polymers
Article
Catalytic Pyrolysis of PET Polymer Using Nonisothermal
Thermogravimetric Analysis Data: Kinetics and Artificial
Neural Networks Studies
Ibrahim Dubdub * and Zaid Alhulaybi *
Citation: Dubdub, I.; Alhulaybi, Z.
Catalytic Pyrolysis of PET Polymer
Using Nonisothermal
Thermogravimetric Analysis Data:
Kinetics and Artificial Neural
Networks Studies. Polymers 2023, 15,
70. https://doi.org/10.3390/
polym15010070
Received: 7 December 2022
Revised: 16 December 2022
Accepted: 21 December 2022
Published: 24 December 2022
Copyright: © 2022 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
Polymers 2023, 15, 70. https://doi.org/10.3390/polym15010070
Department of Chemical Engineering, King Faisal University, Al-Ahsa 31982, Saudi Arabia
* Correspondence: idubdub@kfu.edu.sa (I.D.); zalhulaybi@kfu.edu.sa (Z.A.)
Abstract: This paper presents the catalytic pyrolysis of a constant-composition mixture of zeolite
β and polyethylene terephthalate (PET) polymer by thermogravimetric analysis (TGA) at different
heating rates (2, 5, 10, and 20 K/min). The thermograms showed only one main reaction and shifted
to higher temperatures with increasing heating rate. In addition, at constant heating rate, they moved
to lower temperatures of pure PET pyrolysis when a catalyst was added. Four isoconversional models,
namely, Kissinger–Akahira–Sunose (KAS), Friedman, Flynn–Wall–Qzawa (FWO), and Starink, were
applied to obtain the activation energy (Ea ). Values of Ea acquired by these models were very close
to each other with average value of Ea = 154.0 kJ/mol, which was much lower than that for pure
PET pyrolysis. The Coats–Redfern and Criado methods were employed to set the most convenient
solid-state reaction mechanism. These methods revealed that the experimental data matched those
obtained by different mechanisms depending on the heating rate. Values of Ea obtained by these
two models were within the average values of 157 kJ/mol. An artificial neural network (ANN) was
utilized to predict the remaining weight fraction using two input variables (temperature and heating
rate). The results proved that ANN could predict the experimental value very efficiently (R2 > 0.999)
even with new data.
Keywords: PET; catalytic pyrolysis; zeolite β; kinetics; TGA; ANN
1. Introduction
Plastic waste has become a serious global problem, especially in countries where
a huge quantity of plastic waste is produced and disposed. There are many academic
research centers attempting to find a suitable way to recycle such vast waste [1].
PET is one of six polymers, with the others being low-density polyethylene (LDPE),
high-density polyethylene (HDPE), polystyrene (PS), polypropylene (PP), and polyvinyl
chloride (PVC). PET accounts for almost 7.6 wt% of plastic solid waste (PSW) [2,3]. It is
very commonly used in manufacturing food packaging and fiber cloth and is considered as
the most recycled material [4]. PET is mostly used in manufacturing fiber, staple filaments,
textile, and carpet [3,5].
TGA is a thermal analysis method that is best known for calculating the kinetic
parameters of plastic pyrolysis. Estimation of kinetic parameters from TGA data can be
performed by different methods. These methods use either multiple TGA datasets at
different heating rates (called “model-free” or “isoconversional” method) or one single
TGA dataset (called “model-fitting” method) and generate the final kinetic equation using
either differential method or integral technique.
In the following literature survey, only the kinetics of PET by TGA data and the
kinetics of catalytic cracking of different plastic waste polymers by TGA data are examined.
Yang et al. (2001) [6] initiated the kinetic study of PET using the Arrhenius equation with
Friedman’s method for DTG curves to obtain the kinetic parameters of PET polymers. TG
and DTG curves showed only one single decomposition that started at 370 ◦ C, reached
https://www.mdpi.com/journal/polymers
Polymers 2023, 15, 70 2 of 14

maximum at 441 ◦ C, and ended at 502 ◦ C at 10 ◦ C/min heating rate. The authors obtained
activation energy value of 242 kJ/mol with reaction order (n) of 1. Martín-Gullón et al.
(2001) [2] showed that the pyrolysis reaction could almost be divided into two reactions.
The first one with 80% of weight loss, considered as the main reaction within the range
of the temperature (700–750 K), had activation energy of 256.4 kJ/mol, while the second
one with only 6% weight loss had activation energy of 332.3 kJ/mol. Girija et al. (2005) [7]
highlighted that PET had only single-stage degradation at temperature of 440 ◦ C with
weight loss of 60%. Saha and Ghoshal (2005) [8] studied the pyrolysis kinetics for different
soft drink bottles to obtain the activation energy value using different models. They
reported Ea within the values of 320–339.89 kJ/mol.
Aguado et al. (2007) [9] measured the activity of two catalysts, namely, n-HZMS-5 and
Al-MCM-41, on the degradation of LDPE. This activity was measured by how much the
TG shifted to lower temperature. They stated that the pure polymer started to degrade
at 477 ◦ C and shifted to 419 ◦ C with Al-MCM-41 and then further to 396 ◦ C with zeolite
n-HZSM-5. They attributed this strong activity for zeolite n-HZSM-5, which has weak
acid properties, to the large pore dimension. Moltó et al. (2007) [10] investigated the
kinetics of PET fiber cloth pyrolysis with three different heating rates. They found three
different partial weight loss reactions, with the second reaction being the main one within
the range of 640–730 K temperature. They first suggested three independent reactions with
activation energy values of 121.5, 238.3, and 300.0 kJ/mol but later considered the second
reaction with 238.3 kJ/mol as the main reaction. These results were close to the findings of
Yang et al. (2001) [6], who showed an energy value of 242 kJ/mol.
Halász et al. (2009) [11] repeated the same experiment carried out by Aguado et al.
(2007) [9] but with PET polymer. They found that pure polymer PET had DTG peak
maximum equal to 473 ◦ C, which shifted to 412 ◦ C with H-ZSM-5 and to 458 ◦ C with
Ti-MCM-41.
Çepelioğullar and Pütün (2013) [12] investigated the effect of different biomasses (cot-
ton stalk (CS), hazelnut shell (HS), sunflower residue (SFR), and arid land plant Euphorbia
rigida (ER)) in the copyrolysis with PET. They mentioned that the pyrolysis of PET decom-
posed at two stages, which started at 360 ◦ C with peak temperature of 427.7 ◦ C and weight
loss of about 79.78%. They made unusual observation when the pyrolysis of biomass–PET
mixtures occurred at temperatures higher than pure biomass and PET. Moreover, they
found the range of activation energy values of PET was higher than the biomass used.
This demonstrated that the decomposition of biomass started at lower temperatures. They
attributed this difference to the structural differences between them. For pure PET, they
obtained a reaction in two stages. The first stage had temperature range of 373–443 ◦ C,
conversion range of 0.4–81%, and activation energy value of 347.4 kJ/mol, while the second
stage had a temperature range of 448–503 ◦ C, conversion range of 88–99%, and activation
energy value of 172.6 kJ/mol. They presented two tables of activation energy values for
different pure biomasses (CS, HS, SFR, and ER), PET, and the biomass–PET mixture. The
average activation energy value of CS, PET, and CS–PET decompositions were 71, 260, and
95.7 kJ/mol, respectively.
Lee et al. (2018) [13] studied the catalytic decomposition of mixed torrefied cork
oak (TOak) and HDPE using three different catalysts (mesoporous HY1, mesoporous
HY2, and microporous HY). They stated that the catalytic pyrolysis of HDPE over HY
had two reactions within the range of 200–300 and 300–400 ◦ C with peak temperature
of 392 ◦ C, which was lower than the pure FDPE temperature of 477 ◦ C. This behavior
was different from the catalytic copyrolysis of Oak/HDPE and TOak/HDPE over HY. In
these cases, the peak temperature was shifted higher to 411 and 414 ◦ C, respectively. Diaz
Silvarrey and Phan (2016) [14] found that the decomposition of PET by TGA started earlier
than most other polymers with a thermal decomposition range of 395–520 ◦ C. They used
the KAS method to obtain 197.61 kJ/mol as the average value of activation energy and
4.84 × 1014 s−1 as the pre-exponential factor.
Polymers 2023, 15, 70 3 of 14

