Bioresource Technology 102 (2011) 15561561

Contents lists available at ScienceDirect

Bioresource Technology
journal homepage: www.elsevier.com/locate/biortech

Logistic distributed activation energy model Part 1: Derivation and numerical

parametric study
Junmeng Cai a,, Chuan Jin b, Songyuan Yang b, Yong Chen b
a
School of Agriculture & Biology, Shanghai Jiao Tong University, 800 Dongchuan Road, Shanghai 200240, PR China
b
School of Mechanical Engineering, Shanghai Jiao Tong University, 800 Dongchuan Road, Shanghai 200240, PR China

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

Article history: A new distributed activation energy model is presented using the logistic distribution to mathematically
Received 10 June 2010 represent the pyrolysis kinetics of complex solid fuels. A numerical parametric study of the logistic dis-
Received in revised form 23 August 2010 tributed activation energy model is conducted to evaluate the inuences of the model parameters on the
Accepted 23 August 2010
numerical results of the model. The parameters studied include the heating rate, reaction order, fre-
Available online 26 August 2010
quency factor, mean of the logistic activation energy distribution, standard deviation of the logistic acti-
vation energy distribution. The parametric study addresses the dependence on the forms of the calculated
Keywords:
a  T and da/dT  T curves (a: reaction conversion, T: temperature). The study results would be very
Logistic distribution
Nonisothermal kinetics
helpful to the application of the logistic distributed activation energy model, which is the main subject
Numerical solution of the next part of this series.
Pyrolysis Crown Copyright 2010 Published by Elsevier Ltd. All rights reserved.
Solid fuels

1. Introduction tribution. However, the application of the Gaussian distributed

The pyrolysis kinetics of a solid fuel, such as biomass, coal, oil
shale etc., can be used for design and control of the pyrolysis pro-
cess (Mani et al., 2009). Many kinetic models have been proposed
(Scott et al., 2006). Application of various model-tting methods to
the non-isothermal pyrolysis data results in highly uncertain ki-
netic parameters (Zhou and Grant, 2004). The alternative to mod-
el-tting methods is the consideration of variation of the activation
energy (Leroy et al., 2010; Budrugeac, 2009; Liu et al., 2009). There-
fore, the distributed activation energy model has been widely used
for the pyrolysis kinetic analysis of complex solid fuels (Sonobe
and Worasuwannarak, 2008; Li et al., 2009; Navarro et al., 2009;
Liu et al., 2001).
The distributed activation energy model assumes that the dis-
tribution of reactivity caused by the reaction complexity can be
represented by a set of independent, parallel reactions, each with
their own frequency factor and activation energy (Vrhegyi et al.,
2010; Wu et al., 2009; Yan et al., 2009; Vrhegyi et al., 2009).
Usually, it is further assumed that all reactions share the same
frequency factor, so the reactivity distribution is represented
by a continuous distribution of activation energies. The activa-
tion energy distribution is usually assumed by the Gaussian dis-
Corresponding author. Tel.: +86 21 34206624; fax: +86 21 34205744.
E-mail address: jmcai@sjtu.edu.cn (J. Cai).
activation energy model is not always valid for the kinetic
description of the pyrolysis processes of complex solid fuels,
especial at the initial and nal stages of the pyrolysis processes
(Burnham and Braun, 1999). These problems have prompted the
authors to develop a new distributed activation energy model,
where the logistic distribution is used for the description of
the distribution of reactivity. The derivation and numerical para-
metric study of the new model is the main subject of this part of
the series, and the application of the new model to the kinetic
description of cellulose pyrolysis and the multiplicity in the
kinetic parameters will be represented in the next part of the
series. It is anticipated that the logistic distributed activation en-
ergy model will be useful for a number of process chemical engi-
neering applications. Before proceeding to the development of
the model, it is necessary to consider some characteristics of
the logistic distribution.
2. Characteristics of logistic distribution
The logistic distribution has been applied in studies of popula-
tion growth (Hatis and Kiffe, 2004), of physicochemical phenom-
ena (Menon and Bhandarkar, 2004), bioassay (Li and Majumdar,
2008), of life test data (Srivastava and Mittal, 2010), and of bio-
chemical data (Peleg et al., 2007). Owing to the plentiful applica-
tions of the logistic distribution, the importance of the logistic
distribution is evident (Yeh, 2010).

