A Formal Method for Analysing Field Data and Setting the Design Requirements for Scheduling Tools Peter G. Higgins School of Engineering and Science Swinburne University of Technology Hawthorn, Australia, 3122 phiggins@swin.edu.au Abstract A methodology is presented that can be used to analyse scheduling behaviour by formally structuring field data collected from the manufacturing environment. The method links the tasks used in controlling production to the environmental factors to various levels of abstraction in regard to ends and means. It is an extension of Cognitive Work Analysis (CWA), which incorporates two different types of analysis: Work Domain Analysis (WDA) and Control Task Analysis (CTA). WDA uses an abstraction hierarchy (AH) - a generic framework for describing goal-oriented systems - to describe a system in a way that distinguishes its purposive and physical aspects. WDA is event independent and is quite separate from Control Task Analysis (CTA), which is a subsequent event dependent analysis of the activity that takes place within a work domain. The discussion leads to the expected benefits for tool designers and tool users from of using this methodology. 1 Introduction In intermittent job shops or industries where there is inherent uncertainty or where human judgment is necessary, production scheduling is generally acknowledged to be a skilled craft practised by experienced human schedulers (McKay & Wiers, 1999). They are practical persons who understand the capabilities of machines and work practices within their domain. Their decision-making behaviour is mostly rational and goal directed (Higgins, 1999). Their decision strategies are often complex and hinge on their awareness of the subtle relationships between factors that comes from an intimate knowledge of the plant, products, and processes (Higgins, 2001). Knowledge and intuition, gained through years of first-hand experience, are the principal tools they employ to generate and maintain satisfactory schedules (Rodammer & White, 1988). They produce realistic schedules by balancing many competing and conflicting goals. Numerous constraints imposed by environmental factors restrict the degrees of freedom on decision choices. Their goals and the means they use to assess performance depend upon context. Various cognitive factors also influence decision-making activities. Their behaviour is situated activity embedded in the particular work environment (Suchman, 1987). They have to make effective decisions in circumstances where there is no clear prediction of the state of the system, within an environment in which information regarding jobs, materials and resources are ill defined and scheduling goals are diverse. In Suchman s terms, there is an irremediable incompleteness of instructions. They act pragmatically. They don t generate alternative schedules and then compare their strengths and weaknesses. Instead, they recognise typical situations and ways to respond. Predisposed towards
actions that require little expenditure of time and cognitive effort, they behave like Klein s (1989) proficient decision maker, who evaluates possible responses one at a time. He argues that proficient decision makers try to anticipate what would happen if they carry out a specific action, by imagining its execution in the specific working environment. For simple cases, they easily recognise the situation and know straight away how to act. Klein calls this recognition-based reaction. It is similar to Rasmussen s (1986) rule-based level of performance. For more complex cases, they consciously evaluate feasible choices. To support their decision-making behaviour, schedulers need software tools that help them seek patterns within data, recognise familiar work situations, and explore different decision-making strategies under novel circumstances (Higgins, 1999). How can designers develop software tools to support schedulers, if human behaviour in job-shop scheduling is too complex for complete specification by rules? Make the design focus the support of situated activity. Preserve their initiative to evaluate with utmost control: place them at the centre of the decision architecture. 