Chemical Engineering Science 171 (2017) 459–470

Contents lists available at ScienceDirect

Chemical Engineering Science

journal homepage: www.elsevier.com/locate/ces

Assessment of thermogravimetric methods for calculating coke

combustion-regeneration kinetics of deactivated catalyst
Aitor Ochoa, Álvaro Ibarra, Javier Bilbao, José M. Arandes, Pedro Castaño ⇑
Department of Chemical Engineering, University of the Basque Country UPV/EHU, 644-48080 Bilbao, Spain

h i g h l i g h t s

Guidelines for calculating combustion kinetics of coke combustion.

Data obtained in a thermobalance using a spent FCC catalyst.

Kinetic model-based, isoconversional and modulated methods.

Best practices for each method and its pros-cons.

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

Article history: This work compares different methodologies for calculating the kinetic parameters of coke combustion,
Received 13 January 2017 employed for catalyst regeneration, using thermogravimetric methods. A reference fluid catalytic crack-
Received in revised form 18 May 2017 ing (FCC) spent catalyst was used as a representative example of the deactivated catalyst for the combus-
Accepted 23 May 2017
tion runs, pre-used in the cracking of a vacuum gas oil at 773 K and 3 s. Three different types of
Available online 31 May 2017
approaches have been performed in order to obtain kinetic combustion parameters: (i) kinetic model-
based, (ii) isoconversional and (iii) modulated methods. Additionally, a series of empirical modifications
Keywords:
have been proposed to predict the kinetic behavior at different heating rates for the model-based
Thermogravimetry
Coke combustion
approach. Using the best conditions and methods, the combustion activation energy of coke, deposited
Kinetic modeling after the reaction mentioned, is in the order of 114, 156, and 162 kJ mol1 for the kinetic model-
Isoconversional based, isoconversional and modulated methods, respectively. The recommendations for measuring
Modulation kinetic parameters are reported together with the benefits/disadvantages using the three mentioned
approaches.
Ó 2017 Elsevier Ltd. All rights reserved.

1. Introduction stripper and continuously delivering regenerated catalyst to the

Coke fouling is an inevitable and unwanted problem in many
heterogeneous catalytic processes involving hydrocarbons
(Bartholomew and Farrauto, 2006). Indeed, coke fouling causes
millions of dollars of worldwide investment for a variety of cat-
alytic processes such us hydrogenation, hydrotreating, reforming
and cracking, among many others (Argyle and Bartholomew,
2015; Guisnet and Ribeiro, 2011). In fluid catalytic cracking
(FCC), coke fouling has a critical role in the overall process design
and throughput, as a typical unit ca. 35,000 bpd (219 t h1) fed
with gas oil yields 13 t h1 of coke (Gary et al., 2007). Thus, coke
is eliminated in this process by combustion in the regeneration
stage, continuously withdrawing coked catalyst from the FCC
⇑ Corresponding author.
E-mail address: pedro.castano@ehu.eus (P. Castaño).
cracking inlet (Borio et al., 1992; Čejka et al., 2007). The regenera-
tor consists in a fluidized bed at ca. 973 K with continuous feeding
of air or oxygen-enriched air (Upson and Lomas, 2000).
The coked FCC catalysts, as well as many other catalysts deacti-
vated by coke, have been routinely analyzed by temperature pro-
grammed oxidation (TPO) techniques with three different aims
related with coke properties (Aguayo et al., 1999; Guisnet and
Ribeiro, 2011; Opfermann et al., 2002): (1) quantitative yet fast
analysis of the amount (and yield) of coke, (2) qualitative analysis
of the nature and location of coke deposits and (3) evaluation of the
kinetic parameters of coke combustion. In this sense, TPO allows to
discretize coke into fractions with differentiated composition or
location on the catalytic surface, and additional insights can be
performed analyzing the exhausts with mass spectrometry
(Aguayo et al., 2011) or the solid by FTIR spectroscopy (Ibáñez
et al., 2014). The general criterion is that one type of coke burns

http://dx.doi.org/10.1016/j.ces.2017.05.039
0009-2509/Ó 2017 Elsevier Ltd. All rights reserved.
460 A. Ochoa et al. / Chemical Engineering Science 171 (2017) 459–470

Nomenclature

Symbols Te temperature calculated for the I7 isoconversional meth-

A pre-exponential factor (atm1 s1) od, in Eq. (40) (–)
A 0
constant of the calorimeter in which DSC analysis is per- v1, v2 DTGu value at the same time as vmin and vmax, respec-
formed (m) tively (lg s1)
b total number of chosen temperatures in the I7 isocon- vmax, vmin maximum and minimum local value in DTGm, respec-
versional method, in Eq. (40) (–) tively (lg s1)
Ccoke coke content of the catalyst (lg g1 1
catalyst or mg gcatalyst) Vp total pore volume (cm3 g1)
Deff effective diffusivity (m2 s1) x apparent activation energy at T temperature (–)
DTG derivative weight loss (lg s1 or lg s1 g1catalyst) y total number of heating programs (–)
E activation energy (kJ mol1) z total number of reaction steps (–)
E1, E2 activation energies obtained in the modulated method
from the lower and higher value of the modulation, Greek letters
respectively (kJ mol1) a combustion conversion (–)
F function defined in Eqs. (SI-17) and (SI-18) (–) b heating rate (K min1)
f(a) kinetic model (–), mathematical expression describing kg thermal conductivity of the crucible (W m1 K1)
coke combustion as a function of conversion sHFM characteristic response time value of the heat flowmeter
g(a) integral form of the kinetic model in Eq. (SI-4) (–) (s)
G(a) kinetic parameter defined in Eq. (SI-8) (–) /N-RECc net recorded heat flow during a DSC analysis (W)
H modulation amplitude (K) c1, c1, c3 empirical parameters used in Eqs. (47)–(49), respec-
J function defined in Eq. (35) (I4 method) and Eq. (36) (I5 tively (min K1)
method) u sum of the logarithmic expressions in the right side of
k kinetic parameter (s1) Eq. (SI-13) (–)
m mass (g) j intrinsic kinetic parameter (atm1 s1)
Mc heat capacity (J K1) k1, k2 empirical parameters used in Eq. (45) (min K1)
n reaction order (–) h1, h2 constants defined in Eqs. (51) and (52), respectively (K)
NWP Weisz-Prater modulus (–) x modulation frequency (Hz)
p(x) temperature integral in Eq. (SI-5) (–)
PO oxygen partial pressure (atm) Subscripts and superscripts
R gas constant (8.314 J mol1 K1) 0 initial value of the combustion run (–)
ra combustion rate (s1) calc calculated data (–)
rp particle diameter (m) exp experimental data (–)
SSE sum of square errors (K2) f final value of the combustion run (–)
T temperature (K) h number of the chosen temperature in the I7 isoconver-
t time (s) sional method, in Eq. (40) (–)
Ty reference temperature (K) i number of a reaction step or the fraction of coke (–)
T1, T2 unmodulated temperature at which Tm-Tu is minimum j, k number of a heating program (–)
or maximum, respectively (K) max, min maximum and minimum value, respectively (–)
Tr real temperature on the combustion zone, estimated in m, u modulated and unmodulated signal, respectively (–)
Eqs. (47)–(50) (K)