Xiang et al. (2018) [15] investigated the catalytic decomposition of cobalt-modified

ZSM-5 catalyst on copyrolysis of lignocellulosic biomass (rice straw (RS)) and linear low-
density polyethylene (LLDPE). They stated that there was only one peak for the pyrolysis
of pure RS at 341.39 ◦ C and pure LLDPE at 491.53 ◦ C, two peaks for the copyrolysis of
RS–LLDPE at 339.57 and 470.06 ◦ C, and two peaks for the pyrolysis of RS–LLDPE with
Co/ZSM-5 catalyst at 336.32 and 451.73 ◦ C. They calculated the activation energy of the
pyrolysis of LLDPE, RS, RS–LLDPE, and catalytic RS–LLDPE as 212.23, 79.61, 34.27, and
41.04 kJ/mol, respectively. Das and Tiwari (2019) [5] used PET collected from soft drink
bottles to evaluate the degradation kinetics of nonisothermal TGA runs with five different
heating rates. They obtained activation energy values in the range of 203–355 kJ/mol using
the Vyazovkin method and established the F1 reaction model using Criados’ master plot.
Osman et al. (2020) [16] employed TGA analysis for PET pyrolysis with five heating
rates with a ratio of 16 between the lowest (0.5 K/min) and highest (8 K/min) rates. They
applied different isoconversional methods (integral Friedman, ASTM-E698, and integral
FWO) and concluded that the activation energy was between 165 and 195 kJ/mol.
Ali et al. (2020) [17] studied the catalytic cracking kinetics of polystyrene with syn-
thesized copper oxide as a catalyst. They used Friedman, KAS, FWO, and Coats–Redfern
to obtain the kinetic parameter values. They concluded that the low value of the activa-
tion energy would help to run experiments at lower temperatures. Lai et al. (2021) [18]
examined the performance of coal ash (fly and bottom ash) with different blending ratios
for the catalytic pyrolysis of LLDPE. They showed that 15 wt% bottom ash had higher
effect on catalytic activity compared to 15 wt% fly ash. Yap et al. (2022) [19] used two
different catalysts (HZSM-5 zeolite and natural limestone (LS)) for copyrolysis of HDPE
and Chlorella vulgaris mixture. They concluded that the combined HZSM-5/LS catalyst
had the best effect in the copyrolysis of the HDPE and Chlorella vulgaris mixture, with the
activation energy decreasing from 144.93–225.84 to 75.37–76.90 kJ/mol. Kokuryo et al.
(2022) [20] studied the catalytic degradation of LDPE using two types of catalysts (zeolite β
and zeolite MTES-β) at 5 K/min heating rate. They concluded that zeolite MTES-β pushed
LDPE to decompose at a lower temperature than zeolite β. Dourari et al. (2022) [21] and
Tarchoun et al. (2022) [22] developed new composites with different materials and tested
them using various analytical techniques. They concluded that there was a big drop in the
activation energy of the thermal behavior.
In this study, the activation energy of PET obtained by catalytic thermal decomposition
was calculated by four model-free methods and two model-fitting techniques. In order to
study the effect of the zeolite β catalyst, these kinetic parameter values were compared
with the pure PET pyrolysis values collected from the literature survey. Furthermore,
the catalytic pyrolysis behavior of PET was determined for the first time by a strongly
promising ANN model.

2. Materials and Methods

2.1. Materials
PET polymer samples were collected from Recycled Plastic (Ipoh, Malaysia). The
two main analyses of the polymer (shown in Table 1) were performed using PerkinElmer
thermal analyzer and PerkinElmer elemental analyzer (Waltham, MA, USA). Zeolite β (its
main properties are shown in Table 2) catalyst in powder form was obtained from Alfa
Aesar (Ward Hill, MA, USA).

Table 1. The two main analyses of PET.

Proximate Analysis, wt% Ultimate Analysis, wt%

Plastic
Moisture Volatile Ash C H N O
PET 0.523 88.231 11.246 64.256 4.367 0 31.377
Polymers 2023, 15, 70 4 of 14

Table 2. The main properties of Zeolite β.

S.A. m2 /g SiO2 :Al2 O3 Pore Volume 1.7–300 nm, m2 /g

680 25:1 0.127

All the details related to preparing the powders and sample weighing can be found
in [23].

2.2. Catalytic Thermal Decomposition of PET

The 1020 series TGA7 thermogravimetric analysis was used to meet the objective of
this research. Under the control of the system controller, the changes in weight due to
chemical reaction were measured. PET polymer pellets were ground into powders before
being fed to TGA. A good reproducibility was achieved in the range of 5–100 mg. Therefore,
samples weighing 10 mg were used throughout the study. The samples were heated at 2, 5,
10, and 20 K/min under nitrogen at 20 mL/min within the temperature range of 298–873 K
at constant composition of zeolite β to PET (20/80 wt/wt).

