Image Data Mining for Quality Control of Nanofiber Reinforced Piezoelectric Nanocomposites Based on SEM Images

Recently, portable and wireless nano-scaled devices have been widely used in environmental monitoring, medical implants, defense technology, industrial safety, and personal electronics, such as the nanowire-based gas and chemical sensors and programmable nanowire circuit for nano-processors. There has been an increasing demand for high energy density capacitors that can be engineered for various applications in advanced devices. In it, piezoelectric nanofiber materials play critical role in producing new products. A piezo-ceramic polymer composite contains a polymer of high dielectric strength and high permittivity ceramic inclusions, making it well suited as a high energy density capacitor. Furthermore, the research shows that the well aligned piezoelectric nanofibers generate better dielectric permittivity performance than random distributed ones. Hence, how to evaluate the nanofiber alignment for quality control becomes very critical. However, the extremely small size of nanofiber makes the quality control evaluation process very difficult. Currently, the standard quality inspection technique is the morphology analysis of nanofibers embedded in the base material based on microscopic images, e.g., scanning electron microscope (SEM) images. Visual checking of these SEM images is often adopted to evaluate the nanofiber alignment, which is often subjective, inaccurate and time-consuming. Therefore, how to automatically extract the nanofibers (number, sizes, locations and orientations) is highly desirable. The objective of this research is to fill such need by developing various image data mining methods for automatical nanofiber alignment evaluations through the SEM images. These image data mining methods are proposed to extract the information of the nanofibers through SEM images. The first method use a series of image processing algorithms (thinning, windowing, transforming and convolution process) to get the probability distribution to find the orientation of the nanofibers in SEM images. Second method use Hough transform based algorithms to segment the nanofibers from SEM images. The third method use a statistics method (cost function based multiple changepoints method) base on the boundary data of the nanofibers to extract the parameters (size, location, orientation) of nanofibers in SEM images. Finally, through vi the comparison of these image data mining methods, the advantages and weak points of each method can be summarized to help people improve them in future quality control results