Analysis and simulation of operator's behaviour in controlling continuous baking ovens
This is the complete paper by John Beishon.
The suggestions for simulation mechanisms are outdated but I haven’t done any editing, as the simulation suggestions are closely interwoven with comments on the operator’s cognitive behaviour.
After this bakery study, John Beishon went on to direct the study of furnace power control reported in my papers. I was his research assistant for collecting and first analyses of the steel works data. So those analyses were done within the context of this bakery study. My further analysis was done when possible between teaching commitments after I moved on from working for him to lecturing at the University of Reading.
John Beishon later went on to be Professor of Systems at the Open University, and then head of the Consumers Association.
An Analysis and Simulation of an Operator's Behaviour in Controlling Continuous Baking Ovens
By R. J. Beishon
Department of Psychology, University of Bristol
first published in F. Bresson and M. de Montmollin (eds) 1969. The Simulation of Human Behaviour, Paris : Dunod.
reprinted in E. Edwards and F. P. Lees (1974) The Human Operator in Process Control. Taylor & Francis. pp.79-90.
I. Introduction
The concept of the man-machine system has proved valuable in providing a basis for the analysis of complex situations where men and machines work in close interaction with each other. In many cases these systems can be analysed without difficulty and a number of suitable analytical techniques are available; see, for example, Gagne (1963). However, numerous man-machine complexes are developing in which men have to use a high level of intellectual skill with little or no manual component. These skills usually take many years to acquire and there is relatively little understanding of their fundamental psychological nature. Only a small amount of research has been done on the cognitive activity involved in these real-life tasks and most of the published work for example refers to game-playing tasks undertaken in the laboratory.
Industrial skills appear to be a promising field for research in this area for various reasons. Apart from the practical value of gaining a better knowledge of skilled behaviour for training and design purposes, industrial tasks are well suited for analysis. They usually have clearly defined goals and the machine part of the system to a large extent conveniently constrains the man's activity. Further, the situations which arise in industry are sufficiently complex to be non-trivial, but not so complex that a very high degree of mental sophistication is needed by the operator. The behaviour of people in industrial settings is also 'realistic' and genuine positive and negative rewards operate so that the motivational aspects of the situation are usually under reasonable control or at least known. For these reasons a typical industrial skill was selected for study as part of a general study of cognitive processes and the present paper describes an investigation of a cake-baking production unit controlled by one operator. The study aimed principally at analysing the man-machine system in detail and at gaining an understanding of the underlying cognitive processes involved in the skill. The process and the ovenman's job are described first, followed by an analysis of the operating system. The development of a model to simulate the ovenman and his control behaviour is then described.
2. The Baking of Cakes
In modern [1960s] bakeries of the kind studied here, cakes are produced on a semi continuous batch basis; prepared mixes for each variety of cake are formed into appropriate shapes on specialised machines and delivered on baking trays to ovens. The trays are entered into the ovens by the ovenman who places them onto continuously-running endless belts which carry the trays through the hot zones of the oven. After leaving the oven the cakes are cooled, inspected and packed for sale. The complete production process is illustrated diagrammatically in Figure I.
In the plant studied there are three continuous tunnel ovens in one piece of plant, two upper ovens and one larger lower oven; the diagram in Figure 2 shows the general arrangement. Each oven, which is about 13 metres long, has a layer of transverse gas burners above and below the conveyor belt so that the cakes are carried between them. Each gas burner can be switched on or off independently and the profile of temperature experienced by the cakes can be varied by altering the disposition of the lighted gas burners along the oven. Oven temperatures arc indicated at two points about one-third and two-thirds along the length of each oven, see Figure 2; the range of temperatures available is roughly between 200ºF and 550°F. The other major control variable is the baking time which is determined by the conveyor belt speed; this can be varied from 2 to 5 minutes. Each oven has two inspection ports, sited as shown in Figure 2; at these places the ovenman can see into the oven and inspect the partially cooked cakes.
3. The Ovenman's Job
The ovenman's main task is to arrange for the raw cakes to be converted into finished cooked articles. He has four principal aims:
(1) To ensure that the correct chemical transformations occur in the cakes.
(2) To achieve acceptable top and bottom colouration of the cakes
(3) To obtain an acceptable degree of 'rise' for each type of cake.
(4) To produce cakes in batches which are uniform with respect to the above characteristics within a batch.
To some extent the above aims, or output variables for the system, are independent of each other since each can be influenced separately by altering the burner dispositions and baking times. All four of the variables have to be assessed subjectively by the ovenman who attempts to match them to a 'standard' of acceptable cakes. It should be noted that these four aims are the principal ones laid down by the management but that there are a number of subsidiary aims which have to be considered.
