Discovering modeling errors on large finite element models is complex
and can often go unnoticed, particularly once good ''correlation'' is
achieved. The signs of trouble and/or inaccuracies need to be recogniz
ed and, at the very least, the effects of system simplifications on a
model's accuracy need to be known. One danger in using simplified mode
ls occurs when the models, which have been correlated to test results,
are used for predictive work. When predictive work begins, it is extr
emely difficult to determine if model ''simplifications'' have a small
or large influence on the predictions. Fortunately, there are things
that can be done to help recognize potential trouble. Similar to the e
xamples shown in this paper, a library of small test models should be
created in order to evaluate modeling techniques and code features. Th
e evaluation should include determining limitations of the techniques
and features that may cause modeling errors when applied to larger sys
tems. This is analogous to the design process itself, where components
must first be designed and thoroughly tested and understood before th
ey can be reliably used in larger systems. (C) 1998 Elsevier Science B
.V. All rights reserved.