Novel techniques for characterising graphene nanoplatelets using Raman spectroscopy and machine learning

A significant challenge for graphene nanoplatelet (GNP) suppliers is the characterisation of platelet morphology in industrial environments. This challenge is further exacerbated to platelet surface chemistry when scalable functionalisation processes, such as plasma treatment, are used to modify the GNPs to improve the filler-matrix interphase in nanocomposites. The costly and complex suite of analytical equipment necessary for a complete material description makes quality control and process optimisation difficult. Raman spectroscopy is a facile and accessible characterisation technique, with recent advancements unlocking fast mapping for rapid data collection. In this study, we develop novel techniques to better characterise GNP morphology and changes in surface chemistry using Raman maps of bulk powders. Providing a bespoke algorithmic framework for the analysis of these advanced materials. An unsupervised peak fitting and processing algorithm was used to extract crystallinity data and correlate it with laser-diffraction-derived lateral size values for a commercial set of GNPs rapidly and accurately. Classical machine learning was used to identify the most informative Raman features for classifying the plasma-functionalised GNPs. The initial material properties were found to affect the peak features that were the most useful for classification. In low defect density and low specific surface area GNPs, the D peak full width at half maximum is found to be the most useful, whereas the I-2D/I-G ratio is the most useful in the opposite case. Finally, a convolutional neural network was trained to discern between different GNP grades with 86% accuracy. This work demonstrates how computer vision could be deployed for rapid and accurate quality control on the factory floor.