Planning the training of a complex skill
This paper suggests ideas and issues which need to be considered in devising a training scheme for a complex task, rather than being a fully specified procedure for planning training.
Note : as usual the word ’skill’ is used here with the meaning of development of ability to do a task, from beginner to expert (what training is about), rather than as the label for a specific type of task processing.
Topic areas :
1. Introduction.
1.1. Typical operator thinking activities and knowledge.
1.2. Basic requirements for the training scheme.
2. Understanding the process and the task objectives :
2.1. Understanding how the process works.
2.2. Task objectives : product goals and plant constraints.
3. Main operating tasks :
3.1. Inferring the state.
3.2. Control.
3.3. Sequencing.
4. Fault management :
4.1. Dealing with unfamiliar situations by reference to familiar ones.
4.2. Working from first principles.
4.3. Exercises in dealing with unfamiliar situations.
5. Final comments.
References.
Planning the training of a complex skill
Lisanne Bainbridge
Department of Psychology, University College London
Le Travail Humain, Special Issues in Honour of Jacques Leplat, 1991, 56 (2/3), 211-232.
1. Introduction
Jacques Leplat has often written, with his usual wise combination of theoretical acuteness and practical experience, on the nature of complex skill (e.g. Leplat, 1989), on how skill is acquired, on how behaviour changes during this acquisition, and on training (e.g. Leplat et al, 1970).
The nature of skill and its learning pose a classic training problem : how to break a complex task down into parts small enough for a beginner to learn, and then how to help the trainee integrate these parts into the whole task. In many tasks there is an additional complexity: several different but interdependent types of expertise to be developed. In industrial process operator training (such as in power, steel and chemical works) an operator uses several kinds of knowledge describing the process and the task, and there is no simple relation between them. In earlier approaches to operator training, different aspects were taught separately, without integration or reference to a task context. This paper illustrates how the organisation of an operator training scheme can be based on an analysis of what the operators need to know about the physical and causal structure of the process, the task objectives, and their own thinking strategies, with the aim of making explicit the links between these different types of knowledge.
There have been major developments in organising operator training in terms of task objectives, e.g. Duncan (1974), Shepherd (1985). Such an approach is a considerable improvement on approaches to training which focus mainly on the physical structure of the process, on the basic physics and chemistry underlying the causal effects in the process, or on operating procedures. The latter approaches often leave the operators to work out for themselves what are the practical implications of the knowledge they have been given, and how the different aspects of knowledge interrelate.
But both task and process oriented approaches to training focus primarily on one type of knowledge. Studies of operators suggest that they have various aims in their thinking, which use different types of knowledge. So optimally the operators' training needs to build up simultaneously their knowledge of the process, the task, and how to do their own thinking. This should start at a simple level of basic understanding and practical strategies, then build up to more complex levels by iterating through the different types of knowledge.
The first half of this paper is on normal operation. This introduction will outline how an operator thinks, and the knowledge they use. The first two main sections of the paper will then be on training for understanding the process and its constraints, and how to operate it. Fault management is treated as a special case, in the second half of the paper, which outlines the categorisation and problem solving processes by which operators deal with unfamiliar situations. Any situation in which training is needed is by definition unfamiliar. However the aim of most training schemes (and the first part of the scheme proposed here) is to minimise the extent to which trainees have to think out for themselves what to do in this unfamiliar situation. But in training for fault management the emphasis shifts, to explicit training in dealing with unfamiliarity.
Many of the proposals in this paper have been incorporated in a training simulator and flexible instructional programme developed by Kassianides (1991). There are other practical descriptions of real large-scale simulator-based operator training schemes in Tompsett (1987) and Marshall and Baker (1989).
1.1. Typical operator thinking activities and knowledge
In complex industrial plant, operators do not just react to a signal, such as an alarm, by producing an associated response. They do much sophisticated thinking, to build up overviews of the present and future state of affairs in the process, and of the effects of possible actions, and to make future plans. Except in the simplest processes and situations, the operators need to understand enough about the process and the task to be able to infer the present inner state of a complex multidimensional process from incomplete external information, to predict its behaviour, and to plan how best to operate it. They also need to be adaptable about the methods they use to meet different goals. In more detail, the operators' main thinking activities (which are not done in a fixed order, and are not all used in all processes) are to : review the product demands; infer the present state of the process; predict events, and associated process states and actions; evaluate the present or predicted state against the demands; review the actions available and their effects; choose the best action, using multiple criteria; and to plan future activities (Bainbridge, 1992).
In doing this thinking, the operator uses various types of knowledge such as :
- the physical structure of the plant;
- product targets and plant constraints;
- (interface to process state) inferences;
- (process state to response required) associations;
- cause to effect relations;
- dynamic 'mental models' of how parts of the plant change over time;
- typical sequences of events in the process.
1.2. Basic requirements for the training scheme
A training scheme which focusses on the operator's cognitive goals, or on the operator's knowledge types, independent of the use to which they are put, would be as one dimensional as focussing only on the process or the task demands. These are different perspectives which the operator needs for different purposes, so a training scheme needs to develop them all and their interrelations. The main aspects of normal operation, which will be developed more fully in the main body of the paper, are concerned with understanding the process; assessing the task demands; and operating the process using the interface.
