT
where k0 is the pre-exponential factor, E is the activation energy (J mol1), R is the universal gas
constant (8.31447 J mol1 K1), and T is the absolute temperature (K).
One commonly used model suitable for many chemical reactions is the nth-order kinetic
model,
da
¼ k1 ð1  aÞn ð3Þ
dt
where n is the order of reaction. It describes a reaction that slows down rapidly as the conversion
proceeds toward completion (a  1) because of depletion of the reactants.
However, rubber cure chemistry is often more complex than can be modeled using nth-order
kinetics because of parallel and sequential reaction paths. Autocatalytic models are often preferred.
One specific empirical model, sometimes called the Sestak-Berggren model,3 has been used for
certain thermoset (e.g., epoxy resin) cure reactions. It is expressed in the form
da
¼ k2 am ð1  aÞn ð4Þ
dt
where m and n are reaction orders. The form of Eq. 4 describes a reaction that is slow initially,
increases in rate as the reaction proceeds (autocatalytic), before slowing down again as the reactants
are depleted. It has four adjustable parameters, k02, E, m, and n.
Another popular autocatalytic form is the Kamal4,5 (sometimes Kamal-Sourour) model, given
by
da
¼ ðk1 þ k2 am Þð1  aÞn ð5Þ
dt
where k1 and k2 are rate constants assumed to have independent Arrhenius-type temperature
dependences of the form given in Eq. 2. Expansion of Eq. 5 reveals that the Kamal model is actually
a combination of the nth order and Sestak-Berggren models and hence more general than either. The
Kamal model contains six coefficients, namely, k01, k02, E1, E2, m, and n.
Specifically for rubber vulcanization, Isayev and Deng6 suggested an empirical model
da p 1p 2
¼ t a ð6Þ
dt k3
which may be derived from the equation
k3 t p
a¼ ð7Þ
1 þ k3 t p
484 RUBBER CHEMISTRY AND TECHNOLOGY, Vol. 85, No. 3, pp. 482–492 (2012)

by taking the derivative of a with respect to time, t. This model has three coefficients, k03, E,
and p.
At first, it might appear from Eqs. 6 and 7 that the Deng-Isayev model bears no resemblance to
the other models. However, as has been pointed out,7 by rewriting Eq. 7 to express t as a function of
a and k3 in the form
 1p
a
t¼ ð8Þ
k3 ð1  aÞ
and plugging it into Eq. 6, we get another form of the Deng-Isayev model,
da 1 ðp1Þ ðpþ1Þ
¼ pk3p a p ð1  aÞ p ð9Þ
dt
Comparing Eqs. 4 and 9, we find that the Deng-Isayev model is actually a special case of the Sestak-
Berggren model. The Deng-Isayev model clearly stems from restricting the values of m and n in the
Sestak-Berggren model to a specific relation between them. The Sestak-Berggren model itself is a
special case of the Kamal model. Hence, we have a hierarchy of related models of increasing
complexity. The Deng-Isayev model has three adjustable parameters, the Sestak-Berggren model
has four, and the Kamal model has six.

III. DENG-ISAYEV MODEL

It is worthwhile exploring the form of the Deng-Isayev model further. At any given
temperature, k3 is a constant, so, from Eq. 8, it is seen that t is a function of a/(1 – a) and p. To
illustrate the form, it is convenient to choose k3 ¼ 1. Figure 1 shows how p affects the relationship
between t and a. Note that in Figure 1, the time axis is on a logarithmic scale. It is seen that p affects
the shape of the curve; as p increases, the curve becomes markedly steeper at a ¼ 0.5.
Also, the symmetry of the curves about a ¼ 0.5 is striking. At first, this feature might seem
unexpected. However, if we made the substitution a ¼ (1 þ x)/2, it can be readily shown that
 
