Interviewer: Arushi Gupta

Trivia

• Favorite book: The Name of the Rose by Umberto Eco.
• Favorite place: Home.
• Favorite food: Sushi.
• Favorite example or kind of turbulent flow: Turbulent mixing, where two fluids combine—it’s extremely pretty and visually pleasing to watch!

Could you give an overview of the research you and your lab work on, and its real-world applications?

My research focuses on wall-bounded turbulence, which refers to the chaotic and disorganized movement of fluids such as air, water, or oil when influenced by the presence of a solid surface. I employ computational tools to simulate these turbulent flows to better understand their behavior and how to mitigate or enhance them as needed. For example, on aircraft, we aim to reduce turbulent drag to decrease fuel consumption. On the other hand, in engines, we want to enhance turbulence to improve the mixing of air and fuel for more efficient combustion. Weather predictions also rely on understanding wall-bounded turbulence. The Earth’s atmosphere acts as a wall-bounded turbulent flow, where the “wall” is the Earth’s surface, and so accurately predicting weather depends on predicting the behavior of this turbulence. Some work I’ve personally been involved with recently includes studying ocean-ice interactions in Antarctica, particularly how tidal waves underneath the ice contribute to ice melting, which is an effect driven by turbulence.

My lab uses computational techniques to visualize turbulence because it can be difficult to observe and measure in real-world settings, especially in fluids like air and water. Our simulations provide three-dimensional representations that allow us to see turbulent flows as they would occur in reality. Ultimately, the goal of my research is to deepen our understanding of turbulence and to develop methods to either mitigate or enhance it as necessary for various applications.

What inspired you to pursue research in the areas of turbulence and fluid dynamics?

My journey into turbulence and fluid dynamics was somewhat serendipitous. During the second year of my PhD, I was uncertain about what kind of research I wanted to do. I was in a program for computational math, similar to Caltech’s ACM program, and I knew I wanted to apply mathematical concepts in a more practical, engineering context. Fortunately, I found my PhD advisor who was looking for a new student, and his work in turbulence seemed interesting to me. At first, I was more drawn to the mathematical aspects of the problem. However, as I spent time in the research group, I began to appreciate the complexity and significance of turbulence. Over time, this led to a shift in my focus towards the physics side, rather than the mathematical side of the problem.

Your group has applied machine learning and data-driven methods to problems in turbulence research, such as turbulence modeling, and simulation. Can you describe how machine learning methods have transformed traditional research approaches in your field?

Turbulence is what we call a nonlinear and chaotic problem. The Navier-Stokes equations, which describe the motion of fluids, are nonlinear by nature. They are also chaotic, which means that even slight variations in initial conditions can lead to vastly different outcomes over time. These characteristics make turbulence one of the most challenging problems to solve, despite decades of past and ongoing research.

Traditional mathematical tools, especially linear ones, do not work well due to the nonlinear and chaotic nature of turbulence. As nonlinear problems typically require nonlinear tools, machine learning has emerged as a promising tool. Machine learning can be thought of as a form of nonlinear regression, which makes it a powerful option for navigating the wide state space associated with turbulence. Its application has led to some notable improvements in the field. 

However, one significant challenge with machine learning is the lack of interpretable results. Turbulence is a physical phenomenon, and understanding the underlying mechanics is crucial for effective modeling. The ‘black box’ nature of some machine learning approaches can hinder this understanding.

Enhancing the interpretability of machine learning models for turbulence research remains a challenge. Despite this, machine learning being a nonlinear tool makes it valuable and effective for addressing problems in turbulence. 

Could you describe a specific project you have worked on where you applied machine learning techniques?

One research project in my group is about mitigating lift fluctuations in an airfoil. Imagine an airfoil in flight, especially under gusty and turbulent conditions. Similar to a paper airplane with a fan blowing on it, the airfoil experiences unstable up-and-down movements due to the turbulent flow. This instability is problematic for small drones or aircraft, where maintaining control is crucial.

In our project, we aim to stabilize the airfoil using machine learning. Specifically, we’re employing reinforcement learning techniques. Airplanes have flaps on their wings that allow them to control their movement through any turbulence. Our process involves placing sensors on the wing to monitor dynamics and manipulate its flaps, which help control the aircraft’s stability. The goal is for the reinforcement learning model to learn how to adjust these flaps optimally, minimizing the lift fluctuations caused by turbulence. 

What aspects of being a researcher do you most enjoy?

The part I enjoy the most is that no two days of my job are the same. With research, I’m always encountering new problems, which means there are always new things to tackle, different tools to use, and various approaches to try. I enjoy the creativity required and dealing with different problems every day. I’ve always been a problem solver, and being a researcher allows me to problem-solve for a living.

What are some challenges you encounter as a researcher? How do you address them?

The biggest challenge I’ve encountered is that research is inherently uncertain. Often, it’s unclear whether the problem we’re trying to solve even has a solution, if that solution is of any interest, or if it’s possible to find the solution. 

Now that I’m advising students, both undergraduates and PhD students, I feel an increased responsibility to ensure their projects have meaningful outcomes. This adds to the stress, but also to the importance of the work. Through my experiences, I’ve learned that research projects are dynamic; they evolve. You might start with one question, but as you gather data and insights, you might need to modify your approach or even change the question altogether. Sometimes, you may need to completely reconsider your initial hypothesis.

I’ve found that viewing these adjustments as part of the problem-solving process can make them less daunting. Over time, I’ve learned different techniques for adapting problems so that they are solvable. Ultimately, taking a more flexible approach is how I tackle the issue of the uncertainties and unknowns involved in research.

Outside of research, what are some hobbies or interests you have?

I play tennis at least twice a week, so you might catch me at the tennis courts if you’re passing by! I also enjoy swimming; it’s a great way to clear my mind. When I’m in the water, I’m not thinking about much else because when swimming, you have to move forward to survive. I also enjoy just staying at home and relaxing. On my days off, I am very much a couch potato. I like sitting on my couch and playing video games with my cat. 

What advice do you have for undergraduate students pursuing research, especially in your areas of interest?

If you are interested in research, Caltech provides a lot of opportunities to get involved. Engaging in research early on allows you to explore different topics and decide whether you might want to pursue graduate school, continue with research, or explore other interests. Programs like SURF and academic year research are great opportunities for getting hands-on experience with research. 

It’s important to note that some research areas might require specific prerequisites. For instance, in my group, we do a lot of simulation work, so I like people coming in to have some coding experience. I recommend checking out what each opportunity entails and talking with people involved to understand what’s expected. Starting this process early rather than later can better prepare you for these opportunities as they arise.