at lower temperature when its composition is more hydrogenated
(higher proportion of aliphatics respect to aromatics) (Ibáñez et al.,
2012) and/or it is more accessible (Epelde et al., 2015), e.g. in the
case of FCC catalyst, when it is located in the catalyst matrix
instead of within the micropores.
The coke combustion mechanism involves many different reac-
tion steps in parallel and in series, and the overall mechanism is
not fully understood (Le Minh et al., 1997). Coke also suffers ‘‘ag-
ing” in the process of heating or stripping (made in specific sec-
tions of the FCC unit), leading to compositional and location
changes of coke (Aguayo et al., 2003; Magnoux et al., 2002;
Marcilla et al., 2008; Royo et al., 1996). Moreover, FCC catalyst
comprises zeolites having shape selective microporosity and acid-
ity, among other materials like metal oxides for promoting coke
combustion. Thus, the kinetics of coke combustion depends on
many factors including the complex composition of the catalyst
and a wide spectrum of operational conditions (Moljord et al.,
1995; Royo et al., 1996). Having a reliable coke combustion kinetic
model and its corresponding parameters may have two different
targets: (i) the design of the FCC regenerator or more in general,
choosing the right operational conditions for the regeneration
stage of coked catalysts, and (ii) in order to obtain representative
and intrinsic values of coked catalysts, useful for a systematic com-
parison of the nature or location of coke within the catalytic sur-
face. Besides, for the first target, the model should incorporate
heat and mass transfer rates as well as the proper fluid dynamic
equations. On the other hand, if a mechanistic understanding is
sought, then the chemical events associated with high-
temperature oxidation of hydrocarbons should be considered,
using a modeling software like ReaxFF (Chen et al., 2013).
The standard experimental procedure of the temperature pro-
grammed oxidation (TPO) of coke by thermogravimetric analysis
(TGA or TG) involves a linear heating program:
T ¼ T 0 þ bt ð1Þ
Then, assuming a combustion kinetic model, we can obtain its
parameters with a single experiment and performing a non-
linear fitting of the experimental results. Due to the complex nat-
ure of coke deposited on the spent catalyst, it is commonly
assumed that the combustion profile is the sum of independent
combustion steps of each fraction of coke, each having one kinetic
 A. Ochoa et al. / Chemical Engineering Science 171 (2017) 459–470 461

expression. Thus, the overall combustion kinetics, as well as for relevant conditions (773 K and 3 s). The aim was to present and
degradation, can be expressed in terms of the rate of combustion compare relatively simple models and methods for accounting a
conversion ra (da/dt) as it follows: very complex process (coke combustion), in order to establish
   the grounds for comparing quantitatively the nature or location
da X z X z
Ei 1 1
 y f i ðaÞ
Ty of coke within the catalytic surface. We have incorporated several
r a ¼ ¼ ðra Þi ¼ ki exp ð2Þ
dt i¼1 i¼1
R T T amendments for accounting the influence of the experimental
methodology and the calculation of the kinetic parameters.
where kiTy is the kinetic parameter:

ki ¼ jTy
Ty
i PO ð3Þ 2. Experimental
where a is the combustion conversion calculated dividing the frac-
2.1. Catalytic cracking runs
tion of coke combusted by the total amount of coke, i is the number
of coke fraction (1, 2 ,. . . , z), PO is the partial pressure of oxygen in
The vacuum gasoil (VGO) used as feed has been supplied by Pet-
the thermobalance which could be considered constant along the
ronor S.A. (Somorrostro, Spain). The VGO has been characterized by
experiment and fi(a) is the mathematical expression describing
simulated distillation (following the ASTM D2887 standard in an
coke combustion (kinetic model) as a function of the conversion.
Agilent 6890 Series GC System with a column Simdis D2887
The number of coke fractions must be identified from the TPO pro-
Fast/Ext.) and elemental analysis (LECO TruSpec CHN Macro). The
file assuming that one fraction should have a Gaussian shape in a
characterization results of the VGO have been previously reported
linear temperature programmed combustion. In standard TG exper-
in another work (Ibarra et al., 2016).
iments, the flow of air/oxygen is enough to allow considering a con-
An industrial catalyst based on HY zeolite has been employed,
stant pressure of oxygen along the experiment. The selection of fi(a)
supplied by Petronor S.A. The properties of the catalyst are dis-
to describe the process has been found to influence strongly in the
played in Table 1. The physical properties have been determined
obtained kinetic parameters (Vyazovkin and Wight, 2000). For this
by means of N2 adsorption isotherms in a Micromeritics ASAP
reason, the accuracy in the values of the kinetic parameters by
2010. Total acidity and acid strength distribution have been deter-
assumption of a reaction model has always raised uncertainties
mined by differential scanning calorimetry on a Setaram 111 dur-
(Mamleev and Bourbigot, 2005).
ing the differential adsorption of NH3 at 423 K. Subsequently,
Isoconversional methods exclude the need of assuming a kinetic
temperature programmed desorption (TPD) of absorbed NH3 has
model and are reliable for calculating activation energies in ther-
been performed following a ramp of 5 K min1 up to 823 K (TA
mally distorted processes (Mamleev and Bourbigot, 2002;
Instruments SDT 2960, coupled with a MS Thermostar Balzers
Sbirrazzuoli et al., 2009). There are two groups within the isocon-
Instruments mass spectrometer). The Brönsted/Lewis site ratio
versional methods: differential methods, which consider the
has been determined from the intensity ratio of the bands at
dependence of the activation energy (E) for each differential ele-
1545 and 1455 cm1 measured by pyridine adsorption and FTIR
ment of the combustion conversion (da); integral methods, which
spectroscopy (Nicolet 740 SX).
assume independent E respect to da. These methods assume that
The catalytic cracking runs were carried out in a riser simulator
any combustion conversion element (isoconversion) behaves
reactor consisting in a basket with the catalyst, an internal impeller
intrinsically and independently of the heating rate (Budrugeac,
which moves the gases (containing the reactants and products)
2002; Flynn and Wall, 1966; Friedman, 1964; Ozawa, 1965;
throughout the bed, fluidizing it (de Lasa, 1992). This reactor oper-
Sbirrazzuoli et al., 2009; Vyazovkin and Dollimore, 1996). Thus,
ates in conditions similar to these of industrial FCC risers (short
isoconversional methods fail to quantify the interaction among
contact times and high bed porosity) (Ibarra et al., 2016; Vicente
the different combustion stages (Opfermann et al., 2002), but sur-
et al., 2014). The runs have been carried out at 1 atm and 773 K;
pass other approaches by assuming a discrete distribution of vari-
contact time, 3 s; catalyst/feed ratio in dry base, 6 gcatalyst g1
feed. A
able activation energies along the combustion experiment.
detailed characterization of the coke deposited in the catalytic
Isoconversional methods use experiments performed under sev-
cracking can be found elsewhere (Ibarra et al., 2016).
eral linear heating programs (different b in Eq. (1)). Nevertheless,
isothermal experiments or non-linear non-isothermal experiments
have also proved to reach reliable results (Ortega, 2008). 2.2. Combustion runs
Modulated thermogravimetric (MTG) analysis gives an addi-
tional model-free calculation method for E. Its advantage com- Coke combustion runs were performed in a TA Instruments TGA
pared to the previous isoconversional methods is that it requires Q5000TA thermobalance. Samples followed a protocol involving
a single heating program perturbed by a sinusoidal modulation: the sweeping of impurities with a N2 stream (50 mL min1), at a
heating rate of 10 K min1, reaching 623 K and keeping this tem-
T m ¼ T 0 þ bt þ H sinð2pxtÞ ð4Þ
perature for 30 min. After cooling down to 523 K, a stream consist-
MTG analysis consists of comparing the perturbed and the ing in 50 mL min1 of air was allowed (and 5 mL min1 of N2 as
unperturbed derivative of weight (DTG) signal for calculating E balance gas), then temperature was raised following a protocol
without assuming any single model (Mamleev and Bourbigot,
2002, 2005; Mamleev et al., 2004, 2000). Based on this sinusoidal
Table 1
modulation method, Schawe (2014) proposed two non-periodic Properties of the catalyst.
modulation techniques, namely, the stochastic modulation and
the event controlled modulation, aiming a reduction of uncertain- BET surface area, m2 g1 122
Vp (cm3 g1) 0.15
ties in the standard sinusoidal modulation method. Part of the Micropore volume percentage, % 11.4
mathematical expressions of the modulated methods are Average pore diameter, Å 117.3
explained in the Supporting Information. Unit cell size, Å 24.30
In this work we have compared the three methods of data anal- Zeolite percentage, wt% 15
ysis (model-based, isoconversional and modulated), applying them Total acidity, lmolNH3 g1
cat 30
to the combustion of coke deposited on a FCC catalyst upon the Acid strength, kJ mol1
NH3 100
Brönsted/Lewis ratio, mol mol1 0.75
cracking of VGO in a riser simulator reactor under industrially
462 A. Ochoa et al. / Chemical Engineering Science 171 (2017) 459–470

Table 2
Kinetic models for describing coke combustion.

Name Model f (a) Eq.