Providing the trainee with a learning context. Part-whole training is complex, because a training scheme has to build up simultaneously, from simple low-level components : the operator's understanding of the process, the task and the associated thinking activities. But there is a 'what first ?' problem because, if training starts with low-level parts of the task, the trainees will not see why they need to know them. So training needs to start with a simple overview of the whole process and task, to provide the context for later learning. This gives the trainees motivation. Without this it is difficult for the trainees to direct their attention appropriately, or to make the effort to understand and to learn. This general context also provides an initial framework for learning the links between the different types of knowledge. It is therefore important, at the beginning of training, to outline for the trainees the knowledge and skills which they will have acquired by the end of the training. It is also important, at the beginning of each section of training, to outline for the trainees what they should know by the end of this section of training, why they need to know this, and how it contributes to the whole task. And at each step in training the amount of new material should be restricted, so that the trainees retain a feeling of mastery, because confidence is an important factor in speed of learning.
Choice of training method. The best training method to use depends on the nature of the task, and of the trainees. For example, a different method is used in training someone to make a simple response than in training to solve problems. Different types of cognitive skill need different training methods (Bainbridge, 1989b), and different thinking strategies use different types of knowledge (Bainbridge, 1992). For related reasons, operators who will just be expected to deal with routine situations need different training than higher level operators, such as commissioning engineers, who need to be able to deal with fault and unusual situations.
Knowledge of results and attitudes to errors. Generally, feedback about results during learning gives information about whether the learner’s present method for choosing an action or a thinking strategy is or is not successful. When learning a task, the trainee chooses a method which they think will have the required effect. When they get feedback that what they had chosen to do did not actually have this effect, they can use the difference between intention and achievement as a guide to revising the way they choose what to do next time. This revision process continues until they are able to choose a method which does have the intended and required effect. They can then carry out the method with minimum need to check that it is effective or to revise it, they are 'skilled' (Bainbridge, 1978). The same methods of revision can be used in adapting an existing working method for use in another similar situation.
There are two approaches to the use of error feedback in training (Frese and Altmann, 1989).
When trainees are learning some standard pre-specified behaviour, their exercises should be guided so that they make minimum errors during training.
In complex tasks, both instructors and trainees need a different approach to errors during training, treating them as a source of extra information about the process, the task, and how to do it. Part of the aim of training here is that trainees should learn to select for themselves what is the most appropriate thing to do in particular circumstances. So the results of an inadequate inference or action during training should not be considered as wrong. For example, if a trainee was supposed to increase pressure, and they have actually decreased it, they may not have learned how to increase pressure, but they have discovered how to decrease it. So 'wrong' actions are part of exploring the behaviour of the process. The trainee should be encouraged to make use of feedback information explicitly in this way, rather than just rejecting or forgetting about what happened when an action and its result were not what was originally intended.
Training and test facilities. If the training scheme is being introduced into a company which does not have much ergonomics/ human factors experience, it may be necessary to discuss a range of practical details, such as : the best format in which to present different types of information (e.g. diagram or text); issues in the use of simulators, such as the level of fidelity, and the advantages of modular simulation and of providing additional cues which are not available in the real situation (Marshall et al, 1981); how to give knowledge of results to a trainee; and how to test for understanding.
The body of this paper consists mainly of practical suggestions which follow from this outline. This paper does not aim to provide a complete review of theory and associated literature.
2. Understanding The Process And The Task Objectives
'Understanding' here means knowing in general how the process works, (i.e. the relations between its physical structure and its cause-effects), rather than inferring what its actual status is at a particular time (for which see Section 3.1). The focus here is also on what the operators need to understand, rather than on how they understand.
Teaching 'understanding' raises questions about the use of 'theoretical' or 'practical' approaches in training. Results of studies on this do not indicate clearly which type of training is best. This is probably because few studies which have tested 'theoretical' training have said explicitly what they mean by it : it can cover a wide range of possibilities (Bainbridge, 1989b), and many of the studies have tested different things. For example, 'theoretical' knowledge could be at several levels removed from the knowledge an operator uses in operating a process:
1. In thinking, the operator uses reference knowledge which is directly relevant, such as : product targets and plant constraints; criteria for optimising and compromising in decisions; the size and timing of plant dynamics and sequences of events; as well as associations such as : interface to plant state inferences; actions to the effects they give; actions to enabling actions required; required effect to appropriate action to make.
2. Knowledge about these specific parameters can be backed up by process-specific explanatory knowledge, such as the physical, causal and functional structures of the plant, and mental models of plant dynamic behaviour in particular states, or of sequences of events.
3. This process-specific explanatory knowledge may be backed up by general engineering explanatory knowledge, about categories of physical, causal and functional structures and behaviours.
4. This general engineering explanatory knowledge may be backed up by general physical and chemical explanatory knowledge, such as about nuclear reactions.
The assumption in this paper is that the general engineering explanatory knowledge (3) and the process specific explanatory knowledge (2) provide the understanding, which can be learned in a more abstract setting, while the specific process parameters and associations (1) cannot be generalised or transferred to other processes, and are learned only by directly interacting with and operating the plant (Section 3).
2.1. Understanding how the process works
The physical plant in most processes is to some extent modular. For example, if the operator understands the general properties of one valve, heat exchanger, or PID controller, then they can generalise this knowledge to other items in the same category (Marshall et al, 1981). Such modules may be combined in a hierarchy of physical complexity, such as valve - heat exchanger - whole plant. Part-whole training could start from the simplest lowest level items, and build up to the more complex ones. So some questions in designing a training scheme for understanding are : how should the process be broken down into subparts ? and in what order should these be taught ?