The system has a number of relatively independent inputs which can be listed as follows:
(1) Type of cake. (This includes the kind of cake mix, and the size and shape of the articles.)
(2) Size of the batch.
(3) Time the batch arrives for baking.
(4) Variations in the cake mix.
The ovenman has no control over these inputs and he must take account of any variations in them by adjusting the baking conditions accordingly.
It was necessary in the first instance to record in detail exactly what happened to the system in the course of a number of typical working shifts. Observations were made using a portable tape recorder kept running continuously to provide a convenient time base for subsequent transcription. Observers reported on the movements and actions of the ovenman, read out temperatures and burner conditions at approximately 5-minute intervals, and also recorded question and answer conversations with the ovenman. A radio-microphone technique was also used for the ovenman so that his conversations with other personnel could be picked up on a second recording channel. The data were transcribed onto activity charts in which the concurrent events in the various parts of the system are plotted on a suitable time base so that the behaviour of the ovenman and oven system can be followed graphically. An example of a typical day is shown in Figure 3.
4. Analysis of Observations
The overt behaviour of the ovenman could be broken down into a number of simple categories as set out in Table l below:
Table l. Breakdown of ovenman's activity for a typical shift
Activity. % Time spent in shift
Loading oven 24
Moving racks 7
Control adjustments 8
Waiting 8
Inspections 3
Subsidiary work on another baking system 35
(mainly manual work)
Miscellaneous (including meals) 16
(Figures rounded off)
It can be seen from the table that the percentage of time spent on control and inspection activity is only 11% of the total time spent at the plant. However, it is clear from the nature of the actions and comments of the man that a large part of the time spent on manual activity is also used to think about control of the process.
It became clear that there was a standard baking procedure which consisted of a number of phases normally followed through in sequence. At a general level this procedure describes the job and the way of doing it, but it does not make explicit any of the decision rules nor could the actual task, which involves many different cakes and three ovens working simultaneously, be done by simply following this procedure. When a more detailed decision analysis is carried out it becomes clear that there are a number of fairly distinct routines for dealing with the different phases of the task.
The routines which can readily be identified are listed below:
(I) Recognition: cakes are recognised i.e. identified as a cake-type, and are also inspected and classified as to normality.
(2) Oven allocation: an oven is selected for the cake-type waiting or expected.
(3) Oven adjustment: control actions are made to bring the selected oven to a suitable state for baking the current cakes.
(4) Entry: cakes are entered into a selected oven when conditions are appropriate.
(5) Check or Sample: some items are checked at the inspection ports or exit end of the oven.
(6) Feedback/Adjustment: if inspection (see (5) above) shows faults in cakes, adjustments are made to oven conditions.
(7) New item: new cake-types arc baked according to a trial-and-error procedure.
(8) Abnormal items: If the pre-baking inspection shows any abnormality in cakes, altered baking conditions may be needed.
A detailed examination of these routines shows that in most cases a body of facts associated with each routine is needed to enable the decisions in that routine to be made. For example, for each cake-type there is a preferred baking time and temperature profile; for a specific profile there is a suitable pattern of burners which will give the desired temperatures down the oven. With the preferred baking conditions known and the current state of the oven known, a decision can be made as to which is the most suitable oven to use for baking that batch. Similarly, the decision to enter cakes will be made when the preferred conditions match the current oven conditions.
The existence of baking time and burner pattern information, and the way in which the ovenman can produce it when asked, suggests that he has a number of look-up tables in memory stores to which he has ready access. Further study of these has been made and the following tables appear to be present:
(1) Facts about the preferred times and temperatures for all the usual cake-types.
(2) Facts about the degree of tolerance each cake-type has for departures from the preferred baking conditions. (This will give information about the need for sampling when the cakes have travelled some way down the oven.)
(3) Facts about the burner patterns which will achieve the temperature profiles obtained from (1) above.
(4) Methods or procedures for getting the oven to change from one temperature state to another: this includes information about the times taken for the various changes.
(5) Expectancies concerning the kinds of cakes which will arrive for baking and also the time of day when they are likely to come.
(6) Methods or procedures for adjusting baking conditions to correct specific fault conditions which are detected on inspection of the partially, or completely, baked items.
This set of tables would provide most of the basic factual information needed to do the routines assuming that a procedure for the routine itself was available.
