UROP Project

Automatic Detection of Vortex Events using Hybrid Machine Learning Models

Machine Learning, AI, Image Processing, Fluid Dynamics, Experiment

Research Mentor: Dr. Rafsan Rabbi, He/Him
Department, College, Affiliation: National High Magnetic Field Laboratory, FAMU-FSU College of Engineering
Contact Email: rr26f@fsu.edu
Research Assistant Supervisor (if different from mentor):
Research Assistant Supervisor Email:
Faculty Collaborators:
Faculty Collaborators Email:

Looking for Research Assistants: Yes
Number of Research Assistants: 1
Relevant Majors: Engineering
Project Location: National High Magnetic Field Laboratory, 1800 E Paul Dirac Drive, Tallahassee, FL-32308
Research Assistant Transportation Required: Seminole Express
Remote or In-person: Partially Remote
Approximate Weekly Hours: 5-10, Flexible schedule (Combination of business and outside of business. TBD between student and research mentor.)
Roundtable Times and Zoom Link:

• Day: Monday, August 31
  Start Time: 2:00
  End Time: 2:30
  Zoom Link: https://fsu.zoom.us/j/92880897622
• Day: Wednesday, September 2
  Start Time: 2:00
  End Time: 2:30
  Zoom Link: https://fsu.zoom.us/j/92880897622

Project Description

One of the many interesting things about superfluid Helium (temperature below 2.17 Kelvin, very, very close to absolute zero!) is its ability to generate vortices. A vortex (plural: vortices) is a spinning mass of fluid, where the fluid mass rotates about a single axis. This axis of rotation can be a dangling straight or curved line, or a connected curvature like a circle, giving rise to the famous ring vortex shapes. The existence of vortices is a fascinating aspect of superfluid Helium. At this very low temperature, Helium has no viscosity (no internal resistance to flow), and these vortices are the only way to create rotational motion in a superfluid Helium bath.
As part of our work at the Cryolab here at the National High Magnetic Field Laboratory, we have generated a lot (terabytes!!) of image data of these vortex events in superfluid Helium. We intend to understand the governing fluid dynamics behind these vortices, but to do so, we first need to identify them in the continuous video data we captured with our high-speed cameras. A human can find roughly 2-5 of these events from one frame (if they exist, of course!) in roughly a minute. Hypothetically, that means that to analyze, for example, 1 million frames, to identify and categorize these events, we will need 1 million minutes (694 days, or approximately 2 years) of continuous human work.
This is where this research project comes in. We will use state-of-the-art machine learning models to automate this work, so that instead of spending years identifying these events manually, we will have a library of vortex event data, properly detected and categorized within days. That will involve training many different deep learning-based computer vision models and deploying the best-performing model to detect vortex events from terabytes of image data.

Research Tasks: - Image data analysis
- Learning to script programs in Python
- Data annotation for training
- Building custom deep-learning models
- Writing reports and presenting findings

Skills that research assistant(s) may need: Required:
- Comfortable with computers.
- Preliminary knowledge of linear algebra.

Recommended:
- Some Preliminary knowledge of Python scripting.
- Willingness to learn.