1  a  1 1þx x3 x5 x 7
ln ¼ ln ¼ x þ þ þ þ ::::: ð10Þ
2 1a 2 1x 3 5 7
and consequently the Deng-Isayev model shows first and third quadrant symmetry about a ¼ 0.5.
For purposes of illustrating another important feature of the model, it is convenient to keep p
constant. For example, Figure 2 shows the effect of the rate constant, k3, on the relationship between
t and a for the case of p ¼ 4. It is seen that changing k3 merely shifts the curve along the time axis
without changing its shape. Because k3 is related to temperature by the Arrhenius relationship (Eq.
2), it is clear that the effect of temperature is also to shift the cure curve along the time axis without
changing its basic shape.8,9 Again, it is noteworthy that the time axis in Figure 2 is on a logarithmic
scale. If the same data are plotted on a linear time scale (Figure 3), the shape of the curves changes
with k3 and T. Because a linear time scale is almost always used in presenting cure results, constancy
of shape on a logarithmic time scale usually escapes our attention.
When a ¼ 0.5, Eq. 8 reduces simply to
 1p
1
t50 ¼ ð11Þ
k3
So t50 is an easily indentified marker of the time scale of the reaction.
 MATHEMATICAL MODELING OF VULCANIZATION CHARACTERISTICS 485

FIG. 1. — Effect of p on shape of Deng-Isayev model (k3 ¼ 1).

Eqs. 2 and 8 may be combined and rewritten as

h a i1p  1 1p  E 1p
t¼ exp ð12Þ
1  a k03 RT
Taking natural logarithms yields
   
1 h a i 1 1 1 E
lnt ¼ ln þ ln þ ð13Þ
p 1a p k03 p RT
This means that Arrhenius plots (ln t vs 1/T) for any given a are expected to be straight, parallel lines
with slope E/pR (Figure 4). The slope is an important characteristic of the cure system. It reflects
sensitivity of cure times to temperature. A common rule of thumb used in the rubber industry is to
double cure times for every 10 8C reduction in cure temperature. The slope, E/pR, which varies
depending on details of the cure system, provides a more precise (customized) estimate of time/
temperature adjustments than this general rule of thumb.
Further, plots of ln t versus ln[a /(1 – a)] for any given test temperature are expected to be
straight, parallel lines with slope of 1/p (Figure 5). This plot is a useful test of the validity of the
Deng-Isayev model. Deviations from straight lines indicate that the material does not conform to the
required symmetry about a ¼ 0.5.
The above teaches that the Deng-Isayev cure model is not an arbitrary function. The three
parameters k03, E, and p together define a specific temperature dependence (Arrhenius) and a
specific form of symmetry about a ¼ 0.5. These constraints mean that we may expect deviations
from the Deng-Isayev model for materials whose cure behavior is complex and does not conform to
the special mathematical features of the model. Complex behavior may be expected because
vulcanization reactions are believed to involve sequential reactions with various intermediates and
cross-link precursors as well as simultaneous network maturation reactions involving cross-link
486 RUBBER CHEMISTRY AND TECHNOLOGY, Vol. 85, No. 3, pp. 482–492 (2012)

FIG. 2. — Effect of k3 on Deng-Isayev model (p ¼ 4).

FIG. 3. — Effect of k3 on Deng-Isayev model (p ¼ 4).

MATHEMATICAL MODELING OF VULCANIZATION CHARACTERISTICS 487

FIG. 4. — Arrhenius plots for a hypothetical Deng-Isayev material.

FIG. 5. — Plots of ln(t) vs ln[a/(1 – a)] for a hypothetical Deng-Isayev material.

488 RUBBER CHEMISTRY AND TECHNOLOGY, Vol. 85, No. 3, pp. 482–492 (2012)

reversion and other rearrangements.10,11 Despite these reservations, the simple Deng-Isayev model
may be quite successful in describing cure behavior of certain rubber materials.

