Study on Thermal Degradation Processes of Polyethylene Terephthalate Microplastics Using the Kinetics and Artificial Neural Networks Models

Abstract

Because of its slow rate of disintegration, plastic debris has steadily risen over time and contributed to a host of environmental issues. Recycling the world’s increasing debris has taken on critical importance. Pyrolysis is one of the most practical techniques for recycling plastic because of its intrinsic qualities and environmental friendliness. For scale-up and reactor design, an understanding of the degradation process is essential. Using one model-free kinetic approach (Friedman) and two model-fitting kinetic methods (Arrhenius and Coats-Redfern), the thermal degradation of Polyethylene Terephthalate (PET) microplastics at heating rates of 10, 20, and 30 °C/min was examined in this work. Additionally, a powerful artificial neural network (ANN) model was created to forecast the heat deterioration of PET MPs. At various heating rates, the TG and DTG thermograms from the PET MPs degradation revealed the same patterns and trends. This showed that the heating rates do not impact the decomposition processes. The Friedman model showed activation energy values ranging from 3.31 to 8.79 kJ/mol. The average activation energy value was 1278.88 kJ/mol from the Arrhenius model, while, from the Coats-Redfern model, the average was 1.05 × 10⁴ kJ/mol. The thermodynamics of the degradation process of the PET MPs by thermal treatment were all non-spontaneous and endergonic, and energy was absorbed for the degradation. It was discovered that an ANN, with a two-layer hidden architecture, was the most effective network for predicting the output variable (mass loss%) with a regression coefficient value of (0.951–1.0).

1. Introduction

Plastic waste, or garbage generation and disposal, have grown to be an irksome global problem, especially in many industrialized nations. The waste generated was at 220 Mt per annum in 2020, with its end-of-life destination deemed to be mainly in the sediment, biota, and aquatic ecosystem, such as oceans and rivers, as macro, micro (<5 mm), and nanoplastics (<1.2 μm) [1,2]. Municipal solid waste (MSW) is the primary source of plastic garbage. Most plastic waste is either burned or dumped in landfills across the world [3].

The destruction of waste by incineration is equally expensive and causes problems with high emissions and increasing environmental concerns, while the disposal of waste in landfills is still viewed as an unattractive and expensive operation [3]. In addition, the direct burning of plastic waste produces harmful pollutants in addition to phosgene, dioxins, and carbon monoxide, all of which have been associated with human malignancies and endocrine disruption [4]. However, it is conceivable to increase the value of this waste, while encouraging the circular economy by using procedures such as pyrolysis, solvent dissolution, gasification, and other valorization techniques [5].

In the recycling industry, research is currently more heavily weighted toward tertiary recycling by utilizing cutting-edge technologies, such as pyrolysis, gasification, and catalytic cracking [6]. Pyrolysis is the word used to describe the breakdown of organic polymers when they are exposed to high temperatures without oxygen. For low-volume products (gases, liquids, and char) that may be utilized as fuel, added to petroleum refinery feedstocks, or used as chemical feedstocks, pyrolysis offers certain benefits over other waste-disposal methods.

PET polymers are widely employed across a variety of industrial sectors and make up 40% of the worldwide plastic market, together with polyethylene [7]. In addition, PET is the third-most popular polymer in Europe after PP and LDPE, and, by 2011, PET demand and consumption had reached 60 Mt globally, with annual growth rates of 4.5% [8]. PET often degrades in MPs while in the environment, increasing environmental concerns. For example, they can be ingested by aquatic and terrestrial organisms, and can potentially accumulate in their tissues, leading to injury or death. MP can also act as vectors for toxic pollutants and pathogens, which can then be passed on to organisms that consume them [3]. Additionally, microplastics can also be inhaled by humans, which can lead to respiratory problems and other health issues. Studies have also found microplastics in human organs, such as the liver and brain, which may cause endocrine disruption, genetic damage, and other malignancies [3].

