Multiplexed VDT Display Systems : what is Good Practice
This paper raises a large number of problems about the design of multi-format VDT systems for supporting the human operator, with particular emphasis on :
- the consistency of all the different codes that may be used : such as colour, shape, physical position. There needs to be a consistent 1:1 mapping between code and meaning, for automatic interpretation skills to develop.
- proving all the information needed to do a task, so as not to overload fallible working memory while changing from one display to another.
Note that the emphasis is on a framework for good practice, in the sense of the issues which need to be considered, not on detailed recommendations for how to do it.
Note - as usual the word 'skill' is used to mean level of expertise, not a specific type of cognitive processing.
Topics :
1. Introduction.
1.1. The general tasks and skills in using such a system.
2. Perceptual-motor skills.
2.1. Assigning meaning to simple display codes.
2.2. Executing an action to change the process or the interface format.
2.3. Finding the display or control required, finding the next format required.
2.4. Pattern as an aid to interpreting present system state.
2.5. The number of different display patterns/ formats.
3. Familiar cognitive skills.
3.1. Data disruption.
3.2. Information available at the same time.
4. The knowledge base for problem solving.
4.1. Types of device-task knowledge.
4.2. Use of high-capacity long-term memory.
4.3. The displays needed by the operator.
References
Multiplexed VDT Display Systems : what is Good Practice
Lisanne Bainbridge
Department of Psychology, University College London.
In Weir, G.R.S. and Alty, J.L. (eds.) Human-Computer Interaction and Complex Systems. Academic Press, London, 1991, pp. 189-210.
1. Introduction
Operating a large, complex and changing system such as an industrial process, using an interface of computer generated displays on VDT/VDU screens, is a complex task. The flexibility of the interface technology means that many types of display format are possible, so these formats may be designed specifically to support the user. But there are disadvantages to this technology as well. Typically, a computer based display system has three screens. In a multiplexed system, the user has to choose the three display formats to look at, from perhaps 300 potentially available different formats. The user therefore has only a small 'window' on all the data about the present state of the system. And the user has the extra task, in addition to operating the process, of operating the display system.
This chapter concentrates on the design issues involved in minimising the difficulty of the extra tasks given to the user by this type of interface. The main theme will consider the features of a display system which make it possible to use the most efficient human cognitive processes for doing a task, that is, which encourage the development of [3 types of] skill. A subsidiary theme considers whether skills are best supported by using a minimum number of different display formats, and if so, what these formats should be.
In complex modern systems, most of the simpler well understood tasks of operating the process are done automatically. The human operator is expected to deal with unanticipated situations. These require understanding and problem solving, rather than following predetermined sequences of activity. Classic ergonomics principles for interface layout are concerned with minimising the physical effort of doing pre-specified tasks. Different principles for organising interfaces are needed to support user understanding and problem solving. Unfortunately most of the studies which have been done of real complex interface systems are commercial confidential, not available to the general public. An interesting exception is Kautto (1984).
The question of designing optimum display formats for complex tasks is also discussed by Woods (this volume). In general, this chapter will discuss interfaces displaying basic data, but will not cover displays which depend on computation or on intelligent decision support systems. The aim of the present chapter is to provide a framework within which good ergonomics practice in the design of multiplexed display systems can be understood.
1.1. The general tasks and skills in using such a system
When discussing how to support an operator's tasks by interface design, it is useful to have a simple framework for the types of task involved, because different design principles are relevant to different aspects of the overall task. This chapter will refer to the following:
1. operating the process plant.
* understanding the present plant state from the displays.
* choosing a response.
* understanding and using knowledge about the underlying system function, in unanticipated situations.
2. using any one interface format.
* identifying a particular display or control.
* finding the source of information required,
and assigning meaning to the display.
* finding the means of control required.
and executing the action chosen.
3. 'navigating' in the 'library' of alternative display formats.
* choosing the next format required.
* finding the next format required.
* making the action to obtain the next format required.
The main focus of this chapter is on how to design a multiplexed interface so that the main tasks, of operating the process, are not disrupted because the operator has problems with the subsidiary tasks of using the interface. The approach is to ensure that the interface operating tasks can be done by simple skills.
Types of skill
The general notion of 'skill' is that, after practice, people come to use behaviour which is most appropriate to the context, and most efficient in its use of mental information processing resources. This chapter suggests four main ways in which human information processing can be more efficient, which are summarised by the schema in Figure 1 (Bainbridge, 1989). These types of skill will be described in increasing order of cognitive loading.
Figure 1. A schema for the main mechanisms underlying different types of skill (Bainbridge, 1989).
Perceptual-motor skills. Here, responses are made 'automatically'/'unconsciously'. These responses do not need conscious attention so do not use working memory, which is limited in capacity. So these skills make minimum cognitive demands. For someone to be able to develop this type of skill the environment must be consistent, there must be a 1:1 mapping between display and meaning, display and response, or action and result. And these skills can only be learned by doing them. This chapter will distinguish for convenience between perceptual skills for taking in information, and motor skills for taking action, but they both depend on this basic consistency.
