mokbil

THESIS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY

Levels of Automation in Production Systems

JÖRGEN FROHM

Department of Product and Production Development

Production Systems

CHALMERS UNIVERSITY OF TECHNOLOGY

Göteborg, Sweden, 2008

Levels of Automation in Production Systems JÖRGEN FROHM ISBN 978-91-7385-055-1

© Copyright 2008 JÖRGEN FROHM

Doktorsavhandlingar vid Chalmers tekniska högskola Ny serie nr. 2736 ISSN 0346-718X

Published and Distributed by Department of Product and Production Development Chalmers University of Technology SE-412 96 Göteborg, Sweden Telephone + 46 (0)31-772 1000

Front cover: Picture by Carl-Johan Carlsson in the technical report "Mer flexibel användning av industrirobotar i verkstäder - innefattande enklare sätt att utnyttja sensorer" by Abrahamsson, Carlsson, Widfeldt, 1991/1993, IVF.

Printed in Sweden by Chalmers Reproservice Göteborg, Sweden 2008

To Helena for faith, hope and love

LEVELS OF AUTOMATION IN PRODUCTION SYSTEMS

ABSTRACT

Although automation is often seen as an efficient way to achieve cost-efficient production and to relieve humans from heavy or dangerous tasks, it also has its drawbacks. Earlier research has shown that increasing levels of automation in unforeseen production situations can be related to production disturbances. The human operator that can handle those unforeseen situations does not always have the ability to interpret present and future production situations, based on available information from the production system.

The aim of this thesis is to theoretical and practicable development of the concept of Levels of Automation (LoA) in production systems and to improve the distribution of functions and tasks between humans and automation. A systems approach was adopted and an abductive research approach chosen, since the underlying data are based on qualitative analysis of the literature and observations, as well as individual and consensus views of automation. The empirical studies were conducted as seven case studies in order to develop a LoA taxonomy and a LoA measurement methodology. An exploration of existing taxonomies of LoA was carried out by means of a literature review, and the Swedish industry’s views of automation were explored through a Delphi survey. Also, two reference scales for assessment for LoA was developed.

The results of the research show that the level of information automation, from an industrial perspective, has primarily been seen in terms of an increase in the pace of information and providing decision support in order to help the human in understanding the situation. However, this research also demonstrates that, from a production perspective, it is important to recognise that many automated processes in production involve automation of physical tasks, which are for the most part controlled by computers.

It is also concluded and verified that the two reference scales presented for levels of automation are applicable to production tasks and that the level of automation in production systems can be assessed, measured and analysed with the DYNAMO methodology.

Keywords: production, automation, task allocation, levels of automation, measurement

i


ACKNOWLEDGEMENTS

This work would not have been accomplished without help and support from a number of people whom I would like to take the opportunity to thank.

First and foremost, I thank my supervisor, Assistant Professor Johan Stahre, for always patiently reading and commenting my writing, helping me through with your outstanding energy and problem focused coping, and giving me the opportunity to make this journey. I have appreciated his encouragement, great patience, and not least facilitating financial support throughout my research studies. The financial support were received from two funding agencies: the Swedish Foundation for Strategic Research (SSF) (through the program for Production Engineering Education and Research – PROPER, and through ProViking Graduate School), and the Swedish Governmental Agency for Innovation Systems (VINNOVA), which are gratefully acknowledged. Moreover, I would like to thank the companies involved in the research projects, which enable studies in real production environments.

My deepest gratitude and appreciation also goes to my three co-supervisors, who have guided me along the way during various phases of my research studies: Irma Alm (P.hD.), for inspiring methodological and theoretical discussions. Professor Monica Bellgran, for giving me faith in my work and inspiring discussions, Assistant Professor Mats Winroth, for inspiring discussions and valuable comments during the last daft of the manuscript.

My acknowledgement and appreciation also goes to my co-authors: Veronica Lindström (Lic. Eng.), Arvid Karsvall, Anna Thunberg (Lic. Eng.), Industrial Researcher Ulrika Harlin (Lic. Eng.), Consultant Martina Berglund (Lic. Eng.), Kerstin Dencker (M.Sc.), Jessica Bruch (M.Sc.), Psychologist Matthias Hassnert and professor Erik Hollnagel for helping me view my field of research from slightly different perspectives.

I am also grateful to the friendly and inspiring colleagues, both present and former, I have met as a Ph.D. student, at PROPER, ProViking and the Department of Product and Production Development at Chalmers University of Technology for creating a stimulating environment as well as fun activities on and after work.

iii

iv

My thanks also go to Professor Mats Jackson, Milun Milic (Lic. Eng.), Mikael Hedelind (M.Sc.) and Andreas Ask (Lic. Eng) for help with the cases studies within the Factory-in-a-Box project. I am also grateful to all the help I have got from the participating companies, within the ABA and DYNAMO projects.

Special thanks go to Sabina Fjällström (P.hD.), who has been an appreciated companion throughout my time as a Ph.D. student, Per Gullander (Ph.D.), whom I thank for his special skills in visualisation, as well as for encouraging and fruitful discussions.

A special thanks also goes to Johanna Matti (M.Sc.) and Sara Tiger-Ring (M.Sc.) for showing that the LoA taxonomies and scales worked, and Åsa Fasth (M.Sc.) and her master students for showing that the DYNAMO methodology worked.

My appreciation also goes to Mikael Cederfeldt (PhD), for his innovative and eloquent style, and to language editor Janet Vesterlund for her knowledge of English and constructive suggestions given on late drafts of this thesis. I hope the result of her efforts has made my research more accessible.

Last but not least, I would like to give my warmest appreciation to the ones I love the most, Helena Holm, my mother Ewa, my younger brothers and sister Daniel, Sebastian and Anna for your never-ending support and encouragement, (at last I finished school).

Jörgen Frohm Göteborg, January 2008


PUBLICATIONS

APPENDED PAPERS

The following published, appended papers constitute the basis of this thesis. The paper denotations and references are followed by a description of the author’s contribution and are referred to by Roman numerals in the text.

Paper I Frohm, J., Karsvall, A., and Hassnert, M. (2003). System perspective on task allocation and automation design. In Proceedings of the 8th symposium on automated systems based on human skills and knowledge, Göteborg, Sweden, September 22-24, 2003.

Contribution: Frohm and Karsvall initiated and wrote the paper. Frohm added theoretical references to task allocation. The explorative case study was performed by Frohm, Karsvall and Hassnert. Frohm was the corresponding author and presented the paper.

Paper II Frohm, J., Lindström, V. and Bellgran, M. (2005). A model for parallel levels of automation within manufacturing. In Proceedings of the 18th

International Conference on Production Research, Salerno, Italy, July 31 – August 4, 2005.

Contribution: Frohm initiated and wrote the paper. Frohm was the corresponding author and presented the paper.

Paper III Frohm, J., Granell, V., Winroth, M. and Stahre, J. (2006). The industry’s view on automation in manufacturing. In Proceedings of the 9th symposium IFAC on Automated Systems Based on Human Skills and Knowledge, Nancy, France, May 22-24, 2006.

Contribution: Frohm initiated the Delphi survey together with Granell, Winroth and Stahre. Frohm, Granell and Winroth analysed the data from the two Delphi rounds. Frohm was the corresponding author and presented the paper.

v

Paper IV Frohm, J., Karsvall, A., Thunberg, A. and Stahre, J. (submitted). Possibilities for Automated Decision-making in Industrial Systems. (Submitted for publication in International Journal of Industrial Ergonomics).

Contribution: Frohm, Karsvall and Stahre initiated the paper and wrote it together with Thunberg. Frohm and Karsvall carried out the first case study, Frohm carried out the second and Thunberg the third case study. Frohm contributed references from manufacturing.

Paper V Frohm, J., Lindström, V., Stahre, J. and Winroth, M. (submitted). Levels of Automation in Manufacturing. (Submitted for publication in Ergonomia).

Contribution: Frohm and Lindström initiated the paper and wrote it together with Stahre and Winroth. Frohm and Lindström collected the data. Frohm contributed references from human factors.

Paper VI Granell, V., Frohm, J. Dencker, K. and Bruch, J. (2007). Validation of the Dynamo Methodology for Measuring and Assessing Levels of Automation. In Proceedings of the Swedish Production Symposium, Göteborg, Sweden, August 28-30, 2007.

Contribution: Frohm and Granell initiated the paper and wrote it together with Dencker and Bruch. The case study was conducted by Granell, Dencker and Bruch, while Frohm supervised and collected validation data.

Paper VII Fasth, Å., Stahre, J. and Frohm, J. (2007). Relations between parameters/performers and level of automation. In Proceedings of the IFAC Workshop on Manufacturing Modelling, Management and Control, Budapest, Hungary, November 14-16, 2007

Contribution: Fasth initiated the paper and wrote it together with Stahre and Frohm.

The author’s contribution to the appended papers Paper

IPaper

IIPaper

IIIPaper

IVPaper

VPaper

VIPaper

VIIPlanning 2 3 3 2 3 2 1Data collection 2 3 2 2 2 3 1Analysis 2 2 2 2 2 2 1Writing 2 3 3 3 2 2 13 - responsible 2 - high participation 1 - participation

vi


TABLE OF CONTENTS

CHAPTER 1 INTRODUCTION ......................................................................................1

1.1 BACKGROUND ............................................................................................................1 1.1.1 Definitions of automation ................................................................................ 2

1.2 AIM AND OBJECTIVE ...................................................................................................3 1.3 RESEARCH QUESTIONS................................................................................................3 1.4 DELIMITATIONS..........................................................................................................4 1.5 THESIS OUTLINE .........................................................................................................5

CHAPTER 2 THE RESEARCH PROCESS ...................................................................7

CHAPTER 3 METHOD ..................................................................................................11

3.1 THE SCIENTIFIC APPROACH.......................................................................................11 3.1.1 Systematic combining .................................................................................... 123.1.2 Research strategy, quantitative or qualitative ............................................... 12

3.2 EMPIRICAL METHODS ...............................................................................................12 3.2.1 Case studies ................................................................................................... 123.2.2 Observations .................................................................................................. 153.2.3 Interviews....................................................................................................... 153.2.4 Delphi............................................................................................................. 153.2.5 Hierarchical task analysis ............................................................................. 16

3.3 THEORETICAL METHODS...........................................................................................17 3.3.1 Literature review............................................................................................ 17

3.4 METHODS USED IN APPENDED PAPERS ......................................................................18 3.4.1 Methods used in paper I................................................................................. 183.4.2 Methods used in paper II ............................................................................... 193.4.3 Methods used in paper III .............................................................................. 193.4.4 Methods used in paper IV .............................................................................. 193.4.5 Methods used in paper V................................................................................ 203.4.6 Methods used in Paper VI.............................................................................. 203.4.7 Methods used in Paper VII ............................................................................ 20

CHAPTER 4 FRAME OF REFERENCE......................................................................23

4.1 PRODUCTION SYSTEM AND MANUFACTURING SYSTEM .............................................23 4.2 HUMAN FACTORS IN PRODUCTION SYSTEMS .............................................................24

4.2.1 Information processing .................................................................................. 284.2.2 Mental models................................................................................................ 314.2.3 Mental workload ............................................................................................ 32

4.3 HUMAN ROLES IN AUTOMATED SYSTEMS..................................................................32 4.4 EFFECTS AND CONSEQUENCES OF AUTOMATION ON HUMANS...................................34

vii

4.4.1 Trust in automation........................................................................................ 354.4.2 Situation awareness ....................................................................................... 364.4.3 Skill degradation............................................................................................ 374.4.4 ‘Out-of-the-loop’............................................................................................ 384.4.5 Change in roles .............................................................................................. 38

CHAPTER 5 RESULTS ..................................................................................................39

5.1 RESULTS PAPER I ......................................................................................................39 5.2 RESULTS PAPER II.....................................................................................................40 5.3 RESULTS PAPER III ...................................................................................................41 5.4 RESULTS PAPER IV ...................................................................................................43 5.5 RESULTS PAPER V.....................................................................................................43 5.6 RESULTS PAPER VI ...................................................................................................45

5.6.1 The DYNAMO methodology........................................................................... 475.7 RESULTS PAPER VII ..................................................................................................51 5.8 ADDITIONAL VERIFICATION OF THE DYNAMO METHODOLOGY ..............................52

5.8.1 Truck manufacturing...................................................................................... 525.8.2 Manufacturing of component for indoor climate........................................... 525.8.3 Manufacturing of modules for cooling .......................................................... 535.8.4 Pharmaceutical industry ................................................................................ 535.8.5 Summary of the additional verification of the DYNAMO methodology......... 53

CHAPTER 6 DISCUSSION ............................................................................................55

6.1 DISCUSSION OF RESULTS...........................................................................................55 6.1.1 Is the concept of levels of automation relevant in production? ..................... 556.1.2 Can the level of automation in production be measured? ............................. 566.1.3 What consequences can be expected due to changes in the level of

automation in production?............................................................................ 576.2 DISCUSSION OF METHOD...........................................................................................59 6.3 ASSESSMENT OF RESEARCH QUALITY .......................................................................60

6.3.1 Transferability................................................................................................ 606.3.2 Credibility ...................................................................................................... 606.3.3 Dependability and confirmability .................................................................. 61

6.4 FURTHER RESEARCH ................................................................................................61

CHAPTER 7 CONCLUSIONS .......................................................................................63

7.1 IS THE CONCEPT OF LEVELS OF AUTOMATION RELEVANT IN PRODUCTION?...............63 7.2 CAN THE LEVEL OF AUTOMATION IN PRODUCTION BE MEASURED?...........................64 7.3 WHAT CONSEQUENCES CAN BE EXPECTED DUE TO CHANGES IN THE LEVEL OF

AUTOMATION IN PRODUCTION? ................................................................................64

viii


LIST OF FIGURES

Figure 2-1: The research process....................................................................................... 8

Figure 3-1: An example of a hierarchical task analysis that consists of a task goal and three sub-goals. The operations, required by the operator to reach each sub goal, constitute the most detailed level. ......................................... 17

Figure 3-2: Case studies for development of LoA........................................................... 20

Figure 4-1: Decision matrix for allocation of functions (Price, 1985) ............................ 26

Figure 4-2: Human-implemented tasks may be classified either as mission fulfilment tasks or as information and control tasks, based on the PURDUE Architecture for Manufacturing Systems (PERA) (Williams, 1999) ........................................................................................... 27

Figure 4-3: Extent of automation lines for two functional tasks (Williams, 1999)......... 28

Figure 4-4: The human information processing model presented by Wickens (1992)............................................................................................................ 29

Figure 4-5: Five simplified stages of a human-machine task, derived from the basic perceptual cyclical model of human action by Neisser (1976) ........... 30

Figure 4-6: Initial model describing hypothetical effects on varying the level of automation .................................................................................................... 35

Figure 5-1: Separation of functions into mechanisation and computerisation ................ 41

Figure 5-2: Benefits of automation.................................................................................. 42

Figure 5-3: Automation disadvantages............................................................................ 43

Figure 5-4: The DYNAMO methodology in eight steps ................................................. 46

Figure 5-5: Visualisation of the observed production flow............................................. 48

Figure 5-6: Identification of the main task in the production flow.................................. 48

Figure 5-7: Identification of the sub-tasks in the production flow by HTA.................... 49

Figure 5-8 Analysis of LoA in the Mechanical-Information-LoA diagram (form no. 5) ............................................................................................................. 51

ix

LIST OF TABLES

Table 1-1: Research questions and anticipated contribution of knowledge, inspired by Axelsson (2000) ......................................................................................... 4

Table 3-1: Type of industry and some characteristics of the seven production systems studied ............................................................................................. 14

Table 4-1: Fitts’ List, in Hoffman et al. (2002).............................................................. 25

Table 5-1: LoA-scales for computerised and mechanised tasks in manufacturing ........ 44

Table 5-2: Relevant maximum and minimum LoA for a particular machine or cell ..... 45

Table 5-3: Summery of the four verifications of the DYNAMO methodology............. 53

x


CHAPTER 1

INTRODUCTION

CHAPTER INTRODUCTION

This chapter presents the background and problem area, followed by a specification of objective and research questions. The chapter ends with delimitations and the thesis outline.

1.1 BACKGROUND

Historically, automation has proven to be an efficient way to achieve cost-effective production in discrete parts manufacturing, as well as in the process industry and other industrial areas (Satchell, 1998). In general, automation has also relieved humans from heavy, dangerous, complex, boring and time-consuming tasks (Parasuraman and Mouloua, 1996). Automation has also been extensive not only in the actual production process but also in supportive tasks (e.g. material handling, transportation and storage) according to Reveliotis (1999). Further, automation may offer solutions in highly time-critical situations in which there is insufficient time for a human operator to respond and take appropriate action, or in other types of situations where the human being proves insufficient in one or more aspects (Parasuraman et al., 2000).

Even if the ambition was in some cases to create so-called lights-out factories with complete automation in every step of production (Mital, 1997), most of those automated systems in production are still semi-automatic, where the automated system consists of a combination of automated and manual tasks together. This is especially true in assembly, which has generally been more difficult to automate at a reasonable cost (Boothroyd, 2005).

However, automation also has its limitations. Although it has been very efficient and productive, automated production and assembly systems do have drawbacks. Research has shown that frequent production disturbances related to machines and equipment are common in industries using advanced manufacturing (Järvinen et al., 1996; Ylipää, 2000; Ingemansson, 2004). Since those machines and equipment are vulnerable to unforeseen situations, they sometimes have to act beyond their pre-defined scope of action since many situations have no pre-defined solutions (Winroth et al., 2007), which may result in a loss of system control. Blanchard and Fabrycky (1998) suggested that, although it is possible to find systems or products in use today that are highly sophisticated and will fulfil most expectations when operating, experience shows that the reliability of many of these systems is low and that they are inoperative much of the time.

1

INTRODUCTION

On the other hand, the human actor as a component in the production system has to be involved in technical advancements to be able to handle machines and equipment during unforeseen situations (Karsvall et al., 2003). Consequently, both advanced technical systems and skilled human workers that work together are necessary to achieve flexible and efficient production. Robustness and flexibility are therefore, as Higgins (1999) argued, key issues for further development in the field of Advanced Manufacturing Systems (AMS). However, as product customisation demands have pushed product complexity forward (Youtie et al., 2004), they have also resulted in increasingly complex production systems, as well as an increased level and extent of automation (Sheridan, 2002). This means that automation does not necessarily lead to the expected results (Bainbridge, 1982). In fact, extensive levels of automation may result in poor system performance (Endsley and Kiris, 1995; Endsley, 1997; Parasuraman et al., 2000). Complex production systems are also generally vulnerable to disturbances, which might lead to possible Overall Equipment Efficiency (OEE) degradation (Ylipää, 2000). In parallel, a degradation of operator performance may be caused by e.g. lack of knowledge, gradual loss of specific working skills, degradation of situation awareness (Endsley and Kiris, 1995; Endsley, 1997; Parasuraman et al., 2000) or unexpected increases in the cognitive workload (Connors, 1998).

In other words, complex and partly automated systems place high demands on people to be active parts of the system, to deal with e.g. disturbances, unforeseen situations, non-automated tasks and maintenance of sub-systems. The allocation of tasks between humans and technical systems in an Advanced Manufacturing System (AMS) is therefore a non-trivial task and a central question in this thesis.

1.1.1 Definitions of automation

According to Sheridan (2002), task analysis is commonly regarded as the first step in the human-system design process. Task analysis thereby requires, in line with Sheridan (1997), that each overall task is broken down into its respective elements and tasks and the specifications of how these elements and tasks relate to one other in space, time and function. For this reason as Satchell (1998) argued, the allocation of tasks between humans and automation must be based on an adequate definition of automation.

An introductory definition of automation can be found in Satchell (1998):

“Automation is the replacement of human activity by machine activities.”

The Oxford English Dictionary (2006) defines automation in terms of manufacturing as:

“Automatic control of the manufacture of a product through a number of successive stages; the application of automatic control to any branch of industry or science; by extension, the use of electronic or mechanical devices to replace human labour.”

These two definitions indicate that there might be a tendency to regard automation as a binary feature – you have it or you do not have it. Satchell’s definition is also similar to

2


that of Parasuraman et al. (2000), who define automation as a grey scale, with partial or full automation:

“Automation refers to the full or partial replacement of a function previously carried out by the human operator.”

As seen in the definition by Parasuraman et al. (2000), partial allocation of functions in a human-machine system indicates the presence of intermediate levels of automation. Thus, even if many early definitions of automation, such as those given by Bright (1958) and Marsh and Mannari (1981), emphasise the role of the human in complex systems, the focus is still on technical tools and how they can replace or support the human. The emphasis of this thesis is on automation that partially replaces human beings and optimises the human contribution by focusing on the task allocation between the human and the technical system. Further, the thesis specifically focuses on the interaction and sharing of tasks between humans and automation.

1.2 AIM AND OBJECTIVE

The aim of this thesis is:

To contribute to theoretical and practicable development of the concept of Levels of Automation (LoA) in production systems.

The underlying objective of this thesis is to explore how Levels of Automation can be used in production systems, to gain a better understanding of how the interaction between humans and technology (e.g. automation) can be improved.

1.3 RESEARCH QUESTIONS

Based on the aim and the objective that are presented as a background, the following research questions have been posed:

RQ 1:

Is the concept of Levels of Automation (LoA) relevant in production?

From complex areas such as aviation and telerobotics it can be seen that levels of automation are relevant. The question is whether it is also relevant in production.

RQ 2:

Can the Levels of Automation in production be measured?

If” levels of automation” is relevant in production, how can it be measured?

RQ 3:

What consequences can be expected due to changes in the Levels of Automation in production?

If the levels of automation can be measured, what types of effects can be expected due to the change in levels of automation, not only when increasing LoA but also decreasing it.

3

INTRODUCTION

The contribution of knowledge with regard to the research questions stated above is anticipated in table 1-1.

Table 1-1: Research questions and anticipated contribution of knowledge, inspired by Axelsson (2000)

RESEARCH QUESTION ANTICIPATED KNOWLEDGE CONTRIBUTION

RQ 1: Is the concept of LoA relevant in production?