The pyrolysis and kinetic modelling of PET waste is now being explored in multiple studies [9,10,11,12]. Employing kinetic modelling is very important to better characterize the reaction process during the thermal cracking of plastic polymers, since the operating conditions can change both the product composition and the reaction route [2]. Ganeshan et al. [10] employed the Coats-Redfern technique to evaluate PET thermal degradation and reported activation energy (Ea) values in the range of 133–251 kJ/mol. They obtained a coefficient of determination (R² of 0.8) and concluded that the Coats-Redfern approach is not always appropriate for estimating kinetic parameters. Using the iso-conversional approach, Das and Tiwari [12] examined the kinetic parameters for PET pyrolysis at high heating rates of 5, 10, 20, 40, and 50 °C/min. They provided Ea values between 196 and 217 kJ/mol. For the co-pyrolysis of PET with biomass seeds, Mishra et al. [11] utilized the following models: the Coats-Redfern technique, the Kissinger-Akahira-Sunose (KAS) method, the Flynn-Wall and Ozawa (FWO) method, the Friedman method, and the Starink method. The PET pyrolysis KAS technique revealed that the average Ea was 230.7 kJ/mol. Under identical circumstances, the Ea averages for the FWO, Starink, and Friedman techniques were 230.5, 231.0 kJ/mol, and 225.6 kJ/mol, respectively.

There is a paucity of studies performed on the use of microplastics (MPs) made from polyethylene terephthalate (PET) as the material for thermal degradation. Cho [13] used two analytical methods, TG-FTIR and TED-GC-MS, combined with thermogravimetric analysis, to evaluate the thermal-degradation process of individual and mixed samples of polypropylene (PP), polyethylene terephthalate (PET), and polyvinyl chloride (PVC). However, the kinetics and optimization of the degradation process was not studied. For proper kinetic evaluation of the degradation process, the International Confederation for Thermal Analysis and Calorimetry (ICTAC) recommended the use of thermal degradation data from multiple heating rates, as data measured at a single heating rate can give rise to striking examples of failure when estimating the thermal stability at ambient temperatures [14,15].

To optimize the rate of degradation of PET MPs at different heating rates, it is important to have a thorough understanding of the processes and kinetics involved [14]. Thermogravimetric analysis (TGA) is a common technique used to study the thermal decomposition of solids, and has recently been used to study the pyrolysis of plastic waste as a way to convert it into valuable products or chemicals. In this particular study, the researchers looked at thermal degradation at different heating rates and found that the kinetic model they used could explain the data with a single set of parameters, regardless of the heating rate. Since it shows potent performance to represent linear and non-linear connection and, therefore, saves time, the majority of academics have recently been striving to construct an artificial neural network (ANN) model for the prediction of various data [16,17,18]. As a result, ANN is utilized as a different strategy to assist in the prediction of TGA data [18]. References to the use of artificial neural networks (ANN) for modelling the kinetics of various systems, such as fermentation and catalytic hydrogenation, oily sludge and high-density polyethylene can be found in the literature [19]. Recent studies have also applied ANN in the study of polymer degradation [6,20]. In these studies, results showed good agreement between the ANN predicted and experimental data (R > 0.9999) from HDPE degradation. In terms of PET MPs, there is no prior study applying ANN for its degradation at different heating rates. This study fills this knowledge gap.

This investigation uses TGA to learn more about the kinetics of heat breakdown of PET MPs. One isoconversional approach and two non-isoconversional models have been used to determine the activation energy of the thermally degraded PET MPs as a function of conversion at three different heating rates. A very effective constructed ANN model based on multilayer perceptron has also made the first prediction of the pyrolytic behavior of PET MPs.

2. Materials and Methods

2.1. Plastic Material Preparation

In this experiment, PET plastic water bottles were sourced from the environment (plastic collection bin at Saitama University International House) and used to fabricate PET MPs. The PET plastic was cut with scissors and was then crushed using a high-speed blender. After crushing/blending, it was then sieved through a 1000 μm sieve, washed twice by soaking in methanol, and then dried at 45 °C in an oven. The prepared PET MPs were then kept in a glass bottle (Figure 1). The flowchart of the study design is presented in Figure 2.

2.2. Degradation Study

Using the TG-DTA device (DTG-60 Shimadzu Corporation) presented in Figure 3, thermogravimetric analysis (TGA) techniques were conducted and used to evaluate the degradation process and thermal stability [21]. The furnace’s Pt cell was covered with the produced MPs (5 mg), and the carrier gas (Ar) was inserted to fill the device for 15 min. The PET MPs were warmed from ambient temperature to 700 °C at a different heating rate of 10 to 30 °C/min, with an Ar stream flow rate of 100 mL/min. On the weight loss and related weight loss bends, plotting and analysis were completed. The multiple heating rates were considered following recommendations of the ICTAC [14].

