3.1. Effect of Temperature and Starting Material Composition on the Yield and Chemical Compounds of Pyrolytic Oil
In
Figure 2, the oil yields obtained after pyrolysis with banana peels and/or tires at 400, 450 and 500ºC are shown.
The tests carried out with 100% banana peel (
Figure 2a) and 75% banana peel: 25% tire (
Figure 2c) showed an inversely proportional response in terms of yield as the temperature increased. At 500 ºC the oil yield fell by around 30% when compared to the yield obtained at 400 ºC. The yields observed in this study for the banana peel tests are higher than those reported in the literature, and this may be related to the properties of the raw material used in the experiment. However, the decrease in yield as the temperature increases has also been observed in other studies.
The study by Ozbay et al. (2019) investigated the thermal and catalytic pyrolysis of banana peel (400 to 700 ºC). At 400ºC the yield was 24.94%, at 550º the yield was 28.03% and at 700ºC the yield was 20.19%[
14].
In the study carried out by Daimary et al. (2022), banana peel pyrolysis was carried out at temperatures of 500 ºC, 550 ºC and 600 ºC. From the tests, a yield of 22.18% was observed at 500 ºC, 24.20% at 550 ºC and 20.71 at 600 ºC [
20]. Taib et al. (2021), evaluated the production of bio-oil from banana pseudostem via the fast pyrolysis process (470 ºC to 540ºC), and in their tests also observed a reduction in oil yield as the temperature increased [
12].
Temperature is known as the most important factor in the pyrolysis of lignocellulosic biomass and affects the distribution of the product comprehensively. With increasing temperature, net bio-oil yields tend to decrease as a result of secondary cracking reactions of the pyrolysis vapor, reducing liquid and biochar yields while increasing gas yields [
14,
20].
In the tests where 50% banana peel was used as the starting material: 50% tire (
Figure 2b) and 100% tire (
Figure 2e), it can be seen that as the temperature increases, the oil yield also increases. Kyari et al. (2005) pyrolyzed waste tires from seven different manufacturers and countries at a temperature of 500 ºC. The oil yield varied between 55.4 and 57.8 %, with an average of 56.6 and a standard deviation of 1.2 %. No differences were observed in the yield of the oil produced by mixing the different tires [
21].
Čepić et al. (2021) evaluated the influence of temperature on the yield and composition of tire pyrolysis products, with other process parameters remaining unchanged, the temperatures evaluated were: 400, 450, 500, 550, 600, 650, 700 and 750 °C. From the results obtained, it was observed that with the increase in the final pyrolytic temperature, the oil yield increased simultaneously and reached a maximum value of 43.6% at 500 °C, after this temperature a reduction in yield was verified, at 750 °C the yield was 26.6%. In this study, the gas yield was evaluated, and by monitoring it, it was found that increasing the temperature increases the gas yield and decreases the oil yield [
22].
Similar results were also observed by Cunliffe & Williams (1998), who investigated temperatures between 450ºC and 600ºC. This is due to the secondary cracking reactions that take place at high temperatures to form non-condensable gases [
23].
The co-pyrolysis of tires and banana peels is still a topic that has not been widely explored in the literature. However, the results obtained in this study demonstrate the potential for using these two residues to obtain oil.
The oils obtained from the pyrolysis (400ºC, 450ºC and 550ºC) of the tire and/or banana peel were analyzed by reflectance infrared spectroscopy (FTIR) to assess the functional groups (acids, aromatics and others) present. The infrared spectra are shown in
Figure A1-A3 (
Appendix A).
The two most important areas for preliminary examination of the spectra are 4000-1300 cm
-1 and 900-650 cm
-1. The characteristic peaks of alkanes and alkenes (1645-1500 cm
-1, 1475-1350 cm
-1) characterized in the FTIR spectra of the thermal process at 400°C were repeated for all the pyrolysis processes on the different tire and banana peel matrices, as shown in
Table 2.
The spectra of the oils obtained from the pyrolysis of 100% tires, regardless of temperature, have a similar profile to the spectrum of commercial diesel oil analyzed by Ruschel et al. (2014) [
1].
The double peaks between 2800 and 3000 cm-1 in the stretching mode were identified at all temperatures in the 100% tire and 75% tire: 25 Banana Peel samples, which indicate the presence of alkanes (C-H3, C-H2 and CH).
The peak between 1350 and 1470 cm -1 in the bending mode due to the deformation of CH2 and CH3 denotes the presence of alkanes. The C double bond O stretching vibrations with absorbance between 1650 and 1750 cm -1 indicate the presence of ketones or aldehydes.
Banar et al. (2012) used a fixed bed to produce pyrolytic oils derived from tires at temperatures of 350, 400, 450, 500, 550 and 600 °C and analyzed their properties as a potential fuel. Through FT-IR characterization they observed the presence of alkanes (2800 and 3000 cm
-1, 1350 and 1475 cm
-1), ketones or aldehydes (1575 and 1675 cm
-1) [
24].