Familiar cognitive skills. For these, a well-established successful and goal-related strategy, and the information which it needs for reference, are already available. Familiar cognitive skills can develop when there are stable task situations but not a 1:1 mapping between stimulus and response, so the person has to compare, collate, calculate or choose to find the response. Once a standard strategy for doing this has been learned, the person no longer needs to devise new working methods by problem solving, which is the slowest and most capacity loading type of information processing.
'Prototype' using skills. A person uses this type of skill when they respond to a situation by referring to what is done in a typical situation of the same general type, or by referring to memories of what was done in a previous similar situation. So this is a type of analogical reasoning. Although people do frequently use this method of dealing with new situations it will not be discussed further here. Computer-based displays are not typically designed to support this type of thinking. Perhaps developments in object based programming and interfaces will give some insights in this area.
Problem solving skills. These are used when someone has to devise a new method of working, or to understand some unfamiliar information. People solve problems by working in a 'problem space', the knowledge base used in devising a new method or understanding. So problem solving can become skilled, that is more efficient, when someone has adequate knowledge, organised and accessible in a way appropriate to the type of problems.
Problem solving is not an independent type of cognitive activity, in the sense that it is only involved in distinct types of task. Problem solving may be needed at any level of task complexity, when the person does not know what to do. So it might be used by someone searching for a particular display on an unfamiliar interface, or by a commissioning engineer controlling a new process on the basis of fundamental factors underlying process behaviour. The need for problem solving depends on how stable the environment is, and how familiar the person is with the structure and variability of the working environment, the task, the device, and the interface.
Problem solving is a high workload activity, using limited capacity serial processing. It is also most needed in unfamiliar or unexpected situations. So it is important to minimise workload in these situations by maximising the amount of the task which can be done by perceptual-motor or familiar cognitive skills.
Maximising the possibility of developing one of the first two types of the skill is the basis underlying most ergonomics/human factors interface design principles. The next two sections of this chapter will discuss the nature of these two skill types more fully, and the practical implications for good design of multiplexed display systems to support these skills. The final section of the chapter is concerned with supporting the operator's knowledge base used in problem solving.
2. Perceptual-motor skills
Automatic skills can only develop when there is a 1:1 mapping between a 'code' and its meaning, using the word 'code' to refer to anything, other than language, which has meaning. This means that, for a user to be able to develop perceptual-motor skills, all codes on an interface should be unambiguous (only one code per meaning) and consistent (only one meaning per code). This principle affects all the task types listed in the introduction. These tasks will he discussed in increasing order of complexity, considering first simple one-dimensional codes, then the multidimensional coding involved in locating items and in pattern recognition.
2.1. Assigning meaning to a simple display code
Interpreting the meaning of a simple code can be done automatically, as a perceptual skill, if there is a 1:1 mapping to its identity. There are many reasons why it is difficult to maintain this 1:1 mapping in a complex interface.
Simple codes can be used with many types of meaning.
Examples of colour codes :
The meaning of a code may be an identity, such as :
dark blue steam
light blue water,
or a level of importance, such as :
red danger
blue emergency services,
or a state, such as :
red shut
green open,
or guidance about using the interface, such as :
light blue click here for more information.
Interface users will not be able to interpret an interface automatically if a code has several potential meanings. For example, if the meaning of a shape depends on the particular context, a user has to remember this context before they can interpret the shape, and error rates may be high. This interpretation uses working memory, and so will interfere with other task thinking.
Colour in particular is much overused as a code.
It also introduces another problem. Red and orange appear brighter than other colours (in daylight), and brighter lights attract attention. If these colours are used simply to attract attention, this will detract from their interpretation as having meaning. When easily seen colours are used with unimportant meanings, the eye will be attracted to these colours first, rather than to the more important meanings of less intense colours.
Multiplexed displays systems can increase the coding problem. In these systems there is usually a large number of different display formats, each of which may use shape or colour with a different meaning, or even with several meanings on the same format. Increasing the number of different formats increases the difficulty of using codes with unique, or even few, meanings, and so reduces the possibility of the user learning to make automatic interpretations.
In practice, the designer needs to make a table which lists all the meanings of each code used. The aim is to minimise the number of meanings for each, especially for codes which can indicate danger, urgency or importance. When a code is used with multiple meanings, then if possible there should be other information which makes it clear which of the alternative meanings is intended.
2.2. Executing an action to change the process or the interface format
People learn to make actions automatically, by motor skill, by learning the relation between the effect of an action and the visual, heard, or feel and touch, information they get while making the action. Again learning involves a consistent 1:1 mapping, between the information received while making the action and the result of the action (the display-control ratio). After learning, people know what sensory information implies what result, so they can make effective actions with the minimum of checking and correction. This greatly reduces both their workload and the disturbance to the system being acted on. But again, maintaining the 1:1 mapping necessary for the development of these skills may not be simple in practice.
