Abstract
In the 1960s Crossman performed some classic laboratory studies of process control, requiring the operator to control the temperature of a container of fluid against fluctuations caused by the environment. This led to a model of the novice controller as a damped second order controller, who changed to open loop control with practice.
We have recently repeated and extended this work with a simulated process control system made up of 4 Crossman tasks coupled together, and making the task more complex by requiring temperature, fluid level, and flow rate all to be controlled.
Two aspects of the research will be reported. The first deals with the behavioural data. Changing patterns of skill are revealed as changes in the transition matrices. Weaknesses in display design, and cognitive lockup during fault management are revealed by patterns of eye movements. The second problem is a generic problem in the analysis of complex user-machine system. What kind of summary statistics are appropriate for describing operator skill? The problem here is that the desired performance is specified as a goal, not as behaviour. Since there are many behaviour patterns any of which can satisfy the goal, averaging is not appropriate, since the result frequently approximates noise. But to record merely a catalogue of individual behaviours is clumsy, and also fails to capture the fact that something is in fact present as a common factor in all performance - namely satisfactory achievement of the goal. The problem is how to represent the core of behaviour.
Various methods will be described for overcoming this problem of representing complex dynamic behaviour, and their virtues and drawbacks discussed. The overall aim is to suggest improved ways to investigate process control and supervisory control.
Get full access to this article
View all access options for this article.