There are at least three different ways (which are related to the purpose of the plant) of dividing a complex process into smaller parts which are easier to understand :
* dividing the process into unit operations, subparts of the plant between which the flows are relatively simple,
* grouping parts of the plant together which serve the same function,
* making divisions based on the task rather than the plant (for more on this see Shepherd, 1985).
Dividing the process into unit operations. In training on a specific plant, the plant may be divided into 'units' at different levels of complexity. For example :
1. At the lowest level : valves, fans, pumps, transducers,
2. At a medium level : groupings of low level items into functional units, or 'unit operations', such as :
* storage tank, feed tank,
* condenser, boiler, evaporator,
* absorber, desorber,
* heat exchanger, calandria, super heater,
* PID controller.
(These units have been put into groups within which similar physical principles are used in understanding how the unit works. The sequence is in a suggested order of increasing difficulty of understanding.)
When the trainee first learns about such a unit, this unit should be simulated in isolation, so that its feeds and outputs do not depend on the state of the rest of the plant. The effect of each valve, pump, fan etc., which the operator is expected to use or understand, should be dealt with. At each point in the training scheme, the aim should be to introduce a small amount of new material, and to draw attention to it. If a unit, such as a heat exchanger, is first experienced in the context of its interactions with other parts of the plant, it will seem very confusing.
3. The highest level consists of combinations of unit operations which are interdependent, that is, although it is possible to describe the unit operations separately (and they should be introduced to the trainee separately at first), it does not make sense from the point of view of their contribution to the process to do so. When the unit operations have been understood, then they can be combined into these larger interdependent combinations, and their interdependencies and specific context effects can be demonstrated or explored.
Conveying an understanding of these parts of the process. At this stage the trainee has three aspects to learn in order to understand about all these parts of the process : the physical structure of the plant; the cause-effect relations, i.e. what other variables any one variable affects, and vice versa; and the nature of the transient effects during changes in the plant (for how to operate the plant, and the operating targets and constraints, see Sections 2.2 and 3).
It is probably easier to start with learning the physical and causal structures, and to learn the details of the dynamics later. Operators, supervisors and engineers need to know about the cause-effect relations in the process at increasing levels of detail.
For example, to operate the plant in normal circumstances, it may only be necessary to know 'to control this, turn that'.
But to diagnose faults, knowledge about the underlying plant components and their effects is needed.
While to redesign the plant, information about the properties of the components is needed.
The instruction could start statically, using schematic and cause-effect diagrams of part of the plant, shown together, to teach the general principles, before moving on to a dynamic simulation which shows changes as they happen. A simulator can be used, for example :
1. In elementary training, to demonstrate :
a) if a manipulated variable is changed, what happens to the controlled variable(s) (note that manipulated variables are not necessarily inputs to the unit, e.g. an output flow rate may control upstream levels or flows).
b) what happens when parts of the plant are beyond the limits of their capacity, e.g. an empty feed tank or a flooded receiver tank.
2. In advanced training, to allow (guided) exploration, for example, to find out if a manipulated variable is changed, what effect this has on various controlled variables, or, for a given controlled variable, which manipulated variables can be used to change it.
Within each type of unit there are at least two possible further principles for choosing the sequence of material which the trainees experience : the complexity of the dynamics to be understood, and the importance of the unit.
There are several factors which may influence the difficulty of understanding plant dynamics, such as :
a) the length of the time constants, usually starting by learning about operations with short time constants, and moving on to ones with long time constants,
b) the number of variables, usually starting with operations with a small number of variables, and moving on to ones with a larger number of variables,
c) the complexity of the plant dynamics, usually starting with additive effects, in which the level of one manipulated variable does not affect the effect of another, and going on to interactive effects, in which the level of one manipulated variable does affect the effect of another.
Items could also be learned in order of importance, starting with main functions, going on to subsidiary functions, and ending with maintenance of services.
Understanding the process in terms of maintaining important functions. (This links to Section 2.2 below, on task objectives.) If the plant is sufficiently complex to have main functions which can be maintained in several ways (e.g. alternative methods for maintaining cooling, or drum level) then, at a later stage in training, the process can be subdivided in this way. This is particularly important for trainees who need to learn how to deal with unusual fault situations (see Section 4), as they need to know alternative methods of maintaining the main plant safety functions if the usual methods are not available.
The training could bring together, for each main function : information about all the methods by which a function can be obtained : their relative merits, assessed in terms of efficiency, cost, lead time, etc., and their interactions. If any method of meeting a function also affects other functions in the plant, this will need to be taken into account in using the method, for example, preventive actions may be needed to counteract its effects on other parts of the plant.
2.2. Task objectives : product goals and plant constraints
Before learning to control the plant, an operator needs to know the objectives of operating it. In learning about objectives, the trainees need to know 'why' as well as 'how'. Although these will be described separately, they need to be learned together.
The reasons for the objectives and criteria. There are three main categories of reason why process variables need to be kept within certain limits :
* to attain product quantity and quality, e.g. keep weight within limits, or double the concentration,
* to optimise plant efficiency, e.g. maintain drum at this level because this maximises area for evaporation,
* to maintain plant integrity, e.g. maintain pressure below this level or the plant will explode.
