Thanks for the interest in LAMMPS. You should talk to
Chris Stoltz and P&G (stoltz.c at pg.com). We have talked
about adding CAD boundaries to the code and he has done
some comparisons of LAMMPS to commercial DEM packages.
A couple users have also had some ideas about allowing
regions to be used as walls (imagine several primitives unioned together), and
then rotated for granular models (like a hopper).
Let me know if you have questions about how to do things in
LAMMPS. Using a large CAD geometry (many triangles) for
a wall could pose some challenges for effectively searching
for particle/wall collisions.