Some computer-based interfaces do not have a consistent mapping between an action and its effect. Suppose the same physical device (mouse, tracker-ball, keys) is used both to make control actions on the process, and to access the display system. There is then no consistency, either in the identity of an action, what type of action has what type of effect on the system, or in the size of the effect of an action, the mapping between the displacement, friction or force of an action and its effect. So none of these dimensions can be unambiguously assigned to a meaning in terms of the effect of the movement. The user is unlikely to be able to develop automatic motor skills in using the interface, and the main task thinking will be interrupted by thinking about how to make actions.
A frequently used solution to this problem is to use conventional controls (i.e., physically unique controls in fixed position) for the main process variables, with computer based displays. This is a compromise solution, as it is difficult to make the layout of display and controls compatible (as discussed in the next section). Choosing conventional controls as part of the interface design depends on whether it is important for the operator to have well developed skills for controlling some part of the process, even though these skills may be used rarely.
2.3. Finding the display or control required, finding the next format required
Looking to the display from which information is required, or reaching to the control to be moved, can be done automatically after learning, if each display and control is always in the same place. These 'acquisition' movements again are motor skills, which can develop when there is a 1:1 mapping between location and identity.
This 'location coding' is possible on conventional hardwired interfaces. On computer-based display formats, there is flexibility in the position of items, which has advantages which are discussed below. But this flexibility may be paid for by losing the advantage of automatic search and acquisition. If each item is in a different position on each display format, then it is not possible to develop the motor skills of automatically looking to or reaching to the display or control needed. Instead the user has to search for an item, keeping in working memory what item is wanted, and comparing it with items seen until the required item is found. So working memory for the main task items can be interrupted.
There is a long history of ergonomics studies which have shown that if two layouts which are used together are not the same ('compatible'), then the user takes longer to find an item, and makes more errors in doing so. For example, Lazzari (1988) asked people to search for an item on one VDT graphic format, and then to find the same item in a different position on a different format. People took on average 24 seconds to find the second item in a format which had a structure which was not clear to them.
If the item which varies in position is a control, then reaching to it, once its position has been found, is an action whose accuracy has to be checked visually, so the eyes have to look away from the main task information.
These effects suggest the practical recommendation that, when stable positions of items cannot be maintained, then at least items should have consistent relative positions in as many formats as possible. In this way, they have a stable topological relation to other items, being consistently above or below another, or to the right or left of it. So it is possible for the user to search automatically for the required item in the appropriate direction.
The same reasoning leads to practical recommendations about how an interface user should tell the computer which display format they want next. To be able to use the interface system, the operator needs to know what other display formats are potentially available, where the current format is in relation to the other formats, and how to get to the other formats (see Norman et al, 1986; Canter et al, 1986; Schneiderman, 1988).
Suppose the user wants more detailed information about some part of the process. If they indicate this by moving a cursor on the existing display format to where it shows some information about this item, or by touching this item on a touch panel on which the items are in the same relative positions as on the display formats, then there is a consistent mapping between the layout of the existing displays and the layout of the set of alternative displays. So asking for another display format can be done with the minimum of cognitive effort.
If instead the user has to ask for another display format by pointing to an alpha-numeric menu on which items are in an arbitrary order, or by typing in some (perhaps arbitrary) alphanumeric code, then they will have to 'think about' what they are doing, which will disrupt the main task thinking. Any alphanumeric menus which are used should be explicitly displayed, obvious in meaning, and laid out in a consistent order.
2.4. Pattern as an aid to interpreting present system state
People, as compared with computers, have a high capacity for parallel processing of information presented in patterns. In patterns of items, not only individual items but also the items together have meaning. Patterns are especially good for displaying relational information. Two types of multidimensional meaning are important in interface design: interpreting the present system state, and understanding the underlying system function. People can learn to identify the present system state by pattern recognition, even when the interface has not been specifically designed to support this. Helping people to understand how the process works does require special layouts.
People can learn the meanings of arbitrary patterns of information. Items can be interpreted as a pattern even when their relative positions do not have any meaning. For example, the layout of labelling can affect the ease of searching an alphanumeric format (Mann and Schnetzler, 1986). Or, suppose that in some industrial process : each type of fault appears as a unique pattern of pointer positions on the instruments on a conventional interface. Then there is a 1:1 mapping between a pattern of pointer positions and a process fault, and the operators can learn to do fault diagnosis by perceptual-motor skill (Shepherd et al, 1977). In general, when a process is sufficiently simple for each of its different states to appear as a unique combination of its variable values, then state identification can be done by perceptual-motor skill. Computer generated displays should be designed so that they are not too cluttered for this pattern recognition to be possible.
Although people can learn the meaning of arbitrary consistent patterns of information on a process interface, studies show that operators can identify the overall state of a process plant more easily if information about individual variables is presented in the context of a mimic/ schematic of the plant. The parts of the plant are represented by simple symbols, shapes which either look somewhat like the part, such as distillation columns or heat exchangers, or whose meaning has been established by convention, such as pumps and valves. (These symbols should also be well designed, see e.g., Easterby, 1970; Jervell and Olsen, 1985). The symbols on a mimic are connected by lines representing the flows through pipes between the parts of the process. So the mimic, as a background to the display of process variables, provides a context of reminders about what the parts of the plant do, and how they are connected together. (And these shapes also should not be used with multiple meanings.)