M1 First order (n = 1) ð1  aÞ (6)
M2 Second order (n = 2) ð1  aÞ2 (7)
M3 Third-order (n = 3) ð1  aÞ3 (8)
M4 One third order (n = 1/3) ð1  aÞ1=3 (9)
M5 Three quarters order (n = 3/4) ð1  aÞ3=4 (10)
M6 One and a half order (n = 3/2) ð1  aÞ3=2 (11)
M7 Nucleation (power law, n = 1/4) 4a3=4 (12)
M8 Nucleation (power law, n = 1/3) 3a2=3 (13)
M9 Nucleation (power law, n = 1/2) 2a1=2 (14)
M10 Nucleation (power law, n = 3/2) ð2=3Þa1=2 (15)
M11 Phase boundary, contracting volume (sphere, n = 3) 3ð1  aÞ2=3 (16)
M12 3D diffusion (sphere, n = 3) 3ð1aÞ1=3 (17)
ð1ð1aÞ1=3 Þ
M13 Diffusion (Jander) 1:5ð1aÞ2=3
(18)
ð1ð1aÞ1=3 Þ
M14 Diffusion (Ginstling-Bronstein) 1:5 (19)
ðð1aÞ1=3 1Þ
M15 Diffusion (Zhuravlev-Lesokhin-Tempelman) 1:5ð1aÞ4=3 (20)
ðð1aÞ1=3 1Þ
M16 Diffusion (Anti-Jander) 1:5ð1þaÞ2=3 (21)
ðð1þaÞ1=3 1Þ
M17 Sigmoidal rate (Avrami-Erofeev, n = 1/4) 4ð1  aÞð lnð1  aÞÞ (22)
3=4

M18 Sigmoidal rate (Avrami-Erofeev, n = 1/3) 3ð1  aÞð lnð1  aÞÞ (23)
2=3

M19 Sigmoidal rate (Avrami-Erofeev, n = 1/2) 2ð1  aÞð lnð1  aÞÞ (24)
1=2

M20 Sigmoidal rate (Avrami-Erofeev, n = 2/3) 1:5ð1  aÞð lnð1  aÞÞ (25)
1=3

M21 Sigmoidal rate (Prout-Tomkins, n = 1) ð1  aÞa (26)

M22 Sigmoidal rate (Prout-Tomkins, n = 1/2) ð1  aÞa1=2 (27)

different for each method up to 973 K and finally the sample was
cooled down. The rest of the operational conditions depends on
the method used for calculating the kinetic parameters and will
be discussed later.
In order to calculate the combustion conversion (a), it has been
assumed that the amount of coke at the end of the combustion
experiment, at 973 K, is null (a = 1). The ra (da/dt) is calculated
by means of the following expression, taking the experimental
DTG profile:
the rate-determining mechanism is the branching nuclei (Han
et al., 2012; Koleva et al., 2008; Mamleev and Bourbigot, 2005).
The fitting of those models have been performed by minimizing
the sum of square errors (SSE) using the fminsearch function of
MatlabÒ.
Table 3
Differential and integral isoconversional methods.

da DTG Name Expression E=f Eq.

r a ¼ ¼ R tf ð5Þ (a)a
dt DTGdt  
0 D1 da Ea Yes (29)
ln ¼ ln½Aa f ðaÞP O  
dt a;j RT a;j
In a linear heating program as Eq. (1), the temperature is lin- da 
D2 X yX y exp  RTEaa;j Yes (30)
early dependent with time, so we can calculate ra integrating dt a;j
 ¼ yðy  1Þ
j k–j da
within the temperature range (dT) leading to different units of dt a;k
exp  RTEaa;k
!
ra: K1. I1 bj

Aa P O

Ea No (31)
ln 2
ffi ln  ln½gðaÞ 
T a;j REa RT a;j
 
I2 Aa E a P O Ea No (32)
3. Methods lnðbj Þ ffi ln  ln½gðaÞ  5:3305  1:052
R RT a;j
Z a   Z
I3 da E a da No (33)
ln da ¼  þ GðaÞ
3.1. Combustion model methods 0 dt R 0 T
I4 X y X y J½Ea ; T a;j  No (34)
k–j J½E ; T
¼ yðy  1Þ
The TPO profiles used for combustion models were obtained
j
a a;k 
Z ta  
Ea (35)
using the following heating rates: b = 2.5, 5, 10, 15 and 20 K min1, J½Ea ; T a;j  ¼ exp  dt
RT j
and assuming the reference condition Ty = 823 K. The kinetic mod-
0
I5 X y X y J½Ea ; T a;j  Yes (34)
els fi (a) used in this work are displayed in Table 2, each of which j k–j J½E ; T
a a;k 
¼ yðy  1Þ
Z ta  
represents the rate-determining step during the combustion reac- Ea (36)
J½Ea ; T a;j  ¼ exp  dt
tion. In this way, M1-M6 describe different kinetic models used to t RT j
  aDa
fit TG experiments, where the rate-determining mechanism is the I6 bj Ea Yes (37)
ln ¼u
DT a;j RT a;j
chemical reaction; M7-M10 are kinetic equations describing a
DT a ¼ T a  T aDa (38)
mechanism controlled by the nucleation as the rate-determining  
I7 bj Ea Yes (39)
step; M11 is a rate equation with a phase boundary reaction, ln ¼u
DT a;j RT e
where the rate-determining mechanism is the contracting sphere 1 Xb (40)
Te ¼ T
h¼1 e;h
volume (with spherical symmetry); M12-M16 are based on a b
DT a ¼ T aþDa=2  T aDa=2 (41)
three-dimensional diffusion mechanism; M17-M20 are sigmoidal
rate equations where random nucleation and its subsequent a
The method takes into account the dependence of activation energy on a when
growth is assumed; M21-M22 are sigmoidal rate equations where obtaining the equation.
 A. Ochoa et al. / Chemical Engineering Science 171 (2017) 459–470 463

0 12
X
y X
Tf
r a jj;T;exp r a jj;T;calc
SSE ¼ @R Tf  RT A ð28Þ
T 0 r a jj;exp dT T 0 r a jj;calc dT
f
j¼1 T¼T 0

where j is the number of the heating rate, which in this case corre-
sponds to bj = 2.5, 5, 10, 15 and 20 K min1. The denominator in Eq.
(28) indicates the normalization of the ra profiles at different heat-
ing rates.
The confidence intervals of each calculated parameter have
been calculated considering a 95% interval of confidence, using
the statistical parameters provided by re-calculating the parame-
ters using lsqcurvefit (which provides with the Jacobian, instead
of fminsearch) and nlparci of MatlabÒ.