If the true reason for a criterion is not obvious at the point at which the criterion value is measured (e.g. maintain inflow to tank, to ensure that pump in outflow does not run dry), then it may be necessary to draw attention to the causal chain linking the controlled variable with the criterion variable, and to demonstrate, or allow the trainee to explore (in a simulator!) what happens if the criterion is not maintained.
The operators need to know these reasons for the operating criteria so that they can be 'intelligent' in unusual circumstances, such as faults. Operating procedures also need to include the reasons for quoted criteria, to help the operator in assessing whether the actions recommended by the procedure are appropriate in specific circumstances. If the usual method of meeting a criterion is not available, then operators need to know the purpose of the criterion, as the starting point for thinking of an alternative way of meeting this aim. In a situation in which several objectives must be met at the same time, some of these may conflict, so the operator needs to know their relative importance. This relative importance may depend on the task context. For example, in fault management, only plant integrity criteria may be important.
What needs to be controlled. Hierarchical Task Analysis (e.g. Shepherd, 1985) describes the task goals as a hierarchy. For example, controlling the plant could be broken down into the constituent tasks : start up, operate, shut down. Then 'operate' could be sub divided into 'maintain temperature at x', 'maintain pressure at y', etc.
These values to be controlled may be more or less complex to describe and understand, for example whether the target is :
* a single value on one variable, plus or minus a tolerance limit,
* a combination of values, e.g. pressure plus temperature,
* a profile of values, e.g. across a reactor, or up and down an distillation column or a kiln,
* a ramp of values over time, e.g. bringing up the turbine when starting an electric power plant.
Schematic, and cause-effect, diagrams could be used to draw attention to the location of these critical values in the plant.
Choice of actions to obtain the objectives. If multiple criteria affect what is the best action in the circumstances, this may place considerable limits on the actions which can be used, For example, it may constrain either the actions which are appropriate (e.g. the need to maintain product quality may restrict the ways in which it is possible to increase product quantity), or the sequence in which actions can be done (e.g. to maintain plant integrity during start-up or shutdown). A plant simulator can be used to demonstrate, or allow the trainee to explore, how multiple decision criteria limit the choice or sequencing of actions.
3. Main Operating Tasks
Three aspects of the operators' task will he described here : inferring the process state, controlling it, and sequencing.
3.1. Inferring the state
The operator needs to learn to infer what is happening inside a process from what is displayed on its interface : where given items are on the interface, how to interpret their meaning, and what the relation is between the displayed information and events in the process.
Information to and from the interface. This is another 'which comes first ?' problem, as the trainee needs to know something about the interface before they start to do the task. But knowledge about the interface in detail is probably best learned within the context of doing the task, so that an operator learns how to access the relevant knowledge from the interface while they are thinking. This means that early exercises in thinking about the task need to be done without time pressure, while the trainee works out how to get the required information from the interface as well as how to do the task. The trainees need to learn several things, such as :
a) where every item is on the interface, and how to recognise it. If the operators know where a required item is in space, they can find it automatically by eye-movement, rather than having to search for it (see Bainbridge, 1991a).
b) what the appearance of each display and control 'means', e.g. clockwise movement means increase, or light blue means steam.
Operators or engineers who are expected to diagnose faults need more information about displays, such as where a transducer is on the plant and how likely it is to give an incorrect measure, and about controls, such as whether the interface representation of a control just indicates what has been done on the interface, or whether it directly monitors the state of what is controlled, e.g. whether a valve is open or closed.
When using computer-based multiplexed interfaces, the operator needs to learn what information is available from each of the display formats (what they can be used for), and how to access them. A trainee needs to learn something about how to access the display system before learning to interpret the displays, or they will be confused by having to learn two different tasks at the same time. Learning about the range of display formats available is probably best taught as part of learning to infer the state of the process, because different formats will be useful in different thinking strategies.
Inferring the underlying process state. An operator often needs to infer what is happening behind the displayed information on the interface. When part of the process is not instrumented it may be necessary to infer what is happening there from what is happening upstream and downstream. This is frequently the case in faults. For example, a process interface does not show directly that there is a leak in part of the process : this is inferred by comparing data on in- and outflows, which should balance when there is no leak.
Note that this type of task thinking (inferring the process state) is the inverse of general understanding of the process (Section 2.1). Understanding is concerned with knowing about cause-to-effect. Inference is concerned with working from effect to cause. Unfortunately, if someone can think in one direction, this does not automatically mean that they can think in the other direction. They need experience of both.
There are at least three types of interpretation situation which trainees may need to experience. Two of them are mentioned here and the third, identifying faults, is covered in Section 4.
When a pattern of displayed values on the interface has one, and only one, interpretation, then operators can be trained to associate the interpretation directly with the pattern. For details of the method see Shepherd et al (1977).
When there are several possible, and known, interpretations of a given pattern of information on the interface, then the operator needs to work out which of them is actually the case. This involves knowledge and thinking. The operator needs to learn :
* that alternative explanations are a possibility. A demonstration could be used to show how different process states can underlie the same displayed information,
* what alternative interpretations about the state underlying the displayed information are possible in each case,
* the frequency, importance, and recency, of each of these alternatives,
* what other evidence can be used to check whether or not a given alternative is the case e.g. going to look at the plant, waiting for another symptom to develop, or getting a laboratory test,
* how to obtain this confirming information.
