Novel Techniques for Characterising Graphene Nanoplatelets Using Raman Spectroscopy and Machine Learning

A significant challenge for graphene nanoplatelet (GNP) manufacturers is meaningful characterisation in industrial environments. This has been further exacerbated by the recent push to functionalise GNPs for better compatibility within nanocomposite matrices [1]. Raman Spectroscopy is a fast, facile, and accessible technique for graphene characterisation. The heterogeneity of GNP powders requires that hundreds of Raman spectra are acquired and analysed, to achieve a representative sample [2]. <br/> <br/>In this work commercially available GNPs are plasma oxygenated, aminated and fluorinated. A peak fitting algorithm was developed to reliably fit hundreds of Raman spectra from mappings, and the curve parameters fed into a binary Random Forest classifier to discern between functionalised and unfunctionalised GNPs. This type of classifier can be used to identify which combination of features are most affected by functionalisation, e.g., fluorination primarily alters the I2D/IG ratio and the G peak FWHM. <br/> <br/>Furthermore, computer vision was used to demonstrate a proof of concept for rapid factory floor quality control. A convolutional neural network (CNN) was trained to infer the presence of functional groups directly from Raman spectra with high accuracies. <br/> <br/>[1] Kinloch, I. A., Science 362, 547 553 (2018). <br/>[2] Goldie, S. J., Acs Appl Nano Mater 3, 11229–11239 (2020).