Multilayer perceptron neural network applied to TG dynamic data of biopolymer chitosan – A robust tool to study the kinetics of solid thermal decomposition

Abstract

• New artificial neural network to study kinetics of solid thermal decomposition. • MLP neural network applied in TG non-isothermal data to determine kinetic triplet. • Validation of new MLP is made using simulated data with kinetic model combination. • Accurate kinetic study of chitosan thermal decomposition process. • Multi-step reactions are described by combination of kinetic models. Thermogravimetric analysis (TG) was used to investigate the thermal degradation process of chitosan under dynamic condition. A methodology based on multilayer perceptron neural network (MLP) was proposed and allows quantifying the contribution of each kinetic models to better describe experimental data. Artificial neural networks (ANN) have been applied to chemical problems to reproduce experimental data with higher accuracy, acting in this way as a universal approximation method. In this work, the MLP architecture uses the approximation on the integral temperature by Coats and Redefern, isoconversional KAS methodology and integral kinetic models. The neurons in the hidden layer are activated by kinetic models, not only to reproduce the experimental data, but to bring physical information about the thermal process. The MLP validation was carried out using simulated data, the neural network retrieves the mechanism based on the residual error and consider the curve shape. For the chitosan experimental data, the MLP shows the lowest residual error to fit experimental data, which were performed at N 2 atmosphere under four heating rates: 2.5, 5.0, 7.5, and 10 ° C min - 1 . The activation energy was calculated as 98.1 to 183.3 kJ mol - 1 along the conversion degree. This increasing behavior of activation energy is due to initial polymer dehydration, which foments the enhancement of intramolecular interactions between the monomers and also the stronger intermolecular interactions between the interconnected layers of polymeric chains. Due to decomposition of hydroxyl and acetamido groups, the MLP determined contribution of two-dimensional diffusion model (D2) and for the breakdown of the glycosidic bond, MLP determined high contribution of the volume contraction model (R3) followed by surface area contraction model (R2). These results corroborate the expected behavior as solid interconnected material along the energy supply experiment.