3.2. Control
Manual control. Manual control is a 'feel' skill, which can only be learned by doing it, by hands on experience, not from lectures, verbal descriptions or watching other people or controllers doing the task (Bainbridge, 1989b). Before an operator has learned the gains and lags in the process they will set it into oscillation when they try to control it. The training can include demonstrations, practice, or opportunities to explore what happens.
It is just as important to give tasks of gradually increasing difficulty in learning to control as in learning to understand the plant. Various sequences of learning were described in Section 2.1. An additional dimension in organising control learning is the difficulty of the control task. In likely order of increasing difficulty, typical tasks to be practised are :
* steady-state control,
* making a step-change to a new specification,
* making a ramp change, or a step-change within time limits,
* controlling units which have downstream or upstream effects on other functions, which need to be compensated for,
* controlling states or devices which have to be balanced.
Automatic control. In simple cases, the operator needs to learn how a PID controller works (see Section 2.1) [at the time of writing, this was the main type of 'automatic' controller in use]; where the controllers are in the process and what function they serve; what the usual set-points are and how to choose them; and how to take over manual control. If there is more complex computer control, then the operator may need much more help with exercises in understanding the control system.
If manual control skill is needed in fault situations, which there is not an opportunity to practice during normal operation, then special practice opportunities are needed.
3.3. Sequencing
In many phases of operation a process goes through a sequence of stages. A 'stage' is defined here as a new process configuration, such that the operator needs to change to a different mental model of the process dynamics, in which different variables are important for predicting how the process works, or to a different set of criteria for assessing the process state (see Bainbridge, 1992). For example, the (simplified) stages of operating an electric powered steel furnace are :
* charge : load with metal, no power used.
* melt : melting metal, high power used.
* oxidise : pass oxygen to control steel quality, mid-power used.
* tap : pour out molten steel, no power used.
Sequences of plant activity typically occur in batch processing, during start-up and shutdown, and in events after a fault. The sequencing, the change from one stage to another, may happen inherently or be done by the process, e.g. steam generation starts when temperature reaches 100C, or emergency water injection is started in a pressurised water nuclear reactor when the cooling water falls below a certain level. Or the operator may initiate the change by manual actions.
In either case, the operator needs to know the general form of the sequence. A static simulator could be used to demonstrate this, with a sequence of 'cartoon frames' showing how the sequence develops. A dynamic simulator could be speeded up between stage transitions, and slowed or stopped at stage transitions, with attention drawn to critical features. The trainees' knowledge of the sequence can be tested by asking the operator to predict what will happen, what will appear on the interface, from which it is possible to identify which stage the process is in, and whether it has reached a transition state (Section 3.1); and the reasons for the sequence (Section 2.2).
When the operator has to do the sequencing, there are further aspects to learn : how to control the process to attain the transition state from which the next stage can be initiated (Section 3.2), and how to initiate the next stage.
In advanced training, the operator may need to learn how to devise a sequence of operations for themselves. To do this, they need knowledge of the actions available; the constraints on the choice and sequencing of actions (Section 2.2); and experience with exploring the behaviour of the process (Section 2.1). They also need experience with training exercises in which they practise planning (Section 4.3).
4. Fault Management
Fault management involves both diagnosis and ensuring that the process is in a safe state from which normal operation can be resumed. These are processes of interpretation and control which are not fundamentally different in the nature of their cognitive mechanisms from the thinking that is done during normal operation. However, fault management tends to be trained as a special case for several reasons. The situation is more likely to be unfamiliar, so that the operator needs to think out what to do, rather than using established thinking strategies and knowledge (though problem solving can itself be practised, see below). Methods of producing an answer to a situation might be put in order of difficulty (Bainbridge, 1989a) :
* there is only one answer to the situation, so it can be given (after practice) without thinking,
* an effective working method, with relevantly structured reference knowledge, is readily available (due to recent practice),
* the person has to think out a new (to them) method for doing the task (this is problem solving).
Which of these methods a particular person can use to do a particular task depends on how much experience that person has with this task, and how much regularity there is in the task.
Operators who are expected to deal with faults also need extra knowledge, which is not required by operators who only handle normal situations and call on a superior in any unfamiliar situation. For example, fault management may involve knowledge of : symptom to underlying fault associations; the likelihood that interface information is invalid; sequences of events after faults; changed operating goals; and strategies for dealing with unfamiliar situations.
A basic approach to training many of these aspects has been given in previous sections. This section will focus on unfamiliar situations. It is important to distinguish between situations which are unfamiliar to a particular operator, and situations which are unfamiliar to their industry in general. At the extreme, the operator is there to deal with situations which are unfamiliar to the industry, but it is inherently impossible to practice these. What is possible is for operators to gain experience and skills in dealing with situations which are unfamiliar to them. This can be done by practising situations which are known to the industry but unfamiliar to the operators being trained. This section will outline a simple model of how people deal with unfamiliar situations, and will then suggest the implications for training in diagnosis, control, problem solving and planning.
Although 'problem solving' is frequently mentioned, what this involves is usually not discussed. There are at least two types of problem solving in process situations :
* developing a new structure of knowledge which provides an explanation of what is happening,
* developing a new working method.
It is often assumed that all problem solving involves the sort of reasoning from first principles that is studied in the 'problem solving' literature.
However, observation of people faced with situations in which they do not know how to reach their aims suggests that they may use several strategies, such as trial and error (technical term for kicking the equipment - they learn from experience where best to kick), asking someone else, reading the instructions (not popular), thinking things out by reference to what is done in a similar situation, or thinking things out from first principles. This section will focus on the last two approaches.