2.3. Kinetic Theory

The rate of reactions (r) can be expressed as a product of a temperature (T)-dependent
function, K(T), and a composition- or conversion (α)-dependent term, f (α), as follows [24]:

dα
r= = K ( T ) f (α) (1)
dt
wo − w
α= (2)
wo − w f
The definitions of α, t, T, wo , wo , and wf can be found in [23]. The temperature-
dependent term, which is the reaction rate constant (K), is assumed to obey the
Arrhenius relationship:  
E
K ( T ) = A0 exp − (3)
RT
where E is the activation energy (kJ/mol), A0 is the pre-exponential factor K−1 , and R is
the universal gas constant (8.314 J/mol.K). If it is assumed that a simple nth-order kinetic
relationship holds for the conversion-dependent term, then the following applies:

f ( α ) = (1 − α ) n (4)

Substituting Equations (3) and (4) in (1) yields the following:

 
dα E
r= = A0 exp − (1 − α ) n (5)
dt RT

All the published studies have used either single or multiple thermograms at different
heating rates with either differential or integral methods for calculating kinetic parameters
from TGA data [24]. However, in this work, two different sets of methods were examined
and compared with the published available data. One set used multiple thermograms,
called model-free methods, which included Friedman, FWO, KAS, and Starnik (different
heating rates and constant conversion). The second set used single thermograms, called
model-fitting methods, which included Coat–Redfern and the Criado equation. Derivations
of all these equations with the common solid-state thermal reaction mechanisms g(α) used
in the Coats–Redfern method can be found in [23,25,26].

2.4. Topology of ANNs

There is a common problem in modeling any engineering process, especially if the
final model is very complicated with high nonlinearity and too many parameters. In these
Polymers 2023, 15, 70 5 of 14

cases, ANN can be considered as another alternative option. Muravyev et al. (2021) [27]
reviewed the main concept of ANN for application in pyrolysis, thermal analysis, and
kinetic studies. They highlighted that there is no single mathematical technique to resolve
all pyrolysis problems.
An ANN architecture is ordered in three consecutive layers of neurons: input, hidden,
and output. In each layer, there is a weight matrix, bias vector, and output vector [28]. The
variables influencing the structure of the network should be fixed and the whole problem
area should be represented outside the ANN. ANN architecture should have a number of
layers, with each layer having a transfer function and connections between the layers. The
type of problem presented will determine the best architecture. The number of neurons in
the hidden layer, the number of hidden layers, the momentum rate, and the learning rate
are part of the optimizing parameters.
In this research, the performance of an ANN model was monitored by the following
statistical parameters: average correlation factor (R2 ), root mean square error (RMSE), mean
absolute error (MAE), mean bias error (MBE), and correlation coefficient (R) statistical
correlations [1,29–31].
The remaining mass percentage of PET was measured by the ANN model. There are
some advantages and maybe some disadvantages in using ANN. Some of these advan-
tages include the ability to work with linear and nonlinear relationships and learn these
relationships directly from the data used, while some of these disadvantages including the
need for large memory and computational efforts for fitting [32].

3. Results and Discussion

3.1. TG–DTG Analysis of PET
TG and DTG at different heating rates of the catalytic pyrolysis of PET by zeolite
β are shown in Figure 1. The thermograms were similar for all different heating rates.
Osman et al. (2022) [33] and Farrell et al. (2021) [34] emphasized that TGA experiments
should be run according to the International Confederation for Thermal Analysis and
Calorimetry (ICTAC) instructions in order to collect the proper data for calculation of
kinetic parameters. Therefore, they used 16 as the ratio between the highest and lowest
heating rates. Muravyev and Vyazovkin (2022) [35] reviewed the most proper options that
should be chosen as advised by the International Confederation for Thermal Analysis and
Calorimetry (ICTAC). ICTAC mainly recommends using model-free methods, where the
data should be collected at more than three different heating rates. ICTAC emphasized
that TGA experiments should be run according to its instructions in order to collect the
proper data for kinetic parameter calculations. Because the TG curve showed only one peak
reaction, single-step thermal decomposition as recommended by ICTAC was considered to
derive the proper kinetic parameters [36].
Catalytic thermal decomposition onset, peak, and final temperatures, which were
obtained from the two curves, moved to higher temperature when the heating rate increased
from 2 to 20 K/min. Diaz Silvarrey and Phan (2016) [14] mentioned that as the heating rate
increases, the peak of DTG will be bigger.
It has been shown that all plastic polymers except PVC degrade by only one reaction,
including PET [5,6]. As shown in Table 3, the onset, peak, and final temperatures of catalytic
cracking of PET by zeolite β catalyst were smaller than those of cracking decomposition
of pure PET. For example, the onset, peak, and final temperatures of catalytic cracking at
10 K/min were 625, 710, 740 K in this study, while they were 671, 711, 748 K in [5] and
643, 714, 775 K in [6]. Al-Salem et al. (2017) [4] highlighted that catalytic cracking is more
promising than thermal degradation as it requires lower energy.
Polymers 2023,15,
Polymers2023, 15,70
x FOR PEER REVIEW 66 of 14
14

Figure 1.1. Thermogravimetric

Figure Thermogravimetric analysis
analysis (TGA)
(TGA) and
and derivative
derivative thermogravimetric
thermogravimetric (DTG)
(DTG) curves
curves for
for
catalytic cracking of PET at different heating rates.
catalytic cracking of PET at different heating rates.

Table3.3.Comparison
Table Comparisonbetween
betweencatalytic
catalyticcracking
crackingcharacteristics
characteristicsof
ofPET
PETat
atdifferent
differentheating
heatingrates
rates(this
(this
work)and
work) andpyrolysis
pyrolysisof
ofpure
purePET
PET[5,6,16].
[5,6,16].