The samples with 100% banana peel, 25% tire: 75% banana peel and 50% tire and 50% banana peel showed a profile similar to that observed by Taib et al. (2021) who evaluated the production of bio-oil from banana pseudostem via the fast pyrolysis process [
12]. The strong absorbance between 3550 and 3200 cm
-1 represents the O- functional group, while the absorbance peak located near 3340 cm
-1 is a typical characteristic of cellulose. The absorbance peaks between 1650 and 1580 cm
-1 represent C double bond stretching vibrations which indicate the presence of alkenes. The C-H deformation vibrations between 1470 and 1350 cm -1 indicate the presence of alcohols and phenols. Finally, the peaks between 1300 and 950 cm
-1 indicating the C-O functional group are due to the presence of alcohols and phenols.
3.2. Regression Model with Decision Tree for Predicting CHN Content in Pyrolytic Oil
Elemental analysis was carried out to quantitatively determine the elements carbon, hydrogen and nitrogen in the oils obtained from pyrolysis at 400 ºC with the biomass of: Tire (T, 100%), Banana Peel (BP, 100%), T: BP (50:50), T:BP (25:75), and T:BP (75:25). The results of the elemental analysis are shown in
Table 3, in terms of the percentage of Carbon (C), Hydrogen (H), Nitrogen (N) and the Hydrogen/Carbon ratio (H/C).
Based on the results obtained in the CHN analysis, a regression model based on decision trees was used to predict the carbon (%C), hydrogen (%H) and nitrogen (%N) contents in bio-oils obtained from the thermal pyrolysis of waste tires and banana peels.
Figure 3 shows the decision tree for predicting the chemical parameters of carbon (CHN) content. Analysis of the results revealed that variables such as banana percentage and temperature play critical roles in determining CHN levels, as illustrated in the decision tree diagram generated.
The results of training the model and of a run with the prediction for a bio-oil produced at a temperature of 500°C and a waste mixture ratio of 40% tire and 60% banana are shown in
Figure 4 below.
The Mean Squared Error (MSE) obtained was 7.027, indicating the average level of error in the model's predictions. The predictions for %C varied in accuracy, with some predictions being very close to the actual values, as in sample 1 (Actual: 3.34, Predicted: 3.31) and others with more significant differences, as in sample 9 (Actual: 53.40, Predicted: 37.28).
In the predictions for %H there was a good match in most samples, with a few exceptions, such as sample 12 (Actual: 10.40, Predicted: 8.07), suggesting areas for potential model refinement. On the other hand, the predictions for %N are generally close to the actual values, with small variations observed, such as in sample 5 (Actual: 1.05, Predicted: 1.07).
The larger discrepancies observed, especially in relation to %C, can be attributed to variations in the input data that are not perfectly captured by the decision tree. Adjusting the maximum depth of the tree or implementing ensemble techniques, such as Random Forests, can help improve accuracy.
The results indicate that the decision tree model offers accurate predictions of the chemical parameters of bio-oils, with a significant correlation between the predicted and observed values. The structural simplicity of the model makes it easier to interpret the effects of the input variables, making it an effective tool for analyzing limited experimental data.
Simulations were carried out in the model to predict the behavior of %C, %H and N% in pyrolysis; the simulated conditions and their results are shown in
Table 4.
In the predictions made, the variations in the proportion of waste adopted did not show variations in the percentages of C, H and N, predicting the same values for the ranges of changes in the proportions adopted. The presence of a relatively low MSE suggests that, although the predictions are good, there is still room for improvement. Collecting more data or incorporating additional variables could help refine the model.
The production of bio-oils from the thermal pyrolysis of waste such as tires and banana peels represent a sustainable solution for waste management and energy generation. Accurately predicting the chemical parameters of bio-oils, specifically the carbon (%C), hydrogen (%H) and nitrogen (%N) contents, is essential for optimizing the production process. Several machine learning models are widely used for prediction in similar contexts. Artificial Neural Networks (ANNs), known for their ability to capture complex, non-linear relationships, are often chosen for forecasting tasks due to their flexibility [
25]. However, their effectiveness depends on large volumes of data to effectively train the model's many parameters, which can lead to overfitting in small data sets.
In addition to Neural Networks, Linear Regression is one of the simplest and most interpretable models used to capture linear relationships between variables. Although useful in many contexts, it may not be suitable for data that has complex non-linear relationships, as may be the case with bio-oils. Support Vector Machines (SVM) are effective in high-dimensional spaces but can be computationally expensive and difficult to interpret.
Decision trees, on the other hand, are ideal for smaller data sets. They segment the input space into homogeneous regions, making it easier to understand the interactions between variables and offering robust predictions with less data [
18]. The use of decision trees as a predictive model was based on the need for a robust and interpretable model capable of operating efficiently with a limited data set [
26]. Decision trees not only improved the accuracy of predictions, but also provided valuable insights into the interactions between variables, contributing significantly to the improvement of the thermal pyrolysis process.
The decision tree model implemented proved to be an effective tool for predicting %C, %H and %N in bio-oils, providing useful predictions that can inform improvements in the production process. Future research should consider adjustments to the model and the exploration of complementary techniques to increase the accuracy and robustness of the predictions, contributing to the improvement of the thermal pyrolysis process.