IV. SESTAK-BERGGREN MODEL

The requirement of the Deng-Isayev model for a particular form of symmetry about the
midpoint (a ¼ 0.5) of the cure curve is not met with some rubber compounds resulting in poor fits.
We have seen that the shape of the cure curve in the Deng-Isayev model is controlled by the reaction
order, p. So, symmetry about a ¼ 0.5 in the Deng-Isayev model clearly stems from restricting the
values of m and n in the Sestak-Berggren model to a specific relation between them. Conversely,
using the Sestak-Berggren model itself, with independent m and n values, relaxes the Deng-Isayev
requirement for symmetry about a ¼ 0.5. This additional degree of freedom potentially allows for
better fit.
At the beginning of cure, a is small, so (1 – a) » 1, and Eq. 4 may be written as
da
»k2 am ð14Þ
dt
The coefficient m is seen to be the exponent governing the cure rate curve during the initial part of
the cure curve.
Toward the end of cure, a  1, and am  1. So, Eq. 4 may be written as
da
»k2 ð1  aÞn ð15Þ
dt
The coefficient n is seen to be the exponent governing the cure rate curve during the final part of the
cure curve.
In the middle, a ¼ 0.5, and Eq. 4 reduces to
da
¼ k2 0:5mn ð16Þ
dt
Not surprisingly, we find that m and n together are important in determining the cure rate curve in the
middle portion and hence its overall shape.
We explore the Sestak-Berggren model further by separating the variables in Eq. 4 and
integrating to determine cure times
Z
1 a 1
t¼ da ð17Þ
k2 0 a ð1  aÞn
m

According to Mathematica,12 the solution to Eq. 17 is

að1mÞ
t¼ Fð1  m; n; 2  m; aÞ ð18Þ
k2 ð1  mÞ
where F denotes the hypergeometric function, which can be defined in terms of the series
ab aða þ 1Þbðb þ 1Þ 2 aða þ 1Þða þ 2Þbðb þ 1Þðb þ 2Þ 3
Fða; b; c; zÞ ¼ 1 þ zþ z þ z þ :::
1!c 2!cðc þ 1Þ 3!cðc þ 1Þðc þ 2Þ
ð19Þ
Alternatively, Eq. 17 may be expanded using the Taylor series before integrating to yield the same
result.
 MATHEMATICAL MODELING OF VULCANIZATION CHARACTERISTICS 489

FIG. 6. — Arrhenius plots for a hypothetical Sestak-Breggren material.

The hypergeometric function looks formidable, but it may be evaluated numerically.

Examination of Eq. 19 shows that the hypergeometric series converges rapidly when a is small. If
we take a ¼ 0.05, we can estimate t5 (time to 5% cure) using the first-order term in Eq. 19 without
much error, that is,
 
að1mÞ ð1  mÞn
t5 ¼ 1 þ 0:05 ð20Þ
ð1  mÞk02 eð RT Þ
E
ð2  mÞ

t5 may be taken as an estimate of scorch time, and Eq. 20 allows its calculation for any temperature
once the coefficients of the Sestak-Berggren model have been established.
The hypergeometric series converges much more slowly when a is large, but it can still be
evaluated, and times to higher states of cure, such as t90 (time to 90% cure), may be estimated.
Recalling Eq. 2, it can be shown that the Eq. 17 may be written in the form
 
1 E
lnðtÞ ¼ ln þ þ ln½gðaÞ ð21Þ
k02 RT
where g(a) represents the integral. Eq. 21 indicates that plotting ln(t) against 1/T (called the
Arrhenius plot here) for any particular value of a should yield straight, parallel lines of slope E/R
(Figure 6). This slope reflects the sensitivity of cure times to temperature. This is analogous to the
slope E/pR in the Deng-Isayev model (Figure 4).
Comparing Figures 4 and 6, it is seen that in the Deng-Isayev model (Figure 4), the t10 and t90
lines are equally spaced about the t50 line. This is a consequence of the symmetrical form of the
Deng-Isayev model. However, in the Sestak-Berggren model, the t10 and t90 lines are not
equidistant from the t50 line. For the purposes of illustration only, the values of m and n were chosen
490 RUBBER CHEMISTRY AND TECHNOLOGY, Vol. 85, No. 3, pp. 482–492 (2012)

