Using ART1 neural networks with destructive solid geometry for design retrieving systems

A prototype of a feature-based design retrieval system using a modified ART1 neural network is proposed and implemented in this research. The coding scheme adopted is a destructive solid geometry (DSG) approach in which parts are coded on its respective form features using a difference operator. Then sets of binary vectors representing all designs are fed into an ART1 network to structure design families. This network can create new groups when novel parts fail to match any current grouping instances. A user can also receive a proper number of reference designs by controlling a, similarity threshold, vigilance parameter. This proposed system has been demonstrated as an efficient and robust design retrieving tool. (C) 1997 Elsevier Science B.V.