Research

Our group studies a range of unsteady flow phenomena leveraging data science, nonlinear machine learning, complex network theory, information theory, and computational fluid dynamics. Our ultimate goal is to build a data-oriented foundation for real-time analysis, modeling, and control of unsteady flows ubiquitously appearing in various situations around small air vehicles, airplanes, motor vehicles, and fluid-based industrial machines.

Our poster summarizing recent research activities compiled in Jan 2025 can be found here!

Generalized Super-Resolution with Machine Learning

Actively examined super-resolution that reconstructs a high-resolution signal from its low-resolution counterpart for unsteady fluid flows. Super-resolution analysis is often discussed as a global flow field reconstruction since a pixelized input can be regarded as a set of sparse sensor readings.

Data-Driven Manifold Identification

Seemingly complex unsteady fluid flows in a high-dimensional space often lie on a low-rank manifold capturing the essence of their dynamics. To promote an understanding of flow physics and develop low-cost, better control strategies, various data-driven techniques for manifold identification are being proposed by us.

Machine Learning-Based Dynamical Modeling

Due to high dimensionality and nonlinearities of unsteady fluids, reduced-complexity dynamical models are often helpful for understanding and controlling fluid flows. Data-science techniques for reduced-order modeling that accurately capture the kinematics and dynamics of unsteady flows are actively examined.

Phase-Amplitude Modeling

Once possessing ``coordinates" that describe complex unsteady fluid flows in an interpretable, tractable manner, there exists a chance to understand and control dynamics assisted with applied mathematical tools. We particularly focus on phase and amplitude-based low-order expressions of highly unsteady flows while combining with data-based techniques.

Network Science

Network science enables representing complex interactions across the unsteady flow dynamics based on nodes and edges. What can be extracted would depend upon how you define nodes and edges, which is of interest to us. We aim to construct aerodynamics-informed networks for extracting hidden flow physics from data.

Computational Fluid Dynamics

High-fidelity data is the foundation in accordingly performing data-driven analysis for unsteady flows. Computational fluid dynamics with code developments for direct numerical simulations and large-eddy simulations are investigated to produce reliable data sets.

Data-Driven Discovery of Turbulence and Aerodynamic Theory

High-fidelity fluid flow data must involve their physics. With physics-inspired data science techniques and such high-fidelity turbulence data, we aim to reveal new theories that have not been found by us yet.

Uncertainty Quantification

Once using models and data, there should exist uncertainties associated with them. We aim to accurately identify uncertainties and distinguish their sources, which is essential for interpretable and practical data-driven modeling for fluid flows.

A posteriori Testing of Data-Driven Techniques for Industrial Turbulence

Examined the applicability of our developed techniques to industrial turbulence. Due to inherent challenges such as measurement noise across different frequencies, a posteriori test in industrial scenarios is necessitated.

Providing "Guides" of Data-Driven Fluid Dynamics

Actively disseminate our findings with data-driven techniques for unsteady flows as guideline papers. While documenting, open-source codes are being made available to accelerate research activities over the community. See Let us Machine-Learn!