3.2. Isoconversional methods

Table 3 summarizes the isoconversional methods used in this

work, divided into differential and integral methods. As for the dif-
ferential methods: (D1) Friedman (Friedman, 1964) and (D2) non-
linear differential (Budrugeac, 2002). As for the integral methods:
(I1) Kissinger-Akahira-Sunose (Akahira and Sunose, 1971;
Kissinger, 1956), (I2) Ozawa-Flynn-Wall (Flynn and Wall, 1966;
Ozawa, 1965), (I3) Li-Tang (Li and Tang, 1997, 1999), (I4) non-
linear integral (Vyazovkin, 1997; Vyazovkin and Dollimore,
1996), (I5) modified non-linear integral, (Vyazovkin, 2001), (I6)
average linear integral (Ortega, 2008) and (I7) modified average
linear integral (Han et al., 2013). The meaning of the parameters
in the equations in Table 3, as well as the procedure for obtaining
the equations, are explained in detail in the Supporting
Information.
The differential methods consider the dependence of E on a (Eq.
(29) for D1 and Eq. (30) for D2). Some of the integral methods
assume a constant value of E for obtaining the integrated form of
the rate equation, leading to evident inaccuracies in E calculations,
especially in processes where a great variation of E with a exists
(Sbirrazzuoli et al., 2009). Ortega (2008) evaluated the error for
assuming constant E in the rate equations in the Kissinger-
Akahira-Sunose (I1, Eq. (31)) and Ozawa-Flynn-Wall (I2, Eq. (32))
integral methods. This error was reported to be corrected by means
of other integral methods, such as the average linear (or Ortega)
method (I6) (Ortega, 2008) or the modified non-linear method
(I5) developed by Vyazovkin (2001), in which the change of E with
a is taken into account in the rate equation. The modified non-
linear method (I5) is known to be one of the most accurate integral
isoconversional methods with good tolerance of noise, despite its
increased complexity to perform in contrast with other methods
(Han et al., 2013). The modified average linear integral method
(I7) eliminates systematic errors and reduces the influence of
experimental noise faced in the original I6 average linear integral
method (Han et al., 2013).
The numerical differentiation required for calculating ra in dif-
ferential methods tends to create noise, as instantaneous rate val-
ues are calculated and these are sensitive to experimental noise
(Ortega, 2008). For this reason, the calculation of differential meth-
ods in this work is performed over little segments from five-point,
fourth-order finite difference approximations. On the other hand
and by using integral methods, this differentiation is avoided, lead-
ing to lower scattering of calculated E values (Vyazovkin and
Dollimore, 1996). In this way, integral methods turn out to be more
tolerant to experimental noise, thus being generally more devel-
oped and employed in kinetic analysis (Ortega, 2008).
3.3. Modulated method
The modulated method involves the combination of a set of
three experimental parameters (b, H and x) provided in Eq. (4),
Fig. 1. Calculation procedure of the activation energy (E): (a) Modulated temper-
ature (Tm) and unmodulated temperature (Tu) for a linear heating rate; (b)
difference between the modulated and unmodulated temperatures; (c) DTGm; (d)
maximum (vmax) and minimum (vmin) values in DTGm, and the estimated DTGu.
Operating conditions: b = 5 K min1; H = 5 K; x = 1  102 Hz.
which in this work have been used by combining the following val-
ues: 1/x = 50, 100 and 200 s; H = 2, 5 and 8 K; and b = 5 and
10 K min1. These values are the standard ones provided for the
thermobalances with the modulation feature (Mamleev and
Bourbigot, 2005).
Fig. 1 shows the procedure followed for the calculation of the
activation energy by means of this method during a modulated
thermogravimetric analysis and applied to the combustion of
coke. Fig. 1a shows the modulated temperature (Tm) curve when
sinusoidal modulation is applied to the linear heating rate, along
with the temperature corresponding to an unmodulated analysis
(Tu). The first step of the procedure consists of subtracting the
unmodulated temperature from the modulated one (Tm-Tu), as
shown in Fig. 1b. The next step consists of fitting the obtained
curve to a sinusoidal equation: Tm-Tu = H sin(2pxt). The deriva-
tive of this equation allows calculating the time values in which
the sinusoidal signal is minimum and maximum (black dots in
Fig. 1b). Then, the unmodulated temperature values in which
Tm-Tu (Fig. 1b) shows minima and maxima are calculated, called
T1 and T2 in this work, respectively. Subsequently, the modulated
weight loss derivative (DTGm) is plotted (Fig. 1c) and the time
values in which Tm-Tu (Fig. 1b) shows minima and maxima are
registered, as pointed with black dots in Fig. 1c. These pointed
values are the vmin and vmax displayed in Fig. 1d. Performing a
linear interpolation of adjacent vmin and vmax values, the DTGu
values are estimated (Fig. 4d). The time values at which vmin
and vmax data have been pointed in Fig. 1d allow the calculation
of the v1 and v2 values, which are the unmodulated signal
(DTGu) values at the same time values as vmin and vmax data,
respectively, as shown in Fig. 1d.
Once the parameters mentioned in the procedure have been
calculated, the method calculates E in the points in which the mod-
464 A. Ochoa et al. / Chemical Engineering Science 171 (2017) 459–470

ulated temperature Tm is minimum (E1) and maximum (E2) using 4. Results

the following expressions:
4.1. Combustion model methods
   
T1  H v1
E1 ¼ RT 1 ln ð42Þ
H v min The DTG profiles of coke combustion are shown in Fig. 2a and
the ra profiles used for the model-based methods are shown in
Fig. 2b. The calculated coke content in the catalyst (Ccoke):
   
T2 þ H v max 17.1 ± 0.4 mg g1 catalyst (1.71 ± 0.04 wt%) is hardly influenced by the
E2 ¼ RT 2 ln ð43Þ
H v2 heating rates used. Nevertheless, performing the combustion at
20 K min1 lead to a significantly lower coke content (12.8 mg gcat-
where the parameters within these expressions have been previ- 1
), in agreement with previous works (Altun et al., 2003, 2002;
alyst
ously explained and displayed in Fig. 1. The E in each period of Qing et al., 2013). Therefore, b  20 K min1 are not suitable for
the modulation is defined as the mean value between E1 and E2. combustion kinetic studies of coke. The value of the DTG at which
The whole procedure for obtaining Eqs. (42) and (43) has been the internal mass transfer is controlling the combustion (NWP > 0.3)
explained in detail in the Supporting Information, as well as else- can be calculated substituting in Eq. (44) the following values:
where (Mamleev and Bourbigot, 2002). Ccoke = 17,100 lg g1 6
m; and Deff = 5  108 m2 s1
catalyst; rp = 35  10

3.4. Mass transfer limitations (taking a worst case scenario (Chen et al., 2013)). Thus, for the
kinetic control regime, all experiments should be obtained with
Mass transfer limitation has been pointed to interfere in the cal- values of DTG < 4.19  104 lg s1 g1 catalyst, which is the case as

culation of coke combustion kinetics (Han et al., 2012; Vyazovkin,
2001) and it is one the main disadvantages of model-based
approaches. A typical approach to solve this limitation is crushing
the coked catalyst down to the lowest possible size and/or decreas-
ing the partial pressure of oxygen. However, from the FCC opera-
tional point of view, decreasing particle size is impractical
because the catalyst particles should retain its shape for keeping
the hydrodynamic behavior in the riser and regenerator. Besides,
decreasing the partial pressure of oxygen in the regenerator is
not recommended because it would imply increasing the yield of
CO in the exhausts. Thus, we have estimated the impact of mass
transfer by calculating the Weisz-Prater modulus:
DTGjmax r 2p
NWP ¼
C coke Deff
DTG|max 50 lg s1 g1 catalyst (using b = 20 K min
1
). In view of these
results we can safely discard the effect of internal mass transfer
limitations on the observed kinetic performance.
The temperature at which the maximum DTG or ra is obtained
shifts to lower value as the heating rate decreases. There are two
possible explanations for this behavior (Altun et al., 2003, 2002):
(i) at low heating rates the ignition of coke lead to higher temper-
atures within the solid than that of the gas, leading to apparent fas-
ter combustion whereas (ii) at high heating rates the temperature
of the solid is lower than that of the gas, leading to apparent slower
combustion. The results of Fig. 2 do not enable to specifically rec-
ognize what are those boundary values of heating rates at which
one mechanism is controlling over the kinetic regime of coke com-
bustion. The kinetic models used in this work and summarized in
ð44Þ
Table 2 (M1-M22) fit two parameters (k823 K and E) by means of
the least square approach previously explained in Eq. (28). These
kinetic parameters are related with the profiles of the ra vs. T
in the following manner: k823 K, position of the peak or the T of
maximum ra; and E, width of the peak. Thus, the kinetic models
used in this work (M1-M22, Table 2) are unable to predict this
change of the kinetic parameter values due to the heating rate.
Fig. 3 shows the results of E and the apparent kinetic parameter
(k823 K) for the M1 model from Table 2, as well as the correspond-
ing error bar for each calculated parameter value. The increasing
trend for both k823 K and E contradict the general statement that
kinetic behavior only depends on the reaction itself, and thus
exemplify the observation that not all the TG experiments repre-
sent the intrinsic kinetic regime. The reason behind the deviation
from intrinsic kinetic behavior lays in unrealistic temperature

Fig. 2. (a) DTG and (b) normalized ra profiles of the combustion of coke deposited Fig. 3. Values of activation energy (E) and apparent kinetic constant (k823 K) for the
on deactivated catalysts at different heating rates (b). first order kinetic model (M1) at different heating rates (b).
 A. Ochoa et al. / Chemical Engineering Science 171 (2017) 459–470 465

T r ¼ T 0 þ bð1  ec2 b Þt ð48Þ

T r ¼ T 0 þ bð1  c3 bÞt ð49Þ

where c1, c2 and c3 are empirical parameters, in min K1.

