Predicting the thermal degradation of Agave americana L. biowaste fibers using kinetic analysis and artificial neural networks

Traditional kinetic models are commonly employed to analyze the pyrolysis behavior of biomass fibers. However, their accuracy is often limited because they struggle to capture the complex, nonlinear interactions among thermal factors. This study uses thermogravimetric analysis and derivative thermogravimetric analysis to examine the pyrolysis properties and thermal degradation behavior of flower stalk fibers from Agave americana waste under different heating rates ( β = 5, 10, 15, 20, 25, and 30 °C/min). The results show a clear shift of both devolatilization zones to higher temperatures as β increases, with the first peak moving from 320°C at 5 °C/min to 350°C at 30 °C/min, and the second peak shifting from 410°C to 445°C over the same range. This shift is attributed to a decrease in heat transfer efficiency at higher β , which affects the thermal degradation kinetics. Artificial neural network (ANN) models, especially the architecture (5 × 17 × 1), were developed to model the pyrolysis process. The ANN model achieved a mean bias error of less than 2 %, a mean absolute error of less than 0.03, and a correlation coefficient of over 0.98, demonstrating strong agreement with experimental data. Comparisons with kinetic parameters obtained from Flynn-Wall-Ozawa, Kissinger-Akahira-Sunose, and Starink methods indicated that the ANN slightly overestimated activation energies, with average predicted values of 135 kJ/mol compared to 125–130 kJ/mol from kinetic methods. The ANN also effectively predicted thermodynamic parameters, with enthalpy change values between 120–140 kJ/mol, Gibbs free energy around 110 kJ/mol, and entropy change values that varied slightly with minor deviations from experimental results.