This display method of providing a context for interpreting information about the present state of a process is, within limits, much easier to achieve with computer generated graphic displays. It is important to remember however that understanding a mimic has to be learned. To someone who does not know how a process works and what its parts are, a mimic is simply confusing (Lazzari, 1988). The pattern recognition involved, the perceptual skill, is learned after and still underpinned by, a stage of understanding the separate parts. It does not start by being automatic, but it can become the basis for automatically interpreting the state of the process, or automatically finding the value of a particular variable without searching for it. Only a person who is familiar with a mimic and with what it represents can interpret it easily, especially when mimics are distorted to fit them onto a computer screen. A person with this perceptual skill has developed a 'frame' for interpreting the meaning of patterns.
More usually pattern recognition is not sufficient for understanding the process state, and a process operator can only identify the present process state after doing some 'thinking', comparing and collating data, etc. This involves familiar cognitive skills, as discussed in the next main section.
And in more complex plant, fault diagnosis can only be done by reference to some understanding of how the process works. A pattern-based interface can support this underlying understanding of the general functions of the process, as well as aiding identification of its particular state at any one time. The relative layout of items, and the type of representation, can give information about relationships underlying the variables displayed. For example, a layout to support understanding might be :
understand cause-effect relations : oil heating affects steam generation rate
layout oil steam.
Compare this with a layout to support following a given sequence of behaviour :
task if steam flow changes use oil flow control.
layout steam oil.
This issue of design to support understanding will be approached in the final section of this chapter.
2.5. The number of different display patterns/formats
Although people can learn to deal with pictorial formats automatically, by perceptual skill, we do not know how many different formats people can handle automatically. It is unlikely that the number is large, another reason for reducing the number of different formats. In a display system with 300 different formats, it is unlikely that the users can learn each sufficiently well to be able to:
* automatically find a required item without search.
* automatically remember the correct way of interpreting the codes used.
* automatically integrate the given information into an overview of the system state.
But how many formats can people learn to use easily, is it 5, 10, 20? and how does this change with experience? This sort of information would be very useful, but is difficult to study experimentally.
One aim of increasing the number of different display formats is to make each format task specific. If each sub-task has a specific display, then the argument is that each format can be specially designed so that what to do is 'obvious', i.e., it should be possible to do the task automatically, by perceptual-motor skill.
But if it is not possible to learn a large number of formats sufficiently well to respond to them all using perceptual-motor skill, then many of the formats can only be used by searching and thinking about them, that is by problem solving. Ironically, increasing the number of different formats, with the aim that the task can be done by perceptual-motor skill, actually has the opposite effect. The approach of increasing the number of formats considers the mental workload of doing each sub-task in isolation. If however one considers the cognitive processing required by the sub-task done within the context of a complex display system, this suggests the opposite conclusions about interface design.
At present it seems that the best recommendation is to avoid these problems of reducing the possibility of using an interface by perceptual-motor skill and increasing conscious attention or problem solving. The main approach would be to use a minimum number of different formats, on which codes have consistent meanings and items are laid out so they have a consistent topological relation to each other.
For example, in displaying information bout a plant with a relatively simple structure, it may be possible to represent the process at different levels of detail. (In a complex interacting plant it may be difficult to divide the process into relatively independent parts.) As formats should be similar, to maintain perceptual skills, so displays at different levels of detail should have the same general topographic layout. The question of the best basic format to use is addressed in the final section of this chapter.
3. Familiar cognitive skills
Operators use well-established strategies to understand and control industrial processes in familiar situations. This chapter will not discuss decision support systems, in which the computer does the computations and displays the result (see Hollnagel et al, 1986; Lehner and Zirk (1987); Rouse et al ,1987; and Woods, this volume).
The cognitive strategies use working memory. The problem is that working memory is easily disrupted. The key recommendation for minimising this disruption is to ensure that all the information needed in any one decision is displayed at the same time, and within the same perceptual interpretation. This has several implications for display system design.
Figure 2 : An example strategy. In a steel works, as part of the task of allocating available electric power between five steel-making furnaces, an operator works out which furnace will be least affected by a power change.
Figure 2 shows an example strategy. The task structure is a hierarchy of goals, and the sub-goals for meeting them. Such a strategy can be thought of as providing a 'framework' in working memory, which structures the task information processing, by :
* structuring the task goals and sub-goals.
* structuring the mental picture of the current state of the task and the plant, relative to the task goals.
* specifying the information processing to be carried out.
* directing attention to which information is needed.
This structure could also be called a 'frame', but to clarify the argument this chapter will use the term 'frame' for the structures of pattern recognition done by parallel processing as an automatic perceptual skill, and the term 'framework' for the structure of task goals and ways of meeting them which is built up in familiar cognitive skill.
The practical problems with supporting familiar cognitive skills are that such multilevel frameworks and the data remembered with them are easily disrupted.