� A review of Levels of Automation (LoA) taxonomies

� A definition of “Levels of Automation”

� A reference scale for physical levels of automation

� A reference scale for cognitive levels of information/control automation

RQ 2: Can the Levels of Automation in production be measured?

� A methodology for measuring and assessing LoA

� Limitations on the reference scales based on relevant span of LoA

� An analysis model for LoA based on the relevant span of LoA

RQ 3:

What consequences can be due to changes in Levels of Automation in production?

� A discussion of what effects can be anticipated when increasing or decreasing LoA in production

1.4 DELIMITATIONS

The emphasis of this research is on the levels of automation (LoA) in production systems. Areas that will not be dealt with in detail in this thesis are the psychological background to how humans acquire, analyse and decide appropriate actions. Neither will this thesis cover any organisational, strategic or tactical aspects of LoA in relation to allocation of tasks between humans and technology.

The thesis also excludes any attempts to define or predict, from an automation strategy perspective, what would be the optimal level of automation in production systems.

4


1.5 THESIS OUTLINE

The thesis consists of eight chapters, the contents of which are briefly described below.

Chapter 1

INTRODUCTION

This chapter introduces the area of research by providing a short description of the background and problem area, followed by a specification of objective and research questions. The chapter ends with the delimitations of the thesis and its outline.

Chapter 2

THE RESEARCH PROCESS

This chapter provides a brief description of the research process behind the work reported in this thesis, the order in which the case studies and Delphi study were conducted and when the appended papers here written. The relation between the appended papers and the research questions is also demonstrated.

Chapter 3

METHOD

This chapter presents the methodological research approach. Some issues related to the case studies are presented, as is the concept of systematic combining. Finally, the methods used in the appended papers are described.

Chapter 4

FRAME OF REFERENCE

This chapter provides an overview of the theoretical framework of the thesis. The chapter starts with defining manufacturing and production, followed by a discussion of human factors in production systems, the role of the human and the allocation of tasks in the automated system. The chapter ends with a discussion of the effects and consequences of automation on humans.

Chapter 5

SUMMARY OF RESULTS

This chapter constructs the foundation for answering the research questions. The answers are based on the summary of the results presented in the appended papers. The chapter ends with the interpretations made in four independent cases in which the DYNAMO methodology was used.

Chapter 6

DISCUSSION

This chapter discusses the results of the study. Further, the relations between the results, the overall research aims, and methodological issues are examined. The chapter ends with some recommendations for further research.

Chapter 7

CONCLUSIONS

This chapter gives a discussion of the degree to which the research has answered the research questions and fulfilled the research objective and aim.

5

INTRODUCTION

6


CHAPTER 2



This chapter provides a brief description of the research process in this thesis and the relation between the appended papers and the research questions.

The background in Chapter 1 aimed to define the problem area and recognised industrial needs. The underlying research process followed to achieve the aim of the thesis covered a six-year period. During this process, the specific problem area of Levels of Automation (LoA) was developed and defined, as illustrated in figure 2-1.

7


Cognitive

8-10 levels

Cognitive

10 levels

Physical

10 levels

Th

eo

ry/L

ite

ratu

re

IIIIII

Research

Question

IIIIII

Research

Question Development of LoA scalesPaperMethodType of case study

companyDevelopment of LoA scalesPaperMethod

Type of case study

company

T

I

M

E

= Major contribution = Minor contribution

VI

VII

Case study

Valve

manufacturing

(G)

V

IV

Case study

Furniture

manufacturing

(F)

Case study

Robotic

manufacturing

(E)

Case study

Plant

construction

(D)

III

DELPHI

Case study

Car

manufacturing

(C)

II

Case study

Truck

manufacturing

(B)

I

Case study

Petrochemical(A)

VI

VII

Case study

Valve

manufacturing

(G)

V

IV

Case study

Furniture

manufacturing

(F)

Case study

Robotic

manufacturing

(E)

Case study

Plant

construction

(D)

III

DELPHI

Case study

Car

manufacturing

(C)

II

Case study

Truck

manufacturing

(B)

I

Case study

Petrochemical(A)

Cognitive

9 levels

Physical

9 levels

Cognitive

7 levels

Physical

7levels

Cognitive

7 levels

Physical

7 levels

Cognitive

7 levels

Physical

7 levels

Figure 2-1: The research process

The research process started with an introductory literature review (in appended paper I) in relation to case A concerning task allocation and the role of the operator in an industrial context. The aim was to explore how common models for task allocation handle the complex setting of production systems. Based on the theoretical and empirical findings, a research design was developed, including research questions, planned case studies and more specific methods to assess the expected findings.

During the work of collecting data in cases A and B, it was observed that automation in production, as well in other domains, can be seen as two types of technical support: physical and cognitive. These findings were presented in appended paper II.

In 2004, a pre-study interview was conducted to test some of the questions in the subsequent Delphi survey, which aimed to assess the state-of-the-art picture of the view on automation from an industrial perspective. The results of the two rounds of the Delphi survey were presented in appended paper III.

Based on the findings from the Delphi survey, an analysis was made of the state-of-the-art and a measurement model for assessing and analysing LoA was developed for testing and

8


verification in the planned case studies (case studies C to F). In this work done in real settings, the model for assessing and analysing LoA was continuously updated and re-worked. Appended papers IV and V were written in parallel with cases C-F.

Appended paper IV reports on a literature review done on task allocation and decision-making in relation to three case studies (cases A and B). The focus of appended paper IV was to discuss how decision-making may be automated or at least supported.

Appended paper V describes a literature review of levels of automation based on material from areas such as manufacturing and human factors.

Based on the theories and empirical findings, a measurement model for assessing and analysing LoA was presented and tested in case study G. The model is described in appended paper VI.

On the basis of the taxonomy of LoA presented in appended paper V, a discussion is given in the final appended paper VII of some of the numerous parameters and performance indicators that affect LoA and must be considered if a company wants to achieve an optimised and effective production system.

9


10


CHAPTER 3

METHOD


This chapter presents the overall methodological research approach of the thesis, and how the methods are used in the appended papers.

3.1 THE SCIENTIFIC APPROACH

According to Starrin and Svensson (1994) research is either done in order to answer questions put forward by theoretical considerations, a deductive approach, or the reverse, to view theory as something that occurs after the collection and analysis of the data, an inductive approach.

In a deductive approach, which is often used in the quantitative research area according to Starrin and Svensson (1994), the researcher begins with a theory and then gathers evidence, or data, to assess the validity of the theory. Through an inductive position, which is associated with qualitative research methods, data are collected, hypotheses are formed and possible theories are developed related to what the data uncover (Starrin and Svensson, 1994).

Research is however seldom carried out in a strictly deductive or inductive manner (Starrin and Svensson, 1994), which describes the process in this thesis well. Instead, research can, in line with Alvesson and Sköldberg (2006), be seen as a combination of the inductive and deductive approaches, which is known as abduction.

According to Alvesson and Sköldberg (2006), abduction starts from the empirical data but without rejecting theoretical conceptions. Further, abduction also seeks to find structures through which the empirical patterns found on the surface can be interpreted and understood. The process is therefore characterised according to Alvesson and Sköldberg (2006) through a continuous alteration and interplay between the theoretical and empirical areas. The findings are further continuously verified in line with Starrin and Svensson (1994) by repeated applications of earlier results to new cases. Unexpected empirical findings and new theoretical insights in line with Alvesson and Sköldberg (2006) thereby lead to successive modification of the original conceptions.

Since the research design chosen also involves interpretations of empirical case studies from a production systems context, the approach chosen can also be characterised in line with Walsham (1995) and Wynekoop and Russo (1997) as an interpretive approach. An interpretive approach can according to Wynekoop and Russo (1997) be defined as “an attempt to understand a phenomenon, by studying it in its natural context from

11

METHOD

participants’ perspectives.” Consequently, both the theoretical and empirical areas gradually develop during the abductive and interpretive process.

3.1.1 Systematic combining

Since theory cannot be understood without empirical observations and vice versa (Holme and Solvang, 2001), a continuous confrontation, as in this thesis, takes place between the theory and the empirical areas, between the evolving framework and the evolving cases. This means, in line with Holme and Solvang (2001), that the researcher is constantly moving back and forth between different research activities, as well as between theory and empirical data. Such intertwined research has earlier been presented by Dubois and Gadde (2002) and labelled “systematic combining”.

The process of systematic combining consists according to Dubois and Gadde (2002) on the one hand of matching theory and reality by moving between framework, data sources and analysis and on the other of directing and redirecting the study. Direction and redirection, which are important factors for achieving matching according to Dubois and Gadde (2002), involve the impact of different sources and data collection methods. These different sources and data collection methods are also considered to play an important role, according to Dubois and Gadde (2002), since they contribute to the discovery of new dimensions of the research problem.

3.1.2 Research strategy, quantitative or qualitative

A quantitative research strategy emphasises the data and analysis being able to be achieved by using a consistent device or yardstick to obtain finer distinctions and a basis for a more precise estimate of the degree of relationship between concepts (Holme and Solvang, 2001). Qualitative research, on the other hand, usually emphasises words rather than quantification, according to Holme and Solvang (2001). Although possibilities exist to distinguish things in terms of characteristics, as in a qualitative research strategy, finer distinctions and smaller differences are much more difficult to recognise, as noted by Starrin and Svensson (1994).

In this research, the abductive research approach was chosen since the underlying data are based on qualitative analysis of the literature and observations, as well as on individual and consensus views on automation.

3.2 EMPIRICAL METHODS

This thesis aims to develop the LoA concept in an industrial context. Consequently, empirical observations are essential to the fulfilment of the research objectives in dynamic situations. Methods for analyses in these dynamic situations are described in the following sub-sections.

3.2.1 Case studies

According to Yin (2003), case studies are considered as appropriate when studying contemporary events that may be difficult, or even impossible, to control. A major strength of case studies according to Dubois and Gadde (2002) is that in-depth, case studies provide the best opportunities for understanding the interaction between a phenomenon and its context. Further, according to Yin (2003), case studies can involve multiple cases as well as several levels of analysis in a single study. The selection of cases

12


is, however, as argued by Yin (2003), an important aspect in building theory from case studies. On the other hand, Eisenhardt (1989) for example, rejects random sampling and suggests theoretical selection instead. The selection may thereby be based on specific markets, according to Dubois and Gadde (2002), to allow control of environmental variation or selection of specific cases in order to extend the theory to a broad range of organisations.

Case studies were chosen in papers I, IV and VI from the seven companies (see table 3-1), because case study methodology, as noted above, is suitable when studying activities which may be difficult, or impossible, to control.

13

METHOD

Com

pan

y G

Valv

e m

anuf

actu

ring

Flow

-pac

ed U

-line

Sprin

g 20

07

Obs

erva

tion,

vid

eo

reco

rdin

gs a

nd se

mi

stru

ctur

ed in

terv

iew

Leve

ls of

aut

omat

ion

18 v

arian

ts

5 00

0 pr

oduc

ts/y

ear

Com

pan

y F

Furn

iture

m

anuf

actu

ring

Equ

ipm

ent-p

aced

lin

e

Sprin

g –

autu

mn

2006

Obs

erva

tion

and

sem

i stru

ctur

ed

inte

rview

Task

allo

catio

n an

d lev

els o

f au

tom

atio

n

Cust

om m

ade

230

000

prod

ucts

/yea

r

Com

pan

y E

Robo

tic

man

ufac

turin

g

Ope

rato

r pac

ed

line

Aut

umn

2005

Obs

erva

tion

and

sem

i stru

ctur

ed

inte

rview

Task

allo

catio

n an

d lev

els o

f au

tom

atio

n

Cust

om m

ade

12 0

00

prod

ucts

/yea

r

Com

pan

y D

Plan

t co

nstru

ctio

n

Ord

er to

E

ngin

eerin

g

Sum

mer

200

5

Obs

erva

tion

Task

allo

catio

n an

d lev

els o

f au

tom

atio

n

Cust

om m

ade

500

prod

ucts

/yea

r

Com

pan

y C

Car m

anuf

actu

ring

Equ

ipm

ent-p

aced

lin

e

Sprin

g 20

05 -

autu

mn

2006

Obs

erva

tion,

pa

rtici

pant

ob

serv

atio

n an

d se

mi s

truct

ured

in

terv

iew

Task

allo

catio

n an

d lev

els o

f au

tom

atio

n

4 m

odels

200

000

prod

ucts

/yea

r

Com

pan

y B

Truc

k m

anuf

actu

ring

Equ

ipm

ent-

pace

d lin

e

Sum

mer

-au

tum

n 20

03

Obs

erva

tion

and

sem

i stru

ctur

ed

inte

rview

Task

allo

catio

n an

d de

cisio

n su

ppor

t

1 m

odel

11 0

00

prod

ucts

/yea

r

Com

pan

y A

Petro

chem

ical

Cont

inuo

us

proc

ess

Sum

mer

200

2

Obs

erva

tion,

vid

eo

reco

rdin

gs a

nd

sem

i stru

ctur

ed

inte

rview

Dec

ision

-mak

ing

5 va

riant

s

5 00

0 00

0 to

ns/y

ear

Tab

le 3

-1:

Typ

e of

ind

ustr

y an

d so

me

char

acte

rist

ics

of t

he s

even

pro

duct

ion

syst

ems

stud

ied

Pro

du

ctio

n

syst

em

Typ

e of

in

du

stry

Typ

e of

la

you

t

Stu

dy

per

iod

Typ

e of

stu

dy

Un

it o

f an

alys

is

Nu

mb

er o

f va

rian

ts

Vol

um

e

14


3.2.2 Observations

According to Bryman (2004), observing by merely walking around an operating human-machine system to ‘see what goes on’ is hardly a good observational method. Yet it is an indispensable first step according to Bryman (2004), since the work tasks observed are seldom simple. Further, observations are often a useful method in collecting additional information about how operators behave in actual work situations (Yin, 2003). The collection of observation data can range from indirect, informal and casual data collection to direct and structured observations (Bryman, 2004). More precisely, direct observationis the collection of information on subject performance either directly by the observers themselves or from objective recordings of subject behaviour; indirect observation involves the provision of behaviours, attitudes and opinions by the subjects themselves or by other associated individuals who are not observers (Wilson and Corlett, 1990).

The choice was made to use observations in the case studies conducted, which are presented in appended papers I, IV, V and VI, because observation methodology is a useful method when collecting information and creating an understanding of how operators behave in actual work situations.

3.2.3 Interviews

One of the most important sources of data collection in a case study is the interview, according to Yin (2003). The research method literature shows mainly three broad interview categories: unstructured, semi-structured and structured (Bryman, 2004). Unstructured interviews are appropriate when the researcher has little knowledge of the issues in question. The respondent has an opportunity to turn the discussion towards issues he or she sees as important. Semi-structured interviews usually include an interview guide with determined question areas to ensure that the relevant issues are covered in the interview. However, according to Bryman (2004), the questions in a semi-structured interview do not need to be taken in sequence. In addition, the interviewer usually has some latitude to ask further questions in response to what can be seen as significant replies. The semi-structured approach also provides an opportunity for more systematic analysis than the unstructured, while at the same time still allowing the respondent to raise important issues, according to Bryman (2004).

The choice was made to use interviews in combination with observations in the work presented in papers III, IV and VI because additional information could be collected to gain an understanding of how operators interact in actual work situations by interviewing and listening to them on site.

3.2.4 Delphi

The Delphi method is an appreciated method for eliciting and synthesising expert opinions (Ayyub, 2001). The method has been used extensively by researchers in a wide variety of situations as a tool for expert problem solving (Linstone and Turoff, 2002). The method has also been developed in a variant that can be tailored to specific types of problems and outcome goals. See Linstone and Turoff (2002) for a description of the evolution of the method.

The purposes of this method can primarily be categorised into technological forecasting and policy analysis, according to Ayyub (2001). One variant of the technological forecasting that according to Okoli and Pawlowski (2004) has received widespread use is the “ranking type”, used to develop group consensus on relatively important issues.

15

METHOD

Technological forecasting relies on a group of experts in a subject matter of interest, and where the experts should be the most knowledgeable in specific issues or questions (Ayyub, 2001).

The basic Delphi method consists of the following steps (Williamsson, 2002):

� The initial procedure in a Delphi is to prepare, distribute and synthesise a series of problem statements for evaluation.

� A panel of experts in the field under study is set up.

� The first round of the questionnaire is sent out.

� The results of the first round of the questionnaire, together with a summary of the comments and arguments given by the respondents to support their answers, are incorporated into a second round of the questionnaire, and the original questions are repeated.

� Each respondent thus has an opportunity to reconsider his or her view in the light of other views of the panel and to answer the original questions again. This enables the participants to reaffirm original options, modify some, and “collectively brainstorm” by adding new items to the list.

� Further rounds on the same basis can be conducted if the experts do not reach consensus during the second round.

The Delphi methodology, as presented in appended paper III, made it possible to collect information from Swedish experts in manufacturing about their attitudes towards automation.

3.2.5 Hierarchical task analysis

Hierarchical task analysis (HTA) is a task description method based on the principle of an overall goal that is divided into sub-goals and the operations needed to reach these goals and the conditions under which the operations have to be performed (Kirwan and Ainsworth, 1992).

The starting point for a HTA is the task requirements, known as the task goal. Task goals are expressed in a verb-noun form (Liljegren, 2004), for example assemble hydraulic pump or perform a complete quality check. Progressive re-description is then used to identify the task requirements as sub-goals and operations (Liljegren, 2004). This process according to Liljegren (2004) describes the task behaviour in progressively finer detail. Figure 3-1 shows an example of a HTA, in a hierarchical form, which consists of a task goal and three sub-goals.

16


0. Task goal

1. Sub-goal 1 2. Sub-goal 1 3. Sub-goal 1

1.1. Operation 1 2.1. Operation 1 3.1. Operation 1

1.2. Operation 2 2.2. Operation 2 3.2. Operation 2

1.2.1. Operation 2.1 2.3. … 3.3. …

Figure 3-1: An example of a hierarchical task analysis that consists of a task goal and three sub-goals. The operations, required by the operator to reach each sub goal, constitute the most detailed level.

Data for a HTA is usually gathered either from observations of humans performing the task or by using manuals and operation instructions. If based on observations, the description is of how the tasks really are performed, which might not be as intended. If based on manuals or operation instructions, the description is how the tasks are intended to be performed, which might not be the most suitable way.

The HTA methodology, as presented by Kirwan and Ainsworth (1992), made it possible to identify the task behaviour in progressively finer detail, as done in paper V.

3.3 THEORETICAL METHODS

As noted in chapter 3.1.1, empirical data can not be understood without a connection to theoretical data. Theoretical methods are thereby essential to the fulfilment of the stated research objectives. Methods for analysing the literature on dynamic situations are described by the following sub-section.

3.3.1 Literature review

According to the Psychology Writing Center (2005) are there mainly two different approaches to how to conduct a literature review. The first approach is to choose an area of research, read all relevant articles, books etc., and organise them in a meaningful way. For example, the focus of the review can be to organise the theme as a conflict or controversy in the particular area of research. The discussion focuses first on one side and then on the findings that support the other side. The other approach is to choose an organising theme or a point that supports the researcher’s line of research and then to select the studies according to the chosen line of argumentation.

1.3. … Plan 1: Do 1.1, then …

Plan 2: Do 2.1 or 2.2

Plan 3: Do 3.1

17

METHOD

Regardless of how the literature review is organised, it will have two purposes according to the Psychology Writing Center (2005): (1) to thoroughly describe how the work was done in a specific area of research and (2) to evaluate how the work has been done. Both the descriptive and evaluative elements are important parts of the review. It is naturally not possible to simply describe earlier research without evaluating it, since the information would then just be summarised and not analysed. Moreover, if recent theories in an area are merely discussed without describing the work that has been done to test those theories, then the line of argument may lack supporting empirical evidence.

The first step in conducting a literature review according to the Psychology Writing Center (2005) is to identify and develop the topic of the review as a research question. For example, if the interest is on levels of automation, the question might be “What are levels of automation?” Keywords such as “automation” and “levels” are picked from the question. The next step is to collect, read, and evaluate relevant articles in the research area. This is done by using the keywords in the search in relevant books, journals, conference proceedings, doctoral dissertations etc. in order to obtain information on the research topic in focus. The final step is then to summarise the findings, according to the Psychology Writing Center (2005), and answer the research question in a review article.

The choice was made to use a literature review during the entire research process, and particulate in paper I, II, IV, V and VII, as it is not possible to describe and build new knowledge without evaluating and connecting the empirical data to earlier knowledge in the research area.

3.4 METHODS USED IN APPENDED PAPERS

In the following sub-sections are the generally described methods presented in chapter 3.1–3.3 connected to the seven appended papers.

3.4.1 Methods used in paper I

The method used in paper I was an explorative field study in an automated process control room at a refinery during two nightshifts (case A). The control room included three separate workstations, where each workstation was manned by two console operators. There was also one process supervisor in the control room and an additionally three to four factory floor operators on the plant floor. Each of the workstations in the control room consisted of several process control terminals (displays and keyboards), a set of work protocols and other printed material, and a communication radio. The study consisted of three observations. A first observation was made at the crude distillation workstation, at which two observers focused on the work of one process operator, and this was followed by a semi-structured interview with the same operator. The second day, two more observations were made, one at the previous workstation and one at the desulphurization process control. These observations focused on two different process operators, who were also observed by two researchers each and then interviewed in semi-structured interviews. In all cases, the observations took about six hours to complete, and the observers made video recordings and written protocols of events that were observed. In the protocols, the times for events were logged as a means of enhancing inter-observer reliability. Due to the small sample of subjects, only descriptive analysis and a summary of the results were conducted. The observational context was chosen in cooperation with representatives from the involved company taking consideration to the needs of the project and the company’s need for minimal disruption of production.

18


3.4.2 Methods used in paper II

Paper II consisted of a literature review to compile theories on how LoA taxonomies from areas such as aviation and telerobotics can be used in manufacturing. The literature review was carried out by using Science Direct, Compendex, Inspec, IEEE Xplorer, Ergonomics Abstract and Google Scholar. The following keywords in different combinations (each keyword was used truncated) were used: levels, automation, task, allocation, control, production, and manufacturing. These were selected in order to grasp the most essential aspects of task allocation and automation in manufacturing systems.

