Dec 4, 2024
4:30pm - 4:45pm
Sheraton, Third Floor, Fairfax B
Min Gee Cho1,2,Katherine Sytwu1,Luis Rangel DaCosta2,1,Myoung Hwan Oh3,Mary Scott2,1
Lawrence Berkeley National Laboratory1,University of California, Berkeley2,Korea Institute of Energy Technology3
Min Gee Cho1,2,Katherine Sytwu1,Luis Rangel DaCosta2,1,Myoung Hwan Oh3,Mary Scott2,1
Lawrence Berkeley National Laboratory1,University of California, Berkeley2,Korea Institute of Energy Technology3
The emerging domain of nanomaterials holds the potential to revolutionize crucial industrial technologies, particularly in the areas of nanocatalysis, sensor technology, and devices for energy storage and conversion. This study focuses on the controlled synthesis of nanoparticles, particularly in tailoring their morphology to enhance the efficiency of catalysts made from noble metals such as platinum and palladium. Changing the morphology of nanoparticles alters the surface facets exposed, directly impacting their catalytic performance. Traditionally, the analysis of active sites on nanoparticles has been limited to a few representative particles in a sample. This approach neglects the variance in characteristics within a batch, leading to incomplete understandings of nanoparticle behavior. Our research addresses this gap through a comprehensive, population-wide statistical characterization using high-resolution transmission electron microscopy (HRTEM) images, encompassing a vastly larger dataset of nanoparticles.<br/>We synthesize cubic-shaped cobalt oxide nanoparticles, varying in size and shape descriptors such as circularity and face convexity. We then obtain HRTEM images of hundreds of thousands of these nanoparticles produced under various synthetic conditions. The large scale of this analysis requires automated image processing. Conventional computer vision techniques, such as thresholding or K-means image segmentation, are insufficient for high-resolution images with complex contrast and texture, which exhibit detailed surface boundaries crucial for identifying particle surface characteristics. To resolve these challenges, we apply a convolutional neural network (CNN) for image analysis. This approach allows for precise, pixel-by-pixel segmentation of particles from backgrounds in several hundred 4k TEM images, each containing hundreds of nanoparticles. This method efficiently detects particles, facilitating the correlation of statistical distributions of their size and shape with synthesis conditions. This machine-learning-assisted statistical methodology will open new opportunities for the designed synthesis of nanomaterials with advanced functionality.