3.1. Data disruption
The data is easily disrupted. Suppose that an interface user has to remember something from a previous format, which they will compare with something on the next format, when they have found it. Search studies (e.g., Lazzari, 1988) show that people take 20-25 seconds to find something on an unknown format. Classic memory studies (e.g. Posner and Rossman, 1965) show that if a person has to remember an item for 25 seconds, during which time they are doing something else (in the display case - searching for another item), they will make about 30% errors in remembering the first item. So it is important that all the information needed in one decision can be identified quickly.
3.1.1. Mental task structure disruption
And the mental picture of the task structure is also easily disrupted, particularly by having to use poorly designed interfaces. Operators do not seem to have much difficulty with keeping track of these multiple levels of task goals and information if they are using an interface on which :
* all the information needed when working through one strategy is available at the same time.
* the information and controls are in fixed positions, so can be acquired by perceptual-motor skill.
*the displays are in an unchanging format, so can be interpreted automatically using a single perceptual 'frame'.
But suppose the operator doing the task described in Figure 2 is using a multiplexed interface on which :
* one format shows the electrical power usage.
* another format is used for making control actions.
* there are five furnace formats, one for each furnace.
* to get from one format to another it is necessary to type in an arbitrary code number, which is not given on the format which can be seen.
An operator using such an interface has several difficulties which are added to the difficulties of meeting the main task goals. The operator has to remember charge times (which are trivial from the point of view of the main task decisions) for extended periods of time while looking through the furnaces' formats. This puts an unnecessary load on the limited capacity of working memory for arbitrary data items. The operator also has to change rapidly between task frameworks and perceptual frames. There are two task frameworks, for controlling the furnaces and for using the menu system, and three perceptual frames, for finding the data required on each of the three different formats. As the operator has to alternate between these it is difficult to maintain a stable overview of the main task. For other discussions of the problems of keeping track of where one is in a large library of different formats (see e.g., Woods, 1984; Barker et al, 1987).
Designing to maintain the integrity of working memory therefore has two aspects, which will be discussed in more detail. All the information about the process state which is required in any task should be available at the same time, to minimise the data load on working memory. And supporting familiar cognitive skills leads to the same recommendation as for supporting perceptual-motor skills : there should be a minimum number of different display formats, and the 'frames' for interpreting the display formats and operating the interface system should as far as possible be the same.
3.2. Information available at the same time
If all the information about the process state which is needed in a given task is not available at the same time, the effect of delay and interference on working memory data capacity means that there will be a high probability of human error.
If the information needed at one time is incomplete this is also a source of task stresses, particularly in fault situations, from :
* not knowing what is happening in all parts of the process.
* the risk of giving up the information which is currently available, in order to find out something else.
* the time costs, as well as the disruption, from having to change display formats.
There are two practical aspects to making all the required information available at the same time: what information actually is required at the same time, and how many screens are needed to convey this information.
3.2.1 Information needed at the same time
This chapter suggests three approaches to identifying and grouping task in formation, which depend on the extent to which the task strategy can be pre-specified.
If the task strategy can be pre-specified, as in Figure 2, then classic ergonomics Task Analysis techniques may be used to identify what information is needed for any one sub-task. These displays will be working-method-specific.
If what needs to be done cannot be pre-specified, then the operators instead need help with knowledge about how the process works, from which they can think out what to do. For purposes of the present discussion, there are two main ways of grouping information which are not specific to a particular working method :
* displays which give information about all the parts of the process which might be used in meeting any one main task goal/process function, such as maintaining cooling or pH.
* displays which give general information about the plant, not related to a particular goal or working method.
There are some situations in which the choice of display type is clear :
*When the operator must do specific operations, perhaps in a particular order, as in start-up or shutdown, then method-specific displays are appropriate (though beware how they relate to displays used during ongoing operation).
* In a fault situation, once the main function contributing to plant integrity which has failed has been identified, then it may be appropriate to use function-specific displays, to support achieving that function in other ways.
Otherwise there are good reasons to ask whether it is a good idea to use several different approaches to grouping the information on displays. It does, for example, increase the number of different display formats, and so the number of different perceptual frames needed.
The operator of a complex industrial plant is usually within the control system to deal with unanticipated situations, which essentially require problem solving, thinking of new approaches. If method-specific displays are used for tasks which can be pre-specified, then the operator will have little practice with using the device-specific displays supplied for problem solving, and so will not develop the perceptual frames for interpreting them automatically. Thus when the operators have the most difficult type of process operating task, they will also have problems with using an unfamiliar interface. Also it is often not possible to predict when someone in a complex situation will need to change from using familiar strategies to problem solving.
Another argument in favour of method-specific displays is that people make errors when they think for themselves. But people also make errors in following instructions, and people are only good at recognising and recovering from their errors if they have sufficient task information and understanding to know that the result of their actions is not what it should have been. And people who follow instructions are frequently given the higher level instruction that : they should assess whether it is actually appropriate to follow the given instructions in this particular context. So people who are following instructions actually need to understand their task in order to do so effectively.
Both these arguments suggest that it is better to use a more general display format, even in situations for which task-specific displays could be developed, so that the user develops the perceptual-motor skills of accessing and interpreting this information.