3.4.3 Methods used in paper III

Paper III started with a preliminary interview survey among 16 respondents from seven medium to large sized Swedish manufacturing companies. The data from the interviews were collected in semi-structured interviews that included three main sections. The first part focused on the respondents’ background, experience and especially expertise, since the questions in the subsequent parts were based on the manufacturing practice in which the respondents had most experience.

The second part of the interview was based on questions focusing on the respondents’ view of automation and on the allocation of tasks between humans and technology. In the third and final part of the interview, the topics focused on levels of automation and if and how it could be measured.

Based on the results of the preliminary survey, a second survey was conducted, founded on the Delphi methodology. The purpose of the survey was to partly capture the industry’s view of automation and the perception of terms such as “levels of automation”. The Delphi questionnaire was then put on a website to be answered by the respondents. The sample for the Delphi survey was taken mainly from production and industrial managers from 108 industrial companies in Sweden. Out of the 85 respondents, 62 answered the questionnaire in the first round. The results from the first round of the survey were then fed back to the remaining 62 respondents via e-mail for a second and final round, where 45 respondents replied. This enabled the participants to reaffirm original options, modify some and “collectively brainstorm” by adding new reflections.

3.4.4 Methods used in paper IV

Paper IV consisted of a literature review, similar to the review in paper I, to compile theories on decision-making in real-time practice and task allocation in production. Following the literature review, two case studies were conducted to observe how operators interact with equipment and co-workers in order to handle everyday decision situations. The first study (case A), which is the same as in paper I, was conducted in an automated process control room at a refinery in Sweden.

The second study (case B) was conducted in a manufacturing system that processes complex machined products. The study was done during four-shift operations, and the selected manufacturing system was a highly automated manufacturing line. Its manufacturing was distributed along a line of 17 machines for processing raw material. Since the computer network was integrated into virtually all the manufacturing equipment, each operator could manage and control several machines in a particular part of the line. It was thus possible to follow one operator, and the observations were noted in observation

19

METHOD

protocols and followed up in a semi-structured interview to gain a deeper understanding of the working procedures.

3.4.5 Methods used in paper V

The literature review in paper V, which was an extension of the literature review in paper II, was carried out to grasp the most essential aspects of how levels of automation have been used earlier, in areas such as aviation and telerobotics. The selected definitions and taxonomies were studied and classified into mechanisation and computerisation, and their relevance to the manufacturing industry was discussed.

3.4.6 Methods used in Paper VI

The development and implementation of the DYNAMO measurement methodology took place during the years 2004 to 2007 in seven case studies (cases C to G in figure 3-2, and cases C to G in the appended papers). The inductive research design chosen was a single case study design (Yin, 2003), where the results and findings of each case study were transferred together with theory to the next case

Figure 3-2: Case studies for development of LoA

Based on the findings of the first six cases (C to F), the method was rewritten into the final DYNAMO methodology, which was tested in case study G (case G in the appended paper). The methods used in the case studies were semi-structured interviews, informal discussions and direct and participatory observations

3.4.7 Methods used in Paper VII

The purpose of the literature review in the final appended paper, VII, was to review some of the numerous parameters and performance indicators that affect LoA and must be considered if a company wants to achieve an optimised and effective system. To reduce the complexity caused by multiple relations between different parameters and areas (see figure 1-3 in paper VII), the review was limited to one main area (flexibility) and two

Case D – Plant construction

Case C –Car manufacturing

2006 2007

Case E – Robotic manufacturing

Case F – Furniture manufacturing

Case G – Valve manufacturing

2005

20


parameters (flow and time). This was done because the main area chosen and the parameters were those that were most important according to the four points listed in paper VII. The remaining main areas and parameters mentioned earlier in the paper were considered to be unlimited or non-existent.

21

METHOD

22


CHAPTER 4

FRAME OF REFERENCE


This chapter provides an overview of the theoretical framework of the thesis.

4.1 PRODUCTION SYSTEM AND MANUFACTURING SYSTEM

A pioneer in manufacturing engineering, Eugene Merchant, suggested to the production engineering research community that its previous focus on automation of processes and stand-alone automation should be expanded to cover what he called “the manufacturing system” (Merchant, 1961). Merchant claimed that such a system should include the whole chain of events starting with the product designer’s activities and ending with the delivery of finished parts. Automation efforts should then be applied to the design, programming and manufacturing as well as the machine and its control system, thus covering what we today call the product realisation chain. Further, Chryssolouris (1993) and Chapanis (1996) suggested that an “equipment system” should include both human and technological resources as well as procedures, software and facilities, all intrinsically linked to form an efficient system.

However, the word system, as in manufacturing system, production system and equipmentsystem, is often used in somewhat ambiguous ways and the need for clearer definitions has been realised by researchers and organisations (Winroth, 2004). An often used and common sense definition of a system is “a set of interacting units or elements that form an integrated whole intended to perform some function” according to (Skyttner, 2001). Translated into everyday language, a system can be expressed in line with Skyttner (2001), as any structure that exhibits order, pattern and purpose. This holistic approach to defining a system is also discussed in a socio-technical context by Eijnatten et al. (1993). Eijnatten et al. claimed that a system may consist of three different elements that must be considered as a whole: a technical sub-system, a social sub-system and a work system design comprising an organisational structure and process.

As noted by Winroth (2004), the words production and manufacturing have been interpreted in many different ways, causing many discussions. There are however some internationally accepted definitions. The International Academy for Production Engineering (CIRP) defines production as “the result or output of industrial work in different fields of activity, e.g. agricultural production, oil production, energy production, manufacturing production” (Chisholm, 1990). As a general concept, production can thereby be said to cover all activities that are needed to process raw materials into finished products.

23

FRAME OF REFERENCE

Manufacturing, on the other hand, is defined by CIRP as “a series of interrelated activities and operations involving the design, materials selection, planning, production, quality assurance, management, and marketing of the products of the manufacturing industries.” In contrast to production, manufacturing includes the whole chain of events starting with the product designer’s activities and ending with the delivery of finished parts to the market (Merchant, 1961).

The confusion of the terms production and manufacturing is however connected to the definition of manufacturing production, since production is often used as an abbreviation of manufacturing production (Chisholm, 1990). CIRP defines manufacturing productionas “the act or process of actually physically making a product from its material constituents, as distinct from designing the product, planning and controlling its production, assuring its quality.”

4.2 HUMAN FACTORS IN PRODUCTION SYSTEMS

As noted in the introduction of this thesis, automation has historically proven to be an efficient way to achieve cost-effective production in discrete parts manufacturing, in the process industry and in other industrial areas (Satchell, 1998). However, automated production systems also have drawbacks. This insight into the limitations of automation during the last two decades has resulted in a better understanding of the importance of the human operator as a controller and supervisor (Chapanis, 1996; Billings, 1997). This insight was also facilitated by observations in inter-disciplinary fields such as human-machine interaction (HMI) and other user-centred approaches to system design that acknowledge that the most advanced systems also need human supervision, decision-making and control – given that no automatic solution can be fail-proof in every aspect or all the time (Karsvall et al., 2003)

A central question in the design of physical automation and information technology has therefore not only been how to construct the best possible technical systems but also how to optimise task allocation (TA) between the technology and its users.

The initial contribution to the field of task allocation was made by Fitts in the early 1950s (Fitts, 1951), who presented a list of general tasks covering both humans and machines, illustrating where the performance of one category exceeds that of the other (see table 4-1).

24


Table 4-1: Fitts’ List, in Hoffman et al. (2002)

HUMANS SURPASS MACHINES IN THE: MACHINES SURPASS HUMANS IN THE:

�Ability to detect small amounts of visual or acoustic energy

�Ability to perceive patterns of light or sound

�Ability to improvise and use flexible procedures

�Ability to store very large amounts of information for long periods and to recall relevant facts at the appropriate time

�Ability to reason inductively

�Ability to exercise judgment

� Ability to respond quickly to control signals and to apply great force smoothly and precisely

� Ability to perform repetitive, routine tasks

� Ability to store information briefly and then to erase it completely

� Ability to reason deductively, including computational ability

� Ability to handle highly complex operations, i.e. to do many different things at once

However, since its introduction, a range of criticism has been raised regarding the applicability of Fitts’ list in systems engineering. Jordan (1963), for instance, stated that the attempt to compare the abilities of humans and machines is inappropriate, since only machines can be designed for a specific pre-defined task. According to Jordan, the idea of comparing human and technology should therefore be discarded. Yet system designers should keep in mind what humans and machines do best and embrace the fact that the two are complementary rather than conflicting entities when designing a human-machine system. In order to make use of Jordan’s ideas, Price (1985) represents the allocation of tasks in a decision space, in which the x-axis represents the “performance of humans”, scaled from “unsatisfactory” to “excellent”, and the same for the machine on the y-axis (figure 4-1).

25

FRAME OF REFERENCE

Excellent

Pa Pah

Uh

Figure 4-1: Decision matrix for allocation of functions (Price, 1985)

The x and y coordinates of a point will then represent, according to Price (1985), the predicted future merits of humans, versus the merit of machines, in fulfilling the requirement. As seen in figure 4-1, there are also at least six zones that can be recognised in the decision space for having qualitatively different implications for the allocation of functions. In line with Fitts’ notation, demands in the lower left, in the heavily shaded space (Uah), are performed unacceptably by both humans and machines. The system should, in line with Fitts and Jordan, be redesigned to avoid those functions or tasks. Demands in the shaded area at the left (Uh) are performed poorly by the human and should therefore normally be allocated to a machine (Price, 1985). Conversely, demands in the bottom shaded area (Ua), are performed poorly by the machine (automation) and should consequently normally be allocated to humans. The other areas offer, as noted by Price (1985), more latitude for choices. Areas Pa and Ph, above and below the diagonal area (Pah), are areas in which machines and humans, respectively, have an advantage in performance over each other. Demands in those areas will, as noted by Price (1985) be allocated to machines or to humans, respectively, if no strong reasons exist not to do so. Such reasons might be cost or the need to provide additional tasks to humans for reasons of cognitive support. Finally, there is the centre diagonal area (Pah), an area in which there is little difference between the performance of humans and that of machines (automation). Demands that fall into this area will, as argued by Price (1985), probably be allocated on the basis of criteria other than the relative performance of human and machines.

For example, according to the Purdue Enterprise Integration Reference Architecture suggested by Williams (1999), allocation of tasks can be based on e.g. financial, technical or social factors. These factors can in turn be organised into three categories of implemented tasks or business processes according to Williams:

� Those of the information and control type, which can be “automated” by computers or other “control” devices.

UaUah

Ph

Machine Performance

Unsatisfactory Excellent Human

Performance

26


� Those of the mission tasks or business processes, which can be automated, or rather mechanised by the “mission fulfilment” equipment.

� Those functions carried out by humans, whether of the control or mission fulfilment class.

However, only two of these classes of tasks have to be considered according to Williams (1999) when the functionality is in focus. The first is operating the ‘processes’ (e.g. physical conduct of the task) and the second is ‘controlling’ the mission in an ‘optimal’ manner (e.g. controlling and supervising the accomplishment of the physical task). However, according to Williams, there are three classes of implementation, since humans may be asked to implement any of the tasks from either functional class (figure 4-2).

Mission fulfillment tasks Information and Control tasks

Figure 4-2: Human-implemented tasks may be classified either as mission fulfilment tasks or as information and control tasks, based on the PURDUE Architecture for Manufacturing Systems (PERA) (Williams, 1999)

As noted by Williams (1999) the interpretations of “mechanisation” and “automation” can sometimes be difficult, since “mechanisation” is often referred as “automation”. In line with the argumentation in this thesis, “automation” is referred to the automation of cognitive tasks (e.g. replacement of human sensory and thought processes) and “mechanisation” to the automation of physical tasks (e.g. replacement of human and animal muscle power).

The concept in figure 4-2 is further developed by (Williams, 1999) in figure 4-3, where the dashed lines “Management determined extent of automation line” show the actualdegree of automation carried out or planned. Therefore, it is as noted by Williams (1999) the line which actually defines the boundary between the “Human and Organizational System” and the “Information and Control System” on the one hand, and the boundary between the “Human and Organizational System” and the” Mission Enabling System” on the other hand.

Mechanised tasks Automated tasks Human implemented

tasks

27

FRAME OF REFERENCE

28

MISSION ENABLING TASKS INFORMATION AND CONTROL TASKS

Maximum extent of

mechanisation

Maximum extent of

automation

AU

TO

MA

TIB

ILIT

Y

LIN

E

AU

TO

MA

TIB

ILIT

Y

LIN

E

HU

MA

NIZ

AB

ILIT

Y L

INE

HU

MA

NIZ

AB

ILIT

Y L

INE

Maximum extent of human capability to perform information and control tasks

Maximum extent of human capability for

mission enabling tasks

HUMAN AND

ORGANIZATIONAL SYSTEM

Management determined extent of automation (i.e. mechanisation) line

Management determined extent of automation line

Figure 4-3: Extent of automation lines for two functional tasks (Williams, 1999)

The “automatability line” in figure 4-3 shows the absolute extent of pure technologies in their capability to actually automate physical or cognitive tasks. Williams (1999) also notes that the extent of automatability is limited by the fact that many tasks require human innovation and can thus not be automated with presently available technology. Similarly, the “humanizability line” in figure 4-3 shows the maximum extent to which humans can be used to actually implement the tasks. The human is also limited by its physical and cognitive abilities, such as in speed of response, breadth of comprehension, range of vision, physical strength, and so forth (Williams, 1999).

Making a manufacturing system as robust and flexible as possible is consequently not only a question of how to assign the right task to the right element of the system. It is also a question of how human and automated components can support each other during different LoA, creating a manufacturing system that is as robust and flexible as possible.

One task that seems to attract special interest from an industrial perspective is the task of automating or supporting decision-making (Frohm et al., 2002). A possible reason for this may be that information processing is one of the most critical processes in any production system.

4.2.1 Information processing

To be able to understand how to support the cognitive tasks of controlling and supervising different parts of a production system, it is important to have an understanding of how


operators acquire and analyse information for decision-making and action. In the literature on human information processing, numerous models have been put forth to “explain” how people process information, e.g. Lindsay and Norman (1977), Marr (1982), Sowa (1984) and Flach and Mulder (2004). Most of these models consist of hypothetical black boxes or processing stages that are assumed to be involved in the system or process in question.

One model that opens up the hypothetical black boxes or processing stages is the model proposed by Wickens (1992), shown in figure 4-4, which gives a useful introductory explanation of different functions of human cognition and features of human information processing.

Attention Resources

Figure 4-4: The human information processing model presented by Wickens (1992)

The information process starts, as shown in figure 4-4, by a stimulus being detected by the initial sensor stage (Eyes/Ears). The sensor data are then processed by the brain to form some impressions that may be consciously attended to. In this stage, basic features of human information processing come into play, such as depth perception and pattern recognition (Wickens, 1992). These impressions may then be processed in the working memory, which can be described as the “main thinking” according to Wickens (1992). The working memory interacts with the long-term memory in which our previous knowledge is stored. From this processing a resulting response is formed and executed in the final stage (Wickens, 1992). The response is then fed back into the next turn as a stimulus, and the information in the response is acquired by the initial sensor stage (Eyes/Ears) together with new stimuli from the present situation.

In line with Endsley and Garland (2000) it can also be argued that the information process is limited not only by the working memory but also by the attention resources, which are also limited. According to Sanders and McCormick (1993), attention can have four different modes: focused attention, sustained attention, selective attention and divided attention. Focused attention is the kind of attention we use when we are actively attending to specific visual, auditory or tactile stimuli, or as noted by Logan and Etherton (1994), it is our ability to attend to one thing at the same time that we exclude everything else. Sustained attention refers to the ability to maintain a consistent behavioural response during continuous and repetitive activity (Woods et al., 1995). Further, selective attention

Eyes

Ears

Perception Decision and Response selection

Response and Execution

Workingmemory

Long-termmemory

Stimuli Response

29

FRAME OF REFERENCE

is defined by Sarter (2000) as the level of attention that refers to the capacity to maintain a behavioural or cognitive set in the face of distracting or competing stimuli. The last of the four is divided attention which, according to Woods et al. (1995), is the highest level of attention and refers to the ability to respond simultaneously to multiple tasks or task demands.

Human information processing in relation to human-machine systems often refers to automation of the output functions of a system, e.g. decision and action implementation (Sheridan, 2002). However, automation may also, in line with Parasuraman et al. (2000) and Wickens (1992) as seen in figure 4-5, be applied to input functions that precede decision making and implementation of action.

AcquireInformation

Analyse and Display

Information

Decide Action based on Analysis

Implement Action based on Decision

Feedback and Information

Figure 4-5: Five simplified stages of a human-machine task, derived from the basic perceptual cyclical model of human action by Neisser (1976)

The first stage in figure 4-5, “Acquire Information” refers to the acquisition and registration of multiple sources of information. These operations are equivalent to the initial sensor stages of “Eyes/Ears” and “Perception” in figure 4-4. This first stage also covers the positioning and orienting of sensory receptors, sensory processing, initial pre-processing of data prior to full perception and selective attention (Parasuraman et al.,2000). The second stage, “Analyse and Display Information”, involves cognitive functions such as conscious perception and manipulation of processed and retrieved information from the working memory (Wickens, 1992). This stage also includes cognitive operations such as rehearsal, integration and inference, but these operations occur prior to the point of decision, according to Parasuraman et al. (2000). The third stage, “Decide Action based on Analysis”, involves, in line with Wickens (1992) and Parasuraman et al. (2000) the selection of options from a number of decision alternatives. As with the analogous decision-making stage in human performance in figure 4-4, such systems differ according to Parasuraman et al. (2000) from those involved in analysis automation, because they, in line with Klein (1993), must make explicit or implicit assumptions in a probabilistic world about the costs and values of different possible outcomes of the decision process. The fourth stage, “Implement Action based on Decision”, involves a different implementation of a response or action as in the information processing model by Wickens (1992) in figure 4-4. The fifth and final stage, “Feedback and Information”, involves information transfer based on the results of the implementation in the previous stage. “Feedback and Information” thereby closes the loop by using the output data from the fourth stage as part of the input data in the first stage in the next round.

Figure 4-5 also folds together several relevant dimensions, according to Sheridan (2002), such as:

30


� The degree of specificity required by the human to feed requests to the machine

� The degree of specificity with which the machine communicates decision alternatives or action recommendations to the human

� The degree to which the human has responsibility for initiating implementation of action

� The timing and detail of feedback to the human after machine action is being taken

However, the purpose of the model presented in figure 4-5 is not to discuss the theoretical background of the human cognitive system but to propose a model that is useful as a base for further discussion of how to support mental models by using levels of information automation.

4.2.2 Mental models

A basic description of a mental model is that it can be viewed as an individual’s cognitive representation of how a system operates (Scerbo, 1996). However, the term ‘mental model’ is sometimes also used as a synonym for ‘mental representation’ of the system under control. In other words, as noted by Kantowitz and Campbell (1996), mental models can be viewed as the understanding of the system elements, its dynamics, processes, and inputs and outputs. However, a mental model is also, as noted by Johnson-Laird (1980), a narrower term than mental representation. The differences in the definitions between mental model and mental representation also emphasise the lack of a unified terminology in the research that can be confusing, according to Sasse (1997).

Many of the existing theories on mental models, are according to Sasse (1997), based on the following assumptions:

� Users form a mental model of the internal workings of the computer systems with which they interact.

� The content and structure of the mental model influence how users interact with a system.

� The content and structure of a mental model can be influenced by selecting what information about the system is presented to users and how it is presented.

� More detailed knowledge of how users construct, invoke and adapt mental models could be used to provide guidance for user interfaces and user training, which help users to form appropriate models, and therefore make an important contribution to human-computer interaction (HCI).

Several other researchers, e.g. Endsley and Kiris (1995), Endsley and Garland (2000) and Flach and Mulder (2004), have also argued that it is advantageous for operators to form accurate mental models or representations of complex systems. Doing so enables them to anticipate future system states (Endsley, 1995), formulate planes and troubleshoot more effectively (Wickens, 1992). The operator also uses this model as an organising scheme to plan future activities, derive hypotheses about expected relationships among system components and perform system tasks Kantowitz and Campbell (1996). A key component in designing the system in a manner that is consistent with the operators’ internal model is, in line with Endsley (1995) and Kantowitz and Campbell (1996) to provide the operator with timely and accurate feedback. As noted in figure 4-5, feedback and information help the operator to interpret and understand what automation is doing.

31

FRAME OF REFERENCE

However, it should be noted that the use of mental models has not gone unquestioned. For instance, Norman (1987) argued that mental models need time to evolve since they can be incomplete and even inaccurate. Mental models of particular systems can also be structured from knowledge of other systems, from prototypes or even from faulty information (Scerbo, 1996).

In fact, Wickens (1992) and Woods et al. (1995) claimed that adding more automation to a system not only increases the complexity of systems and the difficulties in creating an accurate mental model or representation but also creates a higher mental workload.

4.2.3 Mental workload

Mental workload has been an important topic in human-machine systems research in the past two decades, according to Sheridan (2002), much because many of the tasks we encounter every day do not seem as demanding, such as supervising machines in production systems, as they might be in abnormal situations with little time for information processing. Much of the research on mental workload has focused on the working conditions for pilots in commercial aviation, personnel in nuclear power plant control rooms and workers in other settings where mental overload poses a serious threat to operator performance (Sheridan, 2002).

However, even though mental workload has been an important topic in human-machine systems research, it still has no universally accepted definition (Averty et al., 2002). One of the more simple definitions of workload is the definition of Waard (1996), who defines mental workload as “the demand that is placed on humans”. Sanders and McCormick (1993) on the other hand defined mental workload in somewhat broader terms, by arguing that “mental workload is the measurable quantity of information processing demands placed on an individual by a task”. However, as argued by Thunberg (2006), it might be even better to define mental workload in terms of experienced workload, since workload is not only task-specific but also person-specific (Rouse, 1988).