4.1. Dealing with unfamiliar situations by reference to similar ones
A popular method of dealing with a situation for which the person has no standard interpretive knowledge or working method is to relate it to a similar situation for which there is knowledge or a method, and to adapt this knowledge or method to fit the new situation (a working method is a type of knowledge, but it is useful to distinguish between them in training). The similar situation might be an episode in the operator's own experience (e.g. what happened last time this difficult start-up procedure was done), or stories circulated in the plant or industry about rare events (e.g. the various nuclear radiation releases), or from other physical worlds (e.g. using how the bath water runs out as an analogy for understanding turbulence in an emptying tank). The technical terms for this sort of thinking are 'case based' reasoning, or thinking by analogy.
There are some key questions about how these thinking processes are done : how does someone identify something as 'similar' ? how do people learn about similarity ? how do people adjust existing strategies to fit different situations ? These questions need to be answered, to give the basis for proposals about aiding this type of thinking by training for it. Much research is still required, so the points made here are based partly on research and mainly on experience.
Categorisation
The ability to recognise that events, situations, and strategies have properties similar to the present requirements depends on the human capacity for categorisation. Categorisation is an aid to efficiency in thinking. Items which have similar properties are grouped together, and the properties which they have in common become the properties of the category. It can be simple for someone to deal with a new item once it has been identified as a member of a category, as a whole range of existing knowledge is then assumed to apply to it.
This categorisation, or ability to generalise, has already been made use of in the above comments on training, for example :
* grouping together several different representations of the same part of the plant,
* dividing the process into standard modules, such as valves or heat-exchangers, and dividing these modules into groups with similar underlying principles of operation,
* grouping together different cause-effect chains in the process which have the same general function, such as to control pressure,
* grouping together types of reason for plant and process operating criteria,
Note that categorisation is purpose oriented. Categories are useful because they help people to deal with situations (Bainbridge, 1993). Some of the properties of an item in a category are related to its primary purpose, some to the secondary aims which can be achieved with it. As an example : using manufactured gas is a quick expensive method of increasing gas supply. From the point of view of goals and working methods, the primary property of this working method is that it is a member of the goal category 'ways of increasing gas supply'. Speed and cost are secondary characteristics from this perspective. However, when someone is choosing between methods of increasing gas supply, there may be aspects of the situation which place limits on cost or speed of operation, they are secondary goals. So the secondary properties may be used in choosing between working methods.
Different properties may also be relevant for different primary purposes. Most items are not members of only one category. Some of their properties are relevant to membership of one category, others to membership of other categories. For example, a pressuriser has some properties which make it suitable for controlling pressure, and others which make it dangerous. New purposes lead to the need for new groupings of items.
The ability to categorise is both powerful and limiting. Once a person has put something in a category, they may find it more difficult to think about this item flexibly ('perceptual set'). But multiple properties and the ability to make new categories are fundamental to the problem of dealing with a new situation by finding a similar one.
Feedback is as important in learning about these general properties of knowledge and working methods, as it is in other aspects of learning. Trainees do not only get feedback about how well the method meets the main goal, they also get information about many other properties of the method. Which of these they notice depends partly on what their attention is drawn to - by feedback about how well the method meets the main and secondary goals. This feedback may be internal, as in personal assessments of the difficulty or pleasantness of the method, or external, as in receiving comments about the cost or speed with which the work has been done.
Noticing that new groupings of items are useful, or that items used for one purpose can also be used for another, often happens in a less coherent way. It may occur as the feeling of 'aha', of realising that a new combination is useful. Often people carry a goal in the back of their minds, perhaps the need for an explanation of something (as in diagnosis) or for a way of achieving something (as in devising a method), and then notice when something arises which fills this need. However, people can be given practice in new ways of doing things, and in being reminded about the need for flexibility (see below).
4.2. Working from first principles
Note that here 'reasoning from first principles' does not necessarily mean reasoning from knowledge of the basic physics and chemistry underlying what is happening in the process, although that may be used (see Section 2). Here this phrase means using basic building blocks of knowledge about the process to build up a new structure of knowledge, or working method, without referring to a similar situation as a guide for how to go about this structuring.
Basic knowledge
Building up a new working method from first principles is not context free, nor independent of knowledge about the process and how to operate it. A new knowledge structure or working method is needed when the trainee operator does not have a standard method of reaching the required objective. However, the new knowledge structure or method that they devise will be built up from familiar sub-components. Defining where problem solving begins and ends is difficult, as most new knowledge structures and working methods are a flexible combination of new and old parts.
Situations which an operator has not previously experienced may vary in their familiarity, in the amount of specifically relevant knowledge the operator has for understanding the process or about working methods. The following list does not make a formal scheme for defining levels of specific or general familiarity, but illustrates the point. It uses air flow in part of a process as an example.
a. Knowledge about the process :
* component specific knowledge, e.g. the factors which make this particular fan more or less effective,
* process specific knowledge, e.g. where this fan is, and what its function is,
* generic knowledge, e.g. what fans are, what they do, and what affects their efficiency.
b. Knowledge about how to operate the process :
* sub-task specific strategies, e.g. how to get at the blades of this fan to clear them,
* task specific strategies, e.g. alternative methods of maintaining air flow in this process,
In many situations for which an operator has not learned a specific response, or a sub-task specific method with associated specific knowledge, they may have sufficient more general (life experience) knowledge from which they can work out what to do.
