CAD interface boosts modeling efficiency

Assessing radiation
transport through materials is easier, thanks to CAD code that
enables more precise geometric modeling. Here, a 40degree slice,
in volume format, of ITER.
(View
larger image) 
or
generating geometric models of such complex systems as the fusion tokamak
ITER (formerly called the International Thermonuclear Experimental Reactor),
a new approach enables researchers to replace combinations of elementary
shapes like spheres, cylinders and bricks with more detailed and precise
CAD representations.
Those old shapes are the bane of engineers using
the popular Monte Carlo transport codes MCNPX or MCNP to assess radiation
transport through materials. A key part of the analysis is specifying
a geometric model for the physical domain—for example, a tokamakstyle
reactor, says Tim
Tautges, an adjunct professor here and a scientist with Sandia National
Laboratories.
But for a system like ITER, where even a simplified
model has 930 separate volumes, using the traditional combinatorial
solid geometry approach to weave spheres, cylinders and blocks into
ITER’s complex shapes is not only tedious and timeconsuming,
it lends itself to error. “The Monte Carlo codes have their own
geometric representation, but it is much less capable of representing
complex models that modern CAD tools can do easily,” says Tautges.
Seeking a more efficient solution, Tautges,
Professor Doug
Henderson and graduate student Mengkuo Wang modified the Monte Carlo
code so that it interfaces directly with a CAD modeling engine and draws
on an external library of CADcreated geometries. Assistant Professor
Paul
Wilson, Research Professor Mohamed
Sawan and several other students and scientists also are helping
to benchmark this code on a fully detailed ITER model.
While other attempts at integrating Monte Carlo
and CAD engine code have resulted in drastically longer computation
run times, the UWMadison group added an extra twist that removed a
timeconsuming step. Most CAD systems view two coincident, or adjoining,
surfaces as separate surfaces, so the raytracing function, which determines
when a particle crosses material boundaries, must be called twice—once
for each adjacent material.
“Part of the reason our approach is faster
than past efforts is that when two volumes have coincident surfaces,
we merged them together so that they’re single surfaces that both
volumes see,” says Tautges.
The group also borrowed a technique from robotics
collision detection and modified it to work on ray tracing, reducing
ray tracing times even more.
As a result of their success, the researchers
drew the attention of the international ITER engineers and, with funding
from the U.S. Department of Energy, are continuing to improve the code,
not only to apply directly to ITER, but also to share with other ITER
partners. In December, these partners met in Rome to share results of
the full ITER neutronics benchmark calculation.