HeatingRate Main
Heating
Reaction (Catalytic Cracking)
Main Reaction (Catalytic Cracking)
Pyrolysis of Pure PET
Pyrolysis of Pure PET
Rate Onset Peak
Peak Final
Final Mass
Mass Onset Peak
Peak References
Final Final Mass Mass Loss References
K/min
K/min Onset (K) Onset (K)
(K) (K)
(K) (K) Loss
(K) Loss
(%)(%) (K) (K)(K) (K) (K)Loss (%) (%)
2 575 675 700 70 623 667 694 100 Osman
Osman et al. (2020)et al.
[16]
52 575
600 675
685 700
710 70 75 623
658 667
700 723 694 80 100
Das and Tiwari (2019)
(2020) [5]
[16]
Das and Tiwari (2019) [5],
10 625 710 740 78 671 (643) 711 (714) 748 (775) 80 Das and Tiwari
5 600 685 710 75 658 700 723 80 Yang et al. (2001) [6]
20 650 720 750 85 681 721 759 80 (2019)
Das and Tiwari [5][5]
(2019)
Das and Tiwari
10 625 710 740 78 671 (643) 711 (714) 748 (775) 80 (2019) [5], Yang et
3.2. Model-Free Kinetics Calculation
al. (2001) [6]
Equation (5) is considered as the starting equation from which all the other Das and equations
Tiwari
20 650 720 750 85 681 721 759
can be attained. Four different types of isoconversional methods80were implemented
(2019) [5] to
calculate the main kinetic parameter (activation energy (Ea ) in kJ/mol). The derivation of
these four methods
3.2. Model-Free can Calculation
Kinetics be found in [23]. Friedman, FWO, KAS and Starnik at conversions
ranging from 0.1 to 0.8 and the activation energies values for the four methods are presented
Equation (5) is considered as the starting equation from which all the other equations
in Figure 2 and Table 4. From these, it can be seen that the values of activation energy
can be attained. Four different types of isoconversional methods were implemented to
with the Friedman’s method (average 167.5 kJ/mol) were slightly higher than and totally
calculate the main kinetic parameter (activation energy (Ea) in kJ/mol). The derivation of
different from the rest three, where the values were homogeneous with an average of
these four methods can be found in [23]. Friedman, FWO, KAS and Starnik at conversions
140.2 kJ/mol. These two values were still much less than what has been published for pure
ranging from 0.1 to 0.8 and the activation energies values for the four methods are pre-
PET pyrolysis. Ea values in the range of 203–355 [5], 165–195 [37], and 185–230 [14] kJ/mol
sented in Figure 2 and Table 4. From these, it can be seen that the values of activation
have previously been reported for pyrolysis of PET. Therefore, the lowest activation for PET
energy with the Friedman’s method (average 167.5 kJ/mol) were slightly higher than and
degradation in the presence of zeolite β catalyst proves the suitability of using a catalyst at
totally different from the rest three, where the values were homogeneous with an average
lower temperatures. This finding has been supported by many researchers [4,9,11,17,20].
of 140.2 kJ/mol. These two values were still much less than what has been published for
Al-Salem et al. (2017) [4] reviewed and confirmed that a catalyst lowers the required
pure PET pyrolysis. Ea values in the range of 203–355 [5], 165–195 [37], and 185–230 [14]
pyrolysis temperature for all polymers. As shown in Table 3, there was no trend in the
kJ/mol have
activation previously
energy, except been
that thereported
Freidmanfor methods
pyrolysisstarted
of PET.atTherefore, the(130
lower values lowest activa-
kJ/mol at
tion for PET degradation in the presence of zeolite β catalyst proves the suitability
α = 0.1), increased over the period of degradation (188 kJ/mol at α = 0.6), and decreasing of using
a catalyst
until at lower
the end temperatures.
(136 kJ/mol at α = 0.8).This
Forfinding has been
FWO, KAS, and supported
Starnik, theby many
trend wasresearchers
the same,
[4,9,11,17,20]. Al-Salem et al. (2017) [4] reviewed and confirmed that a catalyst
with a steady increase from an average of 93.33 kJ/mol (α = 0.1) to an average of 164 lowers the
kJ/mol
required
(α pyrolysis
= 0.7) and then a temperature
decrease to 156for kJ/mol
all polymers. As shown in Table 3, there was no trend
(α = 0.8).
in the activation energy, except that the Freidman methods started at lower values (130
kJ/mol at α = 0.1), increased over the period of degradation (188 kJ/mol at α = 0.6), and
Polymers 2023, 15, x FOR PEER REVIEW 7 of 14

decreasing until the end (136 kJ/mol at α = 0.8). For FWO, KAS, and Starnik, the trend was
Polymers 2023, 15, 70 7 of 14
the same, with a steady increase from an average of 93.33 kJ/mol (α = 0.1) to an average of
164 kJ/mol (α = 0.7) and then a decrease to 156 kJ/mol (α = 0.8).

Figure 2.2.Linear
Figure Linearregression
regression lines
lines of catalytic
of catalytic cracking
cracking of PETofatPET at different
different conversions
conversions by (a)
by (a) Freidman,
Freidman,
(b) FWO, (c)(b)KAS,
FWO, (c)(d)
and KAS, and (d)
Starnik Starnik methods.
methods.
Table 4. Activation
Table 4. Activation energies
energies by
by Friedman,
Friedman, FWO,
FWO, KAS,
KAS, and
and Starnik
Starnikmethods.
methods.

Friedman
Friedman FWO
FWO KAS
KAS Starnik
Conversion
Conversion E (kJ/mol) R2 R2
E (kJ/mol) E (kJ/mol)
E (kJ/mol) R2R2 E E(kJ/mol)
(kJ/mol) RR22 E (kJ/mol) R2
0.10.1 130 130 0.9621
0.9621 97 97 0.9286
0.9286 9191 0.9116
0.9116 92
92 0.9124
0.9124
0.20.2 158 158 0.9271
0.9271 119 119 0.9278
0.9278 115
115 0.9142
0.9142 115
115 0.9148
0.9148
0.30.3 178 178 0.9065
0.9065 137 137 0.9184
0.9184 132
132 0.9052
0.9052 133 0.9058
0.40.4 185 185 0.8493
0.8493 149149 0.898
0.898 145
145 0.8833
0.8833 145 0.844
0.50.5 186 186 0.8041
0.8041 156156 0.8741
0.8741 153
153 0.8573
0.8573 153 0.8581
0.60.6 188 188 0.8034
0.8034 162162 0.8532
0.8532 158
158 0.8347
0.8347 159
159 0.8355
0.8355
0.70.7 179 179 0.7574
0.7574 164164 0.8298
0.8298 161
161 0.8092
0.8092 161
161 0.8101
0.8101
0.8 136 0.5434 158 0.7905 155 0.7654 155 0.7664
0.8 136 0.5434 158 0.7905 155 0.7654 155 0.7664
Average 167.5 0.8192 142.75 0.8775 138.75 0.8601 139.13 0.8559
Average 167.5 0.8192 142.75 0.8775 138.75 0.8601 139.13 0.8559

Generally,
Generally,these
thesemethods
methodsareare
considered
considered more reliable
more compared
reliable to model-fitting
compared meth-
to model-fitting
ods as they obtain consistent activation energy values for the nonisothermal
methods as they obtain consistent activation energy values for the nonisothermal data data [37,38].
They also
[37,38]. helpalso
They to show
help the complexity
to show of multireaction
the complexity as the calculation
of multireaction of the activation
as the calculation of the
energy depends on the conversion [39]. Moreover, Özsin and Pütün (2019)
activation energy depends on the conversion [39]. Moreover, Özsin and Pütün (2019) [40] [40] found
that
foundthethat
activation energyenergy
the activation values values
for the for
four different
the methods
four different depended
methods heavilyheavily
depended on the
conversion
on the conversion value. From the calculations and graphs, it became clear that thereneed
value. From the calculations and graphs, it became clear that there is a is a
in the future to conduct additional experiments with more heating rates to improve some
need in the future to conduct additional experiments with more heating rates to improve
low values of R2 in Table 4 caused by the ratio of 10 between the highest and lowest
some low values of R2 in Table 4 caused by the ratio of 10 between the highest and lowest
heating rates.
heating rates.
3.3. Model-Fitting Kinetics Calculation
3.3. Model-Fitting Kinetics Calculation
Two model-fitting (Criado and Coats–Redfern) methods were implemented to study
the kinetics of catalytic cracking of PET. For more details relating to derivation, please
h refer
i
g(α)
to [26]. The first model-fitting method (Coats–Redfern) was tested by plotting ln
T2
against 1/T. The most suitable formula of function g(α) common solid-state reaction
mechanism was selected and presented in [41,42]. Table S1 lists the kinetic parameter
Polymers 2023, 15, 70 8 of 14