3.2.2. Minimum number of display screens needed
Previous sections mention the importance of displaying at the same time all the information about process state which is needed in one sub-task. If this information will not fit onto one computer screen, then sufficient display surfaces are needed to display all the information. This is not the only reason for increasing the number of screens. The operator may need to refer to more than one type of information, and these types of information may be sufficiently different to be best displayed by different formats (see next section).
The number of VDTs needed can be reduced when any of this information can be displayed in printed form rather than on screens. This is particularly appropriate for alphanumeric information and for stable reference information. When alarm lists are on paper, the whole of the alarm list can be looked at, particularly if the typeface on the printer is easier to read than the typeface on the screen.
Operating procedures are one type of reference information. But there are advantages in using computer based displays for these. Variable values, and facilities for making control actions or accessing other display formats can be included in the listing. The procedures may also be updated relatively easily. The disadvantage is that this does require at least one screen, during fault situations when much other information is needed, and one screen gives access to only a small part of the procedure at any one time. Useful guidelines for displaying procedures can be gained from other work on displaying sequences of instructions, such as Green, 1982; Dewar and Cleary (1986).
Also people operating complex systems often do more than one task at a time, such as diagnosis and control during fault management. Reinartz and Reinartz (1989) found up to 10 task goals being considered at the same time). This means that screens are needed to display all the information needed for all these tasks, for example:
* to cover the number of simultaneous and independent faults which can occur. The basic number of screens needed to describe one part of the process needs to be multiplied by this number.
* to be able to sample information about other parts of the process, which may change while one part of the process is being displayed.
This multitasking is another argument for using general purpose rather than task specific displays. A general purpose display gives the information used in several tasks, and provides a context for seeing any interdependencies between the tasks.
Some of the problems of keeping track of several parts of the plant can be resolved by providing an overview mimic. This could be a dedicated VDT or a wall mimic. A wall mimic has the advantage that it can be hardwired so larger. An overview mimic may also help with the general question, whether operators using an interface of a small number of screens, and therefore having a small amount of information about the total state of the process at any one time, can develop an integrated mental model of the whole process.
If the interactions in the process are such that it is not possible to display simultaneously all the important influences on, or effects of, a particular event on the available VDTs, then it is necessary to consider whether this type of display technology will lead to too high a probability of human error.
And if the number of screens needed is too high for one operator to keep track of simultaneously, then it is necessary to increase the number of operators.
4. The knowledge base for problem solving
To support problem solving, thinking of new working methods in unusual situations, (the final type of mental skill), the operator needs to know how the process works, presented in displays which do not constrain the type of thinking used. These displays cannot be based on a classic type of Task Analysis which concentrates on working methods. Instead it is necessary to analyse the required structure of information about the plant and the operating goals. A major practical problem is that there is a large number of different types of knowledge which could be displayed to the operator. If the aim is, as suggested in previous sections, to reduce the number of different formats, then it is important to ask which formats would best support the operator.
Figure 3 : Some of the links in the knowledge structure about a small part of a complex process.
(thick line = cause-effect relation).
Note the 3 types of relation between items : is-a, part-of, descriptor.
[The original Figure in the paper is an example from process control, this is from flying.]
4.1. Types of device-task knowledge
A list of the different types of knowledge that an operator uses is quite long, and makes it clear that it is not practical to display all the potential information about how the process works. Figure 3 shows some of the knowledge structures that a low level part of a process, such as a pump, can be involved in.
The knowledge involves several 'hierarchies'/networks of networks, which do not have a simple relation to each other.
1. There are at least six 'part-of' hierarchies :
* the task goals and sub-goals : the reasons for using the process, the product criteria, and the human activities by which these are met.
* the process goals and constraints which arise for reasons of safety or efficient productivity.
* the operating procedures.
* the physical structure of the plant.
* the functional structure of the plant : the cause-effect relations, whether implemented by mechanical/chemical means, or by electronics/ software ones.
* the fault tree : and associated failure probabilities.
There are well known techniques for representing most of these types of hierarchy, and for identifying sub-groupings of items which work together so must be displayed at the same time.
There has been recent interest in describing functional effects in a process. Cause-effect relations can be described by signal-flow graphs, as in Systems Theory textbooks on network theory. The technique includes methods of mathematical simplification, for identifying the sub-graphs in a process, and the interactions which cannot be further simplified.
Mass-energy flow techniques (e.g., Lind, this volume; Duncan et al, 1989) may be useful to simplify the description of the functions in a nuclear power station, which essentially consists of a series of heat transfers, but are not rich enough to describe chemical plant in which aspects such as pH, concentration or physical state must be maintained.
In addition, most of these part-of hierarchies are not simple trees. Items at a higher level are not a simple grouping together of items at the next lower level. Instead their organisation is essential. For example, within a cooling circuit there is only a limited range of places where a pump can be located for it to carry out its function adequately.
2. Categories of item :
Most items are also members of another type of hierarchy, an 'is-a' hierarchy. They are members of a general category, such as 'pump' or 'turbine'. The category contains general information about this type of item. This sort of information is usually assumed to be supplied by the operator from memory, and is not displayed.