FIG. 7. — Plots of ln(t) vs ln[a/(1 – a)] for a hypothetical Sestak-Breggren material.

to be 0.85 and 2.0, respectively, in Figure 6. With these specific choices for m and n, the t10 line is
closer to the t50 line than the t90. Thus, the model in Figure 6 represents a material that cures
relatively quickly at the beginning (scorchy) but relatively slowly toward the end (marching cure).
Essentially, the separation between the parallel lines reflects the form of g(a), which in turn depends
on the values of m and n.
Figure 7 shows plots of ln(t) versus ln[a /(1 – a)] for different temperatures. Unlike the
corresponding plots for Deng-Isayev models, which are straight lines (Figure 5), such plots are
curves in the case of Sestak-Berggren models. The shape of the curves depends on the specific
values of m and n.
We point out here that the slope of the Arrhenius plot is an important characteristic of the cure
system. It is noteworthy that both the Sestak-Berggren and Deng-Isayev models have a single
temperature dependence, and this is reflected by the constant slope (E/pR and E/R for the Deng-
Isayev and Sestak-Berggren models, respectively) in the Arrhenius plots. Thus, the Arrhenius plot
is a useful check of the validity of the underlying assumption in the Sestak-Berggren and Deng-
Isayev models regarding temperature sensitivity. If the slopes of the Arrhenius plots are not constant
but depend on a, this implies deviation from the model assumption.

V. KAMAL MODEL
The Kamal model (Eq. 5) has two rate constants, each with its own Arrhenius-type temperature
dependence. Arrhenius plots of Kamal models will not be parallel lines because the temperature
dependence is more complex. This additional degree of freedom in temperature dependence may be
useful for modeling materials that show more complex temperature dependences than can be
accommodated by the Sestak-Berggren model.
 MATHEMATICAL MODELING OF VULCANIZATION CHARACTERISTICS 491

We note that at the beginning of cure, a is small so (1 – a) » 1, and Eq. 5 may be written as
da
»k1 þ k2 am ð22Þ
dt
whereas toward the end of cure, a  1, am  1, and Eq. 5 may be written as
da
»ðk1 þ k2 Þð1  aÞn ð23Þ
dt
Thus, m and n are important in determining the shape of the cure rate curve at the beginning and end,
respectively, as in the Sestak-Berggren model. However, the relative values of k1 and k2 are clearly
important too. Note that k1 and k2 are assumed to have different temperature dependences
complicating the interpretation further.
Unfortunately, integrating the Kamal model to determine times to different states of cure does
not appear to be possible, leaving numerical integration the only practical approach. Despite the
relative complexity of the Kamal model, apparently its versatility is appreciated by practitioners of
computer simulation. It is supported directly in popular mold-filling simulation programs such as
Modelx3D, Moldflow, and 3D-Sigma.

VI. OTHER MODELS

The Kamal model is not the only way to accommodate more complex forms of temperature
sensitivity. Quite early in the literature of vulcanization modeling, it has been suggested that rubber
vulcanization proceeds in two stages. During the first stage, called ‘‘induction,’’ no measurable
reaction takes place. At the end of the induction period, the reaction starts and proceeds at a
rate that varies with time and temperature. It is common to model the second stage using one
of the models discussed above and to model the first induction stage with an equation of the
form6
 
T0
ti ¼ t0 exp ð24Þ
T
where ti is the induction time and t0 and T0 are material constants. Because the two stages are
sequential, elapsed time is the sum of the induction time, ti, and the reaction time, t.
Because t0 and T0 are adjustable parameters, adding the concept of induction time to the Deng-
Isayev model increases the number of parameters from three to five. Similarly, the Sestak-Berggren
model with induction time would have six parameters. Further, because induction time is
temperature dependent according to Eq. 24, such models will show two temperature dependences,
which, in a sense, is similar to the Kamal model.
The concept of induction time may also be invoked with the Kamal model. This would increase
the number of adjustable parameters from six to eight with three temperature dependences.
In our discussions, we have treated the adjustable parameters as being independent of
temperature. But this is not necessarily so. Lopez et al13 have proposed that some of the parameters
may be expressed as power series of temperature. As a result, their Kamal model has 14 adjustable
parameters (when the power series is truncated after T 2) compared with the usual six.

