[Colloquium] Zhuokai Zhao MS Presentation/Oct 13, 2021

[meganwoodward at uchicago.edu]

This is an announcement of Zhuokai Zhao's MS Presentation
===============================================
Candidate: Zhuokai Zhao

Date: Wednesday, October 13, 2021

Time: 12 pm CST

Remote Location: https://urldefense.com/v3/__https://uchicago.zoom.us/j/91973039289?pwd=dlhtMWRDN3Z6RHBia0tFTHpMZ1RLUT09__;!!BpyFHLRN4TMTrA!uQWsn6QoL_XZCvwcYIcRnyESgi1bEd0o6kS4N2F_40bSgj7aGIMEOheo_6FpqwRH65s$  Meeting ID: 919 7303 9289 Passcode: 505266

M.S. Paper Title: Utilizing both past and future: multi-frame memory based network in solving particle imaging velocimetry

Abstract: Research in experimental fluid dynamics requires methods for recovering velocity vector fields from within fluid flow experiments. A widely used method, Particle Image Velocimetry (PIV), analyzes video images of fluorescent particles moving in fluids. Various computational approaches have been applied to PIV, such as traditional cross-correlation, variational analysis, and most recently, machine learning (ML). We describe here a novel ML approach to PIV based on deep learning, with the goal of more accurately and efficiently estimating the dense 2D velocity field for each frame of a PIV video.
Our approach is distinguished by how it flexibly uses multiple frames earlier and later in time, rather than only pairs of frames. We show that on a variety of images and flow fields, our deep learning PIV approach is competitive with other state-of-the-art methods. We also describe a software tool for synthesizing PIV images from known velocity fields for ML training, which will benefit future research on ML-based PIV.

Advisors: Gordon Kindlmann

Committee Members: Michael Maire, Gordon Kindlmann, and William Irvine