(Ea kJ/mol and lnA) and the average correlation factor (R2 ). It can be seen from the Table S1
that there were big differences in the Ea between these 16 different reaction mechanisms.
In the next stage, these data results were connected to the second model-fitting (Criado
method). The Criado method was utilized to calculate the reaction model. In the Criado
method, the left-hand side of the Criado equation is called a reduced theoretical curve,
and the right-hand side represents the experimental data. To use this method, compari-
son between these two sides were carried out to help us define the exact kinetic model
describing the reaction. Figure S1 shows the curves obtained from the Criado method at
different heating rates only for the reaction mechanisms close to the experimental curve.
Figure S1 and Table 5 reveal that the experimental data matched first-order reaction (F1) or
two dimensional phase interfacial reaction (R2) for 2, 5, and 10 K/min heating rates and
(F1) or (R2) or two dimensional nucleation and growth reaction (A2) for 20 K/min. Table 5
reveals that there was a range of activation energy values between 104 to 242 kJ/mol with
an average of 157 kJ/mol depending on different reaction mechanisms. Das and Tiwari
(2019) [5] showed that PET degradation under nitrogen follows many reaction models (A2,
A3, A4, and F1) for all the heating rates.

Table 5. Activation energy of PET catalytic pyrolysis at different heating rates by Coats–Redfern method.

Heating Rate (K/min) E (kJ/mol) Mechanism

146 F1 (first-order chemical reaction)
2
104 R2 (two-dimensional phase interfacial reaction)
179 F1 (first-order chemical reaction)
5
141 R2 (two-dimensional phase interfacial reaction)
139 F1 (first-order chemical reaction)
10
123 R2 (two-dimensional phase interfacial reaction)
242 F1 (first-order chemical reaction)
20 115 A2 (two-dimensional nucleation and growth reaction)
224 R2 (two-dimensional phase interfacial reaction)

3.4. Catalytic Pyrolysis Prediction by ANN Model

ANN model was developed to monitor the catalytic pyrolysis of PET. Feedforward
with backpropagation was developed to predict the remaining weight fraction based on
983 experimental datasets with heating rate and temperature as the input variables.
The collected data is usually divided into three subsets: training set, validation set,
and finally test set [43]. Each set has its own role in the whole process.
The whole data comprising 983 sets were randomly divided into three sets as follows:
70% used for training, 15% used for validation, and 15% used for testing. More details about
the data are shown in Table 6. Osman and Aggour (2002) [44] mentioned that collecting
large sets of data can lead to a model with high accuracy compared to the original data.
To obtain the best performance of ANN structure, it is necessary to fix: (1) the number
of hidden layers, (2) transfer function for output and hidden layer/s, and (3) number of
neurons in each hidden layer. A comparison was then carried out between different ANN
structures with variable number according to the above points.

Table 6. Distribution of 983 datasets (first step: training set) and 12 dataset (second step: simulation
set) for heating rates of 2, 5, 10, 20 K/min.

Training Set No.

Heating Rate (K/min) Simulation Set No.
Training Validation Test
2 248 3
5 248 3
10 246 3
20 241 3
Total 983 12
 Polymers 2023, 15, 70 9 of 14

The value of R was used as the main criterion for judging the most efficient structure
to estimate the remaining weight fraction as the output variable for 23 runs. A total of
13 runs out of the 23 runs resulted in high value of R. These 13 runs were moved to the
second stage comparison, where 12 new simulation sets were tested again against the value
of R. Table 7 shows the 12 new sets that were collected for the four heating rates.

Table 7. ANN-predicted remaining weight values of the new (simulation step) input.

Input Data Output Data

No. Heating Rate
Temperature (K) Remaining Weight (Fraction)
(K/min)
1 2 702.1 0.30153
2 2 682.2 0.46976
3 2 618.9 0.90445
4 5 704.8 0.25160
5 5 670.4 0.62158
6 5 641.3 0.85232
7 10 968.5 0.15028
8 10 687.2 0.69914
9 10 700.4 0.55610
10 20 718.6 0.43629
11 20 697.8 0.79087
12 20 732.4 0.20794

The final and best ANN structure (2-15-15-1), as shown in Figure 3, was selected after
the two comparison steps. It had 15 neurons for each two hidden layers (tangsig–tangsig
transfer functions) with linear transfer function for the output layer. The efficiency and the
accuracy of the ANN output is heavily dependent on the number of neurons in the hidden
layers. To avoid underfitting and the overfitting, one must select the number of neurons in
such a way that the performance function will eventually obtain the optimum value [43,45].
Al-Wahaibi and Mjalli (2014) [38] concluded that the number of neurons in the hidden
layer is a very crucial parameter in predicting the accuracy of the ANN output, and using
very few numbers may result in something called underfitting. Qinghua et al. (2018) [40]
also pointed out that the number of neurons in the hidden layer may cause training
error if the number is too small or overfitting problem if the number is too high. The
Levenberg–Marquardt algorithm is commonly suited for this type of network [46]. The
ANN model was run with different algorithms, namely, scaled conjugate gradient and
Polymers 2023, 15, x FOR PEER REVIEW Bayesian regularization, to check that the test was relevant. As shown in Figure 4, all the 10 of 14
comparative results were close to the diagonal, which proved good agreement between the
ANN values (Y-axis) and real values (X-axis).

Temperature
Remaining weight fraction
Heating Rate

Figure 3. Topology
Figure ofofthe
3. Topology thebest
best selected (2-15-15-1)
selected (2-15-15-1) network.
network.
 Temperature
Remaining weight fraction
Polymers 2023, 15,Heating
70 Rate 10 of 14

Figure 3. Topology of the best selected (2-15-15-1) network.

Figure 4.
Figure 4. Regression
Regression of
of training,
training, validation,
validation, and
and test
test plots
plots for
forthe
theselected
selected(2-15-15-1)
(2-15-15-1)model.
model.