3. Descriptors : Information about the state of the process :
* the actual variable values, or multidimensional states, in the past, present and predicted future.
* patterns of changes over time, the parameters of transient and steady-state responses.
* sequences of events : process phases, event trees.
4. Descriptors : The physical properties of an item :
The colour or geographical position of an item may have nothing to do with its function in the process, but may still be useful to know, if it helps in identifying or finding the item.
5. Events (not represented in Figure 3) :
Memories of past events and typical incidents, which are referred to in thinking about unfamiliar situations.
Obviously Figure 3 cannot be used as a framework for display design. Although it is complex, it does not show all the possible types of complexity. It does illustrate the complexity involved. Any one component of the process, whether at the level of individual pumps, or the level of major groupings such as heat exchangers, is a part of several hierarchies, or networks of interdependent items.
The display problem is that there is not necessarily any simple mapping between these hierarchies. That means that any one two-dimensional display format can only show a small slice through all this information. So what should be displayed ? Two aspects will be discussed: what need not be displayed because it can be supplied from the operator's memory, and what should be emphasised in the primary display formats.
4.2. Use of high capacity long-term memory
Compared with working memory, the capacity of human long-term memory is huge, and access to the stored information is relatively good, if it is accessed fairly frequently, and the person is given cued reminders. Obviously the present state of the process has to be displayed explicitly, people cannot be expected to remember it. But providing the underlying information about how the process works might be approached in three different ways, at 3 levels of explicitness :
* using 'external' memory : displaying explicitly all the information the operator might need about how the process works.
* using 'cued' memory : giving sufficient reminders on the displays, so that experienced users can remember relevant backup information.
* using 'recall' memory : expecting the user to remember without help everything about how the process works, what the operating constraints are, and so on. This is the approach on most conventional interfaces.
Taking advantage of cued memory access to the users' own knowledge means that the number of different display formats can be reduced. So the question becomes : which of the above types of information can the display designer expect an experienced user to supply from memory?
4.3. The displays needed by the operator
Any representation of information should :
- make the most important points the most salient (easily seen),
- provide a structure for thinking about the related task,
- provide cues to remind experienced users about other relevant information,
- handle complexity by providing several levels of detail, and different representations which are optimum for different aspects.
Papers by Shepherd (1985), Mitchell and Miller (1986), Mitchell and Saisi (1987), and de Keyser (1987) discuss the information needed by the operator. Papers by Vermeulen (1987) and by Chechile et al (1989) discuss the usefulness of, and mapping between different types of knowledge and display. Many studies on how to help people to extract technical information from pictorial representations have actually been done in a different context, that of technical drawing, e.g., Rabardel and Weill-Fassina (1987).
In a pilot study of two graphic displays, Brennan (1987) asked people to explain events in a complex system. They were given system information either in a mimic or in a signal-flow-graph representation. The results suggested that the signal-flow-graph was more useful to people who did not already know how a particular system worked, and so needed information about the cause-effect links underlying the events. But the mimic display was better for people who did already understand the system represented.
The mimic diagram can give many cued reminders about other types of information, while the SFG represents only one of the many possible 'hierarchies' of information, and does not directly give cues about the physical structure of the plant.
There has been little research on how easy operators find it to think in terms of higher level abstract functions, but it is important to establish this before displays representing abstract functions are used as the operators' primary display format. The above findings suggest that mimics should be used as the primary display, with SFGs or box diagrams, which focus on function, as backup for dealing with unfamiliar situations. The irony of this suggestion is that while SFGs or box diagrams may be good in principle to support the understanding of unfamiliar situations, in an unanticipated fault situation the problem is that cause-effect relations are unknown, so it is not possible to provide an SFG or other cause-effect display.
The most promising current approaches deal with unfamiliar situations by sidestepping this issue, avoiding expecting the operator to think about diagnosis and fault management from first principles, and concentrating on process behaviour. This is possible because the operators' primary responsibility in fault situations is to maintain process integrity and stability.
'Critical function monitoring systems' identify the critical process functions which maintain system integrity, and then present information to the operator about which methods of maintaining these functions are still available. This is a 'decision support system' involving sophisticated computing, so is outside the scope of this chapter. Though it is relevant to mention that a complete display system for critical functions may raise the same issues of complexity as have been discussed in this chapter.
The 'state' approach to fault management is based on a system analysis. This identifies all the states which the process can get into which require a different response. It also provides general procedures for identifying which of these states the process is in, and for making the appropriate response. This approach may only be applicable to fairly simple processes, not to ones in which there are complex interactions, such as a train of distillation columns. Earlier this chapter pointed out that it is important not to give a restricted view of the state of such a process. The usual difficulty with complex systems appears - the more complex the process, the fewer methods of aiding someone to think about it there are.
This chapter shows that a few simple principles for using efficient cognitive processes, for maximising the use of perceptual-motor skills, and minimising the disruption to working memory, have wide ranging implications for many aspects of complex interface design. In summary, these might support the use of 2-3 levels of mimic/schematic displays, giving a permanent overview and more detailed formats with the same general topology of layout, and reminder cues about functions. Subsidiary formats would be used for goal/function integrity checking, and for procedures.