For several reasons the two components discussed so far, routines and look up tables, are not sufficient to account for all the behaviour of the ovenman. In a system containing time lags such as the baking system, a single routine cannot always be followed through continuously to completion and often a new routine has to be started before the current one has finished and the two then alternated. There are also external stimulus events which can interrupt a current routine and which have to be dealt with before the original routine can be resumed. This suggests that there must be a directing program or executive routine (E.R.) which controls the sequential stream of behaviour in response to external and internal stimuli. In addition to an executive routine, there must also be perceptual mechanisms for dealing with the recognition of cake-type, and for judging the quality of the output variables.
5. A Model for the Ovenman's Behaviour
The aim of this part of the project was to construct a model which would behave in the same way as the ovenman does, but having available only the same information as the man is known, or inferred, to have. The basis of the model is a central program-type procedure called the main procedure which links together the routines discussed in the previous sections. A flow chart for this main procedure is given in Figure 4. This sets out the way in which the different phases of the task might be linked and shows how one routine would follow on from another. Progress through the main procedure depends to a large extent on events taking place in the external system and must be controlled by the E.R. Flow charts for individual routines have been worked out and an example of one is shown in Figure 5.
A major difficulty in constructing a simulation for a human operator's behaviour is in accounting for the sequence of events which occur in the ongoing stream of behaviour. A study of the records from the observations on the ovenman reveals fairly distinct sections which are clearly purposive and directed to a particular goal or end state. In general, these correspond to the routines described above but frequently a routine is not carried through to completion without interruption; also when a routine is completed the ovenman will often proceed to a new routine which appears to be selected arbitrarily. So although parts of the behaviour are understood, in the sense that rules for the steps in the sequence can be identified, the total stream appears disconnected and subject to random jumps from one routine or activity to another. When these cases are examined in detail, however, a pattern begins to emerge. Interrupts, which occur when some on-going stream of activity changes suddenly, appear to arise from two causes: external triggers or signals, and internal triggers. The external triggers are events which occur in the system and which either attract the attention of the ovenman directly, such as a comment from another worker bringing up a rack of cakes, or indirectly, as when he notices that a new rack has appeared as he happens to pass the entry end of the oven. These external events will usually break into the current routine and the ovenman will attend to the new event for a time before returning to his original activity. Internally-initiated interrupts are observed when, for example, the ovenman stops while loading the oven and goes to look through the inspection port at a specific batch of cakes. This latter action implies that there is some internal call-up system which is set to alert the ovenman at some appropriate time in the future.
The jumps from a completed routine to the next one are less easy to account for; in some cases the progression is virtually dictated by the process, as when a batch of cakes arrives which the man has been expecting. He will identify these, check that conditions are in fact suitable and go on to enter them into the oven. In other cases the choice of the next activity is less clear and the ovenman himself may be unable to give a clearcut reason for it. In practice the choices appear to be made in relation to an advance planning routine which is concerned with organising his activity over the next half-hour or hour period. This work is done in anticipation of events and hence much of the activity is done for reasons which may not be apparent before the events occur. The fact that the sequence of routines is not entirely dictated by external events suggests that the executive routine is concerned with handling and resolving the often conflicting demands placed upon the operator by the requirements of the production process.
The Executive Routine (E.R.)
The responsibilities of E.R. can be set out as follows:
(1) Entering and retrieving call-ups for future entry to specific routine.
(2) Handling externally-initiated interrupts.
(3) Searching through future expectancies for anticipatory activities.
(4) Keeping and up-dating a list of activities to be done next.
The main responsibility of the E.R. is centred on (4) above. This 'activities list' will be composed of routines from the main procedure and activities associated with interrupts and anticipations. The items in the list will be positioned according to the current state of the oven system and will be made up of items from the main procedure for each of the cake batches currently being processed. The E.R. extracts the top activity on the list and carries this out, automatically moving the remaining items on the list up one place. The activities list probably covers a time span of between half to one hour ahead since the ovenman plans for this period in practice. Interrupts can either be placed at the top of the list, in which case they are dealt with immediately the current activity is finished, or they can be placed further down the list, depending on the result of conditional tests which would be applied.
To fit items into the list, and to keep track of events, the E.R. must follow some kind of cyclic scan procedure which looks at the internal call-ups and external information sources on a regular basis. This could be done by a clock pulse trigger system giving a priority scan, but it seems more likely that the scan is itself a list item which is put into, say, every third or fourth slot in the list. This means that the search for current items to add to the list, or up-dating of the list, can itself be delayed by series of priority external call-ups which are added to the top of the list. Behaviour of this kind is observed when the ovenman appears to 'forget' a sampling or scanning action for a while when he has to deal with a number of external interrupts. Important scans or interrupts already in the list cannot, however, be kept down indefinitely and there is probably a mechanism which adds a 'tag' to these list items each time they are kept down by the insertion of an item above them; accumulation of such tags eventually enables the suppressed item to operate as a priority interrupt itself.
