Computer vision and machine learning for assessing dispersion quality in carbon nanotube / resin systems

Abstract

The addition of nanomaterials to polymeric resins can enhance a range of bulk material properties, but the nanofiller effectiveness varies strongly on the dispersion quality. The ability to independently, objectively, and quickly assess the dispersion quality of nano-loaded resins based on microscopy is desirable, but current techniques are often subjective and time-consuming. For this paper, we utilize a dispersion metric based on the use of image segmentation of optical microscope images. We then show that by training a computer vision model on a dataset of segmented microscopy images, the model can then quickly and accurately assess the dispersion of nanoparticles in a material. We apply this process to microscope images of carbon nanotube-loaded commercial resins. Our results indicate that this machine-learning methodology can match the accuracy and repeatability of current methods. In principle, this same machine-learning approach can be applied to a broad range of nanomaterials and matrices, allowing for rapid and quantitative analysis of microscope images for in-line quality control.