Center for Data-Driven Computational Physics
The mission of the Center for Data-Driven Computational Physics is to usher in the future of large-scale, data-driven modeling of multi-scale physical systems. The phenomena of interest range from turbulent flows and cardiovascular flow modeling, through materials physics to climate science and the large-scale structure of the universe. All these problems exhibit emergent behavior that can only be understood and predicted by accounting for the underlying complexity.
We focus on data-driven solutions to these problems by combining inference, machine learning and uncertainty quantification with physics-based modeling constraints. The interaction of high performance computing with large-scale data is itself a challenging problem, meriting new hardware configurations, software, and computational methods. Thus, we are exploring various pieces of the puzzle including physical modeling, computational science, data science and computer science. Our funding is anchored via an NSF Major Research Instrumentation grant that led to the establishment of a hardware/software ecosystem called ConFlux. We have several projects that are supported by NSF, NASA, DARPA, NIH and AFOSR.
CDDCP Article from Scientificcomputing.com: ConFlux article
Join us for an exciting symposium on the future of data-enabled computational science: MICDE Symposium
Announcement of Center of Excellence in Rocket Combustor Dynamics: AFOSR/AFRL CoE
Announcement of Toyota Research Institute project on data-enabled modeling of battery materials: TRI Article
CDDCP/NASA turbulence modeling symposium July 11/12/13, 2017 was a great success: Website
Subject-Specific Blood Flow Modeling
Subject specific blood flow modeling has developed notably over the last decades. From the pioneering academic work in the 1990s to today’s landscape in which computer-based simulations to assess the severity of coronary disease have been recently approved by the FDA, the progress has been remarkable. The biggest challenge in the field of blood flow simulation is the lack of physiologic data to inform the boundary conditions (at inflow and outflow branches) as well as the mechanical properties (e.g., distribution of stiffness and perivascular tissue support) of the vascular model.
Climate Systems Interaction
The Earth’s climate system is composed of multiple interacting components that span spatial scales of 13 orders of magnitude and temporal scales that range from microseconds to centuries. Decades of research and development have produced a global multivariate observing system, global numerical process models, sophisticated model-data fusion algorithms, and ever-increasing computational capacity. While the physics that underlie climate system interactions are now well understood, the key responses and feedbacks in the system are controlled by nonlinear and multivariate processes whose interactions are not well characterized.
Turbulence in Fluid Flow
The prediction of turbulent flow is a long standing problem in science and engineering. Some of the applications are flow over aircraft wings, combustion in automobile engines, blood flow in arteries, magnetic confinement in fusion, climate modeling, cosmic structure formation.
COMPUTATIONAL MATERIALS PHYSICS
Computational materials physics aims to model the behavior of metallic alloys, polymers, biological and biologically inspired materials, semiconductors, glasses, battery materials and many other categories of substances (materials). The goal is to identify, explain, predict and ultimately to design the properties and responses of these materials.
Cosmologists understand well the statistical properties of the Universe when it was a mere 380,000 years old. The COBE, WMAP, and PLANCK satellite missions have mapped out the tiny temperature fluctuations which grew from quantum seeds implanted in the first instants of the Big Bang. The Universe then was a plasma described by linear fluid equations and characterized by a small set of cosmological parameters. Comparison of linear calculation predictions to the data yields high precision constraints for many of these parameters, but significant degeneracies remain from this analysis alone.