2.3. Kinetic Theory

Thermogravimetric analysis is used to study the thermal decomposition of solid materials, and the variations in conversion rate with time and temperature under various thermal conditions can provide information about the degradation process, such as the decomposition temperature, the peak temperature, and the average reaction rate [22]. The mathematical formula for the thermal deterioration of materials in non-isothermal conditions is presented in Equation (1). This equation is generally used for describing the single-step thermal decomposition processes [23],

$$\mathrm{r}=\frac{\partial \mathsf{\alpha }}{\partial \mathrm{T}}={\mathrm{Ae}}^{\left(-\frac{E}{RT}\right)} {\left(1-\mathsf{\alpha }\right)}^{\mathrm{n}}$$

(1)

where $\mathsf{\alpha }$ is a conversion (sometimes also termed as degree of conversion, fractional reaction, transformation degree, and degree of reaction [14]), T is time, A is the pre-exponential factor, E is the activation energy, R is the universal gas constant, T is temperature, and n is the reaction order. Conversion can be calculated as follows:

$$\mathsf{\alpha }=\frac{m_o-m}{m_o-m_f}$$

(2)

where m_o is the sample weight at time = 0; m is the sample weight at any specific time and; m_f is the sample weight at the end of the TGA experiment.

Some models generated from Equation (1) may be used to get kinetic triple parameters from TGA data. The published models either make use of many TGA at various heating rates (known as isoconversional or model-free techniques), or only one TGA data (called non-isoconversional or model-fitting methods). The Friedman model, as an isoconversional method, and the Arrhenius and Coats-Redfern (CR) models, as non-isoconversional models, of first-order processes, were utilized in this work to obtain values for the kinetic parameters.

2.3.1. Isoconversional Methods

Friedman Model

The Friedman model equation [24] is presented in Equation (3):

$$In\left(\mathsf{\beta }\frac{\partial \mathsf{\alpha }}{\partial \mathrm{T}}\right)=In\left(A\right)+In\left(1-\mathsf{\alpha }\right)-\frac{E}{RT}$$

(3)

Plotting $In\left(\mathsf{\beta }\frac{\partial \mathsf{\alpha }}{\partial \mathrm{T}}\right)$ vs. [1/T] of Equation (3) gives E and A from the slope and intercept, respectively.

2.3.2. Non-Isoconversional Methods

Arrhenius Model

The Arrhenius equation is a formula for the temperature dependence of reaction rates, given in Equation (4):

$$In\left(\frac{\frac{\partial m}{dt}}{m}\right)=In\left(A\right)-\frac{E}{RT}$$

(4)

Plotting $In\left(\frac{\frac{\partial m}{dt}}{m}\right)$ vs. [1/𝑇] of Equation (4) gives 𝐸 and 𝐴 from the slope and intercept, respectively.

Coats-Redfern Model

In this study, the Coats-Redfern method was used to model the thermogravimetric process. This is an integral model-fitting method that involves fitting various reaction models to the extent of the reaction-temperature curves, and simultaneously determining the activation energy and pre-exponential factor. This model equation is presented in Equation (5):

$$In\left[\frac{-ln\left(1-\mathsf{\alpha }\right)}{T^2}\right]=In\left[\frac{AR}{\beta E} \left(1-\frac{2RT}{E }\right)\right]-\frac{E}{RT}$$

(5)

where 𝑇 is the temperature (K). Plotting $In\left[\frac{-In\left(1-\mathsf{\alpha }\right)}{T^2}\right]$ vs. [1/𝑇] of Equation (5) gives 𝐸 and 𝐴 from the slope and intercept, respectively.

2.4. Thermodynamic Analysis of PET MPs’ Thermal Breakdown Process

Any chemical process’ overall performance may be better understood by thermodynamic analysis, which also makes it possible to finally pinpoint the causes of losses brought on by irreversibility. By using the determined kinetic parameters (from the Coats-Redfern model) to calculate the values of thermodynamic parameters, such as the changes in enthalpy ($\Delta$H, kJ/mol), entropy ($\Delta$S, kJ/mol.K), and Gibbs free energy ($\Delta$G, kJ/mol), we attempted to perform the thermodynamic analysis of the thermal degradation process of PET MPs at various heating rates in our study. Additionally, it was essential to monitor the energy and assess the course and viability of the thermal degradation process using thermodynamic parameters [25]. The following Equations (6)–(8) can be used to get the values of $\Delta$H, $\Delta$G, and $\Delta$S:

$$\Delta H=E-RT$$

(6)

$$\Delta G=E+RTIn\left(\frac{k_{b}T}{h A}\right)$$

(7)

$$\Delta S=\frac{\Delta H-\Delta G}{T}$$

(8)

where 𝑇 represents the temperature (K), 𝑘𝐵 is the Boltzmann constant (1.381 × 10⁻²³ 𝐽/𝐾), h is the Planck constant (6.626 × 10⁻³⁴ 𝐽. 𝑠), and 𝐴 is the pre-exponential factor, respectively.

2.5. Machine-Learning Approach to Model the Thermal Degradation of PET MPs

The machine-learning approach used in this study is based on Multilayer Perceptrons (MLP) Artificial Neutral Networks (ANNs). ANNs are nonlinear models composed of linked “neurons,” or units, that can recognize patterns in a number of ways, including classification and prediction [26]. ANNs pick up new information by seeing patterns in the data, which they store in weights, which are groups of connection strengths matching to regression coefficients. The MLP ANN used in this study learns by using a technique called backpropagation, in which weights are changed after analyzing either the entire data set or each datum. Similar to regression coefficients, ANN weights assess the associations between independent and dependent variables. Regression weights assess the global effects of independent factors on dependent variables across all data, whereas ANN weights measure local effects [21].

Equation (9), in which f is the activation function, N is the number of inputs per neuron, and k is the layer (hidden, output), may be used to represent the ANN system in its simplest form [19]:

$$Y_j^{k+1}=f({\displaystyle \sum }_{i=1}^N{X}_i^k{w}_{ij}^k+b_i^k$$

(9)

The MLP ANN makes an effort to compute the weights connecting the input and output layers in order to represent the relative weights of the inputs into, and outputs from, the ANN hidden layer. The hidden layer transforms the input data into a group of values that are subsequently utilized by the output layer. The hidden layer’s mathematical operations are what give the ANN its nonlinear behavior, making it an essential component of the system [27]. The output layer holds the dependent variable, which, in this study, is binary classifications. The ANN will investigate the particular relationship between the independent variables of the input layer and the dependent variable of the output layer during training. The nodes of the hidden layer include mathematical functions defining the relationships. Testing data will be utilized to validate the linkages (mathematical functions) once they have been constructed. The hyperbolic tangent and identity function, which is frequently utilized in supervised ANN modelling, served as the foundation for the hidden layer activation functions and output function in this study [21].

Table 1. The range of data employed in MLP ANN and % of input data used for training and testing.

Heating Rates (°C/min)	Parameter	Type	Minimum Value	Maximum Value	Training (%)	Testing (%)
10	Mass loss (ML, %)	Output	72.70	100	70.1	29.9
	Time (t, mins)	Input	0	7112
	Heating rate (HR, K/min)	Input	0	297.65
	Temperature (T, °C)	Input	24.65	900.49
20	Mass loss (%)	Output	77.34	100	69.8	30.2
	Time (mins)	Input	0	4433
	Heating rate (K/min)	Input	0	306.77
	Temperature (°C)	Input	33.74	901.3
30	Mass loss (%)	Output	71.62	100	70.4	29.6
	Time (mins)	Input	0	3564
	Heating rate (K/min)	Input	0	302.55
	Temperature (°C)	Input	29.55	896.37

lists the input and output variables that were used in this study for development. The independent variables (time, heating rate, and temperature) are present in the input layer, whereas the dependent variable (mass loss) is present in the output layer. The network was trained using the batch training method, which excels in data sets with few input variables [20].

To examine and assess how effectively an ANN model predicts the output (mass loss), several statistical models, such as the average correlation factor (R²), sum of square error (SSE), and relative error (RE) were utilized, presented in Equations (10)–(12) [28,29]:

$$R^2=1-\frac{\sum {\left(M{L}_{est}-M{L}_{exp}\right)}^2}{\sum {\left(M{L}_{est}-{\overline{ML}}_{exp}\right)}^2}$$

(10)

$$SSE=\sum {\left(M{L}_{est}-M{L}_{exp}\right)}^2$$

(11)

$$RE=\left|\frac{M{L}_{exp}-M{L}_{est}}{M{L}_{est}}\right|\times 100$$

(12)

where:

• (ML)_est: is the estimated value of the PET MPs mass loss (%) by ANN model;
• (ML)_exp, is the experimental value of the PET MPs mass loss (%); and
• ${\overline{ML}}_{exp}$: is the average value of PET MPs mass loss (%).