All these empirical approximations lead to lower SSE values for
each model of Table 2. However, all these empirical approaches
lead to kinetic parameters that are one order of magnitude higher
or lower than these represented in Fig. 3, so again they have been
discarded.
Salvador and Ferrasse (2006) proposed a method for calculating
the real coke combustion temperature considering an inefficient
thermal profile at high heating rates:

T r ¼ T 0  h1 þ bðt  h2 Þ ð50Þ

where h1 (in K) and h2 (in s) are assumed constant, expressed as

follows,

/NREC
h1 ¼ ð51Þ
Fig. 4. (a) Normalized sum of squared errors (SSE) of the combustion kinetic
A0 kg
models (Table 1); (b) assuming the approach with h1 and h2, Eq. (49).
Mc
h2 ¼ sHFM þ ð52Þ
A0 k g
profile within the coke sample, that is to say, a heat transfer limited
process (Salvador and Ferrasse, 2006). being uN-REC the net recorded heat flow during a DSC analysis (in
Despite the limitations explained before, a selection of the best W), A0 a constant of the calorimeter in which DSC analysis is per-
model is performed on the basis of the SSE, calculated using Eq. formed (in m), kg the thermal conductivity of the crucible (in
(28), for each heating rate and all the heating rates jointly. Both W m1 K1), sHFM the characteristic response time value of the heat
treatments lead to the same conclusions and Fig. 4a represents flowmeter (in s), and Mc the heat capacity (in J K1). The detailed
the SSE for the second one (all the heating rates jointly), as an procedure for obtaining Eqs. (50)–(52) is explained elsewhere
example. The SSE values of some of the nucleation (M10), some (Salvador and Ferrasse, 2006).
of the diffusion (M14-M16) and some of the sigmoidal rate Fig. 4b shows the SSE values of the fitted models considering
(M17-M20) models have been removed from the graph as they the approach in Eqs. (46) and (50)–(52), depicting a fitting
are incomparably high respect to the models represented already. improvement for most studied models with respect the SSE values
Besides and more importantly, the results presented in Fig. 4a in Fig. 4a. Once again, despite the good fitting in M21 and M22,
clearly point that the first order (M1), M21 and M22 are the best their corresponding kinetic parameters are one order of magnitude
fitting models. However, the last two models lead to kinetic higher or lower than those represented in Fig. 3, and the models
parameters that are one order of magnitude higher or lower than have been discarded. Thus, M1 in Fig. 4b shows the best fitting
those represented in Fig. 3, so these models have been discarded among all approaches tested in Fig. 4, obtaining the following
in spite of the good fitting. The calculated values of the kinetic kinetic parameters from the fitting: k823 K = 2.40 ± 0.4  103 s1,
parameters for M1 in Fig. 4a are k823 K = 1.8 ± 0.3  103 s1, E = 36.9 ± 0.9 kJ mol1, h1 = 303 ± 20 K; h2 = 4.0 ± 1 s. The correla-
E = 114 ± 2 kJ mol1, which are approximately similar to those pre- tion matrix of these four parameters, shown in the Supporting
viously reported for the model as a function of the heating rate Information (Eq. (SI-24)), indicates that the most related parame-
(Fig. 3). The correlation matrix of these two parameters, shown ters are k823 K and E, with values of 0.39, whereas h1 and h2 do
in the Supporting Information (Eq. (SI-23)), indicates that k823 K not correlate with the rest.
and E are relative correlated with values of 0.42.
In order to consider a unique value of k823 K and E regardless of
the heating rate, some empirical approaches have been tested in
this work, all based on modifications of Eq. (2), as for example
assuming kinetic parameters (k823 K or E) depending on the heating
rate as:
  
Ty Eð1  k2 bÞ 1 1
r a ¼ k ð1  k1 bÞ exp  y f ðaÞ ð45Þ
R T T
being k1 and k2 empirical parameters.
Besides, an inefficient thermal profile TPO experiment has been
considered, leading to:
  
Ty E 1 1
r a ¼ k exp  y f ðaÞ ð46Þ
R Tr T
where Tr is the real combustion temperature, computing its value
from:
 
b
Tr ¼ T0 þ t ð47Þ Fig. 5. Arrhenius plots at different values of degree of conversion (a), equivalent to
1 þ c1 b the Friedman method (D1) using different heating rates (b).
466 A. Ochoa et al. / Chemical Engineering Science 171 (2017) 459–470
Fig. 6. Activation energy (E) as a function of the degree of conversion (a), calculated
by means of several isoconversional methods: (a) Friedman (D1), non-linear
differential (D2), Kissinger-Akahira-Sunose (I1), Ozawa-Flynn-Wall (I2) and Li-Tang
(I3); (b) non-linear integral (I4), modified non-linear integral (I5), average linear
integral or Ortega (I6) and modified average linear integral (I7).
4.2. Isoconversional methods
As an example of the fitting of an isoconversional method, Fig. 5
shows the results obtained with the Friedman method (D1) where
the slope of each line corresponds to E at several values of a.
According to Eq. (29), E is calculated from the slope of Arrhenius
plots at different arbitrary a values. Fig. 5 shows that the points
obtained at b  20 K min1 do not lay on the lines so such high
heating rates should be avoided in any further treatment. The
200
180
E (kJ mol )
-1
160
140
120
D1 D2 I1
Fig. 7. Box plot of the distribution of activation energies (E), in the region a = 0.2–0.8, obtained by means of several isoconversional methods: Friedman (D1), non-linear
differential (D2), Kissinger-Akahira-Sunose (I1), Ozawa-Flynn-Wall (I2), Li-Tang (I3), non-linear integral (I4), modified non-linear integral (I5), average linear integral or
Ortega (I6), modified average linear integral (I7).
suitability of the optimal b range (2.5–15 K min1) has also been
evidenced by using the rest of the isoconversional methods.
The evolution of E along the combustion of coke (a) is shown
in Fig. 6, obtained using the model-free isoconversional methods
summarized in Table 3, and performing runs at the following
heating rates: b = 2.5, 5, 7.5, 10, 12 and 15 K min1. For the cal-
culations with the modified non-linear integral method (I5),
average linear integral method (I6) and modified average linear
integral method (I7), an interval of conversion of Da = 0.025
was selected as suitable (in Eq. (36), Eq. (38) and Eq. (41),
respectively), as well as a value of b = 5 in the modified average
linear integral method of Eq. (40). All the isoconversional meth-
ods in Fig. 6 show an increasing trend in E during the combus-
tion process. This may be due to the fact that the coke burning
at the lowest values of a corresponds to the lightest fraction and
has lower E for its combustion, and the heavier fractions evolve
into more condensed structures, whose combustion requires
higher values of E. This kind of behavior has been proved by
in situ FTIR spectroscopy (of the solid) during coke combustion
(Epelde et al., 2014; Ibáñez et al., 2014).
Fig. 7 depicts the distribution of E values in a box plot in the
region of a = 0.2–0.8. The reason for selecting this region is that
the modulated method studied below (Section 4.3) has shown
important inaccuracies in the determination of E at lower and
higher values of conversion (a < 0.2 and a > 0.8, respectively). Thus,
in order to properly contrast the results of model-free isoconver-
sional and modulated methods, the distribution of E is represented
in the region of a = 0.2–0.8 for all model-free methods.
The E calculated by means of isoconversional differential meth-
ods (D1 and D2) show very similar results, both in the evolution of
E (Fig. 6a) and its distribution (Fig. 7). I1 and I2 integral methods
show similar trends to those of D1 and D2 differential methods,
with slightly lower values of E, as observed in Fig. 6a and Fig. 7.
Compared to the rest of the isoconversional methods, I3 method
shows a slower increase of E with the conversion. I4 method shows
the lowest E values of all studied methods, and its modification (I5)
does not show any improvement, as more noise is observed in its
corresponding curve in Fig. 6b. Interestingly, the average linear
integral method (I6) has the broader data distribution (Fig. 7)
due to its high noise to signal ratio (Fig. 6b). The modified average
linear integral method (I7) improves the results obtained in the
original method, though it is computationally more complex com-
pared to other easier methods, such as differential methods or inte-
gral I1 and I2 methods.
I2 I3 I4 I5 I6 I7
 A. Ochoa et al. / Chemical Engineering Science 171 (2017) 459–470
Fig. 8. Example of the procedure performed on an experiment (b = 5 K min1;
H = 5 K; x = 1  102 Hz) to calculate the activation energy (E): (a) DTGm and Tm; (b)
vmin and vmax calculated from the DTGm of bare spent catalyst; (c) vmin and vmax
calculated from the DTGm of regenerated catalyst as blank; (d) vmin and vmax
calculated from the subtraction of the blank from the bare data.
Fig. 9. Evolution of the activation energy (E) of the coke combustion by means of
the modulated TG method (b = 5 K min1; H = 5 K; x = 1  102 Hz), obtained from
the ‘untreated’ (Fig. 8b) and ‘corrected’ modulated signals (Fig. 8d).
4.3. Modulated method
Fig. 8 shows the procedure for the determination of the activa-
tion energy corresponding to the combustion of coke deposited on
the spent catalyst, by means of the modulated method previously
explained in Section 3.3. First, the DTGm and Tm were obtained
using different heating rates, amplitudes and frequencies. Fig. 8a
shows an example of the ‘‘bare” result obtained for b = 5 K min1;
H = 5 K; x = 1  102 Hz. The equipment used allows to directly
obtain the distribution of activation energies displayed in Fig. 9
as ‘‘bare” results. Note that the grey areas of Fig. 9 correspond to
these where the results are not representative (a < 0.2 and a > 0.8).
Fig. 8a shows that at low (T < 650 K) and high temperatures
(T > 900 K), the DTG signal is oscillating in a region where the com-
bustion is not occurring (as previously proved in Fig. 2). This obser-
vation leads to the need of performing a ‘‘blank” experiment of the
catalyst without coke (after the combustion of Fig. 8a).
467
Fig. 8b and c shows the vmin and vmax of the coke + catalyst and cat-
alyst alone, respectively. The values of vmin and vmax have been
obtained by interpolating the DTG signals to the values of time
(maximums and minimums) obtained by fitting Tm-T signals (as
shown in Fig. 1b) to a sinusoidal function. More importantly,
Fig. 8b and c indicate that there is a contribution of the catalyst
to the DTGm that needs to be subtracted, and the results are shown
in Fig. 8d, whereas the calculated activation energies obtained
from these results are plotted in Fig. 9 as ‘‘corrected”. This kind
of contributions along with the experimental noise, causes scat-
tered results and incorrect determination of E (Blaine and Hahn,
1998; Moukhina, 2014). This effect is much more difficult to
observe in the combustion model and isoconversional methods,
because their heating rates are significantly lower (at local scale)
than those used in the modulated method.
In fact, activation energy values obtained by ‘‘bare” modulated
method are higher than those calculated by isoconversional meth-
ods (140–170 kJ mol1, Figs. 6 and 7) and much higher than than
those calculated with kinetic models (102–117 kJ mol1). How-
ever, sustracting the influence of the catalyst leads to E values com-
parable to those of isoconversional methods.
In order to study the influence of the modulation parameters, as
well as to propose combinations of these parameters aiming a
more accurate calculation of the activation energy, modulated runs
have been performed varying three parameters: (i) b linear heating
rate, 5–15 K min1; (ii) x modulation frequency (or 1/x period),
5  103–2  102 Hz; (iii) H amplitude, 2–8 K. The effect of these
parameters on the obtained E results is depicted in Fig. 10, in order
to determine the most appropriate modulation parameters when
performing the MTG method.
The selection of the heating rate and the frequency are related
with the required number of recorded modulation periods during
the process, thus requiring a minimum number of periods to get
acceptable results (Blaine and Hahn, 1998). Mamleev and
Bourbigot (2002) reported that in order to accurately calculate E
one should choose a heating rate so as to provide at least 10–15
modulation cycles for each decomposition stage, where each stage
corresponds to a differentiated weight loss peak in a DTG signal.
Nevertheless, results in Fig. 10 note that the recommendation
made by Mamleev and Bourbigot (2002) is not enough, as the mod-
ulation frequency is a factor to be considered depending on the
chosen heating rate. In this way, Fig. 10 shows a negative effect
(depicted as a broader distribution of E) when employing high
heating rates (15 K min1) together with high modulation periods
(200 s), as well as low heating rates (5 K min1) at low modulation
periods (50 s). On the one hand, working with high heating rates
and high modulation periods (or low frequency) would mean few
modulation periods in which to calculate E during the whole com-
bustion process, thus leading to inaccuracies (2002). On the other
hand, combining low heating rates with low modulation periods
means a high number of points in which to perform the calculation
of E, leading to little differences in the calculated values and noise
in the trend of E, due to an unavoidable experimental error. All in
all, with regard to the choice of the heating rate, results in Fig. 10
suggest employing a combination of low heating rates (5 K min1)
with high modulation periods (200 s), or low modulation periods
(50 s) when using higher heating rates (15 K min1).
Fig. 10 shows a correlation between the employed modulation
amplitude (H) and the modulation frequency, x (or period, 1/x).
In this way, on the one hand, at low modulation periods (50 s)
the distribution of the calculated E does not follow a straightfor-
ward trend when varying the amplitude. However, at high modu-
lation periods (200 s), the use of higher amplitudes shows a better
distribution (less dispersion) of the calculated E within the studied
region (a = 0.2–0.8), as well as more similar results to those the
majority of isoconversional methods (140–170 kJ mol1). This
468 A. Ochoa et al. / Chemical Engineering Science 171 (2017) 459–470

