Illinois team uses Ember to simulate 1 million-grain cube of steel
06.29.12 - Permalink
by Nicole Schiffer
Researchers from the University of Illinois at Urbana-Champaign used NCSA's Ember to simulate the elastic-plastic transition, or the change from reversible to irreversible deformation, of a 1 million-grain cube of grade-316 steel under pure shear stress. It is thought to be the largest nonlinear simulation in which each grain has different material characteristics.
Professor Martin Ostoja Starzewski from the Department of Mechanical Science and Engineering led the team, which included NCSA's Senior Computational Resources Coordinator Seid Koric and Jun Li, a graduate research assistant. Koric is also an adjunct assistant professor in the Mechanical Science department.
"While specialized models have been developed for particular problems in special settings," Koric says, "there is a potential for fractals to offer an assessment of inelastic state and damage in many common materials and structures."
Simulations of soils and rocks in two dimensions suggest a similar relationship in non-metallic materials, Li says. Adding a third dimension accelerated the transition in steel, probably because heterogeneous grains had more opportunities to interact.
Their experiments work toward a way to quickly and cheaply estimate the potential for breakage in things like steel beams, ice flows, concrete sidewalks, and granodiorite, a type of rock found in mountains and near fault lines like the San Andreas, using finite-element simulations. Finite-element analysis is a numerical method that approximates the solution to the integral and partial differential equations that describe material behavior at different points in the elastic-plastic transition.
Koric says Ember's large shared memory was essential to the project. Preprocessing with Abaqus for a single run involving 1 million grains used 64 cores and hundreds of gigabytes of memory for two weeks.
"Even the colleagues from Simulia Dassault Systemes (Abaqus developers) were pleasantly surprised that their code was able to process such a variety of material properties," says Koric.
This research, funded by the National Science Foundation, was recently published in Philosophical Magazine, one of the oldest scientific journals in the world.