0960-8524/$ - see front matter Crown Copyright 2010 Published by Elsevier Ltd. All rights reserved.
doi:10.1016/j.biortech.2010.08.079
 J. Cai et al. / Bioresource Technology 102 (2011) 15561561 1557

2.1. Logistic distributed activation energy model

The distributed activation energy model may be applied to

either the total amount of volatiles released, or to the amount of
an individual volatile constituent (Cai and Liu, 2008). The descrip-
tion here follows the development for the total amount of volatiles.
Let the pyrolysis of a complex solid fuel be assumed to consist
of a large number of independent, parallel reactions of rst or
nth (n 1) order. The kinetic equation of the ith particular reaction
is given by
8
> V i V i
dV i =V i < ki V i first order
  n 3
dt >
: k V i V i
nth ordern 1
i Vi

where Vi is the mass of released volatiles at time t, Vi* is the total

mass of volatiles originally available for the ith reaction, and ki is
Fig. 1. The probability density function curves of the logistic distribution for the reaction rate constant. The effect of temperature on the rate
various values of l and r. constant, ki, is assumed to follow the Arrhenius law:
 
Ei
The probability density function f(E) (as applied to describe the ki k0i exp  4
RT
distribution of activation energies, E) for the logistic distribution is
given by where k0i is the frequency factor, R is the universal gas constant, Ei
the activation energy, T is the absolute temperature.
b ebEl=r
f E 1 The solution of Eq. (3) is given by
r 1 ebEl=r 2
8  R 
p >
<
t
1  VVi exp  0 k0i eEi =RT dt first order
where b p= 3, l and r are the mean and standard deviation of i

the logistic distribution, respectively. h R i1=1n

>
: 1  Vi 1  1  n t k0i eEi =RT dt nth ordern 1
The cumulative distribution function, F(E), is then V i
0

5
1
FE 2
1 ebEl=r In this paper, it is assumed that the kis for all reactions differ
The cumulative distribution function is similar in form to the only in activation energy, that is, that k0i = k0 for all i, and that
solution to the continuous logistic equation giving the distribution the number of reactions is large enough to permit Ei to be repre-
its name (Balakrishnan, 2010). sented by a continuous distribution f(E). The fraction of potential
Fig. 1 depicts several probability density function curves for total volatiles released that has an activation energy between E
various values of l and r. From this gure, it can be seen that: and E + dE is represented by f(E)dE. Then Vi* and Vi in Eq. (5) are re-
(1) As l decreases, the probability density function is shifted to placed by the following formulae:
the left; (2) As l increases, the probability density function is
shifted to the right; (3) As r decreases, the probability density dV  V  f EdE 6
function gets pushed toward the mean, or it becomes narrower
and taller; (4) As r increases, the probability density function dV Vf EdE 7
spreads out away from the mean, or it becomes broader and shal-
lower (See the website: http://www.weibull.com/LifeDataWeb/ Then, Eq. (5) becomes
the_logistic_distribution.htm).
The main difference between the logistic distribution and the 8 h  R i
dV t E=RT
Gaussian distribution lies in the tails. The logistic distribution has >
< V  1  exp  0 k0 e dt f EdE first order
 h i1=1n 
slightly longer tails compared to the Gaussian distribution. This > R t
: dV 1  1  1  n 0 k0 eE=RT dt f EdE nth ordern 1
fact can be obtained the kurtosis values of the two distributions. V

The kurtosis for the logistic distribution is 1.2, while the Gaussian 8
distribution has a kurtosis of zero (Kotz et al., 2000). Because of
this characteristic, the logistic distribution is more appropriate to
describe the activation energy distribution for the pyrolysis of Eq. (8) is integrated over E to get

8
V
R1 h  R
t
i
< a V  0 1  exp  0 k0 e
> E=RT
dt f EdE first order
 h i  9
> R R 1=1n
: a VV 01 1  1  1  n 0t k0 eE=RT dt f EdE nth ordern 1

complex solid fuels than the Gaussian distribution, which will be where a is the fraction conversion of pyrolysis.
presented in detail in the next part of the series. Substituting Eq. (1) into Eq. (9), yields the following results
1558 J. Cai et al. / Bioresource Technology 102 (2011) 15561561