References
'This Volume' is Weir, G.R.S. and Alty, J.L. (eds.) (1991) Human-Computer Interaction and Complex Systems. Academic Press.
Bainbridge, L. (1989). Development of skill, reduction of workload. In: Bainbridge, L. and Ruiz Quintanilla, S. A., (eds), Developing Skills with New Technology. John Wiley &;Sons.
Barker, P. G., Najah, M., and Manji, K. A. (1987). Pictorial communication with computers. International Journal Of Man-Machine Studies, 27 : 315-336.
Brennan, A. C. C. (1987). Cognitive support for process control : Designing system representations. Master's thesis, University of London.
Canter, D., Powell, J., Wishart, J., and Roderick, C. (1986). User navigation in complex database systems. Behaviour and Information Technology, 5: 249-257.
Chechile, R. A., Eggleston, R. G., Fleischman, R. J., and Sasseville, A. M. (1989). Modelling the cognitive content of displays. Human Factors, 31: 31-43.
de Keyser, V. (1987). How can computer-based visual displays aid operators? International Journal of Man-Machine Studies, 27: 471-478.
Dewar, A. D. and Cleary, J. G. (1986). Graphic display of complex information within a prolog debugger. International Journal of Man-Machine Studies, 25: 503-521.
Duncan, K. D., Praetorius, N., and Milne, A. B. (1989). Flow displays of complex plant processes for fault diagnosis. In: Megaw, E. D., (ed), Contemporary Ergonomics, 1989, pp. 199-206. Taylor &;Francis.
Easterby, R. S. (1970). The perception of symbols for machine displays. Ergonomics, 13: 149-158.
Green, T. R. G. (1982). Pictures of programs and other processes, or how to do things with lines. Behaviour and Information Technology, 1: 3-36.
Hollnagel, E., Mancini, G., and Woods, D. D., (eds) (1986). Intelligent Decision Support in Process Environments. Springer-Verlag, Heidelberg.
Jervell, H. R. and Olsen, K. A. (1985). Icons in man-machine communications. Behaviour and Information Technology, 4: 239-254.
Kautto, A. (1984). Information presentation in power plant control rooms. Research Report 320, Technical Research Centre of Finland, P0 Box 516, SF-00101 Helsinki 10.
Lazzari, C. (1988). Efficacy of VDU graphic displays measured on variable mappings on two formats. Master's Thesis, University of London.
Lehner, P. F. and Zirk, D. A. (1987). Cognitive factors in user/expert system interaction. Human Factors, 29: 97-110.
Mann, T. L. and Schnetzler, L. A. (1986). Evaluation of formats for aircraft control/display units. Applied Ergonomics, 17: 265-270.
Mitchell, C.M. and Miller, R. A. (1986). A discrete control model of operator function : A methodology for information display design. IEEE Transactions on Systems, Man, and Cybernetics, SMC-16: 343-357.
Mitchell, C. M. and Saisi, D. L. (1987). Use of model-based qualitative icons and adaptive windows in work stations for supervisory control systems. IEEE Transactions on Systems, Man, and Cybernetics, SMC-17: 573-593.
Norman, K. L., Weldon, L.J., and Shneiderman, B. (1986). Cognitive layouts of windows and multiple screens for user interfaces. International Journal Of Man-Machine Studies, 25: 229-248.
Posner, M. I. and Rossman, E. (1965). TIle effect of size and location of information transforms upon short-term retention. Journal of Experimental Psychology, 70: 496-505.
Rabardel, P. and Weill-Fassina, A., (eds) (1987). Le Dessin Technique. Hermes, Paris.
Reinartz, S. J. and Reinartz, G. (1989). Analysis of team behaviour during simulated nuclear power plant incidents. In : Megaw, E. D., (ed), Contemporary Ergonomics, 1989, pp. 188 193; Taylor &;Francis.
Rouse, W. B., Geddes, N. D., and Curry, R. F. (1987-8). An architecture for intelligent interfaces: Outline to an approach to supporting operators of complex systems. Human-Computer Interaction, 3: 87-122.
Shepherd, A. (1985). Hierarchical task analysis and training decisions. Programmed Learning and Educational Technology, 22: 162-176.
Shepherd, A., Marshall, E. C., Turner, A., and Duncan, K. D. (1977). Control panel diagnosis: A comparison of three training methods. Ergonomics, 20: 347-361.
Shneiderman, B. (1988). We can design better user interfaces: A review of human-computer interaction styles. Ergonomics, 31: 699-710.
Vermeulen, J. (1987). Effects of functionally or topographically presented process schemes on operator performance. Human Factors, 29: 383-394.
Woods, D. D. (1984). Visual momentum : A concept to improve the cognitive coupling of person and computer. International Journal of Man-Machine Studies 21: 229-244.
Access to other papers via Home page
©1998, 2022 Lisanne Bainbridge
|
|
|
Comments
Post a Comment