VII. DISCUSSION AND CONCLUSION

The Deng-Isayev, Sestak-Berggren, and Kamal models are members of a series of related cure
models of increasing complexity. They are versatile and appear capable of fitting the salient features
of the vulcanization characteristics of rubber compounds. Further enhancement is possible by
492 RUBBER CHEMISTRY AND TECHNOLOGY, Vol. 85, No. 3, pp. 482–492 (2012)

introducing additional adjustable parameters such as suggested by incorporating the concept of

induction time or by making some of the parameters dependent on temperature. However, it is not
always clear if introduction of more adjustable parameters can be easily justified by the
experimental data. For example, Eq. 22 indicates that the initial rate of reaction and hence the scorch
time at any given temperature is influenced by three adjustable parameters, k1, k2, and m, in the
Kamal model. Inclusion of induction time ti adds a fourth parameter. Considering experimental
difficulties in making measurements on reaction kinetics, it is not obvious that the quality of the data
will be sufficient to discriminate between contributions from the respective parameters.
There is a tendency among practitioners of computer simulation to assume that the most
complex and general models are best. The underlying rationale appears to be that computer
regression algorithms used to reduce experimental data into a model would take care of spurious
parameters by returning zero values because of a lack of correlation. Unfortunately, experimental
data inevitably contain errors, and regression algorithms will use all available degrees of freedom to
fit the data including the errors. Thus, the resulting model may contain parameters with little
physical meaning, and such models can lead to misleading conclusions when extrapolated.
This review has identified the special features of some common models for rubber
vulcanization. In general, the use of simpler models is preferred for robustness. Tests of their
validity against experimental data are suggested here for use to guide the choice of models. This
approach appears to be more logical than adopting more complex models by default.

VIII. REFERENCES
1
J. G. Sommer, Elastomer Molding Technology, Elastech, Hudson, OH, 2003.
2
T. A. Osswald and J. P. Hernandez-Ortiz, Polymer Processing: Modeling and Simulation, Hanser, München, Germany,
2006.
3
J. Sestak and G. Berggren, Thermochimica Acta 3, 1 (1971).
4
M. R. Kamal and S. Sourour, Polym. Eng. Sci. 13, 59 (1973).
5
M. R. Kamal, Polym. Eng. Sci. 14, 231 (1974).
6
A. I. Isayev and J. S. Deng, RUBBER CHEM. TECHNOL. 61, 340 (1988).
7
J. S. Deng and A. I. Isayev, RUBBER CHEM. TECHNOL. 64, 296 (1991).
8
T. W. Chan, G. D. Shyu, and A. I. Isayev, RUBBER CHEM. TECHNOL. 66, 849 (1993).
9
G. D. Shyu, T. W. Chan, and A. I. Isayev, RUBBER CHEM. TECHNOL. 67, 314 (1994).
10
C. P. Rader, ‘‘Vulcanization of Rubber – A. Sulfur and Non-peroxides,’’ in Basic Elastomer Technology, K. C. Baranwal
and H. L. Stephens, Eds., Rubber Division, ACS, Akron, 2001.
11
J. B. Class and P. R. Dluzneski, ‘‘Vulcanization of Rubber – B. Peroxides,’’ in Basic Elastomer Technology, K. C.
Baranwal and H. L. Stephens, Eds., Rubber Division, ACS, Akron, 2001.
12
‘‘Mathematica On-line Integrator,’’ http://integrals.wolfram.com/index.jsp.
13
L. M. Lopez, A. B. Cosgrove, J. P. Hernandez-Ortiz, and T. A. Osswald, Poly. Eng. Sci. 47, 675 (2007).

[Presented at the Spring Meeting of the Rubber Division, ACS (San Antonio, TX), 22–25 April 2012]

[Received March 2012, Revised June 2012]