8 R1 h  R
t
i
ebEl=r
> b
< a r 0 1  exp  0 k0 e
E=RT
> dt 2 dE first order
1ebEl=r 
 h i  10
>
> R1 Rt 1=1n
ebEl=r
: a rb 0 1  1  1  n 0 k0 eE=RT dt 2 dE nth ordern 1
1ebEl=r 

which is the general form of the logistic distributed activation en-
ergy model.
Most laboratory experiments involving the pyrolysis of solid
fuels are conducted at temperatures linearly varying with time
from a starting temperature.
T T 0 bt
where b is the heating rate, T0 is the starting temperature of
pyrolysis.
Making use of Eq. (11), Eq. (10) takes the following form
study of the model, which will illustrate how variations in the
model parameters affect the forms of calculated a  T and da/
dT  T curves. Investigated parameters are l, r, k0, b, and reaction
order.
Eq. (12) can be solved by using numerical techniques for certain
parameter values. It is noted that the inner dT integral of Eq. (11) is
11
the temperature integral, which doesnt have an analytical solution
(Cai et al., 2006a; Cai et al., 2006b; Deng et al., 2009). A large num-
ber of temperature integral approximations have been proposed
(Cai et al., 2008). The following approximation presented by

8 R1 h 
k0
R T E=RT i ebEl=r
> b
< a r 0 1  exp  b T 0 e
> dT
1ebEl=r 
2 dE first order
 h i1=1n  12
>
> R1 RT ebEl=r
: a rb 0 1  1  1  n kb0 T 0 eE=RT dT 2 dE nth ordern 1
1ebEl=r 

which is the linear nonisothermal form of the logistic distributed (rfo, 2006), which can approximate very accurately the temper-
activation energy model. In the following sections, this linear non- ature integral, is used in this paper.
isothermal form of the model is considered.
Z T
eE=RT dT gE; T  gE; T 0 13
2.2. Numerical parameter study of the new model T0

Here we will present numerical solutions of the logistic dis- where

tributed activation energy model. Also presented is a parametric

RT 2 u 0:9999936u4 7:573939u3 12:4648922u2 3:690732u

gE; T e 14
E u 9:5733223u3 25:6329561u2 21:0996531u 3:9584969
4

Fig. 2. Variation of l: the mean of the logistic activation energy distribution. Other parameters were kept identical. r = 10 kJ mol1, k0 = 1010 s1, T0 = 473 K, b = 10 K min1
and n = 2.0 (for nth-order).
 J. Cai et al. / Bioresource Technology 102 (2011) 15561561 1559

Fig. 3. Variation of r: the standard deviation of the logistic activation energy distribution. Other parameters were kept identical. l = 150 kJ mol1, k0 = 1010 s1, T0 = 473 K,
b = 10 K min1 and n = 2.0 (for nth-order).

Fig. 4. Variation of k0: the frequency factor. Other parameters were kept identical. l = 150 kJ mol1, r = 10 kJ mol1, T0 = 473 K, b = 10 K min1 and n = 2.0 (for nth-order).

E The effect of l on the numerical results of the logistic distrib-

u 15
RT
To deal with the outer dE integral of Eq. (11), some numerical cal-
culations have been performed with an absolute error tolerance 108
by using the quad function of the MATLAB software system (Karris,
2007; Butt, 2009). Based on the obtained a  T data, the correspond-
ing da/dT  T data can be determined by means of numerical differ-
entiation. In this paper, the implementation of numerical
differentiation has been performed with the help of the Origin soft-
ware system (detailed information about this software system can
be found in the website: http://www.originlab.com/).
The numerical parameter study is carried out by changing each
of the parameters, taking one parameter at a time and keeping the
remaining parameters unchanged. We considered the base values
of the parameters as follows: l = 150 kJ mol1, r = 10 kJ mol1,
k0 = 1010 s1, T0 = 473 K, b = 10 K min1 and n = 2.0 (for nth-order).
uted activation energy model is illustrated in Fig. 2, where it is
seen that conversion curves are shifted up the temperature scale
by an increase in l. At higher l values, the maximum reaction
rate, (da/dT)m, is reached at higher temperatures. And as the l
values increase, (da/dT)m slightly decreases.
The effect of r on the numerical results of the logistic distrib-
uted activation energy model is illustrated in Fig. 3. Increase in r
values causes slopes of conversion curves to get less. As the values
of r increase, (da/dT)m increases and Tm (temperature at the max-
imum reaction rate) remains unchanged.
Fig. 4 shows the effect of k0 on the numerical results of the
logistic distributed activation energy model. According to the
curves included in Fig. 4, the increase in k0 value causes
conversion curves to lead toward lower temperatures. And as
the values of k0 increase, (da/dT)m increases while Tm decreases.
1560 J. Cai et al. / Bioresource Technology 102 (2011) 15561561