It is thereby argued by Karsvall et al. (2003) that there need to be different design solutions at different levels of decision-making and that the purpose of automation and decision support may vary. Technology that aims to support the operative level of decision-making, for example, should, as argued in Karsvall et al. (2003), rather mediate the skills of the operators than present static lists of decision alternatives. Such operative support would otherwise transfer responsibility from the joint system of human and automation to the individual operator without giving him or her any essential means of control. Another example noted in Karsvall et al. (2003) would be an adjustment of the support of cognitive automation to the level of work experience by presenting safer or more basic alternatives to less experienced users, while experienced users would be able to take short-cuts or to question the validity of the support.

4.3 HUMAN ROLES IN AUTOMATED SYSTEMS

Despite technological advances to develop automated production processes that can perform functions more efficiently, reliably or accurately or at a lower cost than human operators (Landes, 2003), automation has still not replaced humans in the production systems (Benhabib, 2003). Even if automation occasionally had this effect in some areas of the production industry (e.g. die casting), it has more generally not completely replaced human workers (Sheridan, 2002). In lay terms, however, it is easy to think of automated

32


systems as not including humans. However, most of those “unmanned” systems, such as automatic circuit board assembly operations, involve human operators in supervisory or monitoring roles (Sheridan, 1992).

Sheridan (1992) stated in his model for Supervisory Control that the human as a supervisor can be seen as one or more of the following five generic and interconnected functions: the first role for the supervisor is to plan what needs to be done over some period before any automation is turned on. The second role is to teach (e.g. instruct, command, program) the computer with what it needs to know in order to perform its assigned function for that period. The third role is to turn on the automation and monitorthe automatic action and detect any deviations or failures. The fourth role of the supervisor is to intervene in the automatic action if necessary and to decide on necessary adjustments to the automation. The fifth and final role is to evaluate performance and learn from observed experience.

Alternative models for describing supervisory control were presented by Stahre (1995), who showed in a literature review of available publications in the early 1990s that, by recognising Sheridan’s original definition of a supervisory control model, a useful paradigm can be created. Alternative models such the Production Activity Control (PAC) by Bauer et al. (1991) describe the human role in automated systems in a way similar to the Supervisory Control in FMS at Georgia Tech (Ammons et al., 1988) in terms of production planning, scheduling and cell management. A more cognitive approach to the role of humans in an automated system is the Human Performance Model by Nakamura and Salvendy (1994), who presented a decision-aiding system for operators in advanced manufacturing systems. The model focused on planning, scheduling and executing tasks. The model of Nakamura and Salvendy (1994) is also according to Stahre (1995) also related to the Rasmussen’s decision ladder model (Rasmussen, 1983). According to Rasmussen (1983) the human control functions can be organised on three levels: skill-based, rule-based and knowledge-based (S-R-K). Each of the three levels incorporates perception, action and information processing. An important aspect of the S-R-K levels according to Rasmussen (1988) is that cognitive control (e.g. knowledge and experience) may shift from one level to another, which can be termed as learning or gradual evolution of experience.

Later models that also acknowledge human roles in automated systems are Task or Function Allocation strategies, which are based on the original work of Fitts (1951) and extended, as noted in appended paper V, by Price (1985) and Williams (1999). One of the task allocation strategies that were presented as a serious alternative, according to Daouk (2000) was the left-over allocation strategy, which proposed that the designers of production systems should automate whatever they could and leave the remaining functions to human operators. Similar allocation strategies are Situation Dependent Allocation (Greenstein and Revesman, 1986), Flexible and Adaptive Allocation (Mouloua et al., 1993) and Adaptive Aiding (Rouse, 1988). An alternative model for describing human roles in an automated system is the model of Joint Cognitive System (JCS) presented by Hollnagel and Woods (2005). As described in paper IV, JCS, which is a part of Cognitive System Engineering (CSE), describes the human and technology as two equal partners involved in a collaborative accomplishment of a shared task.

Thus the human operator can still be seen in line with Brehmer (1993) as an essential, although vulnerable, component of the human-machine system; and, with the emergence and rapid growth of information technology, the relevance of systems that integrate

33

FRAME OF REFERENCE

information technology and mechanical technology increases. So, with such emerging concepts as knowledge workers (Drucker, 1999), the human roles in e.g. production systems have gone from actually conducting physical tasks to also covering cognitive tasks such as problem solving. The roles of the human are thus connected to his or her skill in interpreting the present situation and thus connected to his or her knowledge and experience, in line with Endsley (1996).

However, one of the main considerations in preventing the total removal of human operators from such systems has been the common perception, according to Wickens et al.(2004), that humans are more flexible, adaptable and creative than automation and thereby better suited to respond to changes or unforeseen situations. Inagaki (1993) also noted, for example, that the human operator is often kept in the system, to act and control situations if unanticipated events happen.

Thus in one sense, one might consider the levels of automation as the designer’s trust in the technical system that is designed (Frohm et al., 2002). Given that, no designer of automation can foresee all possibilities in a complex environment such as a production system, and therefore must rely on the human operator’s experience and judgment in using the automated system (Karsvall et al., 2003). This approach, however, tends to define the human operator’s roles and responsibilities in terms of automation (Riley, 1996a).

There is also a tendency according to Sheridan (1992) that we focus automation solutions on the easiest tasks and leave the rest to the operator as described in the left-overallocation approach. This focus may, according to Sheridan (1992), lead to a hodgepodge of partial automation, which will lead to a situation less coherent and meaningful and more complex than it needs to be. Merchant (2000) also noted that technology will only perform at its full potential if the utilisation of the system’s human resources is fully engineered. There is thus little question that the human is seen as an essential element of the system (Sheridan, 1995).

4.4 EFFECTS AND CONSEQUENCES OF AUTOMATION ON HUMANS

The promised and frequently realised benefits of automation, in line with Sarter et al.(1997), are generally sufficient to encourage production systems designers to move away from the lower levels of automation (e.g. manual work with or without hand tools). Many authors, e.g. Merchant (2000) and Saridis (2000), have concluded that the benefits of automation, among others, are that it may led to increases in productivity. The economic benefit of automation is also strong, but is not the only motivation for increasing LoA. Automation also offers substantial benefits in decreasing human workload, improving a worker’s performance and worker safety, when it is properly designed and used (Billings, 1997; Parasuraman et al., 2000).

On the other hand, in the case of not automating everything always, the highest level of technological capability has been harder to achieve (Satchell, 1998). Nevertheless, much research in the area has shown disadvantages in reducing the human role in system control and problem solving (Bainbridge, 1982; Rasmussen, 1986; Sheridan, 1992).

A key problem is therefore, according to Harlin et al. (2006) to establish and maintain an appropriate level of automation where the full potential of automation is realised and where negative automation side effects are avoided. Figure 4-6 describes a suggested model of the hypothesis for the automation balance.

34


Effects of automation Potential

state

Expected state Gap between the observed

and potential state of the

production system

Actualstate

Incr

easi

ng

posi

tive e

ffect

s

Observedstate

Increasing automation Levels of

automation

Figure 4-6: Initial model describing hypothetical effects on varying the level of automation

The potential state in figure 4-6 exemplifies a potential outcome of automation, while the observed state is the realised outcome of the implementation of automation. The gap between the observed and the potential state of the production system indicates, according to Harlin et al. (2006), the potential of increasing automation if critical aspects are sufficiently regarded.

Increasing positive effects of automation are e.g. higher safety, which, in line with Wickens et al. (2004), can be achieved by increasing the distance between human operators and the actual production process. Automation may also reduce the workload by increasing the support of physical and cognitive work tasks (Wickens et al., 2004). On the other hand, a negative effect of automation is e.g. an increased workload due to information overload in critical situations and production disturbances, as noted by Endsley and Garland (2000). An increased use of automation may also lead to a degradation of situation awareness when the human is not involved in sufficient communication with the automation (Endsley and Kiris, 1995). Further, communication mismatch may also lead to an over-reliance on automation with less initiatives for worker’s own decisions or to a distrust in the automation, according to Endsley (1996).

4.4.1 Trust in automation

As indicated in chapter 4.3, a major issue influencing how effectively human operators use automation is trust in automation (Lee and Moray, 1994; Muir, 1994). According to Lee and Moray (1992), have it been documented that operators may not always use well-designed and reliable automation if they believe it to be untrustworthy. According to the authors, operators may continue to rely on automation even when it malfunctions, if they are overconfident in its behaviour.

35

FRAME OF REFERENCE

Muir (1987), for example, argued that individuals’ trust in machines and automation can be affected by the same factors that influence trust between individuals. For example, people often trust others if they are reliable and honest, but they often lose trust when they are let down or betrayed, and a subsequent re-development of trust takes time.

For example, Muir found that use of automated aid to control a simulated soft-drink manufacturing plant was correlated with a simple subjective measurement of trust in that aid. Using similar process control simulation, Lee and Moray (1992) also found a correlation between automation reliance and subjective trust, although their participants also tended to be biased toward manual control and showed “inertia” in their allocation policy in using automated aid. In a subsequent study by Lee and Moray (1992), it was found that participants often chose manual control if their confidence in their own ability to control the plant exceeded their trust in the automation and that they otherwise chose automation.

Another factor in the development of trust is automation reliability. Several studies, e.g. Parasuraman and Mouloua (1996), have shown that operators’ use of automation often reflects automation reliability, although occasional failures of automation do not always seem to be a deterrent to future use of that automation. For example, Riley (1996b) found in one of his studies that college students and pilots did not delay turning on automation after recovery from a failure; in fact, many participants in the study continued to rely on the automation during the failure. Further, Parasuraman et al. (1996) found that, even after a simulated catastrophic failure of an automated engine monitoring system, participants continued to rely on the automation for some time, although to a lesser extent than when the automation was more reliable.

These findings are, according to Lee and Moray (1992), surprisingly in line with earlier studies, which suggested that operator trust in automation is slow to recover following a failure of the automation. Several possible mitigating factors could also be accounted for in the discrepancy, according to Parasuraman and Mouloua (1996).

First, if automation reliability is relatively high, operators may start to rely on the automation, which means that occasional failures, according to Muir (1987) do not necessarily reduce trust in the automated system unless the failures continue. A second factor that may affect trust is the ease with which automation behaviour and status indicators can be detected. Moreover, as noted earlier in the thesis, the overhead of enabling or disengaging automation may be another factor. Finally, the overall complexity of the task may be relevant, according to Endsley and Kaber (1999); complex task domains may prompt different participants to adopt different performance strategies and operator use of automation may be influenced by these task performance strategies as well as by factors directly related to the automation. This depends on the understanding or perception of the situation.

4.4.2 Situation awareness

The term situation awareness was originally used in the aircraft pilot community (Endsley, 1988) as one attempt to explain and understand pilot error. For example, in a study of pilot errors in commercial aviation accidents by Endsley et al. (1997), it was found that as much as 88 % of all accidents attributed to human error had an underlying problem with situation awareness.

However, the term has also developed as a major concern in many other domains in which humans operate complex and dynamic systems (Kaber et al., 2000), including control of

36


nuclear power plants, medical systems, teleoperations and advanced manufacturing systems. Achieving situation awareness is also one of the most challenging aspects of these operators’ jobs according to Endsley and Kaber (1999) and is thereby a central part of good decision-making and performance (Klein, 1998).

Further, situation awareness is formally defined by Endsley (1988) as:

“The perception of the elements in the environment within a volume of time and space, the comprehension of their meaning, and the projection of their status in the near future”

Situation awareness thereby involves according to Endsley (1996), perceiving critical factors in the environment (Level 1 SA). It also involves understanding what those factors mean, particularly when integrated in relation to personal goals (Level 2 SA) and at the highest level, and of what will happen with the system in the near future (Level 3 SA). These three levels of situation awareness are thereby critical, according to Endsley and Garland (2000) to allowing decision-makers to function in a timely and effective manner.

Automation can also be seen to directly impact situation awareness through three major mechanisms according to Endsley and Kiris (1995): (1) changes in vigilance and complacency associated with passive or active monitoring, (2) assumptions of a passive role instead of an active role in controlling the system and (3) changes in the quality or form of feedback provided to the human operator. Each of these factors can contribute to the out-of-the-loop performance problem as well as skill degradation (Endsley, 1996).

4.4.3 Skill degradation

If a higher level of automation can result in complacency and loss of situation awareness, it is perhaps not surprising that it might result in physical and/or cognitive skill degradation, as e.g. Zuboff (1988) and Parasuraman (2000) argue.

Further, as Rasmussen (1983) stated, behavioural patterns from the higher cognitive levels (skill-based) do not automatically become automated skills (see chapter 4.2). Rather, when a state of expertise is achieved on a lower level (knowledge-based or rule-based), the need for a higher level of control (skill-based) will gradually deteriorate (Rasmussen, 1983).

Another view of skill degradation was presented by Form (1987), who stated that changing economic conditions and additional case studies gave rise to four theories on how automation effects on skills. First, Blauner (1964) proposed a theory that mechanisation causes deskilling whereas automation reverses that trend. Second,Braverman (1974) put forward the theory by Bright (1958), who said that automation first upgrades skills and then accelerates the historic deskilling trend. Third, some writers e.g. Danziger (1985), proposed a polarising trend that automation upgrades advanced work tasks that are in need of skilled personnel and downgrades other tasks. Fourth, and finally, Simpson (1985) summarised a contingency theory that the automation effects are contingent to the type of industry, occupation, market and other factors.

However, despite hundreds, if not thousands, of automation studies, no one has tested all four hypotheses for major occupational and industrial groups, as stated by Form (1987).

37

FRAME OF REFERENCE

4.4.4 ‘Out-of-the-loop’

The out-of-the-loop performance problem, presented by Endsley and Kiris (1995), is a major class of errors associated with human interaction and automation. Hence, the modern operator work still represents a significant amount of manual labour, which makes it necessary (and possible) for the operator to gain detailed knowledge about the workplace in general and the production process in particular (Karsvall et al., 2003). This is especially apparent in cases where human operators must ensure that the automated system is performing properly and must detect situations that diverge from the expected performance. In addition to delays in detecting those situations and the occurrence of a problem, has occurred, a significant period of time may be required by operators to develop a sufficient understanding of the state of the system to be able to act appropriately (Endsley and Kiris, 1995). However, this delay may prohibit operators from carrying out the very tasks they are to perform or even diminish the effectiveness of their actions, as argued by Endsley and Kiris (1995). The Three-mile Island incident in 1979 is a classic example of what can occur when the human operator is left out-of-the-loop and has difficulty gaining a proper understanding of the situation so that proper oversight and interventions can occur. This and other frequently reported problems in automated systems can be directly connected to lower levels of situation awareness, according to Endsley and Garland (2000). In the end these changes in automation, can as Endsley and Garland (2000) argue change the entire role of the human and place him or her in the role of a passive monitor of the automated system.

4.4.5 Change in roles

Automation can also create both low and high workload extremes. That high levels of automation can leave the operator bored can perhaps be understandable, but that automation can increase e.g. the cognitive workload is one of the “ironies of automation”, according to Bainbridge (1982). Automation was initially employed to decrease the workload during normal operations (Wickens et al., 2004). Ironically, the technology that aimed to give conceptual simplicity also seems to decrease the sense of direct control in the actual work situation. Inagaki (1993) also noted that, since the human locus of control is required, the role of the human has become more important than before automation was introduced. One reason for this is that, in line with Endsley and Garland (2000), even though automation relieves the user, he or she must still have full awareness over the present and further situations, as Endsley and Garland (2000) argue. This shift from manual and distributed activities to relatively static or passive monitoring of computer screens (Karsvall et al., 2003) has placed the human operator into a role they are ill-suited to handle (Bainbridge, 1982). In addition, as noted by Wickens and Hollands (2000), an increase in the level of technological system complexity, associated with the incorporation of more automation, will eventually correspond to an increased probability of catastrophic failure.

38


CHAPTER 5

RESULTS


This chapter constructs the foundation for answering the research questions. The answers are based on the summary of the results presented in the appended papers.

The results presented in this thesis are based on three projects funded by the Swedish Foundation for Strategic Research (SSF) and one project funded by The Swedish Governmental Agency for Innovation Systems (VINNOVA).

The first research project, ”User-driven Decision Automation” was funded by VINNOVA and dealt with situation awareness and mental models in term of improvements in disturbance handling. Its main focus was on production-disturbance handling from a cognitive systems engineering perspective.

In a supporting project, “Disturbance Handling in Continuous Development of Manufacturing Systems”, within the SSF-programme PROPER (Programme for Production Engineering Research and Education) concepts for disturbance handling was investigated deeply by this author and three other PhD-students.

The third project was called “DYNAMO - Dynamic Levels of Automation for Robust Manufacturing Systems” and funded by the SSF-programme ProViking. The last five case studies in this thesis build strongly on DYNAMO.

Two of the case studies were also conducted within the framework of a fourth SSF/ProViking research project, Factory-in-a-Box, which focused on development of concepts for mobile and flexible production systems.

The two first case studies in this thesis were carried out in collaboration with Linköping University and IVF industrial Research and Development Corporation (IVF). The remaining case studies were carried out in collaboration between Chalmers, IVF, and the School of Engineering at Jönköping University. The last three research partners form an extended and distributed research group called ROMUS.

5.1 RESULTS PAPER I

The first paper, Frohm et al. (2003), focuses on how common models for task allocation handle the complex setting of production systems. From the elaboration of existing models of task allocation in relation to case study A, it can be noted that design models that are directly grounded in the task allocation perspective might not be sufficient to understand

39

SUMMERY OF RESULTS

the general mechanisms behind production systems and the role of automation. By focusing on individual requirements on automation, instead of demands on general system performance, task allocation implies that automation technology per se is more desirable than human labour. An unnecessary polarisation between human and machine is thereby created, which will lead designers in the wrong direction and eventually diminish the potential for truly adaptive production systems. Thus, when it comes to understanding the purpose of automated systems, it is argued in the paper that less time should be spent on allocating tasks between human and machine attributes and more time should initially be given to the process of describing the requirements of the overall system.

So, even if it is true that the operator does what the machine cannot do, and vice versa, operators also have their own work agenda (or joint cognitive system) that must be accounted for in the general picture. The operators are, after all, not employed to be an individual function in the automated system, but instead are a part of the overall solution. It can therefore be discussed whether operator work can really be defined as some sort of leftover allocation. As seen in case study A, the decision-making activities of operators not only contribute to the system performance but are also one of the most crucial events in overall production efficiency. The operator’s role as a problem solver or decision-maker seems therefore just as important for the outcome of the system as it has ever been.

Hence, along with increasingly complex production system, the need for competent professionals will rather increase than decrease. Because of the unpredictability of the human factor, system designers will normally not allocate essential production tasks to specific individuals. Thus, as technology advances, there appears to be limited room for specialised or localised work activities. We are rather headed towards general problem solving tasks for operators, which means less direct manipulation of local sub-systems and more time spent on system overview and interaction with co-workers.

5.2 RESULTS PAPER II

In the second appended paper, Frohm et al. (2005), it is concluded that, even if many manufacturing processes seem only to involve automation of mechanical or electrical tasks, most of those tasks are controlled or supervised by computers for optimal performance over time. It is therefore important to recognise, as seen in figure 5-1, that automation in manufacturing, as well as in other domains, can be seen as two types of technologies: mechanisation, which represents the basic core technologies, such as drilling, grinding etc., and computerisation which deals with the control and support of the mechanised technologies.

40


Technical System (machine) Control System (computer)

Mechanisation Computerisation

(E.g. material handling) (E.g. controlling the process)

Human

Figure 5-1: Separation of functions into mechanisation and computerisation

As discussed in the appended paper, it is important to recognise that automation is not only the process of reconfiguring the working part into a product. It is also the part earlier theories in LoA have put forward, the replacement of human cognitive processes, i.e. the control of physical activities such as mechanisation. Most research on LoA has thereby, in line with Satchell (1998), focused on the cognitive parts, where automation helps to speed up information flow and give decision support to help the operator monitor the situation. However, from a manufacturing perspective, it is also important to recognise that the focus of automation has often been on mechanised tasks.

To be able to make the manufacturing system as robust, flexible and adaptable as possible, the system must be able to handle variations in the process, such as an introduction of new product, a tool change, a product disturbance etc. It is thus important to understand how to obtain a balanced manufacturing system that has the proper mix of operators and machines in order to e.g. obtain the highest profit possible without suffering any losses of product quality. One way to achieve this balanced manufacturing system is to separate the system into the two basic classes of activities, i.e. information/control and physical work. Each of these two activities can then be categorised into ten steps, from total manual control (LoA 1) to fully automated control (LoA 10), according to the original LoA scale proposed by Sheridan (1980).

5.3 RESULTS PAPER III

The third appended paper, Frohm et al. (2006), is based on the pre-study conducted and Delphi study. The appended paper concludes that the general purpose of automated system is that functions are performed more efficiently, more reliably and more accurately than would be done by human operators. There are also expectations that automated systems can perform tasks at a lower cost than would be achieved with human operators. As discussed in the paper, there are few arguments that can be put against the efficiency, reliability and accuracy of automated systems. It could thus be argued that an automated system would be a safer system as well, since a system of that kind would protect a human from his or her actions and thereby improve the overall safety of the system.

On the basis of the results of the pre-study and Delphi study in relation to earlier research, it can be concluded that one of the primary driving forces for automation is providing possibilities for increases in efficiency and productivity (see figure 5-2). The results also

41

SUMMERY OF RESULTS

show that automation would enable cost reductions that, together with increased efficiency and productivity, will lead to increased competitiveness. Many respondents also acknowledge that an automation of physical tasks would provide possibilities for improving the work environment by eliminating or reducing monotonous and physically demanding work situations. However, the results also indicated that an increase in the usage of automation could lead to more production disturbances and, if that did become the case, the increased usage of automation could affect overall production efficiency.

Figure 5-2: Benefits of automation

Many of the respondents also acknowledge that the primary disadvantage of automation (see figure 5-3) is that there were too many products or variants in production. Almost a third of the respondents also acknowledged, in the final round of the Delphi, that they did not allocate enough time for instructing operators in the new automated equipment, which was one of the main difficulties in the process of automating the system.

