Integrated thermo-kinetic and machine learning approach to pistachio shell pyrolysis for renewable energy applications

This study explored the potential of pistachio shells as a renewable energy source and carbon material by combining chemical, thermal, and machine learning (ML) analyses. Thermogravimetric and derivative thermogravimetric analyses were conducted at heating rates of 5–25 K min⁻¹, to investigate pyrolysis behavior. These tests revealed that the decomposition process occurred in four distinct stages. The kinetic evaluation showed that the activation energy (E_a) varied depending on the method used: the Friedman approach gave an average value of 179 kJ mol⁻¹, while the Flynn–Wall–Ozawa and Kissinger–Akahira–Sunose methods yielded 170 kJ mol⁻¹ and 169 kJ mol⁻¹, respectively. For the Coats–Redfern method, the values ranged between 128 and 171 kJ mol⁻¹. Additionally, the calculated enthalpy (ΔH) and Gibbs free energy (ΔG) values from the Friedman, Flynn–Wall–Ozawa, and Kissinger–Akahira–Sunose methods showed a consistent trend across the techniques. To better understand and predict the E_a, ML models were applied. Multiple ML models like Extra Trees Regressor and Extreme Gradient Boost Regressor were used with conversion fraction (α), logarithmic heating rates, and inverse temperatures as inputs. Both models achieved high accuracy (R² ≥ 0.99) in training and testing, and their predicted E_a and Arrhenius constants closely matched the experimental values. This study provides insights into the pyrolysis behavior and kinetics of pistachio shells, demonstrating their promise as a renewable fuel, and highlighting the value of ML in predicting thermochemical properties.