Sparse Sensing and Modal Decomposition for Unsteady Fluid Flows

Thesis (Ph.D.)--University of Washington, 2018 ; This work explores data-driven methods, including sparse sampling, modal decomposition and machine learning techniques, for high-dimensional systems in fluid dynamics. Fluid flows are characterized by their nonlinearity, multi-scale structures and unsteady behaviors. Understanding the patterns and their evolving dynamics is crucial for control purposes. Robust control calls on fast signal processing and real-time decisions made in the online stage. Modern data science enables appropriate basis transformations that facilitate efficient sensing strategies for state-space estimation, prediction and control. This thesis builds models to save tremendous online experimental and computational power, by transferring the burden in solving optimization problems to the offline stage. It applies to a variety of real engineering applications, including, but not limited to PIV/optical data collection in wind/water tunnel and DNS/LES simulation data. The data-driven methods developed here apply broadly to high-dimensional complex systems from experiments and simulations, and offer a paradigm shift in our ability to measure, model, and manipulate fluid flows efficiently. They provide physical interpretability of the data that will hopefully lead to future developments in the use of artificial intelligence in real systems.