2 Design Requirements via Modelling Decision Making Development of the design requirements of software to support scheduling as practised depends upon an analysis of the socio-technical system: the manufacturing system and the problem-solving operations of human decision-makers. The analysis is posited on a systems-thinking context in which production schedulers, whose decisions are perceived to be rational and goal-directed. It uses a formal language that encompasses both the engineering system and the problem-solving operations of the human decision-maker. It has a framework that acts as a template on which to plot the information used in the making of decisions. In forming structural models of scheduling activities on which a human-computer scheduling system can be designed, it is not necessary to detail the actual mental processes engaged by schedulers. A model has to support activities associated with the conceptual models held by domain experts, without necessarily preposing their mental models (Wilson and Rutherford, 1989). Higgins (1999) proposed a formal methodology for representing the sources and of information used by decision-makers. His method extends Cognitive Work Analysis (CWA), developed by Rasmussen (1986) and further elaborated by Sanderson (1998) and Vicente (1999). CWA is a pragmatic systems-based approach to the analysis, design, and evaluation of humancomputer interactive systems. It is a structural model of the activities decision makers may use (Rasmussen, 1986). It is a mechanism for generating descriptions of system purpose and form, explanations of system ing and observed system states, and predictions of future system states. It incorporates two different types of analysis: Work Domain Analysis (WDA) and Control Task Analysis (CTA). Because scheduling is an intentional process, goals vary. Higgins (1999, 2001), therefore, added a goal structure as a third component to the analyses. WDA describes a system of resources in a way that distinguishes its purposive and physical aspects; its concern is what is being acted upon. The description is at different levels of abstraction (see Figure 1). The manufacturing processes are expressed in different languages at the levels of physical and physical device. The construction of a WDA can begin at any level. An expedient starting point is the level of physical. The generic physical s required for processing production orders are depicted diagrammatically. Under this is a depiction of the manufacturing process in terms of physical devices. Arcs drawn from the level of to device show the physical means for achieving the al ends. These physical s can be associated with physical devices. Each job has its own mapping between these levels, as job
shops are characterised by diversity of product. The purpose-related level depicts the purpose of the production facility. It sets the mapping between s and devices to the job s specific requirements. The priority/values are the criteria used to measure the performance of the system in regard to its al purpose. The hierarchy is one of means and ends. For a particular level, the level below depicts the means for achieving its ends. Functional purpose long-term financial return Priority / Value shortterm financial viability repeat custom Purpose related Process to specified attributes jobno Width Cust Parts Holes Depth Paper type Colours Fold Paper Sheet Paper Sheet Perforate Cut into sheet Special Finish Collate & finish Cut device Reel of Stack of AKIRA Trident Hunkeler Minami Sanden Bowe Figure 1: Work Domain Analysis for scheduling presses showing feasible means-ends links for a particular job specification CTA is an event dependent analysis that shows how activities are directed towards specific goals. It has a structure of recognition-action cycles that provides a framework for locating information used in decision-making. Using concepts of skills, rules and knowledge, decision-making
processes can be structured in a framework made up of recognition-action cycles: known as Rasmussen s decision ladder. It focuses on the user s mental context. For each specific goal, a decision ladder is used as a template to represent the control activities associated with decision behaviour that is directed towards the goal. Support for rule-based decisions relates to the shaded region in Figure 2 (Higgins, 1999). On observing cue-patterns in the data, a decision maker (person or software) sees a particular rule as suitable. If the procedural steps for the rule can be recalled, then they are immediately executed, otherwise, the steps have to be first determined. With no rules clearly pertinent, schedulers exercise deep knowledge of the scheduling practice within their domain. This knowledge-based behaviour is the part of the decision ladder above the shaded region. Information processing activities States of knowledge resulting from information processing What are we aiming for? Evaluate performance criteria Uncertain 10 6 5 11 Performance goal Knowledge-based domain What are the, given goal? KNOWLEDGE BASED ANALYSIS What is the effect on this state? Identify state Determine state 8d State 8 8c 5 State 9 9a 9b 9c Choose criterion 4 8a 8b 12 Criterion KNOWLEDGE BASED PLANNING What do these data signify? Observe information and data What is 2 happening? 