This research was aimed at the needs of process plant operators; not designers or process control people. The control of large industrial process plants has always been challenging and in the last 25 years many plants have moved from predominantly mechanical to central computer control of remote circuits. Computer control has in most cases reduced manpower requirements per unit production and raised the quality of the plant product and this has been achieved by the installation of a large number of sensors from which real-time information is transferred to the computer. The process operator who was formerly dependent on hardcopy process schematics chart recorders in a central control room and on analog indication at a number of physically remote points around the process plant is now able to monitor most readings from the process on a computer screen.
This technological evolution has resulted in larger and more process plant being put under the control of a single operator than was formerly possible. Control has in a sense remained manual because the process operator still performs the reasoning and controlling of the plant albeit by remote electronic/electric control elements. The rate of operator decision-making, level of responsibility and work stress levels have concurrently increased with the process operator using the computer as a conduit or tool, thus requiring process operators to be more experienced and better trained than ever before. These increased demands on process operators have in part been compensated for by more automation, but even so, support for operator reasoning has not kept pace with the volume of incoming process information and this needs to be addressed.
Process operators have the greatest need for support when things go wrong and this typically occurs at times of process change, the most common being during the process startup period. On many plants, process automation is able to cater for process change but there are two cases where operators will always require support. One is when a control system fails and the second is where the process cannot be automated. In these cases process operators need to close the logic control loop themselves and may require urgent help via computer displays to visualise faults and grasp complex situations. Consequently, this research concentrated on the provision of advice to the process operator during process plant startup.
Screen presentations originally stemmed from traditional hardcopy process schematics and there has been significant research into improving screen presentations in the operator environment by Rasmussen (1986) and others but they did not specifically support decision making by the provision of advice or expert systems.
Fundamental consideration indicates the need to understand process plant operator thinking. Little is known about the mechanism of thought and the conscious mind (Foder 2002) and this is considered in this work. The field of cognitive science encompasses contributions from psychology, computing, engineering, science and the arts and it also explores the human factors of screen displays. Contributions on mental imagery from the neurosciences also supported the direction taken in this research.
At the time that research for this thesis began, operator support in the form of expert systems was limited by the available software. Consequently the author created a novel advisory expert system with a suitably flexible architecture for research into new advisory systems for operator support. This architecture provided for unrestricted computational modelling and the use of photo images to support the process operator. Browser software for use as an interface only became suitable for this purpose in 1998. Research was conducted on a pilot-sized batch sugar crystalliser pan, utilised as an evaporator. Sugar crystallisation is an example of a process that in the foreseeable future will still be dependent on the human as a sensor to close the process control loop.
The author's work has combined aspects of process engineering, expert systems and human factors to make the following contributions:-
The development of a software architectural framework for a novel advisory expert system framework from separate, commonly available portions of software, each of which was not normally associated with expert systems, to provide predictive support in real time for process operators.
The novel use of photo images in a new ‘train of thought’ advisory expert system, thus named because it was intended to follow the logical train of mental imagery in the thoughts of the process operator. It sought to alleviate complete dependence on process screen schematics that required extensive training and experience to interpret. This new advisory expert system communicated information on demand to process operators, using a predictive model in real-time without the need for a large database and consequent long search times. The use of photo images are closely aligned to mental imagery and are proposed in this thesis as an future alternative to the symbology in use on traditional process screens.
The novel use of a predictive model of the evaporation process that described the dynamic startup phase in a batch sugar crystalliser, including the verification and optimisation of the model.
The assessment in an industrial environment of the proposed new ‘train of thought’ advisory expert system framework.
The new advisory expert system used the predictive model to attempt to convey knowledge of unacceptable process deviations to the process operator before the fault actually manifested itself. Diagnostic information was then presented to the process operator on demand with reduced dependence on formal ‘language’ interpretation. ‘Language’ in this sense refers to graphical illustrations, process charts, terminology, line diagrams and the like, which are used conventionally to communicate process states and conditions to the operator. This work found that the use of photo images is likely to reduce the possibility of misinterpretation and process operator error.
The new advisory expert system could not reasonably be assessed by students or by the lay population because they lacked the experience, background and environment associated with process control. Eighteen subject process plant operators actually working on control systems in industry therefore completed the assessment. Thirteen had more than 5 years experience (eleven had up to 20+ years) in various industries, three had less than 5 years and two had limited experience.
Assessment of the new advisory expert system incorporating photo images used an agree/disagree scale with before and after questionnaires. There was unanimous agreement that the process plant was seen in the ‘mind’s eye’ when making a process control decision and that the new advisory system helped the subject operator to see the plant in the ‘mind’s eye’ as a form of mental imagery (Kosslyn 1995, Pylyshyn and others 2002). This therefore indicated that the ‘train of thought’ advisory system using photo images was appropriate for process operators.
Objectively it could be stated at a 0.05 level of significance that in the opinion of the subject operators, the new advisory system was quicker at process fault diagnosis than the conventional system. Unfortunately, agreement that the predictive model provided adequate process decision-making support could only be obtained because it did not feature prominently in the assessment.
A majority of subject operators agreed that the new advisory system was subjectively more user-friendly than the systems on which each of them currently worked and also that it helped broaden their understanding of the conceptual needs behind screen based control. After viewing the new advisory system, there was also increased realisation that individual experience was currently the process operator’s sole form of decision-making support.
Comments made by some subject operators during assessment indicated that they did not favour change or the ready provision of expert information. Comments made at the supervisor level were cautiously optimistic.