The goal should be to achieve the lowest error, with (RMSE, MAE, SSE), and the greatest, with (R²) correlations, in order to obtain the optimal ANN model [28].

3. Results and Discussion

3.1. TGA and DTG Analysis

The result for TGA and DTG of the PET MPs at different heating rates of 10, 20, and 30 °C/min is presented in Figure 2. The TGA curves of the PET MPs at different heating rates generally revealed four stages of thermal degradation (Figure 4). These four stages likely correspond to the different fragments of compounds in the PET polymer structure as it degrades thermally. This result is similar to previous studies [21]. In the first stages, the loss was probably CO₂ and CO; then, C₂H₆O₂ and C₂H₄O in the second stage, C₄O₂ in the third stage, and RCO-OR in the final stage [21]. However, as the heating rate rose from 10 to 30 °C/min, greater temperatures were recorded for the thermal decomposition onset, peak, and final temperatures that were derived from the TG and DTG curves, as shown in

Table 2. The onset, end-set and peak values of the PET MPs’ thermal degradation at different heating rates.

Heating Rate (°C/min)	Onset (°C)	End- Set (°C)	Peak Temperature (°C)
10	312	583	434
20	330	609	440
30	388	605	443

, which is in agreement with the result of [2]. This is due to the fact that a higher heating rate means more energy is being added to the sample in a shorter amount of time, causing the thermal degradation process to occur at a faster rate and at higher temperatures. Additionally, increasing the heating rate can also change the kinetics of the thermal degradation process, which can affect the onset, peak, and final temperatures as well. The onset temperature is described as the point at which a reaction’s heat output can no longer be entirely withdrawn from the reaction vessel without causing a visible temperature increase. The onset temperature was 312, 330, and 388 °C for heating rates of 10, 20, and 30 °C/min, respectively. At these temperatures, the residual mass of the PET MPs was >90%. However, during the final, or end-set, temperature, the residual mass of PET MPs was 72, 73, and 78% for heating rates of 10, 20, and 30 °C/min, respectively. The final and highest peak temperature occurred at a higher temperature for 20 and 30 °C/min (Table 2). This result is comparable to the onset and end-set temperatures reported by [2] for PET plastic wastes.

3.2. Kinetic Modeling

The Friedman, Arrhenius, and Coats-Redfern plots for the thermal degradation of PET MPs is presented in Figure 5 and Figure 6, while the activation energy values, calculated by the Friedman, Arrhenius, and Coats-Redfern models, at different conversions, is presented in

Table 3. Computed kinetic parameters from the Friedman model.

Conversion (α)	E (kJ/mol)	A (min−1)	R2
0.1	3.31	0.64	0.8978
0.2	5.04	0.72	0.9869
0.3	6.05	0.83	0.9567
0.4	6.77	0.97	0.9306
0.5	7.32	1.16	0.9108
0.6	7.78	1.45	0.8954
0.7	8.16	1.93	0.8832
0.8	8.50	2.89	0.8732
0.9	8.79	5.79	0.8648
Average	6.86	1.82	0.9110

and

Table 4. Arrhenius and Coats-Redfern kinetic parameters of PET MPs degradation at different heating rates.

Heating Rate (°C/mins)	Arrhenius Model			Coats-Redfern Model
	E (kJ/mol)	A (min−1)	R2	E (kJ/mol)	A (min−1)	R2
10	888.75	1.52 × 107	0.9845	1.02 × 104	1.45 × 1010	0.9658
20	1057.96	2.15 × 106	0.9968	1.08 × 104	2.91 × 1010	0.9706
30	1889.94	1.43 × 108	0.9010	1.05 × 104	1.36 × 1010	0.9691
Average	1278.88	5.35 × 107	0.9608	1.05 × 104	1.91 × 1010	0.9685

