National Science Foundation: Dynamic Feature Extraction & Data Mining for the Analysis of Turbulent Flows
Principal Investigator: Professor Ivan Marusic
|Professor Graham Candler||Aerospace Engineering and Mechanics|
|Professor Ellen Longmire||Aerospace Engineering and Mechanics|
|Professor Sean Garrick||Mechanical Engineering|
|Professor Vipin Kumar||Computer Science|
|Professor Victoria Interrante||Computer Science|
|Professor George Karypis||Computer Science|
This research program unites experts from data visualization and data mining with those from experimental and computational fluid dynamics to develop new computational tools for the quantitative visualization and analysis of large data sets. The need for this work has become apparent with the recent rapid advances in computational and experimental capabilities, which have advanced to the point that they have out-stripped our ability to analyze the data.
The objective of the proposed research is to develop a new class of data analysis methods. This will involve on-the-fly identification of physically-important events to selectively store the data. This database will then be interrogated by feature extraction algorithms to yield an additional object database, which is a compact representation of the physically important space-time object trajectories. These data objects will be used as input to novel data mining methods to discover causal relationships between the objects. This approach will result in efficient storage of data, visualization of the important events within the data set, and methods for high-level analysis of relationships between objects in the data.
The test bed problems for the development of these methods are
turbulent boundary layer flows and turbulent combustion. These databases have
been chosen because of their technological importance and because they offer
different challenges to the method. We will consider two specific events in
these flows: drag-producing bursts in boundary layers, and soot formation in
combustion. This integrated approach of feature identification, selective
storage of data, and advanced analysis will hopefully result in a better
physical understanding of fluid turbulence. This would then potentially result
in the development of mechanisms for drag reduction and reduced soot emissions.
More generally, the approach may be applied to a wide range of scientific
simulations and experiments, and represents a new way to analyze large data
Last Modified: 2011-07-01 at 10:45:18 -- this is in International Standard Date and Time Notation