Problem solving frameworks
When people are not able to think of a similar situation to think in reference to, their activities can supported by a general framework for problem solving, as a basis for fitting existing knowledge together in a new way. Such a problem solving framework suggests a sequence of sub-goals to work through (Samurçay and Rogalski, 1988). These general strategies will be discussed in two groups, those for diagnosis (inference), and those for restoring plant integrity (planning actions).
Diagnosis. The main aim of diagnosis is to identify an explanation for what is happening in the process, which can be used both as a basis :
* for anticipating and then acting to counter undesirable process behaviour and
* for planning restoration to normal operation, including repairing the fault.
A general framework for diagnosis has two main sub-goals :
* to suggest hypotheses about what may underlay the fault, and then
* to test these hypotheses, either by mentally simulating their effects and comparing them with what is happening on the process, or by trying out test actions or getting confirming information from the process.
The primary problem in the first step is - how does the operator get hypotheses about what the fault might be ? In cases where there is a single relation between symptom and fault, or the fault frequently occurs and is well known, methods of inferring faults are the same as the methods of inference used in normal operation (see Section 3.1). A fault which has not previously been experienced cannot be dealt with by these pre-specified methods. Operators may use several types of general strategy in this situation, such as :
* taking a general category of fault (such as a leak) and tracing through all the plant parts which might have failed in this way,
* making a best guess about what is wrong, and then using a dynamic mental model to predict what would happen in the process if this were the case, and comparing this mental simulation with what is actually happening,
* tracing back through effect-cause chains to suggest what might be the causal fault.
In each of these strategies the problem is that, for any one symptom, there is likely to be a large number of possible reasons why it has occurred, and the need is to focus the attention instead of reviewing all the possibilities. Kassianides (1991) suggests the following strategy, which depends on a modular understanding of the process (see Section 2.1);
1. Check the inputs and outputs of the major plant subsystems. Find which are not normal.
2. a. If there is no recycle connection between affected subsystems, then start by investigating the one furthest upstream.
b. If there is a recycle connection between two affected subsystems, examine the one upstream, then the one furthest downstream, then the one before, and so on.
3. Within the subsystem :
3.1. Check the inputs and outputs to medium level units within the subsystem. Find which are not normal.
3.2. Starting with the one most upstream :
a. If there is an input problem, examine the low level units that affect it.
b. If there is an output problem, examine the low level units that affect it. If there is not a problem with them, then examine the medium unit itself.
Note that this strategy of focusing on greater and greater detail does not attempt to go straight from symptoms to fault, but progressively restricts the part of the plant that the fault might be in. Indeed, fault training can check, correct, refine and develop the knowledge of the plant that the operator acquired for normal operation.
During training, the operators can practise this sequence of diagnostic steps in turn. In each step, if the operator comes to the wrong conclusion, then the feedback should point out which variables they have not considered in their hypothesis about what underlies the symptoms.
Restoring plant functions. The operator has two control tasks in a major fault situation :
* to maintain plant integrity, and
* to restore the plant to a steady state from which it can be shutdown, repaired, and re-started.
The three sub-goals in a general framework for this could be :
* to find out what needs to be achieved;
* to identify the sequence of states which will lead from the present situation to the required one; and then
* to plan how to implement this, by finding a way of moving through this sequence of states, and then mentally simulating and adjusting this plan.
4.3. Exercises in dealing with unfamiliar situations
In training for problem solving, it is generally assumed that it is best to start after normal training, and to build up from simple to complex experiences. Previous training schemes which have considered planning and problem solving have suggested that, as well as guided discovery exercises, it can be useful to include :
* planning exercises,
* discussion groups in which trainees share and compare the knowledge gained by individuals during the training exercises (Taylor, 1978).
* practise with situations which are unfamiliar to the trainee, so that dealing with novel situations is expected (Munley, 1990).
The above outline on problem solving in unfamiliar situations suggests that training should include :
* formal rather than informal experience of plant anecdotes, and how to adapt them to suggest solutions in other situations,
* practice in using general purpose problem solving frameworks, such as the ones suggested for diagnosis and for developing a working method to reach an objective.
* practice with thinking about the process from first principles,
* practice with thinking of other uses for available working methods,
* practice in thinking in terms of the subsidiary effects of working methods, not just their primary ones. This suggests that it would be useful for trainees to practice :
* explicitly noticing the secondary properties of working methods (speed, accuracy, mental and physical effort required, dirtiness, danger, need for special equipment, etc.) which are used in choosing between them,
* noticing other properties which are not necessarily relevant to the present situation,
* thinking flexibly about working methods, i.e. thinking about alternative ways of reaching any given goal, or alternative goals which can be met using a given method.
5. Final Comments
This paper has discussed the organisation of training in a complex task, using industrial process operation as the example. Training for complex tasks involves the part-whole integration of several types of knowledge, such as the physical and functional structures of the process, the task, and the working methods. The paper suggests that, as the training will be in subsections, the trainee needs to be given a context for the purpose of learning. And to increase flexibility, the trainee needs to be helped to use errors as a source of information rather than a reason for punishment.
Training for understanding the process has been divided into two sections :
* on understanding how the process works, by dividing it into modules, and
* on understanding the reasons for the task goals and constraints.
Training for operation has included how to :
* infer the inner state of the process,
* control it - either manually or automatically, and to sequence the process stages.