Fig. 10. Box plots of the of activation energy (E) obtained using modulated ramps with different heating rates (b), periods (1/x) and amplitudes (H). Note: the studied region
is a = 0.2–0.8.

Table 4 rate of heat transfer (Model 1 and the approximation made by

Results of the kinetic parameters obtained using the best methods. Salvador and Ferrasse (2006)) is able to predict the results of coke
Parameter M1 M1a I7b Modulatedc combustion in a thermobalance.
823K 1 3 3
Model-free isoconversional and modulated methods avoid the
k (s ) 1.8 ± 0.3  10 2.4 ± 0.4  10 – –
E (kJ mol1) 114 ± 2 36.9 ± 0.9 156 ± 4 162 ± 4 need to assume a kinetic model and calculate the evolution of E
h1 (K) – 303 ± 20 – – at different conversions. This approach is very interesting in com-
h2 (s) – 4±1 – – plex combustion processes, as the one studied, where kinetic
a
Model 1 and the approximation made by Salvador and Ferrasse (2006) with Eqs.
behavior varies significantly during the process. In this regard
(50)–(52). and according to the results shown in Fig. 7, the most recently
b
From Fig. 7. developed methods give comparable results (D2, I5, I7) to other
c
From Fig. 10, using b = 5 K min1, 1/x = 200 s and H = 8 K. less sophisticated methods such as D1, I1, I2 and I3, which are
easier to perform and provide with comparable kinetic results.
For example, the Friedman method (I1) has shown to be easy to
improvement in the reliability of results when increasing the mod- interpret and straightforward to be used to obtain the dependence
ulation amplitude is thought to be due to the fact that the ampli- of E on a. The increasing trend of E obtained in the isoconversional
tude (H) appears in the denominator of Eqs. (42) and (43) (Blaine methods (Fig. 6) has the following physical sense: during the pro-
and Hahn, 1998). Results at a medium period value (100 s) do cess of coke combustion, the lightest fraction burns first (Ibáñez
not show a straightforward trend when varying the heating rate et al., 2014), which requires lower E, and the remaining heavier
or the modulation amplitude, as it depends on both parameters. (more condensed) fraction requires higher E.
The observed correlations between heating rate, period and ampli- Model-free modulated method requires additional treatments
tude in the calculated results make it necessary to establish the in order to remove the effect of the catalyst in the combustion of
most suitable parameters to employ for obtaining kinetic parame- coke. Besides, the best practices in these methods are the
ters in future thermogravimetric runs. following:

Low heating rates (5 K min1), along with high periods (200 s)

5. Guidelines for calculating combustion kinetics
Combustion models are reliable and ‘‘solid” methods to com-
pare the kinetic parameters (pre-exponential factors and activation
energies) of cokes with different nature or location along the
heterogeneous surface of the catalyst. Nonetheless, the interfer-
ence of heat transfer, auto-accelerated dynamics and change of
composition of the unconverted material lead to different kinetic
parameters using different heating programs. Different empirical
and non-empirical equations have been incorporated in order to
estimate the evolution of the kinetic parameters with the combus-
tion profile. Those approaches lead to conclude that a rational
model for coke combustion is a first order kinetic one with the
approximation proposed by Salvador and Ferrasse (2006). When
performing a unique TG experiment for being used for fitting a par-
ticular combustion model, a good compromise is obtained using
b = 5 K min1. In this sense, the complete picture of coke combus-
tion in the FCC regenerator would imply a microkinetic scheme
considering the coke evolution during combustion, the rate of heat
transfer and the computational particle fluid dynamic (CPFD)
model, like Barracuda Virtual ReactorÒ. In this work, we have
proved that a first order kinetic model partially controlled by the
and high amplitude values (5–8 K). Slower heating rates ensure
a higher number of modulation periods in which to calculate E
during the reaction, yet they lengthen experiment time and
increase resolution at the expense of sensitivity.

