pat@cs.strath.ac.UK (Pat Prosser) (11/05/88)
>Subject: ES for Scheduling Flexible Manufacturing Systems >Hi!... I am compiling a survey of expert systems for scheduling CIM in general >and FMS in particular. >I am particularly interested in the limitations and effectiveness > ...... and what role a human scheduler would play in the highly complex >and constantly evolving environment of FMS. Does anyone know if ISIS still >being used at Westinghouse? Some ES schedulers worthy of study are: (1) SEMIMAN: a scheduler for ASIC production, in use by Plessey, developed by Peter Ellerby in Reading University. This views scheduling as a dynamic CSP. Peter is currently installing his system ... it exists, I've seen it. (2) DAS: a Distributed Asynchronous Scheduler. This again views scheduling as a dynamic CSP but distributes the task over many processors. It has similarities with the contract net metaphor and blackboard systems. We in Strathclyde have developed/implemented a demonstrator. (3) SONIA: Claude le Pape's system. This is a blackboard architecture, looking much like OPIS (4) OPIS: Stephen Smith. Apparently SS is now installing this for IBM wafer fab plant (I think) (5) YAMS: Parunak's system, a contract net. (6) ISIS: The last I heard of this Nancy Skoogland was re-implementing this in KEE. (7) B.R Fox and Kempf: robot assembly ... opportunistic scheduling (apparently Fox was opportunistic to the degree of walking away from FMC with a Ferrari!) Limitations and Effectiveness: How long is a piece of string? Limitations depends on representations used. Generally representing problem as constraints (temporal, precedence, technological, preference) is rich enough to model most domains. Also knowledge elicitation ... generally no one knows what a good schedule is, generally they do know what a satisfactory schedule is. Therefore in many cases knowledge must be created via simulation. Effectiveness .... depends on what state the company's in. generally most companies don't schedule and even fewer schedule reactively. So ... expect a big win. Human Role: See Peter Ellerby's "junk box approach". Take the human out of the system and you are doomed. Until you can capture all domain knowledge, keep the user in there. The appropriatness of AI technology to scheduling: Want to make loadsamoney? Patrick Prosser University of Strathclyde Dept of Computer Science Glasgow