3 7c Data 7 7a 7b Computer Support Rule-based domain 13 Define policy: operationalise criterion Policy 15 What policy satisfies criterion? 14 How to carry it out? Determine steps: operationalise policy 16 Alert 1 Steps 17 Activate Skill-based domain Carry out Figure 2: Computer support for rule-based decisions (Higgins, 1999) The details in the goal structure and the decision ladders vary between decision makers, as the particular problem-solving technique that a person applies depends on experiential familiarity with the task. A structural relationship exists between the various goals that may become activated at different times in decision-making activity (top left of F igure 3). It is found by mapping the actual operational objectives, which form the ultimate goals in the various decision ladders, to goals at higher levels of abstraction. The higher a goal is up the hierarchy, the less directly it relates to immediate operational activity. High-level goals tend to be attained through satisfaction of lowlevel goals, rather than being directly linked to ultimate goals of decision makers. The relationship between a decision ladder, the goal structure and the abstraction hierarchy is shown in Figure 3. There are links between the high-level goals and the al purpose and the priority/values in the means-ends abstraction hierarchy. The apex of the goal structure coincides
Which goal to choose? What is the effect on this s tate? What lies beind? What s going on? IDENTIFY p resent state of the system SET OF OBSERV. OBSERVE information & data ALERT ACTIVATION Detection of need for data processing SYSTEM STATE EVALUATE performance criteria AMBIGUITY INTERPRET for current tas k, safety, efficiency, etc. ULTIMATE GOAL Which is then then target s tate? TA R GET STATE DEFINE TA SK Select appropriate change of syst. cond. Which is the appropriate chage in operating cond.? TASK How to do it? FORMULATE PROCEDURE plan sequence of actions PR OCEDUR E EXEC UTE Coordinate manipulations with the al purpose level of the abstraction hierarchy, and the level immediately below coincides with the highest-level priorities in the abstraction hierarchy. Functional purpose Maxi mise long-term financial return 1D long - term financial return 2D Priority / Value Purpose related shortterm financi al vi ability Process to jobno specified attributes Width repeat custom Cust Holes Depth Colours Parts Paper type 1C 2C 3C Fold Special Finish 1B 2B 3B 4B 5B 6B 7B 8B 9B Paper Perforate Cut into sheet Collate & finish Cut Sheet Paper Sheet 1A 2A 3A 4A 5A 6A 7A 8A 9A 10A 11A 12A 13A 14A 15A 1 2 3 4 device Reel of AKIRA Hunkeler Minami Sanden Bowe Stack of Trident Figure 3: The relationship between the goal structure, decision ladder and abstraction hierarchy. References Higgins, P. G. (1999). Job Shop Scheduling: Hybrid Intelligent Human-Computer Paradigm. Ph.D. Thesis, The University of Melbourne, Australia. Higgins, P. G. (2001). Architecture and Interface Aspects of Scheduling Decision Support. In B. MacCarthy and John Wilson (Eds.) Human Performance in Planning and Scheduling (pp. 245-281). London: Taylor & Francis. Klein, G. (1989) Recognition-primed decisions. In W. B. Rouse (Ed.) Advances in Man-Machine Systems Research, Vol. 5 (pp. 47-92). Greenwich, Connecticut: JAI Press. McKay, K.N., & Wiers, V.C.S. (1999). Unifying the Theory and Practice of Production Scheduling. Journal of Manufacturing Systems, 18 (4), 241 255. Rasmussen, J. (1986). Information Processing and Human Machine Interaction: An Approach to Cognitive Engineering. New York: North-Holland. Sanderson, P. M. (1998). Cognitive work analysis and the analysis, design, and evaluation of human-computer interactive systems. In P. Calder and B. Thomas (Eds.) Proceedings 1998 Australian Computer Human Interaction Conference, OzCHI 98 (pp. 220-227). IEEE. Suchman, L. A. (1987). Plans and Situated Actions: The Problem of Human-Machine Communication. Cambridge UK: Cambridge University Press. Wilson, J. R., & Rutherford, A. (1989). Mental Models: Theory and Application in Human Factors. Human Factors, 31 (6), 617-634. Vicente, K. J. (1999). Cognitive Work Analysis: Towards Safe, Productive, and Healthy Computer-based Work, Hillsdale, NJ: Lawrence Erlbaum Associates.