Backpropagation DNN and Thermokinetic Analysis of the Thermal Devolatilization of Dried Pulverized Musa sapientum (Banana) Peel

This study examined the thermal degradation of pulverized <i>Musa sapientum</i> (banana) peel waste through thermogravimetric measurements and thermokinetic modelling. For the first time, it also incorporated backpropagation deep learning to model pyrolysis traces, enabling the prediction and optimization of the process. Physicochemical characterization confirmed the material's lignocellulosic composition. TGA was performed between 30 and 950 °C at heating rates of 5, 10, 20, and 40 °C min^-1, identifying a primary devolatilization range of 190 to 660 °C. The application of a backpropagation machine learning technique to the processed TGA data enabled the estimation of arbitrary constants that accurately captured the characteristic behaviour of the experimental data (R2~0.99). This modelling and simulation approach achieved a significant reduction in training loss-decreasing from 35.9 to 0.07-over 47,688 epochs and 1.4 computational hours. Sensitivity analysis identified degradation temperature as the primary parameter influencing the thermochemical conversion of BP biomass. Furthermore, analyzing deconvoluted DTG traces via Criado master plots revealed that the 3D diffusion model (Jander [D3]) is the most suitable reaction model for the hemicellulose, cellulose, and lignin components, followed by the R2 and R3 geometrical contraction models. The estimated overall activation energy values obtained through the Starink (STK) and Friedman (FR) model-free isoconversional kinetic methods were 82.8 ± 3.3 kJ.mol^-1 and 97.6 ± 3.9 kJ.mol^-1, respectively. The thermodynamic parameters estimated for the pyrolysis of BP indicate that the formation of activated complexes is endothermic, endergonic, and characterized by reduced disorder, thereby establishing BP as a potential candidate material for bioenergy generation.