The Knowledge Engineering Group was involved in a variety of small projects both within and outside the Laboratory to assess the potential uses of knowledge-based systems. Some are listed below.
Informatics and the OPAL Group in Particle Physics developed a working expert system for configuring a complex physics experiment. The OPAL detector at CERN contains several hundred devices controlled by several MicroVAXes, about fifty M68020 microprocessors and many local devices, the entire complex being controlled by a central VAX minicomputer. There was a need for coordinating the activities in the different parts of OPAL. It was felt that an approach using knowledge-based systems would be effective because the resulting system would be easier to maintain and offer better management of complexity as the system was scaled up, a recognised problem. Knowledge engineers worked with experimentalists to develop a prototype which, after enhancement, realised the expected benefits.
The problem was defined to be the development of a tool which would configure the devices required to perform a particular type of physics and permit a new type of physics or, more specifically, to configure the processes such that state transitions happened in an orderly way. The system was intended to be used by the single operator of the detector, a physicist who would specify the desired goal state.
The heuristic rules often employed in knowledge based systems were not applicable because there was no relevant body of expertise available. Instead, a more general method was chosen based on transition networks which had the advantage of being close to the physicists' own schematic representation of a process diagram showing dependencies. Each process was equated to a device with a set of states and associated conditions and actions, so that transitions could take place when conditions were satisfied and would be initiated by the actions. Easy maintainability of the knowledge base was achieved by having a special language for describing the characteristics of devices. In this way it was simple to add new devices or modify existing ones.
A prototype system was developed using a commercial expert system shell on an IBM PC. The output was a file of DCL commands to the VAX to execute the required configuration. The prototype was then rewritten in Prolog and mounted on a VAX. A graphical interface was added and the system was in routine use to control data collection within the experiment.
See also:
To gain familiarity with both shells and to compare different Expert System techniques, two attempts have been made to build a system to advise on the SERC Travel Regulations.
An Expert System with about 220 rules on a section of the Conditions of Employment Memorandum related to Travel Regulations. was implemented in the Expert System shell Xi.
The ESP Advisor shell used the full text of the Travel CEMs to develop a novel full text database plus expert advisor system. Although originally envisaged as a comparison with the application of the Xi shell to the same problem this approach (provisionally called Interactive Annotation) was so different that it could not be directly compared. In this approach the user was shown the relevant text from the CEM and then asked if a dialogue was required to clarify it. This dialogue took the form of a series of questions and a final decision on the applicability of the rules contained in the text to the users current problem.
This was a planning task concerned with the scheduling of the Electron Beam Lithography facility at RAL. A core program was written by Cliff Pavelin using a very simple algorithm for comparing jobs in a job queue; it was written in (Edinburgh) PROLOG, which proved its worth as a tool for writing elegantly brief logical conditions and its disadvantages in trying to understand control procedures. Martin Dunn extended the program to cover the characteristics of real job queues on the EBL. The problem was interesting in that a simple algorithm seemed to work quite adequately whereas a true AI approach lead rapidly to a combinatorial explosion and the need for heuristics elicited from the expert human schedulers.
A small system was built using the expert system shell SAVOIR to aid in short listing applicants for computing jobs at RAL at SO/HSO level. Huge numbers of such applications were frequently received in response to job of the month advertisements and the aim was to provide a system which a clerk or secretary could use to help decide which candidates should be shortlisted. The completed system was rather limited in scope but undertaking it did provide useful experience both in the use of SAVOIR and in building expert systems.
Martin Dunn built an expert system to advise inexpert staff in running up (conditioning) a new ion source for ISIS. Ion sources are capricious devices and the expertise of the two experts involved was something of a moving target. Nonetheless a system was created, first of all using the Xi Plus expert system shell and finally using the Crystal shell. The use of Crystal proved an interesting experience because of its very high productivity, and it was this system which was eventually tested by the ISIS staff. It proved valuable except at a particularly difficult stage of the operation when events moved quickly.
A book, Approaches to Knowledge Representation: An Introduction edited by David Duce and Gordon Ringland was published in April 1988, containing contribution from many Informatics staff.
A feasibility study was undertaken on the role of a knowledge-based system in assisting in the Lyman Space Mission (the Far Ultraviolet Spectroscopic Explorer). The role examined was that of an Operator's Adviser, which would capture some of the knowledge of the resident astronomer. The benefits would be savings in cost and time, consistency of performance and archiving of knowledge. The study resulted in a contribution to the proposal for UK participation in the Lyman mission.
BT were conducting a study on the theory of decision making, and the IKBS Group's experience in knowledge representation and knowledge acquisition both in theory and practice were valuable. A number of documents produced by the study team were reviewed, and some BT review meetings attended. The study was comprehensive in scope, dealing with both single-person and group decisions and normative and descriptive approaches.
An ESPRIT III Project to develop a self-explaining system. The reasoning and output should be understandable to the user. The initial demonstrator was a sewage plant monitoring system. The system was able to explain to the user what their knowledge was and why this was the appropriate action.