. In the tables, the results for the activation energy (E) and pre-exponential factor (A) is presented. The bare minimum additional energy needed by a reactive molecule to transform into a product is known as activation energy. It is also known as the minimal energy required to energize or activate molecules or atoms in order for them to engage in a chemical reaction or transformation. The result from the Friedman model showed very low E values at different conversions, which ranged from 3.31. kJ/mol to 8.79 kJ/mol with an average of 6.86 kJ/mol, although the coefficient of determination ranged from 0.8648 to 0.9567 (Table 3). The obtained results in this study show lower than previously reported E (activation energy) values for pyrolyzed PET waste using the same model. Mishra et al. [11] studied the co-pyrolysis of PET with biomass seeds using the Friedman model and reported a variation in E values from 208.6 to 236.0 kJ/mol for conversion of 0.1–0.8. They performed the kinetic modelling based on three heating rates of 10, 30, and 50 °C/min. The co-pyrolysis process influenced the E values. However, when Das and Tiwari [12] used the isoconversional approach to test the kinetic parameters for PET pyrolysis at high heating rates of 5, 10, 20, 40, and 50 °C/min, they found that the E values ranged from 196–217 kJ/mol.

However, the average E and A values obtained by two non-isoconversional models were close. Generally, from the Arrhenius model, the E values increase as the heating rate increases, which agrees with the study of Zhang et al. [30] and Al-Yaari and Dubdub [6] for HDPE pyrolysis. This suggests that the increase in the reaction rate causes the degradation of the PET MPs to occur at high activation energy. The E values ranged from 888.75 to 1889.94 kJ/mol from the Arrhenius model, while, from the Coats-Redfern model, it ranged from 1.02 × 10⁴ to 1.05 × 10⁴ kJ/mol, with an average of 1.02 × 10⁵ kJ/mol. Based on the average R², the Coats-Redfern model showed the best fit for the degradation process as it described the entire process efficiently for all temperature ranges. Compared with other studies, the results obtained were higher than those reported for HDPE and LDPE pyrolysis [6,23], and for PET wastes [10].

3.3. Thermodynamic Analysis

In addition to determining the kinetic parameters, thermodynamic data must also be calculated in order to establish the process’s viability and carry out energy calculations. Since the main goal of the thermal degradation process is to produce “energy”, understanding how thermodynamic parameters vary on the circumstances of the process is crucial. Understanding the fluctuation of enthalpy, entropy, and free energy at different heating rates requires taking thermodynamic consideration [21,31].

The thermodynamic parameters were computed based on the kinetic parameters obtained from the Arrhenius and Coats-Redfern models. The results are shown in

Table 5. Thermodynamic parameters for PET MPs degradation under thermal treatment at different heating rates.

Heating Rate (°C/mins)	ΔH (kJ/mol)		ΔG (kJ/mol)		ΔS (kJ/mol.K)
	Arrhenius	Coats-Redfern	Arrhenius	Coats-Redfern	Arrhenius	Coats-Redfern
10	−1380.97	7930.28	30,015.88	23,755.46	−115	−58.0
20	−1211.76	8530.28	34,624.28	22,774.4	−131	−52.2
30	−379.78	8230.28	25,929.38	24,200.9	−96.4	−58.5
Average	−990.84	8230.28	27,549.83	23,430.06	−105	−55.7

. When ΔG° of the degradation process is positive, it indicates that the process is not spontaneously, and may not be thermodynamically, favorable. The ΔG° of the degradation process of the PET MPs by thermal treatment was all positive. The process was all non-spontaneous, endergonic, and energy was absorbed for the degradation; however, a higher magnitude of ΔG° was obtained from the Arrhenius model (27,549.83 kJ/mol) compared to the Coats-Redfern model (23,430.06 kJ/mol). The values of ΔH° and ΔS° can provide further information on the spontaneity of the thermal degradation process. The energy difference between the reactants and the activated complex is represented by the change in enthalpy (ΔH), which may be used to determine if the thermal degradation process is advantageous [21]. The ΔH° from the Coats-Redfern was positive, indicating that the degradation reaction absorbs or uses more energy than it releases, while it was negative from the Arrhenius model. The Coats-Redfern model had the best fit for the degradation, and may provide a more correct description of the PET MPs’ degradation process. In this work, the thermal deterioration and the change in entropy ΔS of a process was utilized to quantify the degree of disorder in the system. ΔS values below a certain threshold suggest that the process can achieve thermodynamic equilibrium and have a reduced reactivity. From both models, the entropy change was negative, but the Coats-Redfern had lower values ranging from −52.2 to −58.5 kJ/mol.K, suggesting that that process is less disordered and can attain thermodynamic equilibrium and low reactivity. Similar thermodynamic results were also obtained in the study of Dhyani et al. [25] for the degradation of sorghum straw.