Better working environment

Possibilities for increased volume capacity

Better product disturbance handling

Producing with a minimum of employees

Improvement of productivity

Cost cuts

Increased competitiveness

Improved quality

Increased efficiency

High Very high

24%

34%

19%

35%

32%

47%

36%

42%

48%

74%

57%

77%

65%

61%

40%

16%

44%

45%

0% 20% 40% 60% 80% 100%

cy

ity

ss

ng

ed

ce

vity

of

uts

42


Figure 5-3: Automation disadvantages

5.4 RESULTS PAPER IV

The fourth appended paper, Frohm et al. (submitted A), discusses whether it is possible to automate decision-making, in the light of modern industrial requirements for stable and robust, yet flexible production. The paper first focuses on previous findings in the field of decision-making (DM) in real-time practice and task allocation (TA). These theoretical findings are then related to observations made in three industrial case studies in three different branches of industry (cases A and B). The paper concludes that a non-routine event in operative decision-making is, per definition, not possible to automate in the classical sense. Machines rather implement ready-made operations in reaction to pre- or user-defined decision situations. Consequently, fully automated systems will not make decisions in action. The human, in contrast, has the ability to make tactical decisions in action, which is also one of the main reasons why human decision-makers must be an active part of the system. However, as system organisation and behaviour become more complex, human cognitive capability and skills are not sufficient to assess all the information or practical skills necessary for reliable or efficient decision-making. It is in these cases the operator needs decision support from automated information systems.

The paper thus concludes that the primary function of technology should be to support short-term operative decision-making and to mediate and amplify operator skills, rather then present static lists of decision alternatives. It was also concluded that tactical and strategic decision-making could be supported by automated information systems. However, it is questionable whether operative decision-making can be fully automated.

In terms of understanding the automated system, less time should be spent on allocating tasks between human and machine attributes; as also argued in paper I, more time should initially be spent on describing the requirements of the overall system.

5.5 RESULTS PAPER V

The background of the Frohm et al. (submitted B) was the need of a theoretical ground for the definition and taxonomy of LoA. Based on the results reported in key publications in

Too many products or variants

Difficulties in ramping up the new product

Knowledge and know-how from the user’s

Union

Investment costs

Sufficient time for planning the automation

Sufficient knowledge for ordering automation

Adapt the product to automatic manufacturing

High Very high

42%

34%

34%

37%

3%

42%

24%

34%

34%

27%

7%

8%

23%

3%

5%

7%

8%

39%

0% 20% 40% 60% 80%

ew

ts

Train and educate the user in the new automationon

on

m

he

ts

ng

tic

43

SUMMERY OF RESULTS

the areas of advanced manufacturing technologies (AMT), human factors, and teleoperations and in relation to earlier case studies, such as in appended paper I, it can be concluded that most tasks in manufacturing often involve a mix of both mechanised (physical) and computerised (cognitive) tasks.

From the review of the academic literature on “Levels of Automation” and the division into mechanised and computerised activities, it is shown that most recent research on LoA (from areas such as aviation and telerobotics) has primarily focused on the cognitive side of the human-machine system. The purpose of cognitive automation was to speed up the information flow and provide decision support, thereby helping the human to monitor the situation. Further, the paper also demonstrates that, from a manufacturing perspective, it is important to recognise that manufacturing processes not only consist of physical tasks such as replacement or support of human muscle power, but also of cognitive support for carrying out control and information tasks, which means that there is a potential to transform experience and knowledge from areas such as aviation and telerobotics.

It can thus be concluded that level of automation in production, as well as in other domains, is not a single progression of tasks from the lowest level of fully manual performance to the highest level of full automation, but a parallel continuum expressed in two seven-step reference scales of physical and cognitive support (see table 5-1).

Table 5-1: LoA-scales for computerised and mechanised tasks in manufacturing

LoA Mechanical and Equipment Information and Control

1 Totally manual - Totally manual work, no tools are used, only the users own muscle power. E.g. The user’s own muscle power

Totally manual - The user creates his/her own understanding of the situation and develops his/her course of action based on his/her earlier experience and knowledge. E.g. The user’s earlier experience and knowledge

2 Static hand tool - Manual work with support of a static tool. E.g. Screwdriver

Decision giving - The user gets information about what to do or a proposal for how the task can be achieved. E.g. Work order

3 Flexible hand tool - Manual work with the support of a flexible tool. E.g. Adjustable spanner

Teaching - The user gets instruction about how the task can be achieved. E.g. Checklists, manuals

4 Automated hand tool - Manual work with the support of an automated tool. E.g. Hydraulic bolt driver

Questioning - The technology questions the execution, if the execution deviates from what the technology considers suitable. E.g. Verification before action

5 Static machine/workstation - Automatic work by a machine that is designed for a specific task. E.g. Lathe

Supervision - The technology calls for the users’ attention, and directs it to the present task. E.g. Alarms

6 Flexible machine/workstation - Automatic work by a machine that can be reconfigured for different tasks. E.g. CNC machine

Intervene - The technology takes over and corrects the action, if the executions deviate from what the technology considers suitable. E.g. Thermostat

7 Totally automatic - Totally automatic work. The machine solves all deviations or problems that occur by itself. E.g. Autonomous systems

Totally automatic - All information and control are handled by the technology. The user is never involved. E.g. Autonomous systems

44


However, as stated in the paper, some tasks can not be allocated to either the human or the technical system. Each of the reference scales can thus individually be delimited by the relevant maximum and minimum LoA for each task (see table 5-2).

Table 5-2: Relevant maximum and minimum LoA for a particular machine or cell

Mechanical and Equipment Information and Control

LoA Workstn. 1 Workstn. 2 Workstn. n Workstn. 1 Workstn. 2 Workstn. n LoA

7 Relevant max 7

6 Relevant max 6

5 Relevant max Relevant max Relevant max Relevant max 5

4 4

3 Relevant min Relevant min Relevant min 3

2 Relevant min 2

1 Relevant min Relevant min 1

5.6 RESULTS PAPER VI

The sixth appended paper, Granell et al. (2007), presents the DYNAMO methodology for assessment, measurement and analysis of LoA.

The development and implementation of the methodology are based on six case studies (A to F in the appended papers) done during the years 2004 to 2007, where the results of each case lay the foundation for the next case (see figure 3-2). Based on the results of the first six cases (A to F), the methodology was rewritten into the “final” eight-step DYNAMO methodology (see figure 5-4), which was tested for validation in case study G.

45

SUMMERY OF RESULTS

Figure 5-4: The DYNAMO methodology in eight steps

Based on the discussion of observed deviations, problems and comments given by the three researchers that conducted the LoA analysis, the validation resulted in some changes of the DYNAMO methodology and its instructions.

The experience gained in validating the LoA measurement and assessment showed that, by moving more activities to the preparation and planning step, step 1, the methodology would benefit in efficiency. Moreover, informing operative personnel about the measurement in step 2 will increase motivation and may be advantageous to the LoA measurement made in step 6. Giving information about the measurement method will also make the dialogue in step 7 easier when assessing LoA. Documentation of sub-tasks is better when done with a

1. Plan ahead before the measurement

2. On-site, start with a pre-study to identify the process

3. Visualise and document the production flow

4. Identify the main task for each section/cell

5. Identify sub-tasks for each section/cell

6. Measure LoA

7. Assess LoA, set relevant maximum and minimum levels

8. Analysis of results

46


video recorder. One experience in the validation is that the observation and LoA measurement can be very time-consuming, especially if there is poor conformity with operation instructions. Both estimating LoA based on the operation instructions off-site as well as measuring LoA by observation in step 6 will broaden the data in the analysis in step 8. Relevant maximum and minimum levels are set by using dialogue, and the person involved in this step should be trained in the LoA measurement methodology before assessing LoA.

Based on the findings and results of the validation in case study G, the final DYNAMO methodology was updated and is described in the following subsection.

5.6.1 The DYNAMO methodology

Step 1: The first step in the DYNAMO methodology, which is conducted off-site, is to discuss the goal and purpose of the measurement and the delimitations of the production flow with the company. For example, the goal of the LoA measurement can be to maximise or minimise the level of automation in the flow. At the same time that the background data, such as goal and purpose, are collected, the operation instructions for the flow in focus are requested from the company.

By conducting a pre-LoA measurement (similar to subsequent steps 4-6), based on the operation instructions from the company and the two reference scales for LoA (see table 5-1), it is possible to gain a pre-understanding of how the tasks are intended to be conducted. Further, by conducting a pre-LoA measurement off-site and identifying the tasks in the flow that will be measured, it is also possible to estimate the time needed for observation on-site and the time needed for the subsequent workshop.

At the same time that the operation instructions are requested from the company, form no. 1 with its questions is sent to the company.

When form no. 1, which will be answered by a company representative on-site in advance of step 2, is sent the company is given time to collect the requested data before conducting step 2.

Step 2: The second step, which is conducted on-site, is to carry out a pre-study to identify and document the purpose of the production flow from the company perspective. At the same time as the purpose is identified and documented, the delimitations of the production flow are documented by identify where it starts and ends. The data are documented in form no. 1. The measurement starts with a run-through of the method with the company so that it is informed about how the measurement will be conducted. After the company personnel involved in the study have been informed about the method, the data collection starts by going through form no. 1 with the company representatives. The number of products and variants produced in the production flow are identified and documented, as are the work organisation and the tasks of machines and humans. During this first visit on-site, it can also be advantageous to inform the personnel on-site about the measurement, since the measurement focuses on how the task is conducted and people can get an incorrect impression of the purpose of the measurement.

Step 3: After the basic data for the production flow have been documented and understood, the next step is to visualise the production flow (see figure 5-5). This is done by “walking the process” and defining the sections/cells of the production flow. The documentation is done by visualising the production flow by means of the symbols shown in figure 5-5, using form no. 2.

47

SUMMERY OF RESULTS

Buffer1

Buffer2Sub

task 1 Sub

task 2 Sub

task 3 Sub

task 4

Buffer1.1 Sub

task 1.1

Subtask 1.2

Figure 5-5: Visualisation of the observed production flow

When the production flow has been visualised and documented, each section/cell and buffer is analysed to identify its purpose and main task (e.g. assembly of X components).

Data, such as the number of products and variants that pass through the section/cell or buffer, are documented in form no. 2, as well as the physical and cognitive tasks that have been allocated to the technology or to the human. The number of operators that are allocated to the section/cell or buffer is also documented and whether the operator is responsible for more than one section/cell or buffer.

In this stage data can be collected in parallel with the visualisation of the production flow. However, it can be an advantage to conduct the identification of each station/cell after finishing the visualisation of the production flow. By walking the process a second time, the observer might gain a better understanding of the process and thereby improve the observer’s focus in identifying the main task at each section/cell or buffer.

Step 4: After the production flow has been understood, visualised and documented, the identification of the main task is done (see figure 5-6). Based on the data collected in steps 2 and 3, the main task of each section/cell or buffer is identified and documented (e.g. mounting X electronic components on a circuit board).

Figure 5-6: Identification of the main task in the production flow

Main task

Buffer1

Buffer2Sub

task 2 Sub

task 3 Sub

task 4 Sub

task 1

Subtask 2.1

Subtask 3

Subtask 4

Buffer3

Buffer2.1

= Workstation/-cell with

operation instructions

= Workstation/-cell without

operation instructions

48


Step 5: After the production flow has been understood, visualised and documented, the measurement continues with the identification of the sub-tasks.

Identification of sub-tasks is done by observing how the main task is achieved, which is done by breaking down the task until it reaches a level of operations where either the human or the technology is solely responsible for achieving the task.

Since the degradation process of finding the sub-task can be difficult if the task lacks operation instructions or if it is conducted in another order than prescribed in the operation instructions, it can be advantageous to use video recordings.

To support the observation, the operation instructions are used as a starting point and an explanation of what is going to be observed. By using the documentation of the task from the company, which is noted in form no. 3 in step 1, the measurement crew can easily identify the sub-tasks and deviations from how the task is intended to be carried out. It should be noted however, that the order in which the tasks are conducted, is not important for the LoA measurement since the tasks are measured individually. Nevertheless, the information about the order in which the tasks were conducted according to the observation can serve as a foundation for improving the operation instructions.

Hierarchical Task Analysis (HTA) is used to simplify and structure the breaking down of the main tasks into sub-tasks (see figure 5-7). HTA is a method (as noted in 3.2.5) for describing the activities under analysis in terms of a hierarchy of goals, sub-goals, operations and plans (Kirwan and Ainsworth, 1992).

Figure 5-7: Identification of the sub-tasks in the production flow by HTA

Step 6: After the tasks have been broken down and identified, LoA is judged based on the two reference scales for “Mechanical and Equipment” LoA and “Information and Control” LoA (see table 5-1). The judged LoA for each task is then based on how the task is conducted and what type of interaction that is observed in fulfilling the task. The type of

1. Sub-task

LoA

Mech. & Equip.

LoA

Info. & Control

LoA

Mech. & Equip.

LoA

Info. & Control

2. Sub-task 3. Sub-task 4. Sub-task

1.1

Sub- task

1.2

Sub-task

3.1

Sub- task

3.2

Sub-task

3.2.1

Sub-task

3.2.2

Sub-task

0. Main task

3.2.2

Sub-task

49

SUMMERY OF RESULTS

interaction for each sub-task is then mapped against the two reference scales. The measurement data from the observation are documented in form no. 3. By observing more than one operator, it is possible to increase the strength of the observations, and to identify whether tasks are conducted with different LoAs depending on which operator conducts the task. If the case is conducted under different LoAs, the task can then be classified as dynamic LoA.

Step 7: After the measurement of the observed LoA values has been judged, the observer, together with the operator and/or production engineer on-site, estimates the relevant maximum and minimum LoA for each task measured. By using respondents that have an understanding of how the tasks that have been observed are conducted, a good estimation can be made of the relevant maximum and minimum LoA during the discussion.

Before giving the respondents the results of the measurement, a run-through of the method should be done to inform the respondent (e.g. the production technician or the operator).

In the discussion with the respondents about the assessed LoAs, the LoA values should not be used but instead translated into hypothetical tasks on-site to which the operator can relate. For example, if the task is to assemble a component by clips and the task is observed as having been done by an electrical screwdriver (LoA-M: 4), the question might be “wouldit be possible to assemble the component with a screwdriver (LoA-M: 2) or by a robot (LoA-M: 6)?” The data from the discussion are then documented in form no. 4, together with the two reference scales for LoA.

Step 8: The final step in the DYNAMO measurement methodology is to analyse the automation potential based on the preliminary and LoA measurement (form no. 3) and the data collected from the LoA measurement on-site (form no. 3), together with the assessment of the relevant maximum and minimum LoA (form no. 4).

50


Figure 5-8 Analysis of LoA in the Mechanical-Information-LoA diagram (form no. 5)

The analysis starts by placing the preliminary LoA value (3;2 in figure 5-8) and the assessed LoA value from observation (4;4 in figure 5-8) in the Mechanical-Information-LoA diagram. Plotting both the preliminary and the observed LoA value for all documented sub-tasks in each individual workstation or cell shows the actual flexibility and dynamics of the automation. Further, drawing the borders of the relevant maximum and minimum of each LoA gives a potential area of automation of the task (see figure 5-8). Depending on the purpose of the LoA measurement, an analysis of the Mechanical-Information-LoA diagram indicates how to take advantage of the automation potential. For example, if the LoA assessed from observation is placed to the left or at the bottom of the shaded area in figure 5-8, it indicates that there is a shaded area also shows the flexibility in the present workstation or cell and the potential for dynamic LoA. This scenario can be advantageous when the company is unsure about the technical solution and there is a need for some built-in flexibility. After the analysis has been made, an action plan is worked out by deciding actions and dates for activation.

5.7 RESULTS PAPER VII

The final appended paper, Fasth et al. (2007), focuses on the means for finding new tactics to decrease the time and cost parameters in a system, i.e. cycle time, ramp-up time, investments etc. The paper presents some of the parameters that exist in a manufacturing

Rel. max

Rel. min

LoA-info.Observed

LoA-mech. Observed

Estimated LoA from operation instructions

LoA-info.

7

6

Assessed LoA from observation

5

4

3

2

1

1 2 3 4 5 6 7 LoA-mech. Rel. min Rel. max

51

SUMMERY OF RESULTS

system and the relationships between them. This was done to be able to identify the parameters that are most important with respect to simulating and visualising different LoA.

Based on the literature review, six different flexibilities were identified in the paper as important parameters for decreasing cycle time, ramp-up time, investments etc. The effects on those six different flexibilities are then discussed in the hypothetical light of changes in LoA.

It can be concluded that companies today must know what parameters are important for their system. On the basis of the assumptions made according to flexibility, time and flow parameters in relation to LoA, it can be argued that the assumptions presented in the paper must be validated beyond theoretical analysis with empirical data.

5.8 ADDITIONAL VERIFICATION OF THE DYNAMO METHODOLOGY

To be able to increase the validity of the presented DYNAMO methodology in the preceding chapter (5.6.1), four additional case studies were conducted by fellow researchers and students without the assistance of the researchers who developed the methodology.

5.8.1 Truck manufacturing

The first additional verification was done by a group of students (Elmgren et al., 2007), to evaluate the automation potential at four workstations based on an earlier bottleneck study by the company. Since the methodology at that time had not yet been presented (appended paper VI), instructions for how to use the methodology were given during preparations for the second company visit.

The realisation of the DYNAMO methodology followed the eight steps that the methodology prescribes. However, the group of students did not have access to the operation instructions before the first visit, as the methodology prescribes. Instead, the group used a time and motion study in combination with observations to form the basis for the identification of sub-tasks in step 5 of the DYNAMO methodology.

The experience from the measurement shows that even though the identification of sub-tasks is based on a time and motion study in combination with observations, the expected results of the methodology are obtained. It can thus be argued that the DYNAMO methodology works in this case

5.8.2 Manufacturing of component for indoor climate

The second case study that used the DYNAMO methodology also obtained the results that the methodology prescribes (Dencker, 2007). In this case, the researchers received the operation instructions during their first visit to the company. The first visit started with a guided tour through the production process to gain a better understanding of the assembly line. At the same time that the operation instructions were obtained from the company, form no. 1 with its questions was given to the company. Even though the pre-LoA measurement could not be made before the first visit, it was conducted before step 2. In relation to the previous verification, the operators also participated in the workshop and could thereby give additional information on their view of how the task could be automated.

52


5.8.3 Manufacturing of modules for cooling

The third verification of the DYNAMO methodology was done as a part of a Master’s thesis, to investigate whether it is possible to automate the manufacturing of modules for cooling (Grundström and Ottosson, 2008). Also in this case, as in all the first verification cases, the operation instructions were not obtained and thus no pre-LoA measurement could be made.

Despite the pre-LoA measurement not being able to be done, the remaining steps of the methodology are followed without any deviations from how the methodology was intended to be conducted. The results of the case study (Grundström and Ottosson, 2008) showed that the methodology gave the results that the methodology prescribes.

5.8.4 Pharmaceutical industry

The last of four cases that were used for the final verification (Jonsson, 2007), also had difficulty assessing the operation instructions, and the pre-LoA measurement could thus not be made. As in the previous cases, the methodology was followed, starting with the pre-study in steps 1-3 to get as thorough a mapping of the production system as possible. Subsequent steps 4-6 and 7-8 also followed the methodology without any deviations according to Jonsson (2007). It could thus be concluded that the fourth and final verification case also gave the results that the methodology prescribes, and thus shows that the methodology works.

5.8.5 Summary of the additional verification of the DYNAMO methodology

On the basis of the results from the four additional case studies (table 5-3), it can be concluded that it is generally difficult to obtain operation instructions before the first visit, and thereby, to conduct a pre-LoA measurement. The results of the four case studies also indicated that the composition of the participants at the workshop in step 7 is important. As seen in table 5-3, the company’s view at the workshop is often represented by production technicians and managers, who have a more tactical and strategic view of production. It is thus, as was seen in the four verification cases, difficult to get the operators’ operative view of the tasks that have been measured. One reason for this is that it is difficult to take operators out of production. However, the operators’ view is still crucial since they are the persons closest to the process and tasks in question.

Table 5-3: Summery of the four verifications of the DYNAMO methodology

53

SUMMERY OF RESULTS

Ver

ific

atio

n 4

Pha

rmac

euti

cal

indu

stry

Sep

tem

ber

– O

ctob

er 2

007

X - X X X X X X X X

- O

bser

vati

on

Pho

to

Obs

erva

tion

X

- P

rodu

ctio

n te

chni

cian

-

Pro

duct

ion

man

ager

X X

Ope

rati

on o

ptim

isat

ion

Ver

ific

atio

n 3

Mod

ules

man

ufac

turi

ng

May

200

7

X - X X X X X X X X

- O

bser

vati

on

Pho

to a

nd v

ideo

Obs

erva

tion

X

- P

rodu

ctio

n te

chni

cian

-

Pro

duct

ion

man

ager

X X

Tas

k an

d op

erat

ion

opti

mis

atio

n

Ver

ific

atio

n 2

Com

pone

nt m

anuf

actu

ring

Apr

il -

May

200

7

X X X X X X X X X X

- O

pera

tion

ins

truc

tion

s

Pho

to a

nd v

ideo

Obs

erva

tion

X

- O

pera

tor

- P

rodu

ctio

n te

chni

cian

-

Pro

duct

ion

man

ager

X X

Tas

k an

d op

erat

ion

opti

mis

atio

n

Ver

ific

atio

n 1

Tru

ck m

anuf

actu

ring

Mar

s -

Apr

il 2

007

X - X X X X X X X X

- E

arli

er b

ottl

enec

k st

udy

- O

bser

vati

on -

Obs

erva

tion

X

- P

rodu

ctio

n te

chni

cian

X X

Tas

k an

d op

erat

ion

opti

mis

atio

n

Dis

cuss

ion

and

iden

tifi

cati

on o

f th

e go

al a

nd p

urpo

se o

f th

e m

easu

rem

ent

Acc

ess

to o

pera

tion

ins

truc

tion

s to

con

duct

a p

re-L

oA m

easu

rem

ent

Iden

tifi

cati

on o

f th

e pu

rpos

e of

the

pro

duct

ion

flow

Del

imit

atio

n of

the

pro

duct

ion

flow

(w

here

it

star

ts a

nd e

nds)

The

dat

a in

for

m n

o. 1

wer

e co

llec

ted

Doc

umen

tati

on o

f pr

oduc

ts a

nd v

aria

nts

Doc

umen

tati

on o

f w

ork

orga

nisa

tion

and

tas

k al

loca

tion

(fo

rm n

o. 1

)

“Wal

k th

e pr

oces

s” a

nd d

efin

e th

e se

ctio

ns/c

ells

of

the

prod

ucti

on f

low

(f

orm

no.

