Four N-doped graphene materials with a nitrogen content ranging from 8.34 to 13.1 wt.% are prepared by the ball milling method. This method represents an eco-friendly mechanochemical process that can be easily adapted for industrial-scale productivity and allows both the exfoliation of graphite and the synthesis of large quantities of functionalized graphene. These materials are characterized by transmission and scanning electron microscopy, thermogravimetry measurements, X-ray powder diffraction, X-ray photoelectron and Raman spectroscopy, and then, are tested towards the oxygen reduction reaction by cyclic voltammetry and rotating disk electrode methods. Their responses towards ORR are analysed in correlation with their properties and use for the best ORR catalyst identification. However, even though the mechanochemical procedure and the characterization techniques are clean and green methods (i.e., water is the only solvent used for these syntheses and investigations), they are time consuming and, generally, a low number of materials can be prepared, characterized and tested. In order to eliminate some of these limitations, the use of regression learner and reverse engineering methods are proposed for facilitating the optimization of the synthesis conditions and the materials' design. Thus, the machine learning algorithms are applied to data containing the synthesis parameters, the results obtained from different characterization techniques and the materials response towards ORR to quickly provide predictions that allow the best synthesis conditions or the best electrocatalysts' identification.
Keywords: ball milling; functionalized graphene; machine learning; oxygen reduction performance; reverse engineering.