The performance
The performance of of the
the current
current (2-15-15-1)
(2-15-15-1) model
model in in predicting
predicting remaining
remaining weight
weight
fraction
fraction waswas examined
examined by by calculating
calculating the
the four statistical parameters. As shown in Table 8,
all measurementsofofdeviations
all statistical measurements deviationswerewere notably
notably lowlow
for for
the the
bestbest
ANN ANN
model.model.
This
This meant the selected ANN model could efficiently predict the output
meant the selected ANN model could efficiently predict the output variable within an variable within
an acceptable
acceptable limitlimit of error.
of error. AfterAfter training
training the ANN
the ANN with with 983 the
983 sets, sets,recommended
the recommendedarchi-
architecture (2-15-15-1)
tecture (2-15-15-1) was was simulated
simulated withwith 12 new
12 new inputinput datasets
datasets as delineated
as delineated in Table
in Table 7. In7.
In
thisthis stage,
stage, newnew inputdata
input datawere
wereentered,
entered,and
andthe
thenetwork
network produced
produced the new simulated
simulated
output.
output. Figure
Figure 55 compares
compares thethe simulated
simulated ANNANN with
with the measured
measured output
output and
and Table
Table 99 lists
lists
the R 22, RMSE, MAE, and MBE values. As can be seen, R22 was very high (>0.9999), while
the R , RMSE, MAE, and MBE values. As can be seen, R was very high (>0.9999), while
RMSE,
RMSE, MAE, MAE,and andMBEMBEwerewerevery
verylow.
low.Figure
Figure6 presents
6 presentsthethe
error histogram
error histogramgenerated
generatedfor
the data. As can be seen, the error was again only in a narrow range (
for the data. As can be seen, the error was again only in a narrow range (−0.00091 to− 0.00091 to 0.000402).
0.000402).
Table 8. Statistical parameters of the best (2-15-15-1) model.
Table 8. Statistical parameters of the best (2-15-15-1) model.Parameters
Statistical
Set 2
R RMSE MAE MBE
Training 1.000 1.820 × 10−4 9.800 × 10−5 −4.505079 × 10−7
Validation 1.000 3.680 × 10−4 1.680 × 10−4 −0.0000295
Test 1.000 2.580 × 10−4 1.390 × 10−4 8.015 × 10−6
All 1.000 2.310 × 10−4 1.140 × 10−4 −0.00000353
 All 1.000 2.310 × 10−4 1.140 × 10−4 −0.00000353
Statistical Parameters
Table Set
9. Statistical parameters
R2 of the (2-10-10-1)
RMSE model for the simulated
MAE data. MBE
Training 1.000 1.820Statistical
× 10−4 9.800 × 10−5
Parameters −4.505079 × 10−7
Polymers 2023, 15, 70 Set
Validation 1.000 3.680 × 10 −4 1.680 × 10−4 −0.0000295
R2 RMSE −4 MAE MBE 11 of 14
Test 1.000 2.580 × 10 1.390 × 10−4 8.015 × 10−6
simulated
All 1.000
1.000 3.903
2.310 ×× 10
10−4−4 2.843
1.140 ×× 10
10−4−4 6.360 × 10−5
−0.00000353

Table 9. Statistical parameters of the (2-10-10-1) model for the simulated data.

Statistical Parameters
Set
R 2 RMSE MAE MBE
simulated 1.000 3.903 × 10−4 2.843 × 10−4 6.360 × 10−5

Figure5.
Figure 5. Regression
Regression of
of simulated
simulateddata
datafor
forthe
theselected
selected(2-15-15-1)
(2-15-15-1)model.
model.

Table 9. Statistical parameters of the (2-10-10-1) model for the simulated data.

Statistical Parameters
Set
R2 RMSE MAE MBE
simulated 1.000 3.903 × 10−4 2.843 × 10−4 6.360 × 10−5
Figure 5. Regression of simulated data for the selected (2-15-15-1) model.

Figure 6. Error histogram of simulation data.

4. Conclusions
As expected, TG and DTG showed the same shapes and trends for different heating
rates, with only one main reaction. To obtain the kinetic parameters, six methods, namely,
four model-free (Friedman, FWO, KAS, and Starnik) and two model-fitting (Coats–Red-
Figure
Figure 6. Error
fern and
6. Error histogram
Criado) methods
histogram of simulation
of simulation data.
were useddata.as the main approach. A highly efficient ANN model
was developed as a second approach to predict the catalytic decomposition of PET.
4.
4. Conclusions
Conclusions
The two approaches were implemented to model the TGA kinetics data. In the first
As
As expected,
approach, the sixTG
expected, TG and
and DTG
differentDTG showed
showed
methods the
the same
were used shapes
same shapes and
and trends
trends
to approximate for
the different
forTGA dataheating
different heating
with a
rates, with only one main reaction. To obtain the kinetic parameters, six methods,
rates, with only one main reaction. To obtain the kinetic parameters, six methods, namely, namely,
four
four model-free
model-free(Friedman,
(Friedman,FWO,FWO,KAS, KAS, and Starnik)
and and
Starnik) and two model-fitting
two (Coats–Redfern
model-fitting (Coats–Red-
and Criado) methods were used as the main approach. A highly efficient
fern and Criado) methods were used as the main approach. A highly efficient ANN ANN model was
model
developed as a second approach to predict the catalytic decomposition
was developed as a second approach to predict the catalytic decomposition of PET. of PET.
The
The two
two approaches
approaches were were implemented
implemented to to model
model thethe TGA
TGA kinetics
kinetics data.
data. In
In the
the first
first
approach, the six different methods were used to approximate the TGA
approach, the six different methods were used to approximate the TGA data with a data with a straight
line. Values of the activation energies obtained by Friedman, FWO, KAS, and Starnik
methods were 167.5, 142.75, 138.75, and 139.13 kJ/mol, respectively. The average calculated
values of activation energies by Coats–Redfern and Criado methods were in the range
of 104–242 kJ/mol. For future work, more heating rates are needed to fill the gap in the
results. Natural polymers, such as cellulose, are also recommended to enhance the catalytic
cracking of polymer plastic waste.
Polymers 2023, 15, 70 12 of 14

An ANN with 2-15-15-1 structure and tangsig–tangsig hidden layers was determined
as the most promising network as it could predict the remaining weight fraction very
precisely with high regression coefficient value. The recommended network was reused
with new input and still showed results that were close to the experimental values with
high R and very low RMSE, MAE, and MBE.

