Artificial Intelligence in Process Control

Until about five years ago, Artificial Intelligence had primarily been in the domain of the academic, the majority of research being carried out, principally in the US, in the development of languages and tools. At this time a number of expert system "shells' were emerging and applications using such shells were being publicised. These shells were usually targeted on the PC user and were to act on non-extensive data bases to provide a desk-top advisory facility. Interest in Intelligent Knowledge Based Systems (IKBS) in the UK was stimulated by the Alvey initiative in 1983. Nine demonstration projects were set up under the Alvey IKBS Progranlme but only one was concerned with the application of real-tinle process control to a chemical batch process. This was RESCU, an acronym for Real-Time Expert Systems Club of Users, and the aim of the project was to demonstrate the potential for the use of expert systenls in this field. Government funding was also given to support the Sira Expert Systems Club and one of the sub-groups took a linguistic controller expert system which had been developed for cement kilns and applied it to an oil lubricating plant. Both these projects were successful but its significant that neither of them utilised existing shells but rather developed their own environments based on conventional AI languages. accessed in some intelligent manner using an "inference engine'. Such shells normally either use forward chaining, ie they are event driven, or backward chaining, ie they are goal driven, or occasionally a combination of the two. Without going into too much detail, the toolkit contains a series of aids to the system developer, including forward and backward chaining, franles, inheritance of values, and hypothetical reasoning, any combination of which can be chosen to solve the particular problem in hand. There are penalties to pay of course in return for having such comprehensive facilities. Firstly, cost; software is typically of the order of £40,000 whereas a PC-based shell may cost less than one tenth of that. Secondly, the hardware required to host the software; for a development system, specialist AI machines with a large amount of core and disc memory are advisable, such as