1)

Iden

tifi

cati

on o

f th

e pu

rpos

e of

the

pro

duct

ion

flow

for

the

mea

sure

men

t (f

orm

no.

2)

Iden

tifi

cati

on o

f th

e m

ain

task

of

the

prod

ucti

on f

low

for

the

mea

sure

men

t (f

orm

no.

2)

Bas

is f

or i

dent

ific

atio

n of

sub

-tas

ks

Pho

to a

nd v

ideo

doc

umen

tati

on

Bas

is f

or m

easu

rem

ent

of L

oA

Doc

umen

tati

on o

f da

ta (

form

no.

3)

Par

tici

pant

s at

the

wor

ksho

p

Doc

umen

tati

on o

f re

leva

nt m

ax/m

in L

oA (

both

phy

sica

l as

cog

niti

ve)

for

each

mea

sure

d ta

sk (

form

no.

4)

Con

stru

ctio

n of

the

Mec

hani

cal-

Info

rmat

ion-

LoA

dia

gram

Res

ult

of a

naly

sis

Step 1 Step 2 Step 3 Step 4 Step 5 Step 6 Step 7 Step 8

Tab

le 5

-3:

Sum

mer

y of

the

fou

r ve

rifi

cati

ons

of t

he D

YN

AM

O m

etho

dolo

gy

Typ

e of

ind

ust

ry

Stu

dy

per

iod

Results

54


CHAPTER 6

DISCUSSION


This chapter discusses the results of the study. Further, the relations between the results and the overall research aim, and methodological issues are examined.

6.1 DISCUSSION OF RESULTS

The background of this thesis is a discernible trend towards increasing automation in production industries and the risk that past mistakes in automation efforts will be made again.

The empirical results, summarised in chapter 5, are further discussed in relation to the theoretical findings in chapter 4. The discussion of the results is organised according to the research questions.

6.1.1 Is the concept of levels of automation relevant in production?

The task of designing and operating production systems has been approached in many different ways (Brehmer, 1993; Endsley, 1997; Williams, 1999; Merchant, 2000). One of those approaches is the technical, which has mainly focused on increasing the amount of automation in the system (Satchell, 1998). In paper V, it is argued that the simplest form of automation operates in one of two possible modes, manual or automatic. The progression from manual to automatic is further viewed as one single step, i.e. when the operators are replaced by robots or advanced machinery. However, this is not completely true. As noted by Kern and Schumann (1985) and Duncheon (2002) in paper V, most physical tasks in manufacturing are semi-automated, a mix where both humans and advanced technology co-operate to achieve the task. In other domains with complex systems, such as aviation and nuclear power control-rooms, it is also not uncommon to find different levels or stages of automation, between fully automatic and manual.

It can thus be argued, with the support of earlier research by Sheridan (1992), Inagaki (1993), Endsley (1997) and Parasuraman et al. (2000), and the case studies in this thesis, that automation is not all or nothing. Instead, automation should be seen as a continuum of levels, from the lowest level of fully manual performance (where the tasks are only based on humans’ capabilities) to the highest level of full automation (without any human involvement).

Further, as noted by Billings (1997), Endsley (1997), Endsley et al. (1997) and Satchell (1998), much of the previous research in LoA, in areas such as aviation and telerobotics,

55

DISCUSSION

has primarily been concerned with cognitive automation. The aim of cognitive automation has often been to speed up information flows and to provide decision support, thus supporting the operator in monitoring the situation. However, from a manufacturing perspective, it is important to recognise, as noted in paper II and in line with Williams (1999), that manufacturing processes in general consist of both physical and cognitive tasks. As noted in paper V, physical tasks are in line with the basic core technologies (Langmann, 2004), such as drilling and grinding, while cognitive tasks are what Langmann (2004) defines as the integration and support technologies, e.g. computer technology. Further, this approach is in line with Bright (1958) and Marsh and Mannari (1981), who concluded that the manufacturing context consists of a mix of both mechanised (physical) and computerised (cognitive) tasks.

Breaking down activities in line with Williams (1999) into physical and cognitive tasks, while still acknowledging co-operation between the human and the technical system, it can be relevant to consider the allocation of tasks as two independent reference scales of LoA relating to the cognitive and physical tasks. Each of these can then be ordered in seven steps, ranging from totally manual to fully automatic.

The advantage of using the two proposed reference scales, in relation to previous taxonomies of LoA by e.g. Sheridan (1992), Billings (1997) and Endsley (1997), is that the levels of both physical and cognitive support can be assessed in the same taxonomy. Since the two reference scales for the assessment of physical and cognitive support are independent, the level of automation of physical and cognitive tasks can vary depending on individual human needs.

However, as noted by Price (1985) and Williams (1999), some tasks can not be allocated to either the human or to the technological system. As showed by Williams (1999), there is a maximum and a minimum ability of the human as well as the technology to handle different tasks. Bearing in mind the model of limitations by Williams (1999) and the case studies in this thesis, it can be noted that, in terms of e.g. investment cost, it is unrealistic in practical industrial situations to consider the entire span of each of the two reference scales. Therefore, and in line with Williams (1999), it is not relevant to discuss the assessment of LoA in terms of absolute maximum and minimum levels; it should instead be seen in terms of relevant levels of automation for each of the two reference scales.

6.1.2 Can the level of automation in production be measured?

The results of the Delphi survey (Lindström, 2005) and interviews with production experts working in Swedish industry showed that Swedish manufacturing firms are not acquainted with the term levels of automation. Measurement or assessment of automation levels is also uncommon, according to Granell (2007), mainly because prerequisites are not being met and barriers must be overcome.

Since automated production systems are often regarded as highly productive, which can potentially improve a company’s competitiveness (Winroth et al., 2006), an understanding of how to balance the mix of operators and machines in the production system requires that the balance can be assessed or measured. The purpose of measuring and assessing LoA, as noted in paper VI, is to find an appropriate level or span of automation and, by that, maintain high productivity by reducing production disturbances.

To design a production system that has the appropriate level or span of automation, it is argued in appended paper I in line with Meister (1987), that the designer of the production system should base the allocation of tasks between the human and technology on the

56


process flow. As noted by Fitts (1951) and visualised by Price (1985) in the papers II and V, some operations or activities can be too hazardous or fatiguing for the human operator to handle. In the same way, some operations or activities can be too complex or need to be so flexible that it is not possible for technology to handle them.

Forming a methodology for measuring and assessing LoA in production flows makes it possible to analyse whether the present level of automation is too low, too high or too static in relation to the company’s intention with its automation. Further, by also forming a platform, some of the expressed prerequisites in earlier case studies can be fulfilled and thereby some of the barriers noted by Granell (2007) can be overcome.

The results of the validation presented in paper VI show that, by moving more activities to the off-site planning of the measurement in step 1 of the DYNAMO methodology, the efficiency of the on-site work could be improved. Conducting a LoA measurement and assessment based on operation instructions prior to the visit to the company, makes it easier for the person who conducts the measurement to understand the production situation better. Further, by informing operative personnel about the measuring at the beginning of the on-site step 2, the motivation for and understanding of the measurement would be easier to communicate. This understanding can be advantageous in the actual LoA measurement in step 6, since the participation and information of the operators in question are crucial to the observers’ understanding of the situation being measured. By also getting the approval of the company and the participating operators to use video recordings of the observed task, the observer/researcher has the possibility to go back to the measurement at a later stage, if something is unclear after the observer/researcher has finished the measurement. Combining the estimated LoA assessment in the early off-site stage (step 1) with the assessment and observation of a number of operators in step 6 will broaden the analysis in step 8. Another advantage of conducting a pre-LoA measurement and assessment based on operation instructions is that it will be possible to estimate the time that the company must allocate for the actual measurement and the subsequent discussion of relevant maximum and minimum levels of automation.

However, as noted in paper VI, the final step (step 8) was not validated in the last case study (case G). Instead, four additional case studies were conducted by fellow researchers and students, who used and extended the methodology at companies that, had not participated in the development of the DYNAMO methodology. The result of the verifications showed that the DYNAMO methodology worked and that the methodology gave the expected results. It can thus be concluded that it is possible to measure levels of automation in production systems with the DYNAMO methodology.

6.1.3 What consequences can be expected due to changes in the level of automation in production?

In line with Ylipää (2000), each change in a complex technical system introduces new risks of causing disturbances in the system as well as risks of decreasing product quality. Further, as noted in paper VII, flexibility, complexity and robustness are three areas that have been indentified by industry as important in terms of decreasing time and cost parameters.

As noted by Brehmer (1993), Endsley and Kiris (1995) and Wickens et al. (2004), complex and partly automated systems place high demands on human operators to be an active part of the system, dealing with disturbances, unforeseen situations, non-automated tasks and maintenance of sub-systems.

57

DISCUSSION

As noted in the literature review in paper V and supported by (Billings, 1997; Endsley, 1997; Endsley et al., 1997; Satchell, 1998), levels of automation have primarily been concerned with cognitive automation. In this perspective, automation of cognitive processes (e.g. decision-making) can be seen as a way of achieving more rationality and reliability in the process of dynamic decision-making by suppressing decision biases based on human cognition (Frohm et al., 2002). However, as noted by Bainbridge (1982), it is not possible to fully exclude human interaction from automated systems. Automated systems lack many of the cognitive skills that are necessary for monitoring the overall system and handling complex production disturbances. This is supported by Brehmer (1993), who claims that the working task of operators in automated systems mainly consists of dynamic decision-making. However, as put forth by Hollnagel (1997), the operator and the automation can also be described as a mutual system, a Joint Cognitive System (JCS), involved in collaborative accomplishments of a shared task. It could thus be argued that these objectives represent the same two extremes of the automation continuum defined by Sheridan (1992).

If automation is to replace human operators, the result can be that we risk losing an important information exchange that is needed to uphold tacit working skills and knowledge. But we may also create a gap between the physical production and the human operator (Frohm et al., 2002). However, it is also possible that e.g. the process operators in case study A, who are already distant from the physical process, could benefit from increased levels of cognitive decision support as this may give them a greater possibility to monitor and operate all parts of the process. Automation may thus also close the gap between the operator and the physical process by basically changing the whole setting of the production system (Frohm et al., 2002; Karsvall et al., 2003).

This imbalance between demands of the work situations, which according to Karsvall et al. (2003) is increasing, creates a decrease in individuals’ sense of control in the work situation as a result of less routine work situations.

Excessive levels of automation can therefore result in a degradation in operator performance e.g. due to a lack of knowledge, gradual loss of a special working skill and a degradation in situation awareness, as noted by Endsley and Kiris (1995), Endsley (1997), Endsley et al. (1997) and Parasuraman et al. (2000). In line with Zuboff (1988) and Connors (1998), it has been shown among operators that an increase in automation and computerisation may result in an increased cognitive workload.

It can also be seen that the increasing usage of computer technology (e.g. information and control) has not necessarily resulted in a decrease in the manual contribution, even if the requirement for cognitive skills in supervising the computerised system has increased. One reason for this is, in line with Endsley and Garland (2000), that even if automation relieves the user, he or she still must have a full awareness of the present and future situations. This shift from manual and distributed activities to relatively static or passive monitoring of computer screens has placed human operators in a role that they cannot handle. Ironically, the same technology that aimed to give conceptual simplicity also seems to decrease the sense of direct control in the actual work situation.

Kaber et al. (2000) also note that taxonomies of LoA that are designed for specific tasks might result in a reduction of the human operator’s understanding of automated systems during an automation failure. This conclusion was already foreseen by Bainbridge (1982) in her classical statement on the ironies of automation.

58


It can therefore be argued that even the most advanced systems need human supervision, decision-making and control – given that no automated system can be 100 % reliable in every aspect or all the time.

The effect of human intervention in advanced manufacturing systems thus depends on the operators’ capabilities in handling their technical complexity. These capabilities include the combination of the accumulated skills and knowledge of the operators involved, e.g. education levels, training, work environment, support tools, system interfaces, work procedures etc. (Harlin et al., 2006).

Thus, as a non-routine event, operative decision-making is, per definition, not possible to automate in the classical sense. As long as we are not talking about (future) AI technology, machines implement ready-made operations in reaction to pre- or user-defined decision situations and, consequently, fully automatic systems will not make decisions in action. The machine operator, in contrast, has the ability to make tactical decisions in action. In fact, as Billings (1997) argues, this is one of the main reasons why we employ humans in control systems at all.

6.2 DISCUSSION OF METHOD

The methodology used in this thesis has mainly followed an interpretive and abductive approach of qualitative research, and the chosen research design involves both observations from empirical case studies and literature reviews. The results of the literature reviews presented in papers I, II, IV V and VII were, together with the empirical results of observations in papers I, IV, VI and the Delphi survey in paper III, interpreted and analysed by the researchers. This means that the general scientific requirements of objectivity, observability and generalisability have different roles (Bryman, 2004).

Since interpretation is also assumed to be inherent in qualitative research, a discussion of objectivity is relevant according to Kirk and Miller (1986). Further, these authors describe how qualitative researchers use the concept to mean the quest of finding out how best to describe the empirical (social) world, and to explain the consequences of the choices made by people that lead to a particular construction of what the world “is”. However, they also argue that it is not a search for an absolute truth. Instead, they argue that new views are generally taken as complementary to old views, rather than replacing them. Objectivity is thus a problematic issue according to Lipshitz et al. (2001), who discuss the differences between the observer as recorder and as interpreter. For example, as a recorder, the observer collects and reports ‘raw data’ that are assumed to be objective. However, as Lipshitz et al. (2001) claim, when an interpreter is observing an interpretation, the interpretation itself is inseparable from the observation. Looking at it in this way, and thereby creating a distinction between collecting and interpreting, objectivity may very well be a quality that is necessary to achieve before being able to show validity and reliability , as Kirk and Miller (1986) suggest. Firestone (1993) and Fishman (1999) also argue that, to achieve confirmability, the researcher has to show that his or her data, interpretations and their outcome are rooted in a context and in other persons apart from the researcher.

Further, as noted by Lipshitz et al. (2001), interpretation also seeks to understand the meaning of what is observed rather than verifying hypotheses, since the interpretation is based on whether the results that correspond to reality can be considered to be “true”. Truth can also, according to Alvesson and Sköldberg (2006), refer to the pragmatic

59

DISCUSSION

usefulness or meaning of a certain phenomenon. And, as Rubenowitz (1980) notes understanding the meaning of what is interpreted also diminishes the importance of determining causes and explaining causal relationships.

6.3 ASSESSMENT OF RESEARCH QUALITY

In the research method literature, it is argued that research design is supposed to represent a logical set of statements to judge the quality of any given design according to certain logical tests. However, it has also been argued by e.g. Bryman (2004) that qualitative studies should not be evaluated using the same criteria as are used in quantitative studies. In order to establish trustworthiness, the conventional criteria that are often used for evaluation qualitative studies have, in line with Lincoln and Guba (1985), been proposed to replace conventional ones: transferability replaces external validity and generalisation, credibility replaces internal validity, dependability replaces reliability and confirmabilityreplaces objectivity.

6.3.1 Transferability

Transferability describes the process of applying the results of research from one situation to other similar situations. To do this effectively, researchers need to know as much as possible about the original research situation in order to determine whether it is similar to their own situation (Bryman, 2004). Therefore, researchers must supply, as Lincoln and Guba (1985) argue, a highly detailed description of their research situation and methods, which is achieved in the early steps (step 1-2) of the DYNAMO methodology. In the first step (step 1) the goal and purpose of the LoA analysis is documented in detail, together with the delimitations and focus of the target system. In the next step (step 2) documentation is followed up by an identification of the purpose of the production system and of the machines in the production flow being analysed. The characteristics of a product or variant are also documented in detail. Further, since the focus in the assessment and measurement made with the DYNAMO methodology is on general physical tasks, such as e.g. pounding, screwing and attaching, transferability can be achieved. The cognitive task, such as obtaining information and decision-making is context independent and thus the transferability of cognition is also achieved.

6.3.2 Credibility

Credibility refers, according to Bryman (2004), to showing how valid the empirical results are from a non-statistical approach. Similar to Bryman, Lincoln and Guba (1985) argue that credibility refers to the “truth value” and therefore involves two tasks: 1) carrying out the investigation in such way as to increase the probability that the findings will be credible and 2) demonstrating the credibility of the findings by having them approved by the respondents. According to Lincoln and Guba (1985) credibility can be strengthened in different ways. One is through triangulation. The material from the first six empirical studies (case studies A to F), which forms the foundation of the DYNAMO methodology, was collected from several sources by observations (papers IV, V and VI) and by interviewing people in different positions (paper III). To further increase the credibility of the actual assessment and measurement with the DYNAMO methodology, the empirical data collected in step 6 are always presented and verified against some of the participants in step 7. By presenting the reference scales and the data from the measurement and assessment, the respondent has the possibility to verify that the researchers’ interpretations

60


and conclusions are correct. An alternative way to increase credibility is by using peer debriefing (Lincoln and Guba, 1985). This type of technique provides an opportunity to test the method by exposing the methodology and its reference scales to a group of researchers to explore aspects of the inquiry that might otherwise remain implicit in the inquirer’s mind. This was done by permitting three colleagues to conduct the assessment and measurement of LoA in case study G, as well as allowing fellow researchers and three groups of students to use the methodology in four additional case studies for further verification of the DYNAMO methodology.

6.3.3 Dependability and confirmability

Dependability and confirmability refer to how consistent and neutral the results are (Bryman, 2004). This means that repeating the inquiry would result in similar findings and a similar degree to which the findings are determined by the respondents and not by the bias, aims interests and perspectives of the inquirer (Lincoln and Guba, 1985). This was addressed by having co-researchers consider alternative interpretations of the empirical material in discussions and by permitting persons that were not involved in the development of the reference scales and the DYNAMO methodology conduct individual case studies. Moreover, parts of the empirical material from the DYNAMO methodology were presented at different workshops and conferences to allow the examination of other researchers and practitioners, as in e.g. papers III, IV and VI.

6.4 FURTHER RESEARCH

This research has explored how levels of automation (LoA) in production systems can be used to gain a better understanding of how the interaction between humans and technology (e.g. automation) can be improved from a task perspective. Since the DYNAMO methodology only gives recommendations for improving the potential of automation based on the analysis of the Mechanical-Information-LoA diagram (figure 5-8), more research is needed both in terms of what parameters and performance indicators should be considered in order to achieve an optimised and effective system. These parameters and performance indicators must be verified by additional empirical studies. This will be done as a part of the ongoing research project ProAct - Proactive assembly systems.

Due to the complex interaction between parameters such as flexibility, complexity, robustness and efficiency, it would be interesting to conduct additional research in simulation and visualisation tools to analyse the parameters. This will be done as a part of the ongoing research project SIMTER - Simulation-based production development tool for traditional manufacturing industries.

Additional research would be of interest for testing the DYNAMO methodology in other human-machine areas, such as nuclear control and telerobotics.

61

DISCUSSION

62


CHAPTER 7

CONCLUSIONS


This chapter contains the conclusions which thereby has answered the research questions and fulfilled the research objective and aim.

The aim of this thesis was to contribute to theoretical and practicable development of the concept of Levels of Automation (LoA) in production systems.This is done by presenting a cognitive and physical approach to the effect and measurement of automation. The objective was to explore how Levels of Automation can be used in production, to gain a better understanding of how the interaction between humans and technology (e.g. automation) can be improved.

7.1 IS THE CONCEPT OF LEVELS OF AUTOMATION RELEVANT IN PRODUCTION?

This thesis shows that the simplest form of automation operates in two modes, manual and automatic. In more complex systems, there may be several different levels and/or modes of operations. In a general perspective, it can be seen that automation can vary across a continuum of levels, from the lowest level of fully manually performance to the highest level of full automation.

In complex systems, e.g. aviation and control rooms in nuclear power plants, it is not uncommon to find multiple automation modes of control and information. This thesis shows that most research on LoA has focused on the cognitive parts, where automation helps to speed up the information flow and gives decision support to help the operator monitor the situation. However, this research also demonstrates that, from a manufacturing perspective, it is important to recognise that the process of manufacturing not only consists of control and information tasks but also of physical tasks, such as replacement or support of human muscle power - tasks that have often been viewed from a production perspective.

Based on the research in this thesis, the level of automation can be defined as:

‘The allocation of physical and cognitive tasks between humans and technology, described as a continuum ranging between totally manual and totally automatic’

63

CONCLUSION

Breaking down the activities into physical and cognitive tasks, but still acknowledging the co-operation between humans and technology, it is useful to see allocation as two parallel LoA scales of cognitive and physical tasks. Each of these two tasks can then be ordered in seven steps, from totally manual performance (without any tools) to fully automated performance (without any human involvement).

7.2 CAN THE LEVEL OF AUTOMATION IN PRODUCTION BE MEASURED?

Measurement of automation is not a trivial task. From the research reported in this thesis, it can be seen that many other researchers have developed different taxonomies for classifications of LoA. Nevertheless, to this time, none of the reviewed taxonomies are designed for measurement of LoA.

With 15 measurements at ten companies, it can be argued that the definition of LoA given in research question I is applicable to measurements. It can further be concluded from the research in this thesis that the seven levels of automation on each of the two reference scales are applicable to most manufacturing tasks. The level of automation in can thus be measured by identifying the main tasks and the sub-tasks with Hierarchical Task Analysis (HTA). When the task is identified, the measurement is conducted by separating the physical and cognitive tasks, and thereby individually comparing each cognitive and physical task against the reference scales. However, for many tasks in manufacturing, as well as in other domains, it is not always relevant to use the entire LoA span. For different practical and economic reasons, each of the two reference scales for LoA can be delimited to the relevant maximum or minimum for a specific task. The research in this thesis also demonstrates that the two reference scales should not be seen as interlinked, but depends on the task and the experience and knowledge of the operator; it can thus be relevant to have different LoA spans for each of the two reference scales.