Supplementary Materials: The following supporting information can be downloaded at: https://www.
mdpi.com/article/10.3390/polym15010070/s1, Figure S1: Criado model master plots of catalytic
cracking of PET at different heating rate: (a) 2 K/min, (b) 5 K/min, (c) 10 K/min, and (d) 20 K/min;
Table S1: Kinetic parameters obtained by Coats-Redfern model.
Author Contributions: Methodology, I.D.; Software, I.D.; Formal analysis, I.D.; Investigation, I.D.;
Resources, I.D.; Data curation, I.D.; Writing—original draft, I.D. and Z.A.; Writing—review & editing,
I.D., Z.A.; Visualization, I.D.; Supervision, I.D. Project administration, I.D. All authors have read and
agreed to the published version of the manuscript.
Funding: The author are grateful to the Deanship of Scientific Research at King Faisal University
(Saudi Arabia) for supporting this research as part of the Research Grants Program (Old No.: AN00029,
New No.: GRANT964).
Institutional Review Board Statement: Not applicable.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Dubdub, I.; Al-Yaari, M. Pyrolysis of Low Density Polyethylene: Kinetic Study using TGA Data and ANN Prediction. Polymers
2020, 12, 891. [CrossRef] [PubMed]
2. Martín-Gullón, I.; Esperanza, M.; Font, R. Kinetic model for the pyrolysis and combustion of poly-(ethylene terephthalate) (PET).
J. Anal. Appl. Pyrolysis 2001, 58–59, 635–650. [CrossRef]
3. Diaz-Silvarrey, L.S.; McMahon, A.; Phan, A.N. Benzoic acid recovery via waste poly(ethylene terephthalate) (PET) catalytic
pyrolysis using sulphated zirconia catalyst. J. Anal. Appl. Pyrolysis 2018, 134, 621–631. [CrossRef]
4. Al-Salem, S.M.; Antelava, A.; Constantinou, A.; Manos, G.; Dutta, A. A review on thermal and catalytic pyrolysis of plastic solid
waste (PSW). J. Environ. Manag. 2017, 197, 177–198. [CrossRef] [PubMed]
5. Das, P.; Tiwari, P. Thermal degradation study of waste polyethylene terephthalate (PET) under inert and oxidative environments.
Thermochim. Acta 2019, 679, 178340. [CrossRef]
6. Yang, J.; Miranda, R.; Roy, C. Using the DTG curve fitting method to determine the apparent kinetic parameters of thermal
decomposition of polymers. Polym. Degrad. Stab. 2001, 73, 455–461. [CrossRef]
7. Girija, B.G.; Sailaja, R.R.N.; Madras, G. Thermal degradation and mechanical properties of PET blends. Polym. Degrad. Stab. 2005,
90, 147–153. [CrossRef]
8. Saha, B.; Ghoshal, A.K. Thermal degradation kinetics of poly (ethylene terephthalate) from waste soft drinks bottles. Chem. Eng. J.
2005, 111, 39–43. [CrossRef]
9. Aguado, J.; Serrano, D.P.; San Miguel, G.; Castro, M.C.; Madrid, S. Feedstock recycling of polyethylene in a two-step thermo-
catalytic reaction system. J. Anal. Appl. Pyrolysis 2007, 79, 415–423. [CrossRef]
10. Moltó, J.; Font, R.; Conesa, J.A. Kinetic model of the decomposition of a PET fibre cloth in an inert and air environment. J. Anal.
Appl. Pyrolysis 2007, 79, 289–296. [CrossRef]
11. Halász, J.; Vőneki, K.; Krajkó, J. Degradation OF Pure and Waste PVC and PET in the Presence of Modified Porous Catalysts.
J. Eng. Ann. Fac. Eng. Hunedoara 2009, 174, 1021–1026.
12. Çepelioğullar, Ö.; Pütün, A. Thermal and kinetic behaviors of biomass and plastic wastes in co-pyrolysis. Energy Convers. Manag.
2013, 75, 263–270. [CrossRef]
13. Lee, H.W.; Kim, Y.; Lee, B.; Kim, S.; Jae, J.; Jung, S.; Kim, T.; Park, Y. Catalytic copyrolysis of torrefied cork oak and high density
polyethylene over a mesoporous HY catalyst. Catal. Today 2018, 307, 301–307. [CrossRef]
14. Diaz Silvarrey, L.S.; Phan, A.N. Kinetic study of municipal plastic waste. Int. J. Hydrogen Energy 2016, 41, 16352–16364. [CrossRef]
15. Xiang, Z.; Liang, J.; Morgan, H.M.J.; Liu, Y.; Mao, H.; Bu, Q. Thermal behavior and kinetic study for co-pyrolysis of lignocellulosic
biomass with polyethylene over Cobalt modified ZSM-5 catalyst by thermogravimetric analysis. Bioresour. Technol. 2018, 247,
804–811. [CrossRef]
16. Osman, A.I.; Farrell, C.; Al-Muhtaseb, A.H.; Al-Fatesh, A.S.; Harrison, J.; Rooney, D.W. Pyrolysis kinetic modelling of abundant
plastic waste (PET) and in-situ emission monitoring. Environ. Sci. Eur. 2020, 32, 112. [CrossRef]
17. Ali, G.; Nisar, J.; MIqbal, M.; Shah, A.; Abbas, M.; Shah, M.R.; Rashid, U.; Bhatti, I.A.; Khan, R.A.; Shah, F. Thermo-catalytic
decomposition of polystyrene waste: Comparative analysis using different kinetic models. Waste Manag. Res. 2020, 38, 202–212.
[CrossRef]
Polymers 2023, 15, 70 13 of 14

