S. Sarkar (PI), S. Baden (co-PI) and Y. Bazilevs (co-PI) UC San Diego
Large scale simulation data sets that describe spatial and temporal variation of physical phenomena over a multitude of scales often contain characteristic flow features that are infrequent but critical to understanding the natural phenomenon or engineered system in question. The research will lead to a novel data discovery framework, to flexibly identify, extract and interrogate features of interest in space-time, leading to both a reduction in size of stored data sets and identification of time periods where high-fidelity simulation of the system is necessary. The data discovery framework will provide algorithmic and data structure optimizations to enable the user to flexibly query the data over space and time. It will be implemented on Gordon, a new data-intensive platform with non-volatile storage - flash memory. The proposed research will bring together investigators with expertise in computer science, fluid mechanics and structural mechanics to advance computational and data-enabled science and engineering.
The motivating engineering application is offshore wind turbines. The wind turbine simulations proposed here are ambitious and will couple computation of flow-structure interaction at the temporal resolution of seconds with simulations of realistic atmospheric turbulence. Such simulations of a full-scale wind turbine are unprecedented and will be enabled by a computational framework that brings together the following components: advanced large eddy simulation for atmospheric flow, isogeometric analysis for fluid-structure interaction, and a finite element method with weak boundary conditions for the blade aerodynamics. Atmospheric conditions in a marine boundary layer, particularly under conditions of stable stratification, lead to intermittent bursts of wind shear and pressure that can have an unusually large effect on the rotor blade structural response. Quantification and understanding of such large-impact events using the data discovery framework will enable better and more cost-effective blade design, better operational performance, and more accurate wind resource forecasting. There will also be better understanding of extreme atmospheric events that lead to wind turbine blade failure. The data discovery framework to be developed here will have broad impact since it can be employed in any field involving multi-scale, time-dependent data containing dynamically important objects of lower-order complexity. Some examples are cardiovascular imaging and modeling, hurricane prediction, epidemiological modeling, and cloud dynamics.
