Computer vision-based analysis for high temperature annealing and dropwise condensation

Abstract

Rapid advancements in deep learning fostered by recent developments of superior graphical processing units has had the most contributions to an exciting development of AI and computer science in the last decade. One of the domains that has benefitted the most is the field of computer vision, which has been able to tackle challenges in image analysis that were once considered formidable such as real-time object detection, instance segmentation, image annotation, and more. These technical accomplishments, however, have been made in limited domains such as driving scene, human-scale everyday objects, or biomedical images. Therefore, there are still many fields of science and engineering where the potential of these latest technologies has not yet been tested. In this thesis, the proven competence of deep learning-based computer vision is applied to two specific engineering domains of interest: manufacturing and heat transfer. For field of manufacturing, annealing fabrication-based Germanium-on-Nothing (GON) structures are analyzed to predict their sub-surface morphology based on its corresponding surface pattern. In addition, their transformation of both surface and sub-surface morphology during annealing is simulated, and Atomic Force Microscopy (AFM) topographies are predicted from Optical Microscopes (OM) using machine learning and deep learning. For heat transfer domain, dropwise condensation phenomenon is quantitatively analyzed its temporal heat transfer performance based on droplet morphologies recorded by a CCD camera. Furthermore, factors contributing to high heat transfer performance are analyzed by tracking dynamic droplet behavior on surfaces with different wettability.