ABSTRACT

Particle image velocimetry (PIV) is a powerful tool in the arsenal of measurement techniques available to fluid flow experimentalists.
The resulting velocity data are both spatially distributed and quantitative, enabling rigorous global analysis.
However, when studying fluid flows with complex temporal dynamics, the use of PIV is often constrained by limitations on the maximum sampling rate.
In this talk, I will present two data-driven approaches to dealing with this issue, leveraging techniques from control theory and signal processing to extend the capabilities of existing equipment.
I will focus specifically on the challenge of identifying characteristic flow frequencies and their associated spatial structures using sub-Nyquist-rate PIV.
The first approach that I will discuss uses slowly sampled PIV data in combination with a fast probe signal.
These data are post-processed with a Kalman smoother, which produces a time-resolved estimate of the flow field evolution.
Spectral analysis can then be done using the estimated velocity fields, for instance using dynamic mode decomposition.
I will also describe a compressed sensing approach that takes advantage of sparsity and random sampling strategies to avoid aliasing, making it possible to circumvent the Nyquist-Shannon sampling criterion.
The capabilities of each approach will be demonstrated using data collected from bluff-body wake experiments.

BIO

Jonathan Tu is currently a postdoctoral scholar at the University of California, Berkeley, where he studies the hydrodynamics of flagellar locomotion under Profs. Murat Arcak and Michel Maharbiz.
His work is part of a collaborative effort to design and manufacture millimeter-scale microbiorobots.
Previously, Jonathan studied at Princeton University, where he earned a Ph. D. working on flow control problems with Prof. Clancy Rowley.
Prior to that Jonathan attended the University of Washington, where he earned Bachelor's degrees in both Aeronautics/Astronautics and Mathematics.
His research looks to bring advanced tools from control and dynamical systems theory to bear in fluid mechanics.
In particular, he is interested in using data-driven methods such as modal decomposition, compressed sensing, and machine learning to model and control fluid mechanical systems.