An improved ART neural net for machine cell formation

Abstract

Several researchers have applied the Adaptive Resonance Theory (ART) neural network for solving the machine cell formation problem. The standard ART1 algorithm is efficient, but its solution quality is dependent on the input order of the machine-part matrix, especially in the presence of ill-structured data. To alleviate this problem, pre-processing and/or post-processing heuristics have been added to the standard algorithm. This paper presents an alternative algorithm which modifies the standard ART1 in two ways: (1) it inputs bipolar vectors instead of binary vectors as given in the original machine-part matrix, and (2) it incorporates performance criteria into the algorithm. The results show that the modified algorithm yields an equally good or better solution than do the existing ones on a set of published examples.