Amphiphilic, Siloxane-Based Fouling-Release Coatings for Oil Boom Applications and Comprehensive, Biological Laboratory Efficacy Testing
Abstract
Numerical methods for the simulation of solid and uid mechanics have a long history in applied elds like structural engineering and aircraft design. However, as modern computing architectures progress and as computational power increases, numerical methods will nd an increasingly wide range of revolutionary applications that would have previously been impractical or impossible. Other revolutionary applications of improved computational capability will result from the increasingly high level of available detail. This is particularly true in biomechanics related problems, where for example computational models that characterize the structure and function of human anatomy have emerged as powerful tools in biomechanical engineering and medicine in recent years. However, these tools are still heavily limited by current performance, and the di erence seems unlikely to be lled by newer generation hardware. While single core computers have traditionally been expected to deliver increased clock rates every few years, this trend ended recently due to strict heat and power constraints, with single-core performance not improving much since 2005, with subsequent improvements mostly through adding more cores on a chip. This increases the overall processing power available, but only if the parallelism can be exploited e ectively. Improvements in memory performance continue to lag signi cantly behind, further limiting the ability to harness the additional power. Exploiting the SIMD parallelization opportunities available on these cores produces a signi cant speedup at all problem sizes. Performance gains through many-core parallelism are signi cant for large problems but have been relatively modest for problems with moderate to low degree of freedom counts. Since the problem sizes that run near real time are too small to bene t from these advances, an alternate approach is required. STI s Cell architecture has proven e ective at computational work ows due to its very high memory bandwidth and raw processing power. This architecture has one general-purpose computing core and eight special-purpose heavily-vectorized processing units. Each special- ized unit has an explicitly-maintained local store instead of the more common implicitly- maintained cache hierarchy, which allows for better utilization of available on-chip storage with less overhead. Coupled with its high bandwidth and computing power, this makes it a promising candidate for advancing performance on near-real-time simulation sizes. The PI proposes to investigate this unique architecture to improve the real-time limit from a few tens of thousands of degrees of freedom to a hundred thousand degrees of freedom while utilizing inexpensive commodity hardware. At the same time, the PI proposes to utilize the same architecture to improve the performance of large simulations. Another approach that has lead to signi cant improvements at large problem sizes is multi- grid preconditioning, but these bene ts are minimal for problems small enough to run near real time. Since multigrid preconditioners are of little bene t for Krylov solvers on small sizes but profoundly important for large sizes, the PI will continue to investigate these solvers and preconditioners as well as e ective transition strategies between them. In addition to these new directions, the PI will continue to pursue improvements in the ability to handle the large deformation in the complex materials typical of biological systems. The PI will also continue to improve the state-of-the-art for simulation algorithms designed to handle complex and irregular geometry as well as multiple interacting materials.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Sep 26, 2018
- Source ID
- N000141712153
Entities
People
- Dean C Webster
Organizations
- North Dakota State University
- Office of Naval Research
- United States Navy