Abstract :
A new approach, based on neural networks, was developed to describe thermal decomposition process. In this method the activation function for the neurons in the hidden layer are substituted by kinetic models functions. Within this framework it was possible to measure the individual importance of the models under consideration.
The rhodium (II) acetate system was used as a prototype to test the efficiency of the neural network. Four models, the Prout–Tompkins model and the Avrami–Erofeev model with m=2, 3 and 4, were selected in a preliminary least square analysis. This will provide the present neural network architecture with an important chemical aspect.
The competition between models was possible to be quantified by the weights in the output layer. Although this thermal decomposition process was, in general, dominated by the Prout–Tompkins model, other models were also important to correctly describe the mechanism. The accuracy of the computed values of decomposition fraction is shown to be greater when compared with the models separately. The present method is of general applicability proposing an alternative efficient way to describe solid thermal decomposition data.
