Multiple Tools for Automated Nanofiber Characterization by Image Processing

Abstract

Nanofibers have been widely used in many engineering applications, including air filtration, energy storage, and biomedical engineering. Their performances largely depend on the morphology of nanofibers. The key morphological parameters include fiber diameter, pore size, porosity, and thickness homogeneity, which are often manually determined at this moment. There is a need of automated tools for fast determination of nanofiber fiber diameters, pore size, porosity, and surface/thickness homogeneity. Researchers have developed automated tools to determine nanofiber diameters, primarily using MATLAB. However, none of the tools reported earlier can automatically process multiple images, which is essential to the accuracy of results. Regarding porosity, the most accurate approach to pore size determination is Brunauer-Emmett-Teller (BET) surface area analysis. This experimental approach is precise but time-consuming, costly, and destructive. Alternatively, the image processing method may offer a quick estimation of the porosity of the nanofiber mat. In addition, many researchers consider the surface homogeneity with even fiber diameters as good a homogeneity. However, the diameters shown in a SEM image only indicate the local homogeneity with a minimal dimension bounding the SEM image. Alternatively, the thickness of the entire nanofiber sample is a more reliable criterion. However, experimental determination of the thickness throughout the entire nanofiber mat is challenging because of its fragility and thinness. If the thickness of only a few places is measured, the local unevenness may be overlooked during the sampling. The main objective of this research is therefore to develop a set of automated tools for the characterization of diameter, inter-fiber and intra-fiber pores, porosity, and thickness homogeneity of nanofiber mats. Among them, three different approaches are used to determine the nanofiber diameter. Specifically, the following tools are developed to achieve the preceding goals. First, multi-image processing tools are developed to determine the diameters of nanofiber mats using MATLAB and machine learning based on UNet and residual neural network. Second, serval image processing tools using different image segmentation methods are proposed to determine the area of inter-fiber pores, intra-fiber pores, and the porosity. The most accurate one is identified by comparing their performances with experimental data. Finally, a characterization tool is proposed to quantitatively compare the nanofiber homogeneity by analyzing the light transmittance.