Turbulent Flow Physics and Control: The Role of Big Data Analyses Tools

Abstract

We are studying several problems involving turbulence and big data that range from more efficient and lower noise in next generation jet propulsion systems to bio-inspired concepts for energy production. Specific examples include flows over airfoils (flapping and stationary) and other complex bodies such as turrets and high-speed jet flows. These research activities involve the collection of massive amounts of data from multi-scale computer simulations and/or large-scale experiments. Such experiments/simulations routinely produce terabytes of multi-modal data (velocity, pressure, acoustics, etc.) in fractions of a second. Time-resolved particle image velocimetry data, for example, has requirements of 10 kHz or higher sampling rates in time along with spatial resolution requirements over a broad range of spatial scales observed in high Reynolds and Mach number turbulent flows. Common questions that arise include: How do we compare and contrast data that have different levels of granularity, density (or sparseness), and distribution (e.g. uniform, checkered, lattice, random, etc.)? Can we combine such fields that span in space and time to develop a holistic systems-level understanding? This is important in linking numerical simulations that apply lenses with varying magnifications to the same system, as well as integrating qualitative and quantitative experimental observations with computer simulations. We will discuss our general efforts to apply big data analyses/modeling tools (the “right filters”) to identify patterns and predictive models rather than just a posteriori trends, statistics, and distributions. Our advanced tools include proper orthogonal decomposition, stochastic estimation, optimal inferred decomposition, wavelet analysis, and Lagrangian coherent structure methods which we are using for understanding, modeling, and controlling such flows. In this paper we focus on high Reynolds and Mach number jets, both axisymmetric and more complex. We have also utilized compressive sensing to examine high-dimensional airfoil data but will not discuss those results here and instead refer the reader to other papers in this volume which focus on this approach in some detail.