7.3 WHAT CONSEQUENCES CAN BE EXPECTED DUE TO CHANGES IN THE LEVEL OF AUTOMATION IN PRODUCTION?

Based on the research presented in this thesis, it can be concluded that every change in a complex technical system, such as a production system, introduces consequences for humans and for the output of the system. As noted in paper VII, flexibility, complexity and robustness are three areas that have been identified as important from an industrial perspective in terms of reducing the time and cost parameters in a system. As noted in paper III, with an increased number of variants in production, the demands on the automated production system will increase. And, with more flexible machines, the complexity increases, as do the demands for handling the automatic equipment.

Further, from a human factor perspective, those three areas, flexibility, complexity and robustness, can be linked to expected consequences for humans, as a consequence of changes in LoA. For example, as argued in papers I and IV, if automation is to replace human operators, the result can be that we risk losing an important information exchange that is needed to uphold tacit working skills and knowledge. As seen in case A, by increasing LoA in order to increase the robustness of the process, we also risk increasing the complexity of the system and thereby create a gap between the physical process and human operators. This increased complexity can also lead to longer time and larger difficulty in diagnosing failures and the possibilities to increase manpower when needed. It can thus be concluded that each change of LoA in a complex system, such as a

64


production system, introduces new risks of causing disturbances in the system and of decreasing product quality. This research also indicates that, if the investment in more advanced production equipment is not followed by necessary training and education, there might be a risk that the production system is not utilised to its full potential.

Intermediate LoA can therefore promote higher operator situation awareness and enhanced performance during production disturbances. Further, the investment and development costs in automation must be put in relation to demands for flexibility and reduction in cost and time.

Finally, as noted by many other researchers, one of the main reasons that an automation project ends in failure is unrealistic or undefined objectives. It is therefore important to understand of what physical and cognitive tasks the production consists, as well as what range of possibilities the automation has, before making the decision of investment and implementation of a new production system. This can be done by assessing, measuring, and analysing the automation needs at the early stages of an automation project, thereby getting a thorough investigation of the needs and tasks of the production system.

65

CONCLUSION

66


REFERENCES

Alvesson, M. and Sköldberg, K. (2006) Tolkning och reflektion: vetenskapsfilosofi och kvalitativ metod. Studentlitteratur, Lund, Sweden (in Swedish).

Ammons, J. C., Govindaraj, T. and Mitchell, C. M. (1988) Decision Models for Aiding FMS Scheduling and Control. IEEE Transactions on System, Man, and Cybernetics, Vol. 5, No. 18, pp. 744-756.

Averty, P., Athenes, S., Collet, C. and Dittmar, A. (2002) Evaluating a new index of mental workload in real ATC situation using psychophysiological measures. In: Proceedings of the 21st Digital Avionics Systems Conference, Irvine, CA, USA.

Axelsson, J. R. (2000) Quality and Ergonomics - towards successful integration. Quality and Human Systems Engineering, Department of Mechanical Engineering, Linköping University, Linköping, Sweden (Ph.D. Thesis).

Ayyub, B. M. (2001) Elicitation of Expert Opinions for Uncertainty and Risks. CRC Press, Washington, DC, USA.

Bainbridge, L. (1982) Ironies of Automation. Automatica, Vol. 19, pp. 775-779.

Bauer, A., Bowden, R., Brown, J., Duggan, J. and Lyons, G. (1991) Shop Floor Control Systems: From Design to Implementation. Chapman & Hall, London, England.

Benhabib, B. (2003) Manufacturing: Design, Production, Automation, and Integration.Marcel-Dekker, New York, USA.

Billings, C. E. (1997) Aviation Automation: The Search for a Human-Centered Approach.Lawrence Erlbaum Associates, Mahwah, New Jersey, USA.

Blanchard, B. S. and Fabrycky, W. J. (1998) Systems engineering and analysis. Third edition. Prentice Hall, Inc., Upper Saddle River, NJ, USA.

Blauner, R. (1964) Alienation and Freedom. University of Chicago Press, Chicago, USA.

Boothroyd, G. (2005) Assembly Automation and Product Design. CRC Press, New York, USA.

Braverman, H. (1974) Labour and Monopoly Capital: the degradation of work in the twentieth century. Monthly Review Press, New York, USA.

67

REFERENCES

Brehmer, B. (1993) Processoperatörernas arbete i moderna kontrollrum. In MÄNNISKOR - DATORTEKNIK - ARBETSLIV. Lennerlöf, L. (Eds.), Stockholm, Sweden (in Swedish).

Bright, J. R. (1958) Automation and management. Maxwell Reprint Company, USA, Boston, USA.

Bryman, A. (2004) Social research methods. Oxford University Press, Oxford, England.

Chapanis, A. (1996) Human Factors in Systems Engineering. John Wiley & Sons, Inc., New York, USA.

Chisholm, A. W. J. (1990) Nomenclature and Definitions for Manufacturing System (English Language Version). Annuals of CIRP, Vol. 39, No. 2, pp. 735-742.

Chryssolouris, G. (1993) Manufacturing systems: theory and practice. Springer, New York, USA.

Connors, M. M. (1998) Teaming Humans and Automated Systems in Safely Engineered Environments. Life Support & Biosphere Science, Vol. 5, No. 4, pp. 453-460.

Danziger, J. N. (1985) Social science and the social impacts of computer technology. Social Science Quarterly, Vol. 66, No. 1, pp. 3-21.

Daouk, M. (2000) Air Traffic Control Task Allocation and Decision Processes. Department of Aeronautics and Astronautics, MIT, Cambridge, MA, USA.

Dencker, K. (2007) LoA mätning med DYNAMO metoden på Siemens Building Technology. IVF Industriforskning och utveckling AB, Stockholm, Sweden.

Drucker, P. F. (1999) Management Challenges for the 21st Century. Butterworth Heinemann, New York, USA.

Dubois, A. and Gadde, L.-E. (2002) Systematic combining: an abductive approach to case research. Journal of Business Research, Vol. 55, No. 7, pp. 553-560.

Duncheon, C. (2002) Product miniaturization requires automation - but with a strategy. Assembly Automation, Vol. 22, No. 1, pp. 16-20.

Eijnatten, F. M. V., Sitter, U. D., Gustavsen, B., Emery, F. and Beinum, H. V. (1993) Theparadigm that changed the work place. Arbetslivscentrum, Stockholm, Sweden.

Eisenhardt, K. M. (1989) Building Theories from Case Study Research. The Academy of Management Review, Vol. 14, No. 4, pp. 532-550.

Elmgren, S., Höglund, M., Lenhagen, L., Rydén, L. and Sjögren, A. (2007) Automatisering - Utredning av automationspotetialen i manuella monteringsprocesser vid Volvo Lastvagnar, Tuve. Chalmers, avd. för produkt och produktionsutveckling, Göteborg, Sweden.

68


Endsley, M. R. (1988) Design an evaluation for situation awareness enhancement. In: Proceedings of the Human Factors Society 32nd Annual Meeting, Santa Monica, USA.

Endsley, M. R. (1995) Toward a theory of situation awareness in dynamic systems. Human Factors, Vol. 37, No. 1, pp. 32-64.

Endsley, M. R. and Kiris, E. O. (1995) Out-of-the-loop performance problem and level of control in automation. Human Factors, Vol. 37, No. 2, pp. 381-394.

Endsley, M. R. (1996) Automation and situation awareness. In: Automation and human performance: Theory and applications. Parasuraman, R. and Mouloua, M. (Eds.), Mahawah, NJ, USA.

Endsley, M. R. (1997) Level of automation: Integrating humans and automated systems. In: Proceedings of the 41st Annual Meeting of the Human Factors and Ergonomics Society, Albuquerque, NM, USA.

Endsley, M. R., Onal, E. and Kaber, D. B. (1997) The Impact of intermediate levels of automation on situation awareness and performance in dynamic control systems. In: Proceedings of the 6th IEEE Conference on Human Factors and Power Plants,Orlando, FL, USA.

Endsley, M. R. and Kaber, D. B. (1999) Level of automation effects on performance, situation awareness and workload in a dynamic control task. Ergonomics, Vol. 42, No. 3, pp. 462-492.

Endsley, M. R. and Garland, D. J. (2000) Situation Awareness Analysis and Measurement.Lawrence Erlbaum Associates, London, England.

Fasth, Å., Stahre, J. and Frohm, J. (2007) Relations between parameters/performers and level of automation. In: Monostori, L. (Eds.) Proceedings of the IFAC Workshop on Manufacturing Modelling, Management and Control, November 14-16, 2007, Budapest, Hungary.

Firestone, W. A. (1993) Alternative Arguments for Generalizing From Data as Applied to Qualitative Research. Educational Researcher, Vol. 22, No. 4, pp. 16-23.

Fishman, D. B. (1999) The Case for Pragmatic Psychology. NYU Press, New York, USA.

Fitts, P. M. (1951) Human Engineering for an Effective Air-navigation and Traffic-control system. National Research Council, London, England.

Flach, J. and Mulder, M. (2004) Putting Humans Into Control. The IT University of Copenhagen, Copenhagen, Denmark.

Form, W. (1987) On the Degradation of Skills. Annual Review of Sociology, Vol. 13, No. 1, pp. 29-47.

Frohm, J., Karsvall, A., Hassnert, M., Stahre, J. and Hollnagel, E. (2002) User-Driven Decision Automation within Industrial Production Systems. In: Caldenfors, D.,

69

REFERENCES

Eklund, J. and Kiviloog, L. (Eds.) Proceedings of the 34th Annual Congress of the Nordic Ergonomics Society, Vildmarkshotellet, Kolmården, Sweden.

Frohm, J., Karsvall, A., Hassnert, M., Stahre, J. and Hollnagel, E. (2003) System perspective on task allocation and automation design. In: Stahre, J. and Mårtensson, L. (Eds.) Proceedings of the 8th symposium on "Automated Systems Based on Human Skills and Knowledge", Göteborg, Sweden.

Frohm, J., Lindström, V. and Bellgran, M. (2005) A model for parallel levels of automation within manufacturing. In: Pasquino, R. (Eds.) Proceedings of the 18th International Conference on Production Research, Fisciano, Italy.

Frohm, J., Granell, V., Winroth, M. and Stahre, J. (2006) The industry’s view on automation in manufacturing. In: Mayer, F. and Stahre, J. (Eds.) Proceedings of the 9th symposium IFAC on "Automated Systems Based on Human Skills and Knowledge", 22-24 May, 2006, Nancy, France.

Frohm, J., Karsvall, A., Thunberg, A. and Stahre, J. (submitted A) Possibilities for Automated Decision-making in Industrial Systems. (Submitted for publication in International Journal of Industrial Ergonomics).

Frohm, J., Lindström, V., Winroth, M. and Stahre, J. (submitted B) Levels of Automation in Manufacturing. (Submitted for publication in Ergonomia).

Granell, V. (2007) Levels of automation in manufacturing systems: aligning strategic and tactical decisions by means of operational measurement. Department of Product and Production Development, Chalmers University of Technology, Göteborg, Sweden (Licentiate Thesis).

Granell, V., Frohm, J., Bruch, J. and Dencker, K. (2007) Validation of the DYNAMO methodology for measuring and assessing levels of automation. In: Stahre, J. (Eds.) Proceedings of the Swedish Production Symposium, Göteborg, Sweden.

Greenstein, J. S. and Revesman, M. E. (1986) Human-Computer Communication for Dynamic Task Allocation. In: Human-Computer Dialogue Design. Williges, R. W. (Eds.), Amsterdam, Holland.

Grundström, G. and Ottosson, P. (2008) Flödesanalys och simulering av kylmodulsproduktion. Institutionen för Produkt- och Produktionsutveckling, Chalmers tekniska högskola, Göteborg, Sweden (Master Thesis).

Harlin, U., Frohm, J., Berglund, M. and Stahre, J. (2006) Towards efficient automation implementation. In: Mayer, F. and Stahre, J. (Eds.) Proceedings of the 9th symposium IFAC on "Automated Systems Based on Human Skills and Knowledge",22-24 May, 2006, Nancy, France.

Higgins, P. G. (1999) Job Shop Scheduling: Hybrid Intelligent Human-Computer Paradigm. Department of Mechanical and Manufacturing Engineering, University of Melbourne, Melbourne, Australia (Ph.D. Thesis).

70


Hoffman, R. R., Feltovich, P. J., Ford, K. M., Woods, D. D., Klein, G. and Feltovich, A. (2002) A Rose by Any Other Name... Would Probably Be Given an Acronym. IEEE Intelligent Systems and Their Applications, Vol. 17, No. 4, pp. 72-80.

Hollnagel, E. (1997) It is all in the mind. Cognitive ergonomics, Vol. 40, No. 19, pp. 1170-1182.

Hollnagel, E. and Woods, D. D. (2005) Joint cognitive systems: Foundations of cognitive systems engineering. CRC Press, Boca Raton, FL, USA.

Holme, I. M. and Solvang, B. K. (2001) Forskningsmetodik - Om kvalitativa och kvantitativa metoder. Studentlitteratur, Lund, Sweden (in Swedish).

Inagaki, T. (1993) Situation-adaptive degree of automation for system safety. In: Proceedings of the 2nd IEEE International Workshop on Robot and Human Communication, Tokyo, Japan.

Ingemansson, A. (2004) On Reduction of Production Disturbances in Manufacturing Systems Based on Discrete Event Simulation. Division of Robotics, Lund University, Lund, Sweden (Ph.D. Thesis).

Johnson-Laird, P. N. (1980) Mental models in cognitive science. Cognitive Science: A Multidisciplinary Journal, Vol. 4, No. 1, pp. 71-115.

Jonsson, K. (2007) Flödesanalys och simulering av ny och gammal produktionsanläggning på EKA Chemicals AB. Institutionen för Produkt- och Produktionsutveckling, Chalmers tekniska högskola, Göteborg, Sweden (Master Thesis).

Jordan, N. (1963) Allocation of functions between man and machines in automated systems. Journal of Applied Psychology, 47, 3, pp. 161-165.

Järvinen, J., Vannas, V., Mattila, M. and Karwowski, W. (1996) Causes and Safety Effects of Production Disturbances in FMS Installations: A Comparison of Field Survey Studies in the USA and Finland. The International Journal of Human Factors in Manufacturing, Vol. 6, No. 1, pp. 57-72.

Kaber, D. B., Onal, E. and Endsley, M. R. (2000) Design of automation for telerobots and the effect on performance, operator situation awareness, and subjective workload. Human Factors and Ergonomics in Manufacturing, Vol. 10, No. 4, pp. 409-430.

Kantowitz, B. H. and Campbell, J. L. (1996) Pilot Workload and Flightdeck Automation. In: Automation and Human Performance: Theory and Applications. Parasuraman, R. and Mouloua, M. (Eds.), Mahwah, NJ, USA.

Karsvall, A., Frohm, J., Hassnert, M. and Stahre, J. (2003) On Decision-Making: Making Sense of System Performance. In: Lützhöft, M. H. (Eds.) Proceedings of the 22nd European Conference on Human Decision Making and Manual Control,Linköping, Sweden.

71

REFERENCES

Kern, H. and Schumann, M. (1985) Industriearbeit und Arbeiterbewusstsein : eine empirische Untersuchung über den Einfluss der aktuellen technischen Entwicklung auf die industrielle Arbeit und das Arbeiterbewusstsein. Suhrkamp, Frankfurt am Main, Germany (in German).

Kirk, J. and Miller, M. L. (1986) Reliability and Validity in Qualitative Research. Sage Publications, New York, USA.

Kirwan, B. and Ainsworth, L. K. (1992) A guide to Task Analysis. Taylor & Francis, London, England.

Klein, G. (1993) Decision making in action: models and methods. Ablex, Norwood, NJ, USA.

Klein, G. (1998) Sources of Power : How People Make Decisions. MIT Press, New York, USA.

Landes, D. S. (2003) The Unbound Prometheus: Technological Change and Industrial Development in Western Europe from 1750 to the Present. Cambridge University Press, London, England.

Langmann, R. (2004) Taschenbuch der Automatisierung: mit 110 Tab. Carl-Hanser-Verl., Leipzig, Germany (in German).

Lee, J. D. and Moray, N. (1992) Trust, control strategies and allocation of function in human-machine systems. Ergonomics, Vol. 35, No. 10, pp. 1243-1270.

Lee, J. D. and Moray, N. (1994) Trust, self-confidence, and operators' adaptation to automation. Human Factors, 40, 1, pp. 153-184.

Liljegren, E. (2004) Increasing the Usability of Medical Technology. Department of Product and Production Development, Chalmers University of Technology, Göteborg, Sweden (Ph.D. Thesis).

Lincoln, Y. S. and Guba, E. G. (1985) Naturalistic Inquiry. Sage Publishing, Thousand Oaks, USA.

Lindsay, P. H. and Norman, D. A. (1977) Human information processing: an introduction to psychology. Academic Press, New York, USA.

Lindström, V. (2005) Individuella bedömningar av dynamisk automationsnivå. Chalmers, Produkt och produktionsutveckling, Göteborg, Sweden.

Linstone, H. A. and Turoff, M. (2002) The Delphi Method: Techniques and Applications.New Jersey Institute of Technology, New Jersey, USA.

Lipshitz, R., Klein, G., Orasanu, J. and Salas, E. (2001) Taking stock of naturalistic decision making. Journal of Behavioral Decision Making, Vol. 14, No. 5, pp. 331-352.

72


Logan, G. D. and Etherton, J. L. (1994) What is learned during automatization? The role of attention in constructing an instance. Journal of experimental psychology. Learning, memory, and cognition, Vol. 20, No. 5, pp. 1022-1050.

Marr, D. C. (1982) Vision: a computational investigation into the human representation and processing of visual information. Freeman, San Francisco, USA.

Marsh, R. M. and Mannari, H. (1981) Technology and Size as Determinants of the Organizational Structure of Japanese Factories. Administrative Science Quarterly,Vol. 26, No. 1, pp. 33-57.

Matti, J. and Tiger-Ring, S (2006) Automationsgradens effekt på svensk produktion - En studio av svenska tillverkningssystem. Institutionen för Produkt- och Produktionsutveckling, Chalmers tekniska högskola, Göteborg, Sweden (Master Thesis).

Meister, D. (1987) Systemsdesign, development and testing. In: Handbook of Human Factors. Salvendy, G. (Eds.), New York, USA.

Merchant, M. E. (1961) The Manufacturing System Concept in Production Engineering Research. Annals of CIRP, Vol. 10, No. 2, pp. 77-83.

Merchant, M. E. (2000) The Future of Manufacturing. In Handbook of Industrial Automation. Shell, R. L. and Hall, E. L. (Eds.), Cincinnati, USA.

Mital, A. (1997) What role for humans in computer integrated manufacturing? International Journal of Computer Integrated Manufacturing, Vol. 10, No. 1-4, pp. 190-198.

Mouloua, M., Parasuraman, R. and Molloy, R. (1993) Monitoring automation failures: effects of single and multi-adaptive function allocation. In: Proceedings of the 37th Annual Meeting of the Human Factors and Ergonomics Society Seattle, WA, USA.

Muir, B. M. (1987) Trust between humans and machines, and the design of decision aids. International Journal of Man-Machine Studies, Vol. 27, No. 5-6, pp. 527-539.

Muir, B. M. (1994) Trust in automation: Part I. Theoretical issues in the study of trust and human intervention in automated systems. Ergonomics, 37, 11, pp. 1905 - 1922.

Nakamura, N. and Salvendy, G. (1994,) Human Planner and Scheduler. In: Design of Work and Development of Personnel In Advanced Manufacturing. Salvendy, G. and Karwowski, W. (Eds.), New York, USA.

Neisser, U. (1976) Cognition and reality principles and implications of cognitive psychology. Freeman, San Francisco, USA.

Norman, D. A. (1987) Some observations on mental models Morgan Kaufmann Publishers Inc., San Francisco, CA, USA

73

REFERENCES

Okoli, C. and Pawlowski, S. D. (2004) The Delphi method as a research tool: an example, design considerations and applications. Information & Management, Vol. 42, pp. 15-29.

Parasuraman, R. and Mouloua, M. (1996) Automation and Human Performance: Theory and Applications. Lawrence Erlbaum Associates, Mahwah, NJ, USA.

Parasuraman, R., Mouloua, M. and Molloy, R. (1996) Effects of Adaptive Task Allocation on Monitoring of Automated Systems. Human Factors, Vol. 38, No. 4, pp. 665-679.

Parasuraman, R. (2000) Designing automation for human use: empirical studies and quantitative models. Ergonomics, Vol. 43, No. 7, pp. 931-951.

Parasuraman, R., Sheridan, T. B. and Wickens, C. D. (2000) A model for types and levels of human interaction with automation. IEEE transactions on system, man, and cybernetics - Part A: Systems and humans, Vol. 30, No. 3, pp. 286-296.

Price, H. E. (1985) The Allocation of Functions in Systems. Human Factors, Vol. 27, No. 1, pp. 33-45.

Psychology Writing Center (2005) Writing a Psychology Literature Review. University of Washington, Washington, USA.

Rasmussen, J. (1983) Skills, rules and knowledge: Signals, signs and symbols, and other distinctions in human performance models. IEEE transactions on system, man, and cybernetics, Vol. 13, No. 3, pp. 257-266.

Rasmussen, J. (1986) Information Processing and Human-Machine Interaction: An Approach to Cognitive Engineering. Elsevier Science Inc., New York, NY, USA.

Rasmussen, J. (1988) Human factors in the high-risk systems. In: Proceedings of the 4th IEEE Conference on Human Factors and Power Plants, Monterey, CA, USA.

Reveliotis, S. A. (1999) Production planning and control in flexibly automated manufacturing systems: current status and future requirements. In: Proceedings of the IEEE International Conference on Robotics and Automation, Marriott Hotel, Renaissance Center, Detroit, Michigan, USA.

Riley, V. (1996a) What avionics engineers should know about pilots and automation. Aerospace and Electronic Systems Magazine, IEEE, Vol. 11, No. 5, pp. 3-8.

Riley, V. (1996b) Operator Reliance on Automation: Theory and Data. In: Automationand Human Performance: Theory and Applications. Parasuraman, R. and Mouloua, M. (Eds.), Mahwah, NJ, USA.

