Parts clustering by self-organizing map neural network in a fuzzy environment

The description of the attributes or characteristics of the individual part s in a feature-based clustering system is frequently vague, and linguistic, fuzzy number or fuzzy coding is ideally suited to represent these attribut es. However, due to the vagueness of the description, the resulting fuzzy m embership functions are usually very approximate. Neural network learning t o improve the fuzzy representation was used in this investigation to overco me these difficulties. In particular, Kohonen's self-organizing map network combined with fuzzy membership functions was used to classify the differen t parts based on their various attributes. The network can simultaneously d eal with crisp attributes, interval attributes, and fuzzy attributes. Due t o the fuzzy input and fuzzy weights, a revised weight updating rule was pro posed. Various approaches have been proposed to define the distance or rank ing of fuzzy numbers, which is essential in order to use the Kohonen map. T he overall existence measurement was used in the present investigation. To illustrate the approach, parts based on two attributes were classified and discussed. (C) 2001 Elsevier Science Ltd. All rights reserved.