Multiscale Space-Time Computational Analysis and Visualization System
Abstract
This DURIP proposal is in connection with the ARO project ÒMultiscale SpaceÐTime Methods for FluidÐStructure Interaction Analysis with Topology Change, Slip Interfaces and Thermal Effects.Ó The objectives set in that project and the classes of fluidÐstructure interaction (FSI) problems targeted will require computations over complex geometries, with multiscale methods that need deep understanding and judicious use, and with the number of unknowns reaching tens of millions. This makes analysis and visualization of the computed data a critical step. It is critical not only in making sure that the computational methods are functioning as expected and the mesh resolutions are sufficient, but also in understanding the significance of the results obtained and, if needed, altering the physical and geometric parameters of the FSI system to increase its performance. This step is too interactive for the supercomputers used in the computations, but too intensive for a typical local computer system. Integration of the methods used in the main project with isogeometric (IGA) discretization, which will enhance the computational scope and accuracy, will further increase the computational analysis and visualization load and complexity. While we will continue to perform the computations on the supercomputers we have access to at Rice and elsewhere, we are proposing here a powerful local computational analysis and visualization system that will substantially enhance the quality and impact of the multiscale spaceÐtime FSI methods we are developing in the main project. The physical components of the system will include 10 high-performance servers with graphics processing units (GPUs) that will provide the power we need for the analysis and visualization of the large computed data, a high-performance disk array with large capacity and high-speed access that will be used for storing the large computed data, and two high-speed networks that will provide high-speed communication between the 10 servers and the disk array. Graduate and undergraduate students using this powerful computational analysis and visualization system will learn to be very effective in visualizing, interpreting and understanding what they compute. They will gain skills in understanding how advanced multiscale FSI computational methods work, what spatial and temporal refinement levels are sufficient in what parts of a very complex FSI problem domain, how the accuracy can be improved, what the limitations of the mathematical model or computational method might be, and how the physical and geometric parameters of the FSI system can possibly be altered to increase its performance. This educational component is critical in making sure that the students learn well how to compute complex multiscale FSI problems with sophisticated methods and at high resolutions. We plan to provide that education both in the context of the research the students are conducting and in the three classes the PI is teaching related to computational FSI. The instructional, interactive and computationally-intensive nature of the educational component makes it essential that the computational analysis and visualization system is local, provides easy, on-demand access, and is powerful.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Feb 14, 2019
- Source ID
- W911NF1810234
Entities
People
- Tayfun Tezduyar
Organizations
- Army Contracting Command
- Rice University
- United States Army