Training for fault management has focussed on unfamiliar situations, in which the operator needs to develop :
* new diagnoses, i.e. new knowledge structures for describing the state of the process, and
* new working methods for recovering the process state.
This paper suggests that dealing with the unfamiliar is done either by reference to categories of similar situations, or by using a general framework or strategy which specifies the sub-goals to be met by integrating elements of existing knowledge.
Evidently there is a large number of issues to consider in organising the training of an extensive task, and these issues have not yet been much discussed. This paper has attempted some preliminary suggestions, which are not complete and are mainly based on generalisation from other areas of experience, rather than on directly relevant research. There is a great deal of research still to be done, in at least three areas.
First, there are many questions about optimum methods for training people to do cognitive complex tasks. Some of these are issues which overlap with educational methods, although education is not usually concerned with dynamic tasks.
Secondly, there are many remaining considerations about what to train. For example, this paper has not addressed all the operator activities mentioned in Section 2.1, nor such important aspects of complex tasks as the allocation of effort between different responsibilities in the task (Amalberti, 1992), or the temporal organisation of behaviour (de Keyser, 1991)
Thirdly, there is as yet little basic research on the actual cognitive processes involved in the development of cognitive skill, such as how people develop new working methods in rich contexts. The existing problem solving literature mostly examines problem solving in task contexts in which the task elements are both simple and similar in nature. However, in real task situations, people need to integrate different types of knowledge, such as structural, causal, temporal, and probabilistic knowledge, each of which may have a network of implications, while taking into account constraints of many different types, including task and personal goals and capacities. Presumably research on such cognitive mechanisms could have important implications for the design of optimum training methods.
References
Amalberti, R. (1992). Models d'activité en conduite de processus rapides : implications pour l'assistance a la conduite. These de Psychologie, Université de Paris VIII.
Bainbridge, L. (1978). Forgotten alternatives in skill and workload, Ergonomics, 21, 169-185. French translation in Le Travail Humain, 40, 203-224.
Bainbridge, L. (1989a). Development of skill, reduction of workload, in L. Bainbridge and S.A. Ruiz Quintanilla (Eds.), Developing Skills with New Technology, Chichester, Wiley.
Bainbridge, L. (1989b). Cognitive processes and training methods : a summary, in L. Bainbridge and S.A. Ruiz Quintanilla (Eds.), Developing Skills with New Technology, Chichester, Wiley.
Bainbridge, L. (1991). Multiplexed VDT display systems, in G.R.S.Weir and J.L.Alty (Eds.), Human-Computer Interaction and Complex Systems, London, Academic Press.
Bainbridge, L. (1992). - Mental models in cognitive skill the example of industrial process operation, in P.Bibby, A.Rutherford and Y.Rogers (Eds.), Models in the Mind, London, Academic Press.
Bainbridge, L. (1993). Types of hierarchy imply types of model, Ergonomics, 36, 1399 1412..
de Keyser, V. (1991). Temporal reasoning in continuous processes : segmentation and temporal reference systems, in Proceedings of the Third European Conference on Cognitive Science Approaches to Process Control, School of Psychology, University of Wales College of Cardiff, 2-6 September.
Duncan, K.D. (1974). Analytical techniques in training design, in E.Edwards and F.P.Lees (Eds.), The Human Operator in Process Control, London, Taylor &;Francis.
Frese, M. and Altmann, A. (1989). The treatment of errors in learning and training, in L. Bainbridge and S.A. Ruiz Quintanilla (Eds.), Developing Skills with New Technology, Chichester, Wiley.
Kassianides, S.C. (1991). An integrated system for computer based training of process operators. Ph.D. Thesis, Department of Chemical Engineering, Imperial College, London.
Leplat, J. (1989). Cognitive skills at work, in L. Bainbridge and S.A. Ruiz Quintanilla (Eds.), Developing Skills with New Technology, Chichester, Wiley.
Leplat, J., Enard, C. and Weill-Fassina, A. (1970). La Formation par l'Apprentissage, Paris, PUF.
Marshall, E.C. and Baker, 5. (1989). Training operators to use advanced display systems for process control, in L. Bainbridge and S.A. Ruiz Quintanilla (Eds.), Developing Skills with New Technology, Chichester, Wiley
Marshall, E.C., Scanlon, K.E., Shepherd, A. and Duncan, K.D. (1981). Panel diagnosis training for major hazard continuous process installations, The Chemical Engineer, 365, 66-69
Munley, G. (1990). Training and transfer of faultfinding skill Ph.D. Thesis, School of Psychology, University of Wales College of Cardiff.
Samurçay, R. and Rogalski,]. (1988). Analysis of operator's cognitive activities in learning and using a method for decision making in public safety, in J.Patrick and K.D.Duncan (Eds.) Training, Human Decision Making and Control, Amsterdam, North Holland.
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.
Taylor, P.G. (1978). The metal working machine-tool operator, in W.T.Singleton (Ed.) The Analysis of Practical Skills, Lancaster, MTP Press.
Tompsett, P. (1987). An integrated initial training programme for a CEGB operations engineer, in Proceedings of the CSNI Specialist Meeting on Training of Nuclear Reactor Personnel, Orlando, Florida, 21-24 April, NUREG/CP-0089.
Access to other papers via the Home page
©1998, 2022 Lisanne Bainbridge
Comments
Post a Comment