18. Lai, J.; Meng, Y.; Yan, Y.; Lester, E.; Wu, T.; Pang, C.H. Catalytic pyrolysis of linear low-density polyethylene using recycled coal
ash: Kinetic study and environmental evaluation. Korean J. Chem. Eng. 2021, 38, 2235–2246. [CrossRef]
19. Yap, T.L.; Loy, A.C.M.; Chin, B.L.F.; Lim, J.Y.; Alhamzi, H.; Chai, Y.H.; Yiin, C.L.; Cheah, K.W.; Wee, M.X.J.; Lam, M.K.; et al. Synergistic
effects of catalytic co-pyrolysis Chlorella vulgaris and polyethylene mixtures using artificial neuron network: Thermodynamic
and empirical kinetic analyses. J. Environ. Chem. Eng. 2022, 10, 107391. [CrossRef]
20. Kokuryo, S.; Miyake, K.; Uchida, Y.; Mizusawa, A.; Kubo, T.; Nishiyama, N. Defect engineering to boost catalytic activity of Beta
zeolite on low density polyethylene cracking. Mater. Today Sustain. 2022, 17, 100098. [CrossRef]
21. Dourari, M.; Tarchoun, A.F.; Trache, D.; Abdelaziz, A.; Bekhouche, S.; Harrat, A.; Boukeciat, H.; Matmat, N. Unraveling the Effect
of MgAl/CuO Nanothermite on the Characteristics and Thermo-Catalytic Decomposition of Nanoenergetic Formulation Based
on Nanostructured Nitrocellulose and Hydrazinium Nitro-Triazolone. Catalysts 2022, 12, 1573. [CrossRef]
22. Tarchoun, A.F.; Trache, D.; Abdelaziz, A.; Harrat, A.; Boukecha, W.O.; Hamouche, M.A.; Boukeciat, H.; Dourari, M. Elaboration,
Characterization and Thermal Decomposition Kinetics of New Nanoenergetic Composite Based on Hydrazine 3-Nitro-1,2,4-
triazol-5-one and Nanostructured Cellulose Nitrate. Molecules 2022, 27, 6945. [CrossRef] [PubMed]
23. Dubdub, I.; Al-Yaari, M. Thermal Behavior of Mixed Plastics at Different Heating Rates: I. Pyrolysis Kinetics. Polymers 2021,
13, 3413. [CrossRef] [PubMed]
24. Chan, J.H.; Balke, S.T. The thermal degradation kinetics of polypropylene: Part III. thermogravimetric analyses. Polym. Degrad.
Stabil. 1997, 57, 135–149. [CrossRef]
25. Xu, F.; Wang, B.; Yang, D.; Hao, J.; Qiao, Y.; Tian, Y. Thermal degradation of typical plastics under high heating rate conditions by
TG-FTIR: Pyrolysis behaviors and kinetic analysis. Energy Convers. Manag. 2018, 171, 1106–1115. [CrossRef]
26. Dubdub, I.; Al-Yaari, M. Pyrolysis of Mixed Waste Polymers: I: Kinetics Study. Materials 2020, 13, 4912. [CrossRef]
27. Muravyev, N.V.; Luciano, G.; Ornaghi, H.L.J.; Svoboda, R.; Vyazovkin, S. Artificial Neural Networks for Pyrolysis, Thermal
Analysis, and Thermokinetic Studies: The Status Quo. Molecules 2021, 26, 3727. [CrossRef]
28. Quantrille, T.E.; Liu, Y.A. Artificial Intelligence in Chemical Engineering; Elsevier Science: Amsterdam, The Netherlands, 1992.
29. Al-Yaari, M.; Dubdub, I. Application of Artificial Neural Networks to Predict the Catalytic Pyrolysis of HDPE Using Non-
Isothermal TGA Data. Polymers 2020, 12, 1813. [CrossRef]
30. Dubdub, I.; Al-Yaari, M. Pyrolysis of High Density Polyethylene: II. Artificial Neural Networks Modeling. In Proceedings of the
9th Jordan International Chemical Engineering Conference (JICHEC9), Amman, Jordan, 12–14 October 2021.
31. Dubdub, I.; Al-Yaari, M. Pyrolysis of Mixed Plastic Waste: II. Artificial Neural Networks Prediction and Sensitivity Analysis.
Appl. Sci. 2021, 11, 8456. [CrossRef]
32. Bar, N.; Bandyopadhyay, T.K.; Biswas, M.N.; Das, S.K. Prediction of pressure drop using artificial neural network for non-
Newtonian liquid flow through piping components. J. Pet. Sci. Eng. 2010, 71, 187. [CrossRef]
33. Osman, A.I.; Fawzy, S.; Farrell, C.; Al-Muhtaseb, A.H.; Harrison, J.; Al-Mawali, S.; Rooney, D.W. Comprehensive thermokinetic
modelling and predictions of cellulose decomposition in isothermal, non-isothermal, and stepwise heating modes. J. Anal. Appl.
Pyrolysis 2022, 161, 105427. [CrossRef]
34. Farrell, C.; Osman, A.I.; Harrison, J.; Vennard, A.; Murphy, A.; Doherty, R.; Russell, M.; Kumaravel, V.; Al-Muhtaseb, A.H.;
Zhang, X.; et al. Pyrolysis Kinetic Modeling of a Poly(ethylene-co-vinyl acetate) Encapsulant Found in Waste Photovoltaic
Modules. Ind. Eng. Chem. Res. 2021, 60, 13492–13504. [CrossRef]
35. Muravyev, N.V.; Vyazovkin, S. The Status of Pyrolysis Kinetics Studies by Thermal Analysis: Quality Is Not as Good as It Should
and Can Readily Be. Thermo 2022, 2, 435–452. [CrossRef]
36. Koga, N.; Vyazovkin, S.; Burnham, A.K.; Favergeon, L.; Muravyev, N.V.; Pérez-Maqueda, L.A.; Saggese, C.; Sánchez-Jiménez, P.E.
ICTAC Kinetics Committee recommendations for analysis of thermal decomposition kinetics. Thermochim. Acta 2023, 719, 179384.
[CrossRef]
37. Osman, A.I.; Young, T.J.; Farrell, C.; Harrison, J.; Al-Muhtaseb, A.H.; Rooney, D.W. Physicochemical Characterization and Kinetic
Modeling Concerning Combustion of Waste Berry Pomace. ACS Sustain. Chem. Eng. 2020, 8, 17573–17586. [CrossRef]
38. Osman, A.I.; Abdelkader, A.; Johnston, C.R.; Morgan, K.; Rooney, D.W. Thermal Investigation and Kinetic Modeling of
Lignocellulosic Biomass Combustion for Energy Production and Other Applications. Ind. Eng. Chem. Res. 2017, 56, 12119–12130.
[CrossRef]
39. Vyazovkin, S.; Wight, C.A. Model-free and model-fitting approaches to kinetic analysis of isothermal and nonisothermal data.
Thermochim. Acta 1999, 340–341, 53–68. [CrossRef]
40. Özsin, G.; Pütün, A.E. TGA/MS/FT-IR study for kinetic evaluation and evolved gas analysis of a biomass/PVC co-pyrolysis
process. Energy Convers. Manag. 2019, 182, 143–153. [CrossRef]
41. Al-Yaari, M.; Dubdub, I. Pyrolysis of High Density Polyethylene: I. Kinetic Study. In Proceedings of the 9th Jordan International
Chemical Engineering Conference (JICHEC9), Amman, Jordan, 12–14 October 2021.
42. Al-Yaari, M.; Dubdub, I. Pyrolytic Behavior of Polyvinyl Chloride: Kinetics, Mechanisms, Thermodynamics, and Artificial Neural
Network Application. Polymers 2021, 13, 4359. [CrossRef]
43. Al-Wahaibi, T.; Mjalli, F.S. Prediction of Horizontal Oil-Water Flow Pressure Gradient Using Artificial Intelligence Techniques.
Chem. Eng. Commun. 2014, 201, 209. [CrossRef]
44. Osman, E.A.; Aggour, M.A. Artificial Neural Network Model for Accurate Prediction of Pressure Drop in Horizontal and
Near-Horizontal-Multiphase Flow. Pet. Sci. Technol. 2002, 20, 1–15. [CrossRef]
Polymers 2023, 15, 70 14 of 14

45. Qinghua, W.; Honglan, Z.; Wei, L.; Junzheng, Y.; Xiaohong, W.; Yan, W. Experimental Study of Horizontal Gas-liquid Two-phase
Flow in Two Medium-diameter Pipes and Prediction of Pressure Drop through BP Neural Networks. Int. J. Fluid Mach. Syst. 2018,
11, 255–264. [CrossRef]
46. Beale, M.; Hagan, T.; Demuth, H.B. Neural Network Toolbox TM User’s Guide; MathWorks: Natick, MA, USA, 2018.

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.