Using an unsupervised neural network and decision tree as knowledge acquisition tools for FMS scheduling

This paper applies neural network learning techniques and a decision tree method to obtain scheduling knowledge for flexible manufacturing systems. A search is made for the reverse relationship between the FMS system performance and its corresponding decision parameters. More specifically, the learning function is to acquire the scheduling knowledge which determines the decision-mix to control adequately the system in fulfilling a specified system performance. This paper first configures a typical flexible manufacturing system and defines a set of decision parameters along with a set of system performance evaluation criteria. The learning procedure initiates accumulating simulated data through the system. Next, the ART2 (adaptive resonance theory) neural model takes the system performance data as inputs and forms performance classes according to their similarities. Finally, the decision tree construction method is applied to extract the definition of each class. The definitions are converted into rule-type knowledge associated with certainty factors, and used as the scheduling knowledge.