The primary goal of this research is to predict the cyclic voltammetry and polari- zation curves of proton exchange membrane fuel cells (PEMFC) without con- ducting any experiments. For the first time ever, artificial neural network (ANN) is applied to introduce a framework for PEMFC that is composed of vari- ous catalyst layers. Carbon-based cathode materials, such as reduced graphene oxide, graphene oxide, graphene nanoplatelets, and carbon black and their hybrids, including various Pt catalyst content, are being investigated. Important properties of cathode materials, such as surface area, Pt percentage, and Pt nanoparticle size were investigated for the classification of various groups. Results showed that total cathode surface area and Pt content are suitable for more precise data classification and are selected as input variables (features), whereas electrochemically active surface area, cyclic voltammetry, and polariza- tion curves are selected as output responses of ANN. In this framework, experi- mental data for various cathode materials is initially classified using support vector machines and then ANN models are applied to predict the cyclic voltam- metry and polarization curves. Results indicate that data are well classified into four main groups, allowing an ANN to achieve the best prediction of curves with a mean square error of less than 0.3% and a relative error of 0.5%. Also, with the help of the polarization curve, the maximum production power vs different volt- ages can be evaluated. By applying this model, it will be possible to get the nec- essary electrochemical data for an unknown carbon-based cathode material of a PEMFC. Finally, ANN applications can be proposed as a useful tool for predict- ing the main cyclic voltammetry and polarization curves of fuel cells.

Keywords