On Impact-Contact Algorithms for Parallel Distributed-Memory Computers

Abstract

Structural vulnerability applications are highly compute intensive. In studies of crashworthiness, impact and penetration, it is not unusual for an analysis to require 100 h of CPU time on current generation production supercomputers, despite the relative simplicity of the models being studied. The difficulty in obtaining blocks of time this large in a production environment severely impairs the number of design options that can be investigated. The lack of software capable of simultaneously, accurately capturing the physics of a crash event and exploiting the power of high-performance computer architectures necessitates costly experimental testing for design verification and certification. For example, in the design of automobiles for crashworthiness, hundreds of sled tests and dozens of full-vehicle crash tests are conducted for each new vehicle program. A sled test may cost $5000 and full-scale prototypes may cost as much as $750,000 each. The cost can be much greater (of the order of $10M) if a redesign results in retooling for a structural component late in the program. Attention has focused on high-performance computer architectures as an effective avenue to bridge the gap between computational needs and the power of computational hardware. New high-performance computer architectures promise order-of-magnitude increases in computational performance, thereby allowing the numerical laboratory to replace physical experiments to a much greater degree.