Page 134 - The Combined Finite-Discrete Element Method

SELECTION OF CONTACT DETECTION ALGORITHM 117

1 Detection time -- N = 10000 - packing D
Time for 10 loops (s) - (R4600 133.0 MHz) 0.6 Legend
0.8

0.4
Experiment
Approximation
0.2

0
0 20 40 60 80 100 120 140 160 180 200
Spacing (S/D)
Figure 3.53 Example IV: total CPU time as a function of packing density for pack D
(A. Munjiza and K.R.F. Andrews, International Journal for Numerical Methods in Engineering,
43/1). (Reproduced by permission of John Wiley & Sons, Ltd).

• In terms of CPU requirements, the Munjiza-NBS contact detection algorithm has a
better performance than either binary search based contact detection algorithms or
sorting contact detection algorithms.
• The Munjiza-NBS contact detection algorithm uses less RAM space than the binary
search based contact detection algorithm.
• The Munjiza-NBS contact detection algorithm uses slightly more RAM space than the
sorting contact detection algorithm.
• Both the Munjiza-NBS contact detection algorithm and sorting contact detection algo-
rithm have RAM requirements proportional to the total number of discrete elements.

3.9 SELECTION OF CONTACT DETECTION ALGORITHM

A whole range of contact detection algorithms is available for large scale combined finite-
discrete element simulations. The most efficient algorithm in terms of CPU time required
to detect all contacts is the screening array based contact detection algorithm. The problem
with this algorithm is that RAM requirements are most often prohibitive.
The most efficient algorithm in terms of RAM requirements is the sorting contact
detection algorithm. The problem with this algorithm is that the CPU requirements are
not a linear function of the total number of discrete elements. This algorithm belongs to
the hyper-linear category of contact detection algorithms, which means that for very large
scale problems, CPU times can be prohibitive