If high heating rates are employed (15 K min1), low period val-
ues (50 s) are recommended, at high amplitude values (5–8 K).
However, not too low amplitude values are recommended (2 K),
in order to avoid the masking of the thermogravimetric curves
with experimental noise, which is of the utmost importance
to consider in a process with high experimental noise. Lower
experimental time is required if high heating rates are used,
and sensitivity of the registered thermogravimetric signals is
increased, at the expense of resolution.

If various combustion steps are taking place during the thermo-
gravimetric run, low heating rates are preferred rather than
higher ones, in order to register with an acceptable resolution
of the events taking place during the reaction, thus improving
the precision of the obtained kinetic parameters.
All this considered, the following combinations would be suit-
able to be employed, as an example, for the calculation of E by
MTG method: (i) b = 5 K min1; H = 8 K and x = 5  103 Hz; (ii)
 A. Ochoa et al. / Chemical Engineering Science 171 (2017) 459–470
b = 15 K min1; H = 8 K and x = 2  102 Hz. The suitability of these
proposed sets of parameters is supported by the similarity of their
results and distribution (Fig. 10) with those of the selected isocon-
versional methods (D1, I1 and I2, among others, in Fig. 7). Both
model-free isoconversional and modulated methods tend to mag-
nify the valued of E for a < 0.2 and a > 0.8, so that the results
obtained in these areas should be neglected.
The straight comparison of the results obtained here and those
of the literature is inadequate due to the divergence of reactors,
operational conditions and feeds used. However, we can point that
all methods, but particularly those selected and displayed in
Table 4, lead to rational values of activation energy compared with
these of the literature: 150 kJ mol1 (Zhang et al., 2015, 2014),
110–150 kJ mol1 (Kanervo et al., 2001), or 130–140 kJ mol1
(Leistner et al., 2012).
On the basis of the different values obtained for each method
and each model, the significance of the kinetic parameters summa-
rized in Table 4 lays on having qualitative values for the systematic
comparison of the nature or location of coke deposited on catalytic
surfaces at different conditions. Moreover, we have not compared
the isoconversional methods on the grounds of a statistical analysis
(unlike for the model-based methods summarized in Fig. 4), due to
the fact that the results of E do not vary that match (Fig. 7), and
additional aspects as computational ease could be a key factor
for choosing the particular isoconversional method. On the other
hand, the comparison of the modulated methods has been per-
formed on the basis the narrower distribution of calculated E.
6. Conclusions
Model-based methods, and first order kinetic model in particu-
lar, are attractive for quickly obtaining apparent kinetic parameters
of the combustion of cokes deposited under different conditions, or
with different nature or location within the catalytic surface. These
methods allow to obtain a simple model for the regenerator design.
However, model-free (isoconversional and modulated) methods
are advantageous for obtaining more reliable kinetic parameters,
without the need of assuming a particular model, and able to pre-
dict the real dynamics of coke combustion.
The isoconversional methods such as the ones proposed by
Friedman (D1), Kissinger-Akahira-Sunose (I1), Ozawa-Flynn-Wall
(I2) or Li-Tang (I3), are relative easy to be applied while giving
acceptable results. On the other hand, the modulated method
requires a single experiment (instead of a series of 4–5 required
for the isoconversional methods) in order to obtain the kinetic
parameters, but its results are not completely reliable and require
a further experiment with the bare catalyst and additional mathe-
matical treatment in order to obtain intrinsic kinetic parameters.
The results show rational combinations of process conditions for
modulated method: (i) heating rates of 5 K min1, along with mod-
ulation periods of 200 s and amplitudes of 5–8 K; or (ii) heating
rates of 15 K min1, modulation periods of 50 s and amplitudes
of 5–8 K.
Additional insights of microkinetic modeling of the combustion
process considering the phenomena of coke evolution during the
combustion like aging and the parallel degradation (pyrolysis)
could yield a better interpretation of the kinetic parameters
involved.
Acknowledgments
The founding sources acknowledged for this work are the fol-
lowing: Ministry of Economy and Competitiveness of the Spanish
Government (MINECO), cofounded with EFDR funds (projects
CTQ2013-46172-P, CTQ2015-67425-R and CTQ2016-79646-P),
469
Basque Government (Spain, project IT748-13) and University of
the Basque Country (Spain, UPV/EHU, UFI 11/39).
Appendix A. Supplementary material
Supplementary data associated with this article can be found, in
the online version, at http://dx.doi.org/10.1016/j.ces.2017.05.039.
References
Aguayo, A.T., Castaño, P., Mier, D., Gayubo, A.G., Olazar, M., Bilbao, J., 2011. Effect of
cofeeding butane with methanol on the deactivation by coke of a HZSM-5
zeolite catalyst. Ind. Eng. Chem. Res. 50, 9980–9988.
Aguayo, A.T., Gayubo, A.G., Atutxa, A., Olazar, M., Bilbao, J., 1999. Regeneration of a
catalyst based on a SAPO-34 used in the transformation of methanol into
olefins. J. Chem. Technol. Biotechnol. 74, 1082–1088.
Aguayo, A.T., Gayubo, A.G., Ereña, J., Atutxa, A., Bilbao, J., 2003. Coke aging and its
incidence on catalyst regeneration. Ind. Eng. Chem. Res. 42, 3914–3921.
Akahira, T., Sunose, T., 1971. Res. Report Chiba Inst. Technol. (Sci. Technol.) 16, 22–
31.
Altun, N.E., Hicyilmaz, C., Kök, M.V., 2003. Effect of particle size and heating rate on
the pyrolysis of Silopi asphaltite. J. Anal. Appl. Pyrol. 67, 369–379.
Altun, N.E., Kok, M.V., Hicyilmaz, C., 2002. Effect of particle size and heating rate on
the combustion of Silopi asphaltite. Energy Fuels 16, 785–790.
Argyle, M.D., Bartholomew, C.H., 2015. Heterogeneous catalyst deactivation and
regeneration: a review. Catalysts 5, 145–269.
Bartholomew, C.H., Farrauto, R.J., 2006. Fundamentals of Industrial Catalytic
Processes. Wiley-Interscience, Hoboken, New Jersey.
Blaine, R.L., Hahn, B.K., 1998. Obtaining kinetic parameters by modulated
thermogravimetry. J. Therm. Anal. Calorim. 54, 695–704.
Borio, D.O., Menendez, M., Santamaria, J., 1992. Simulation and optimization of a
fixed bed reactor operating in coking-regeneration cycles. Ind. Eng. Chem. Res.
31, 2699–2707.
Budrugeac, P., 2002. Differential non-linear isoconversional procedure for
evaluating the activation energy of non-isothermal reactions. J. Therm. Anal.
Calorim. 68, 131–139.
Čejka, J., van Bekkum, H., Corma, A., Schüth, F., 2007. Introduction to Zeolite
Molecular Sieves. Elsevier Science, Amsterdam.
Chen, G.Q., Luo, Z.H., Lan, X.Y., Xu, C.M., Gao, J.S., 2013. Evaluating the role of
intraparticle mass and heat transfers in a commercial FCC riser: a meso-scale
study. Chem. Eng. J. (Lausanne) 228, 352–365.
de Lasa, H.I., 1992. Riser Simulator. US5102628A – CA1284017C. The University Of
Western Ontario, Canada.
Epelde, E., Ibañez, M., Aguayo, A.T., Gayubo, A.G., Bilbao, J., Castaño, P., 2014.
Differences among the deactivation pathway of HZSM-5 zeolite and SAPO-34 in
the transformation of ethylene or 1-butene to propylene. Microporous
