FEATURE-BASED MEMORY ASSOCIATION FOR GROUP TECHNOLOGY

This paper presents a new approach for group technology (GT) part fami ly formation and multiple-application set formation using feature-base d memory association performed by neural networks. The drawbacks of tw o GT approaches, production flow analysis (PFA) and coding and classif ication systems (CCS) are discussed. CCS is useful for similar part re trieval and new part assignment but not adequate for forming machine c ells. PFA can form part families and machine cells simultaneously; how ever, routeing sheets are required and PFA does not provide specific m ethods for part information retrieval. These drawbacks are rooted in t he fact that CCS depends on the relationships between parts and featur es and PFA relies on the relationships between parts and machines. The presented approach emphasizes the relationships between parts. featur es and machines together and incorporates memory association into fami ly and cell formation. The feature-based memory association networks ( FBMAN) system contains three sub-systems. The FBMAN-I system uses auto associative memory and relationships between features and parts to clu ster parts into families. It also overcomes the problems caused by exc eptional parts and the presentation order of seed parts. The FBMAN-II system uses heteroassociative memory and relationship between parts fe atures and machines to form machine cells without routeing sheets bein g provided. The FBMAN-III system, extended from the FBMAN-II system, c an form multiple-application sets at the same time. A case study shows that the FBMAN-III system can also perform part information retrieval , similar part retrieval and new part assignment with autoassociative memory.