Artificial neural networks modeling ethanol oxidation reaction kinetics catalyzed by polyaniline-manganese ferrite supported platinum-ruthenium nanohybrid electrocatalyst

The electrochemical and physicochemical properties of the anode catalyst used in alcohol fuel cells affect the efficiency of the fuel cell due to both its effect on the cell potential and its direct relationship with the reaction stoichiometry. It is possible to determine these parameters from polarization curves (current vs. cell potential) of the cells. At low potentials, Tafel plots offer kinetic information, whereas currents at high potentials reveal the average number of electrons released per ethanol molecule and their potential dependency. Herein, bearing the chemical engineering research and design aspects, it was aimed to model the kinetics of ethanol electrooxidation reaction catalyzed by PANI-MnFe2O4/Pt/Ru nanocomposite by an artificial neural network (ANN) approach, specifically by Differential Evolution (DE) algorithm for the first time. The different parts of the Tafel plots were condensed into a single and efficient model to illustrate the generalization potential of ANN. The findings demonstrated that the best model with a single hidden layer with 18 neurons offered the highest correlation metrics of 0.997476, and the lowest mean-square-error value of 0.000428 at the testing phase. Furthermore, the average absolute error was calculated, with 0.773% in the training phase and 0.66% in the testing phase. These outstanding results indicated that the best model is capable of accurately capturing the dynamics of all instances analyzed. (C) 2022 Institution of Chemical Engineers. Published by Elsevier Ltd. All rights reserved.

Automated summary

This study presents a novel approach using an artificial neural network, specifically the differential evolution algorithm, to model the kinetics of ethanol electrooxidation reaction catalyzed by PANI-MnFe2O4/Pt/Ru nanocomposite. By condensing different parts of the Tafel plots into a single and efficient model, the generalization potential of the artificial neural network is demonstrated.