Mesoporous Mater. 195, 284–293.
Epelde, E., Santos, J.I., Florian, P., Aguayo, A.T., Gayubo, A.G., Bilbao, J., Castaño, P.,
2015. Controlling coke deactivation and cracking selectivity of MFI zeolite by
H3PO4 or KOH modification. Appl. Catal. A 505, 105–115.
Flynn, J.H., Wall, L.A., 1966. General treatment of the thermogravimetry of
polymers. J. Res. Natl. Bureau Stand. 70A, 487–523.
Friedman, H.L., 1964. Kinetics of thermal degradation of char-forming plastics from
thermogravimetry. Application to a phenolic plastic. J. Polym. Sci. Part C: Polym.
Symp. 6, 183–195.
Gary, J.H., Handwerk, G.E., Kaiser, M.J., 2007. Peroleum Refinning. Technology and
Economics. CRC Press, Boca Raton.
Guisnet, M., Ribeiro, F.R., 2011. Deactivation and Regeneration of Zeolite Catalysts.
Imperial Collegue Press, London.
Han, Y., Li, T., Saito, K., 2013. A modified Ortega method to evaluate the activation
energies of solid state reactions. J. Therm. Anal. Calorim. 112, 683–687.
Han, Y., Xi, T., Saito, K., 2012. Comprehensive method based on model free method
and IKP method for evaluating kinetic parameters of solid state reactions. J.
Comput. Chem. 33, 2516–2525.
Ibáñez, M., Artetxe, M., Lopez, G., Elordi, G., Bilbao, J., Olazar, M., Castaño, P., 2014.
Identification of the coke deposited on an HZSM-5 zeolite catalyst during the
sequenced pyrolysis–cracking of HDPE. Appl. Catal. B 148–149, 436–445.
Ibáñez, M., Valle, B., Bilbao, J., Gayubo, A.G., Castaño, P., 2012. Effect of operating
conditions on the coke nature and HZSM-5 catalysts deactivation in the
transformation of crude bio-oil into hydrocarbons. Catal. Today 195, 106–113.
Ibarra, Á., Veloso, A., Bilbao, J., Arandes, J.M., Castaño, P., 2016. Dual coke
deactivation pathways during the catalytic cracking of raw bio-oil and
vacuum gasoil in FCC conditions. Appl. Catal. B 182, 336–346.
Kanervo, J.M., Krause, A.O.I., Aittamaa, J.R., Hagelberg, P.H., Lipiäinen, K.J.T., Eilos, I.
H., Hiltunen, J.S., Niemi, V.M., 2001. Kinetics of the regeneration of a cracking
catalyst derived from TPO measurements. Chem. Eng. Sci. 56, 1221–1227.
Kissinger, H.E., 1956. Variation of peak temperature with heating rate in differential
thermal analysis. J. Res. Natl. Bureau Stand. 57, 217–221.
Koleva, D., Atanassov, A., Nedelchev, N., 2008. Nonisothermal degradation kinetics
of ultra-high molecular weight polyethene composites filled with carbon or
aramid fibers. Int. J. Polym. Mater. Polym. Biomater. 57, 841–851.
470 A. Ochoa et al. / Chemical Engineering Science 171 (2017) 459–470
Le Minh, C., Li, C., Brown, T.C., Bartholomew, C.H., Fuentes, G.A., 1997. Kinetics of
coke combustion during temperature-programmed oxidation of deactivated
cracking catalysts. Stud. Surface Sci. Catal, 383–390.
Leistner, K., Nicolle, A., Berthout, D., da Costa, P., 2012. Kinetic modelling of the
oxidation of a wide range of carbon materials. Combust. Flame 159, 64–76.
Li, C.-R., Tang, T., 1997. Dynamic thermal analysis of solid-state reactions. J. Therm.
Anal. 49, 1243–1248.
Li, C.-R., Tang, T.B., 1999. A new method for analysing non-isothermal
thermoanalytical data from solid-state reactions. Thermochim. Acta 325, 43–
46.
Magnoux, P., Cerqueira, H.S., Guisnet, M., 2002. Evolution of coke composition
during ageing under nitrogen. Appl. Catal. A 235, 93–99.
Mamleev, V., Bourbigot, S., 2002. Calculation of activation energies using the
sinusoidally modulated temperature. J. Therm. Anal. Calorim. 70, 565–579.
Mamleev, V., Bourbigot, S., 2005. Modulated thermogravimetry in analysis of
decomposition kinetics. Chem. Eng. Sci. 60, 747–766.
Mamleev, V., Bourbigot, S., Bras, M., Lefebvre, J., 2004. Three model-free methods for
calculation of activation energy in TG. J. Therm. Anal. Calorim. 78, 1009–1027.
Mamleev, V., Bourbigot, S., Le Bras, M., Duquesne, S., Sestak, J., 2000.
Thermogravimetric analysis of multistage decomposition of materials. Phys.
Chem. Chem. Phys. 2, 4796–4803.
Marcilla, A., Gomez-Siurana, A., Valdes, F.J., 2008. Influence of the final ‘‘ageing”
temperature on the regeneration behaviour and location of the coke obtained in
the HZSM-5 and USY zeolites during the LDPE cracking. Appl. Catal. A 334, 20–
25.
Moljord, K., Magnoux, P., Guisnet, M., 1995. Coking, aging and regeneration of
zeolites. XVI. Influence of the composition of HY zeolites on the removal of coke
through oxidative treatment. Appl. Catal. A 121, 245–259.
Moukhina, E., 2014. Direct analysis in modulated thermogravimetry. Thermochim.
Acta 576, 75–83.
Opfermann, J.R., Kaisersberger, E., Flammersheim, H.J., 2002. Model-free analysis of
thermoanalytical data-advantages and limitations. Thermochim. Acta 391, 119–
127.
Ortega, A., 2008. A simple and precise linear integral method for isoconversional
data. Thermochim. Acta 474, 81–86.
Ozawa, T., 1965. A new method of analyzing thermogravimetric data. Bull. Chem.
Soc. Jpn. 38, 1881–1886.
Qing, W., Xudong, W., Hongpeng, L., Chunxia, J., 2013. Study of the combustion
mechanism of oil shale semi-coke with rice straw based on Gaussian multi-
peak fitting and peak-to-peak methods. Oil Shale 30, 157–172.
Royo, C., Perdices, J.M., Monzón, A., Santamaría, J., 1996. Regeneration of fixed-bed
catalytic reactors deactivated by coke: influence of operating conditions and of
different pretreatments of the coke deposits. Ind. Eng. Chem. Res. 35, 1813–
1823.
Salvador, S., Ferrasse, J.H., 2006. A practical method to derive sample temperature
during nonisothermal coupled thermogravimetry analysis and differential
scanning calorimetry experiments. Chem. Eng. Technol. 29, 696–702.
Sbirrazzuoli, N., Vincent, L., Mija, A., Guigo, N., 2009. Integral, differential and
advanced isoconversional methods: complex mechanisms and isothermal
predicted conversion-time curves. Chemometrics Intell. Lab. Syst. 96, 219–226.
Schawe, J.E.K., 2014. A general approach for temperature modulated
thermogravimetry: extension to non-periodical and event-controlled
modulation. Thermochim. Acta 593, 65–70.
Upson, L.L., Lomas, D.A., 2000. Catalyst Regeneration, FCC Units, Kirk-Othmer
Encyclopedia of Chemical Technology. John Wiley & Sons, Inc..
Vicente, J., Ereña, J., Montero, C., Azkoiti, M.J., Bilbao, J., Gayubo, A.G., 2014. Reaction
pathway for ethanol steam reforming on a Ni/SiO2 catalyst including coke
formation. Int. J. Hydrogen Energy 39, 18820–18834.
Vyazovkin, S., 1997. Evaluation of activation energy of thermally stimulated solid-
state reactions under arbitrary variation of temperature. J. Comput. Chem. 18,
393–402.
Vyazovkin, S., 2001. Modification of the integral isoconversional method to account
for variation in the activation energy. J. Comput. Chem. 22, 178–183.
Vyazovkin, S., Dollimore, D., 1996. Linear and nonlinear procedures in
isoconversional computations of the activation energy of nonisothermal
reactions in solids. J. Chem. Inf. Comput. Sci. 36, 42–45.
Vyazovkin, S., Wight, C.A., 2000. Estimating realistic confidence intervals for the
activation energy determined from thermoanalytical measurements. Anal.
Chem. 72, 3171–3175.
Zhang, Y., Sun, G., Gao, S., Xu, G., 2015. Regeneration kinetics of spent FCC catalyst
via coke gasification in a micro fluidized bed. Procedia Eng., 1758–1765
Zhang, Y., Yao, M., Sun, G., Gao, S., Xu, G., 2014. Characteristics and kinetics of coked
catalyst regeneration via steam gasification in a micro fluidized bed. Ind. Eng.
Chem. Res. 53, 6316–6324.