Fig. 5. Variation of b: the heating rate. Other parameters were kept identical. l = 150 kJ mol1, r = 10 kJ mol1, k0 = 1010 s1, T0 = 473 K and n = 2.0 (for nth-order).

(2) The solutions of the model are numerically calculated by using

two software packages (Matlab, Origin), and the inuences of the
model parameters on the numerical solutions of the logistic dis-
tributed activation energy model are presented. The parametric
study results have shown that the changing of the heating rate,
reaction order, frequency factor, mean and standard deviation of
the logistic activation energy distribution affects the forms of cal-
culated a  T and da/dT  T curves.

Acknowledgements

The work described in this paper was nancially supported by

National Natural Science Foundation of China (Project No.
50806048). This support is gratefully acknowledged.

References

Fig. 6. Variation of n: the reaction order. Other parameters were kept identical.
l = 150 kJ mol1, r = 10 kJ mol1, k0 = 1010 s1, T0 = 473 K and b = 10 K min1.
The effect of b on the numerical results of the logistic distrib-
uted activation energy model is illustrated in Fig. 5. As the heating
rate is increased, Fig. 5 shows that there is a lateral shift to higher
temperatures for conversion curves. The lateral shift is also illus-
trated in Fig. 5 for the da/dT  T curves.
The effect of the reaction order on the numerical results of the
logistic distributed activation energy model is shown in Fig. 6. As
the reaction order increases, conversion curves are shifted up the
temperature scale, and the values of (da/dT)m and Tm decrease.
3. Conclusions
(1) The logistic distributed activation energy model has been
mathematically derived. The linear nonisothermal form of the
model is given below:
Balakrishnan, N., 2010. Handbook of Logistic Distribution, second ed. Taylor and
Francis.
Budrugeac, P., 2009. Application of model-free and multivariate non-linear
regression methods for evaluation of the thermo-oxidative endurance of a
recent manufactured parchment. J. Therm. Anal. Calorim. 97 (2), 443451.
Burnham, A.K., Braun, R.L., 1999. Global kinetic analysis of complex materials.
Energy Fuels 13 (1), 122.
Butt, R., 2009. Introduction to Numerical Analysis Using Matlab. Jones and Bartlett
Publishers, LLC.
Cai, J.M., Liu, R.H., 2008. Numerical analysis of new distributed activation energy
model for the representation of biomass pyrolysis kinetics. J. Energy Inst. 81,
149152.
Cai, J., He, F., Yi, W., Yao, F., 2006a. A new formula approximating the Arrhenius
integral to perform the nonisothermal kinetics. Chem. Eng. J. 124, 1518.
Cai, J., Liu, R., Huang, C., 2008. Kinetic analysis of nonisothermal solid-state
reactions: determination of the kinetic parameters by means of a nonlinear
regression method. J. Math. Chem. 44 (2), 551558.
Cai, J., Yao, F., Yi, W., He, F., 2006b. New temperature integral approximation for
nonisothermal kinetics. AIChE J. 52 (4), 15541557.
Deng, C., Cai, J., Liu, R., 2009. Kinetic analysis of solid-state reactions: evaluation of
approximations to temperature integral and their applications. Solid State Sci.
11, 13751379.
Hatis, J.H., Kiffe, T.R., 2004. On stochastic logistic population growth models with
immigration and multiple births. Theor. Popul. Biol. 65, 89104.
Karris, S.T., 2007. Numerical Analysis Using MATLAB and Excel, thrid ed. Orchard
Publications.

