Thermogravimetric Analysis and Artificial Neural Network Modeling of Syagrus Romanzoffiana Bio-Fibers: A Comparative Study of Kinetic and Thermodynamic Parameters Using Model-Free Methods

Abstract

This work uses thermogravimetric analysis to analyze the thermal degradation and pyrolysis behavior of Syagrus romanzoffiana fibers (SRFs) in a nitrogen environment at heating speeds ranging from 5°C to 30°C/min (30–800°C). Thermal breakdown phases were identified: hemicellulose and a portion of cellulose degrade at 220–315°C, residual cellulose and some lignin at 225–390°C, and total lignin degradation at 500°C. An artificial neural network (ANN) model with two input parameters (temperature and heating rate) was created to predict mass loss. Among 27 ANN configurations, the ANN26 (5 × 17 * 1) model performed better (R2 > 0.99997). Kinetic and thermodynamic parameters were estimated and compared to three model-free methods: Flynn-Wall-Ozawa, Kissinger-Akahira-Sunose, and Starink. The ANN properly calculated enthalpy (ΔH) and Gibbs free energy (ΔG). However, it overestimated activation energy (Ea) and revealed differences in pre-exponential factor (A), notably using the Starink approach. These findings show that ANN have a high potential for modeling SRF pyrolysis, allowing for optimal thermal processing and mass loss prediction at different heating rates.