Application of neural networks to group technology

摘要

Abstract: Adaptive resonance theory (ART) neural networks arebeing developed for application to the industrialengineering problem of group technology - the reuse ofengineering designs. Two- and three-dimensionalrepresentations of engineering designs are input toART-1 neural networks to produce groups or families ofsimilar parts. These representations, in their basicform, amount to bit maps of the part, and can becomevery large when the part is represented in highresolution. This paper describes an enhancement to analgorithmic form of ART-1 that allows it to operatedirectly on compressed input representations and togenerate compressed memory templates. The performanceof this compressed algorithm is compared to that of theregular algorithm on real engineering designs and asignificant savings in memory storage as well as aspeed up in execution is observed. In additions, a`neural database' system under development isdescribed. This system demonstrates the feasibility oftraining an ART-1 network to first cluster designs intofamilies, and then to recall the family when presenteda similar design. This application is of largepractical value to industry, making it possible toavoid duplication of design efforts.!