  i1=1n 
R1 h  RT i
ebEl=r
R1 h RT ebEl=r
a rb 0
1  exp  kb0 T 0 eE=RT dT 2 dE first orderalpha rb 0
1  1  1  n kb0 T 0 eE=RT dT 2 dEnth ordern 1
1ebEl=r  1ebEl=r 
 J. Cai et al. / Bioresource Technology 102 (2011) 15561561
Kotz, S., Johnson, N.L., Balakrishnan, N., 2000. Continuous multivariate distribution,
vol. 1, Models and applications. John Wiley & Sons, Inc., New York.
Leroy, V., Cancellieri, D., Leoni, E., Rossi, J., 2010. Kinetic study of forest fuels by TGA:
model-free kinetic approach for the prediction of phenomena. Thermochim.
Acta 497, 16.
Li, G., Majumdar, D., 2008. D-optimal designs for logistic models with three and four
parameters. J. Stat. Plan. Infer. 138, 19501959.
Li, Z., Liu, C., Chen, Z., Qian, J., Zhao, W., Zhu, Q., 2009. Analysis of coals and biomass
pyrolysis using the distributed activation energy model. Bioresour. Technol. 100
(2), 948952.
Liu, J., Jiang, X., Zhou, L., Han, X., Cui, Z., 2009. Pyrolysis treatment of oil sludge and
model-free kinetics analysis. J. Hazard. Mater. 161, 12081215.
Liu, X., Li, B., Miura, K., 2001. Analysis of pyrolysis and gasication reactions of
hydrothermally and super critically upgraded low-rank coal by using a new
distributed activation energy model. Fuel Process. Technol. 69 (1), 112.
Mani, T., Murugan, P., Mahinpey, N., 2009. Determination of distributed activation
energy model kinetic parameters using simulated annealing optimization
method for nonisothermal pyrolysis of lignin. Ind. Eng. Chem. Res. 48, 1464
1467.
Menon, A., Bhandarkar, S., 2004. Predicting polymorphic transformation curves
using a logistic equation. Int. J. Pharm. 286, 125129.
Navarro, M.V., Murillo, R., Mastral, A.M., Puy, N., Bartroli, J., 2009. Application of the
distributed activation energy model to biomass and biomass constituent
devolatilization. AIChE J. 55 (10), 27002715.
rfo, J.J.M., 2006. Review and evaluation of the approximations to the temperature
integral. AIChE J. 53 (11), 29052915.
1561
Peleg, M., Corradini, M.G., Normand, M.D., 2007. The logistic (Verhulst)
model for sigmoid microbial growth curves revisited. Food Res. Int. 40, 808
818.
Scott, S.A., Dennis, J.S., Davidson, J.F., Hayhurst, A.N., 2006. An algorithm for
determining the kinetics of devolatilisation of complex solid fuels from
thermogravimetric experiments. Chem. Eng. J. 61, 23392348.
Sonobe, T., Worasuwannarak, N., 2008. Kinetic analyses of biomass pyrolysis using
the distributed activation energy model. Fuel 87 (3), 414421.
Srivastava, P.W., Mittal, N., 2010. Optimum step-stress partially accelerated life
tests for the truncated logistic distribution with censoring. Appl. Math. Model.
34, 31663178.
Vrhegyi, G., Chen, H., Godoy, S., 2009. Thermal decomposition of wheat, oat,
barley, and brassica carinata straw. A kinetic study. Energy Fuels 23, 646
652.
Vrhegyi, G., Czgny, Z., Liu, C., Mcadam, K., 2010. Thermogravimetric analysis of
tobacco combustion assuming DAEM devolatilisation and empirical char-
burnoff kinetics. Ind. Eng. Chem. Res. 49, 15911599.
Wu, M., Vrhegyi, G., Zha, Q., 2009. Kinetics of cellulose pyrolysis after a pressurized
heat treatment. Thermochim. Acta 496, 5965.
Yan, J.H., Zhu, H.M., Jiang, X.G., Chi, Y., Cen, K.F., 2009. Analysis of volatile species
kinetics during typical medical waste materials pyrolysis using a distributed
activation energy model. J. Hazard. Mater. 162, 646651.
Yeh, H.C., 2010. Multivariate semi-logistic distributions. J. Multivar. Anal. 101, 893
908.
Zhou, D., Grant, D.J.W., 2004. Model dependence of the activation energy derived
from nonisothermal kinetic data. J. Phys. Chem. A 108, 42394246.