3.4. Thermal Degradation Prediction by ANN

For the first time, as far as we are aware, a very effective ANN model was intended to be created, in order to forecast the thermal disintegration of PET MPs. Based on experimental data sets collected at various heating rates, an MLP ANN was created in the present work to forecast the mass loss percentage.

The ANN network information and framework for the different heating rates is presented in Tables S1–S3 and Figure S1. The network topology was NN-3-2-1. Because models with fewer hidden layers produced higher errors, all ANN featured two hidden layers. Without hidden layers, ANN functions as a linear regression model that cannot detect nonlinearity, rendering it difficult for ANN to accurately replicate (model) nonlinear patterns in data. In the hidden layer, which is a mapping to a higher dimensional space, classification boundaries are therefore easier to establish than they are in the original space. There are several nodes in the two hidden layers of the model that were employed in this study. For instance, the first node from the first layer is indicated by H(1:1), while the first node from the second layer is shown by H(2:1).

The major criterion for choosing the most effective network topology to estimate the percentage mass loss as the output variable is the value of the coefficients of determination (R²) [23]. With R², which ranged from 0.951 to 1, the correlation between the forecast deterioration value from the ANN, and the actual values acquired from the TGA, was significant (Figure 7a(i), b(i) and c(i)), and a comparative plot of the mass loss can be seen in Figure 7a(ii, 7b(ii) and 7c(ii). Correlation coefficient values above 0.9 are regarded as acceptable and very significant [32].

The values of SSE and RE were then used to assess the prediction performance of the present ANN model. These values for the training and testing are listed in

Table 6. Statistical parameters for the ANN model.

	SSE			RE
	10 °C/min	20 °C/min	30 °C/min	10 °C/min	20 °C/min	30 °C/min
Training	4.151	8.009	124.250	0.002	0.003	0.050
Testing	1.704	5.161	51.160	0.002	0.005	0.047
Mean	2.928	6.585	87.705	0.002	0.004	0.049
SDV	1.730	2.014	51.682	0.000	0.001	0.002

. The chosen ANN model has statistically significant low variances across the board, suggesting that it can accurately forecast the output parameter. The fact that the SSE and RE for 30 °C/min were high, however, shows that the ANN model is ineffective for predicting deterioration at the heating rate. As a result, it appears that the ANN model is effective at lower heating rates.

The interpretability of the ANN model was also examined, since, for chemical applications, knowing how a model generates a particular prediction may be just as crucial as the forecast’s accuracy. A method comparable to the SHapley Additive exPlanations (SHAP) was used to assess the interpretability of the relevance of input variables to the ANN model. A thorough framework for examining predictions, SHAP is especially useful for complex black-box models, such as ANN [33]. A relevance value is assigned to each characteristic for each prediction, indicating how much the model depends on that variable and how much it contributes to the forecast [34]. This makes it simpler to understand how input variables impact the forecast of mass loss. The results for the input variable importance are presented in Figure 8. The temperature was identified as the most important input variable in the prediction of mass loss from the ANN, with normalized importance of 100%.

4. Conclusions

The study found that increasing the heating rate in TGA analysis of polymers does not affect the decomposition patterns of PET microplastics. The thermal degradation of PET MPs was studied using three kinetic models: Friedman, Arrhenius, and Coats-Redfern, which showed that the activation energy values increased with an increase in heating rate. The Coats-Redfern model had the best fit for the degradation process, as it could describe the entire process efficiently for all temperature ranges. This study has shown that, in addition to determining kinetic parameters, thermodynamic data must also be calculated to establish the viability and energy calculations of the thermal degradation process of PET MPs. The results obtained from the Arrhenius and Coats-Redfern models indicate that the thermal degradation process is not thermodynamically favorable, as the ΔG° values were all positive. However, the Coats-Redfern model had the best fit for the degradation and provided a more accurate description of the process. The values of ΔH° and ΔS° also provided further information on the spontaneity of the thermal degradation process, with the Coats-Redfern model showing that the degradation reaction absorbs or uses more energy than it releases, and the process is less disordered and less able to attain thermodynamic equilibrium. It was discovered that an ANN with a two-layer hidden architecture was the best effective network for predicting the output variable (mass loss %), with a regression coefficient value of (0.951–1.0).