An outline of a flow chart for E.R. is shown in Figure 6. Clearly a great deal more data would be needed before the priority orderings for the various interrupts and components of the routines could be established.
The Rule Book concept
The usual way to test an information processing model is to program it on a digital computer and to simulate the physical system on the same computer so that the model's control performance can be studied. There are objections to this method in the present case: for example, the perceptual judgments required in many of the routines could not be incorporated into a computer program at present.
A possible alternative test procedure is to use a method which can be called ’the control game’, which is similar in some respects to the Turing 'Imitation Game '. This involves presenting a novice with the task and giving him the model to follow to see whether he will achieve a similar performance to that of an experienced man. The model would have to be set out in a special program form containing the look-up tables, routines and various executive scans so that it could be followed by a human subject. Essentially, this would be a book of rules for action in all the circumstances which arise on the plant and it would have to have a clock system to generate the sampling acts. The rule book would in fact be the model program, but presented in the form of a book with a linking page reference system which would take the game-player through the routines like a branching programmed textbook. The perceptual judgment problem could be overcome by giving the game-player, on request, perceptual classifications made by an experienced man who is standing by but taking no other part in the test. This test procedure has the merit that it takes place in the real situation and observations of the way the novice performs would immediately disclose any deficiencies in the rule book if the subject could not take a particular decision at any time.
A key concept underlying this rule book system is the assumption that human operators categorise continuous, and to some extent discontinuous, variables into a limited number of categories. For example, speeds of moving objects will normally be put into 5 or 7 perceptual categories.
If all the variables in a system are categorised in this way, the state of the system at one point in time can be described in terms of a bounded space in a multi-dimensional system space. For example, if a system had only two variables, say temperature and humidity, and these were each categorised into three classes, the system as seen by the observer would be in one of nine possible system states which form a two-dimensional matrix or array. It should be noted that humans do not necessarily categorise variables into equal intervals and some categories will expand in range as the absolute value of the variable increases. This means that the system states will not always be of the same 'volume' in a multi-dimensioned space. It is clear from observation of the ovenman that he categorises the variables in the oven system and it is possible to construct a multi-dimensioned space which represents his view of the oven system. The variables put into the system space will be temperature, baking times, rates of change of temperature and the like. This concept of the system space with its separate states is important because without it the ovenman is faced with an infinite, or almost infinite, number of conditions or states in which the oven system could be. It is unlikely that the ovenman will carry an actual multi-dimensioned system space for the complete system in his head, more likely there will be a large number of two- or three-dimensioned subsystem spaces available. The position is probably even more complicated since the categorisations used by humans are not fixed in time and immediate past experience can change the category structure as adaptation level studies show.
There will be rules for transforming one system state to another, or rather, for moving from one state to another. These rules will be actions or patterns of actions which operate on the system to change it in a desired direction. So for each system state there will be a set of actions or action states, which if applied will take the system to a pre-determined new state. Where the system states represent a system which has an operating goal there will be an action state for each system which will move the system towards the goal. It is suggested that for a skill of the kind described here, the ovenman builds up a system state model together with a matching action state matrix which incorporates the decision rules necessary to achieve control. Further exploration of the system state/action state concepts can be found in Cooke (1965) and Beishon (1966).
6. Concluding Remarks
This attempt to analyse the skill of the ovenman illustrates the previously mentioned finding that what appear to be simple skills in fact involve much greater complexities than may be expected. The attempt to build a model at least produces a structure which, although far from perfect, is productive in suggesting concepts and specific mechanisms which can be tested against the data or by further experiment. It also suggests reasons why the skill might be difficult for a novice to perform and from this methods for aiding or training new operators can be devised. Improvements to the present set-up can be put forward: for example, much of the memory load taken up in tracking batches of cakes down the oven could be removed by using simple tag devices or mimic models geared to the conveyor speed.
The author would like to express his thanks to the management of the bakery and to the ovenman for his patient cooperation. Thanks are also due to Mr. J.E. Crawley, Mr. E. Foster and Miss N. Petty for assistance in collecting data. The project was supported by a grant from the Social Science Research Council.
-
References
Beishon, R. J. (1966) A study of some aspects of mental skill in the performance of laboratory and industrial tasks. D. Phil. thesis, University of Oxford.
Cooke, J.E. (1965) Human decisions in the control of a slow response system. D. Phil. thesis, University of Oxford.
Gagne, R.M. (1963) Psychological Principles in System Development. New York : Holt, Rinehart & Winston.
Access to other papers from the Home page.
Comments
Post a Comment