Knowledge-guided Computation for Robust CAD

Abstract

AbstractRobust computations have been haunting CAD system builders for decades. The common belief among researchers is that the source of the problem lies in rounded arithmetic and that new forms of computations may be necessary to ensure consistency and accuracy in commercial systems. This paper argues that, although floating point arithmetic is not perfect, the problem is not with the tool, but rather how it is used. The first part of the paper presents several techniques to increase the reliability of computations within the realm of the old-fashioned floating point computation. Many of these techniques have been applied in industrial systems, however, few of them have been publicized. The second part focuses on a knowledge-guided system mostly appropriate in a NURBS-based environment. It will be argued that the more knowledge is available about the entities to be computed on, the more intelligent decisions can be made on how to proceed with the computation to achieve a high level of reliability.