Rouse, W. B. (1988) The Human Role in Advanced Manufacturing Systems. In: Designand Analysis of Integrated Manufacturing Systems. Compton, W. D. (Eds.), Washington D.C, USA.

74


Rubenowitz, S. (1980) Utrednings- och forskningsmetodik Akademiförlaget, Göteborg, Sweden (in Swedish).

Sanders, M. S. and McCormick, E. J. (1993) Human Factors in Engineering and Design.McGraw-Hill, Inc., New York, USA.

Saridis, G. N. (2000) Intelligent Manufacturing in Industrial Automation. In: Handbook of Industrial Automation. Shell, R. L. and Hall, E. L. (Eds.), Cincinnati, USA.

Sarter, N. B., Woods, D. D. and Billings, C. E. (1997) AUTOMATION SURPRISES. In: Handbook of Human Factors & Ergonomics. Salvendy, G. (Eds.), New York, USA.

Sarter, N. B. (2000) The Need for Multisensory Interfaces in Support of Effective Attention Allocation in Highly Dynamic Event-Driven Domains: The Case of Cockpit Automation. International Journal of Aviation Psychology, Vol. 10, No. 3, pp. 231-245.

Sasse, M. A. (1997) Eliciting and Describing Users' Models of Computer Systems. School of Computer Science, University of Birmingham, Birmingham, England (Ph.D. Thesis).

Satchell, P. (1998) Innovation and Automation. Ashgate Publishing, Aldershot, England.

Scerbo, M. W. (1996) Theoretical Perspectives an Adaptive Automation. In: Automation and Human Performance: Theory and Applications. Parasuraman, R. and Mouloua, M. (Eds.), Mahwah, NJ, USA.

Sheridan, T. B. (1980) Computer Control and Human Alienation. Technology Review,Vol. 83, No. 1, pp. 60-73.

Sheridan, T. B. (1992) Telerobotics, Automation, and Human Supervisory Control. MIT Press, Cambridge, MA, USA.

Sheridan, T. B. (1995) Human centered automation: oxymoron or common sense? In: Proceedings of the IEEE International Conference on Systems, Man and Cybernetics, Vancouver, Canada.

Sheridan, T. B. (1997) Task allocation, task analysis and supervisory control. In: Handbook of human-computer interaction. Helander, M., Landauer, T. K. and Prabhu, P. (Eds.), Amsterdam, Holland.

Sheridan, T. B. (2002) Humans and Automation: System Design and Research Issues.John Wiley & Sons, Inc., Santa Monica, USA.

Simpson, R. L. (1985) Social control of occupations and work. Annual Review of Sociology, Vol. 11, pp. 415-436.

Skyttner, L. (2001) General Systems Theory - Ideas and Applications. World Scientific Publishing Co. Pte. Ltd., Singapore.

75

REFERENCES

Sowa, J. F. (1984) Conceptual structures: information processing in mind and machine.Addison-Wesley, London, England.

Stahre, J. (1995) Towards Human Supervisory Control in Advanced Manufacturing Systems. Department of Production Engineering, Chalmers University of Technology, Göteborg, Sweden (Ph.D. Thesis).

Starrin, B. and Svensson, P.-G. (1994) Kvalitativ metod och vetenskapsteori.Studentlitteratur, Lund, Sweden (in Swedish).

The Oxfords English Dictionary. (2006) "The Oxfords English Dictionary " Retrieved December 28, 2007, from http://dictionary.oed.com/.

Thunberg, A. (2006) A Cognitive Approach to the Design of Alarm Systems for Nuclear Power Plant Control Rooms. Department of Product and Production Development, Chalmers University of Technology, Göteborg, Sweden (Licentiate Thesis).

Waard, D. d. (1996) The measurement of drivers’ mental workload. The Traffic Research Centre, University of Groningen, Groningen, Holland (Ph.D. Thesis).

Walsham, G. (1995) The Emergence of Interpretivism in IS Research. Information Systems Research, Vol. 6, No. 4, pp. 376-394.

Wickens, C. D. (1992) Engineering psychology and human performance. HarperCollins, New York, USA.

Wickens, C. D., Lee, J. D., Liu, Y. and Gordon Becker, S. E. (2004) An introduction to human factors engineering. Prentice Hall International, Upper Saddle River, N.J., USA.

Williams, T. J. (1999) PERA AND GERAM:. ESTABLISHMENT OF THE PLACE OF THE. HUMAN IN ENTERPRISE INTEGRATION. In: Proceedings of the IFAC Congress, Beijing, China.

Williamsson, K. (2002) The Delphi method, Research methods for students and professionals. Wagga wagga, New York, USA.

Wilson, J. R. and Corlett, E. N. (1990) Evaluation of human work - A practical ergonomics methodology. Taylor & Francis Inc., Philadelphia, USA.

Winroth, M. (2004) On Manufacturing Strategies - Competing Through Inter-organizational Collaboration. Department of Mechanical Engineering, Linköpings Universitet, Linköping, Sweden (Ph.D. Thesis).

Winroth, M., Säfsten, K., Lindström, V., Frohm, J. and Stahre, J. (2006) Automation strategies - refinement of manufacturing strategy content. In: Proceedings of the 17th Annual Conference of POMS, Boston, USA.

Winroth, M., Säfsten, K. and Stahre, J. (2007 ) Automation strategies: existing theory or ad hoc decisions? International Journal of Manufacturing Technology and Management Vol. 11, No. 1, pp. 98 - 114

76


Woods, D. D., Cook, R. I. and Billings, C. E. (1995) The impact of technology on physician cognition and performance. Journal of Clinical Monitoring and Computing, Vol. 11, No. 1, pp. 5-8.

Wynekoop, J. L. and Russo, N. L. (1997) Studying system development methodologies: an examination of research methods. Information Systems Journal, Vol. 7, No. 1, pp. 47-65.

Yin, R. K. (2003) Case study research: design and methods. Sage Publications Inc., New York, USA.

Ylipää, T. (2000) High-Reliability Manufacturing Systems. Department of Human Factors Engineering, Chalmers University of Technology, Göteborg, Sweden (Licentiate Thesis).

Youtie, J., Shapira, P., Urmanbetova, A. and Wang, J. (2004) Advanced Technology and the Future of U.S. Manufacturing. In: Shapira, P., Youtie, J. and Urmanbetova, A. (Eds.) Proceedings of the Georgia Tech research and policy workshop, School of Public Policy and the Georgia Tech Economic Development Institute, Georgia, USA.

Zuboff, S. (1988) In the age of the smart machine : the future of work and power. Basic Books, New York, USA.

77

REFERENCES

78

APPENDED PAPERS

Paper I Frohm, J., Karsvall, A., and Hassnert, M. (2003). System perspective on task allocation and automation design. In Proceedings of the 8th symposium on automated systems based on human skills and knowledge, Göteborg, Sweden, September 22-24, 2003.

Paper II Frohm, J., Lindström, V. and Bellgran, M. (2005). A model for parallel levels of automation within manufacturing. In Proceedings of the 18th International Conference on Production Research, Salerno, Italy, July 31 – August 4, 2005.

Paper III Frohm, J., Granell, V., Winroth, M. and Stahre, J. (2006). The industry’s view on automation in manufacturing. In Proceedings of the 9th symposium IFAC on Automated Systems Based on Human Skills and Knowledge, Nancy, France, May 22-24, 2006.

Paper IV Frohm, J., Karsvall, A., Thunberg, A. and Stahre, J. (submitted). Possibilities for Automated Decision-making in Industrial Systems. (Submitted for publication in International Journal of Industrial Ergonomics).

Paper V Frohm, J., Lindström, V., Stahre, J. and Winroth, M. (submitted). Levels of Automation in Manufacturing. (Submitted for publication in Ergonomia).

Paper VI Granell, V., Frohm, J. Dencker, K. and Bruch, J. (2007). Validation of the Dynamo Methodology for Measuring and Assessing Levels of Automation. In Proceedings of the Swedish Production Symposium, Göteborg, Sweden, August 28-30, 2007.

Paper VII Fasth, Å., Stahre, J. and Frohm, J. (2007). Relations between parameters/ performers and level of automation. In Proceedings of the IFAC Workshop on Manufacturing Modelling, Management and Control, Budapest, Hungary, November 14-16, 2007.

PAPER I

Frohm, J., Karsvall, A., and Hassnert, M. (2003). System perspective on task allocation and automation design. In Proceedings of the 8th symposium on automated systems based on human skills and knowledge, Göteborg, Sweden, September 22-24, 2003.

SYSTEM PERSPECTIVES ON TASK ALLOCATION AND OPERATOR WORK

Frohm, J.1 Karsvall, A.

2 Hassnert, M.

3

1Division of Human Factors Engineering, Chalmers University of Technology, S-412 96 Göteborg, Sweden

2Division of Computer and Information Science, Linköping University, S-581 83 Linköping, Sweden

3IVF Industrial Research and Development Corporation, Argongatan 30, S-436 54 Mölndal, Sweden

E-mail: [email protected]

One of the main concerns in engineering has been to increase system performance by

the use of automation and information technology. This paper addresses the concept of

system task allocation and the role of the machine operator by exploring how common

models for task allocation handle the complex setting of manufacturing systems.

Conclusions are that we are headed towards more general roles for the operators.

Copyright � 2003 IFAC

Keywords: Automation, Decision making, Tasks, Resource allocation, Systems design

1. INTRODUCTION

In the past century, one of the main concerns for

design of manufacturing systems was to increase the

system performance by the use of automation

technology. In some cases, even, the ambition was to

create so-called lights-out factories with complete

automation of the production units. When the trend

had its peak in the 1970ies and 1980ies, a wave of

automation efforts swept through the industry under

such names as computer-integrated manufacturing

(CIM) and artificial intelligence (AI), and the

everyday worker was more or less seen as an

expendable resource. In the last two decades,

however, this form of scientific management was

challenged by an increasing insight about the

importance of the machine operators. This

development was also facilitated by inter-disciplinary

fields as human-machine interaction (HMI), a user-

centred approach on system design which

acknowledged that also the most advanced systems

need human supervision, decision-making, and

control – given that no automation can be fail-proof

in every aspect or all the time.

Still, in the user-centred view on system design, it is

seldom asked why or how people are put into the

production system in the first place, but taken for

granted that human operators are in place to interact

with the technology. The industry nonetheless spends

considerable resources on employment, training, and

support for machine operators, which implies that

there are several functional and practical reasons for

letting humans do the job, rather than machines. A

central question for the design of automation and

information technology has therefore not only been

how to construct the best possible technical systems,

or to optimise the human-machine interaction, but

also how to optimise task allocation (TA) between

the technology and its users.

Models for task allocation are often criticized for

being too static or too context independent, and

thereby to be insufficient in handling the complex

setting of manufacturing systems and their need for

flexible, yet efficient, production. This paper

addresses the concept of task allocation and the role

of the machine operator, in the context of industrial

production systems. For this purpose, we present a

field study on operator work in highly automated

environments, undertaken in the control-room at a

petroleum refinery, and we discuss the implications

of different task allocation theories on design of

manufacturing systems and operator work.

2. TASK ALLOCATION AND SYSTEM

PERFORMANCE

The initial contributions to the field of task allocation

were made by Fitts in the 1950ies (Fitts, 1951), who

put together a list of general tasks for both humans

and machines, illustrating where the performance of

one category exceeds that of the other. The list, and

later versions of the same concept of comparison

between different parts of the production system, is

often referred to as Fitts List, or the MABA-MABA List; the latter derived from what “men are better at”

and what “machine are better at”. Since its

introduction, however, a range of criticism has arisen

regarding the applicability in system engineering.

Jordan (1963), for instance, stated that the attempt to

compare the abilities of human and machines is

inappropriate since only machines can be designed

for a specific pre-defined task. According to Jordan,

the idea of comparing human and technology should

therefore be discarded, yet system designers should

also keep in mind what humans and machines do

best, and embrace that the two are complimentary

rather than conflicting entities.

As technology advanced, however, the number of

factors in Fitt’s List, in which the abilities of humans

exceed those of machines degreased (Goom, 1993);

and as a result, the importance of human labour and

the role for the machine operator became more and

more out of focus. The leftover allocation strategy

was therefore proposed as a serious alternative,

which sums-up a notion that system engineers should

automate what ever they can and then leave the

remaining functions to the human element (Daouk,

2000). A straightforward task allocation model was

presented by Meister (1987), who suggested that the

system engineers should write down all allocation

possibilities that could be identified, together with

applicable requirements for the task. Following this,

one should rank the order of all combinations of

allocation mix and criteria, thus determining a rank

order score. There are however several difficulties

with any direct allocation method of this kind. It

usually addresses performance differences in short-

term or in laboratory settings, with no consideration

for long-term consequences such as job satisfaction

or whether the machine is capable of successively

perform a given task. A number of unknown factors

must also be accounted for; such as hidden

assumptions, unanticipated situations, and non-

independence of tasks (Sheridan, 1992).

So, although many of the common automation design

models acknowledge that the ideal task allocation is

only plausible in theory, studies of automated

systems tend to focus on stable task allocation

between human and machine, and thereby

emphasising static and individual attributes of the

system components (Billings, 1997; Sheridan, 1997;

Sheridan, 2002). Task allocation models that

standardise the relationship between human and

machine can, in other words, be criticised for

manifesting an artificial distance between the

operator and his/her equipment, as they avoid the

complex interaction between system components. A

development of task allocation that tries to

incorporate the idea of an interactive production

system was the concept of dynamic task allocation

(DTA). The idea behind DTA was that not all

allocations possibilities are applicable to every

situation encountered by the system. To alter the

allocation according to certain situational factors can

therefore be an advantage – in particular when a task

can be performed equally well by humans as by

technology.

There is a variety of dynamic allocation models in

the literature, such as situation dependent (Greenstein and Revesman, 1986), flexible and adaptive allocation (Mouloua, et al., 1993), and

adaptive aiding (Rouse, 1988). Each of these may be

interpreted in different ways. Gramopadhye, et al. (1992), for example, refer to adaptive allocation and

adaptive aiding as the processes in which the

machine is used only when the human requires help

(e.g. during high workload). Greenstein and Lam

(1985), on the other hand, propose that adaptive

allocation may involve implicit and/or explicit

communication, based on models and predictions of

human behaviour. Explicit communication may be

initiated by the human (using commands) or by

machine (using prompts). There are however a

number of trade-offs between the additional

workload for the human related to explicit

communication, and the (lack of) reliability in

implicit communication; with some situations

favouring one over the other (Greenstein and

Revesman, 1986). Greenstein and Lam (1985)

proposes a mixed-initiated allocation, in which the

human is primarily responsible for initiating the

communication, but the machine may also take over

responsibility if the human omits part of a task.

DTA can also be seen as part of the broader term of

job allocation. That is, the procedure does not only

concern the balance between man and machine, and

how to distribute functionalities throughout the

system, but also the inter-personal co-ordination and

communication of employees (Rouse, 1988;Pacaux-

Lemoine and Debernard, 2002). In its framework,

however, it is assumed that the tasks to be allocated

can be generically predefined and then be

transformed in real-time to identified sub-tasks;

which accounts for a clearly defined system, in spite

of the fact that the boundaries are most often not

easily defined and that they are themselves

dynamically set in relation to the system

environment. A model that attempts to explain this

complex setting of human-machine systems is the

theory of Joint Cognitive Systems (Hollnagel, 2003).

As put forth by Hollnagel, the boundaries of a system

can be defined pragmatically, by describing the sets

of objects in functional relation with each other.

Components that are important for fulfilling the goal

of the JCS and that are possible for the JCS to

effectively control are considered as a part of the

JCS. Yet, difficulties in this notion are that both the

goals and composition of a production system might

change rapidly, and that system performance is not

the only criterion for allocating tasks between man

and machine. Meister (1987), for example, suggested

that we should consider variables as cost, reliability,

maintainability, personal requirements, and safety.

The impact of such criteria is apparent when humans

and machines can perform task equally well (or

badly), and by considering these trade-offs, the

optimum level of overall system performance may be

achieved, even if the optimum levels of each criterion

cannot be attained (Chapanis, 1996).

3. DIFFERENT ROLES AND DIFFERENT

ACTIVITIES

Although each system component is employed for a

general purpose, and contributes to the general

outcome, the different roles and abilities of man and

machine should not be forgotten. The operator

clearly has a different (personal or social) agenda,

when compared to the automatic system. For

example, to use decision-support systems (DSS) in

order to give more time for the operators, will result

in an entirely different time-frame for the operators

and generating a new kind of decision-making

process, where the operators are given strategic

capabilities, and responsibilities, rather than tactical

(Karsvall, et al., 2003). In Karsvall, et al. (2003) we

therefore argued that there need to be different design

solutions at different level of operator activity, and

that the purpose of the technical system may vary in

these tasks. We also concluded that system

engineering should focus more on system autonomy,

than on system automation, given that the latter refers

to the technological components in isolation. System

autonomy implies a focus on task performance from

an overall system perspective. In their manifesto for

Autonomic Computing (2001) IBM defines the term

as: “An approach to self-managed computing systems with a minimum of human interference.” Autonomic

computing, they say, derives from the autonomic

nervous system that controls crucial body functions

without conscious awareness or involvement. In the

same manner, technology can be employed to serve

functions that the human workers need not to be

engaged with. She or he would then only be

reminded of its presence when the system – or the

human – needs assistance.

Viewing the technological system as autonomic

points out that the users have different functions or

aims in the general system, and that there often is a

strive for parallel activities between man and

machine rather than interaction per se. From the user-

driven perspective, operators can for instance be set

to solve unanticipated (but likely) problems that the

present technology cannot handle (e.g. maintaining

the machinery, or to keep track on problematic

events in the production process). There are

furthermore professionals that possess expert or implicit skills, which cannot be automated, or are

expected to be too expensive or complicated to

automate. If considering that engineering is naturally

centred on technology, on the other hand, present

design models still tend to treat the human factor as

an odd feature of the productions system, and as a

potential threat to system efficiency (Rouse, 1988;

Parasuraman, et al., 2000). As a result, the machine

operators, and other employees, sometimes only get

passive roles in the production system, in which they

stands for external credibility to, or responsibility for,

the automatic system.

One example of the complexity of achieving

meaningful task allocation between human and

machine can be found in a field study conducted at

the control-room of a petroleum refinery, which we

accounted for in (Frohm, et al., 2002). The aim of the

study was to examine how process-operators

interacted with equipment and co-workers, in order

to handle every day decision situations. In the study,

the operator activities were recorded on videotape

and in written protocols. At the end of the work shift,

a semi-structured interview was conducted with the

participating operators in order to get a better

understanding of the activities that hade taken place

during the observation. The work tasks that were

conducted by the operators were mainly activities

involving adjustment of parameters in the production

system, in order to secure a stable refinery process;

for example, to compensate for the outdoor temperate

during the morning hours, or to re-order new lab

samples from the produced product.

It was observed that the operators spent most of their

time monitoring the system by looking at the

different screens; possibly in order to access relevant

information for a solid understanding of the present

situation, and to be able to predict further states of

the system. The operators also took different causes

of action in order to handle process adjustments, but

the observations indicated that the operators did not

use any explicit routine solutions for handling

process adjustments. In the interviews that followed

with the operators, however, it was clear that most of

the interaction was built on current knowledge and

experience, i.e. the kind of implicit personal expertise

that is not always possible to automate. Moreover, it

was observed that the tasks at hand were performed

very much in a nested and parallel way, where single

activities related to the tasks were extended over

entire work shift, both in terms of time and in terms

of system actors. These findings are right in

conjunction with the notions on the complexity of

task allocation put forth by for instance (Billings,

1997) and (Sheridan, 1997). From the case of

handling the outdoor temperature, it was e.g. seen

that the operator interacted both with the

technological system and with other sources of

information, such as colleagues or checklists, during

the first two hours of the work shift, in order to

handle tasks. We therefore concluded that the

observed tasks were performed in such a dynamic

and interwoven way, that no clear-cut allocation or

easy-to-define tasks could be identified, in the way

needed for meaningful task allocation in automation

design.

4. DISCUSSION

The elaboration of existing models of task allocation

have lead us to the conclusion that design models that

are directly grounded in the task allocation

perspective, might not be sufficient for understanding

the general mechanisms behind production systems

and the role for automation. By focusing on the

individual requirements on the automation, instead of

the demands on general system performance, task

allocation implies that automation technology per se

is more desirable than human labour. An unnecessary

polarisation between human and machine is thereby

created, which will lead designers in the wrong

direction and eventually diminish the potential for

truly adaptive production systems. Thus, when it

comes to understand the purpose of automated

systems, less time could be spent on allocating tasks

between human and machine attributes, and more

time should initially be given to the process of

describing requirements of the overall system.

On the other hand, we can moreover se implications

in another direction. Given that automation refers to

the technological components in isolation, designers

should rather focus on system and user autonomy in

task allocation (i.e., the parallel activities). This shift

in focus has the potential to enhance the ecological

validity of research; by emphasising that system

models should not only explain internal functionalities, but also consider open and flexible

work settings that involve interpersonal interaction as

well as human-machine interaction and system

automation. So, even if it is true that the operator

does what the machine cannot do, and vice versa, the

operators also have their own work agenda (or joint

cognitive system) that need to be accounted for in the

general picture. The operators are, after all, not

employed in order to be a function in the automated

system, but to be part of the overall solution.

It can thus be discussed if operator work really can

be defined as some sort of leftover allocation. As

seen in our field study the decision-making activities

of the operators are not only contributing to the

system performance, they are also one of the most

crucial events for overall production efficiency. The

operator’s role as problem-solver or decision-maker

therefore seems just as important for the system

outcome as it has ever been. Hence, along with

increasingly complex production system, the need for

competent professionals will rather increase than

decrease. Due to the unpredictability of the human

factor, however, system designers would normally

not allocate essential production tasks to specific

individuals. Thus, as technology advances, it appears

to be a limited room for specialised or localised work

activities. We are rather headed towards general

problem-solving tasks for the operators, which mean

less direct manipulation of local sub-systems and

more time spend on system overview and interaction

with co-workers.

5. ACKNOWLEDGEMENTS

The work presented is partial results from the project

User-Driven Decision Automation, which is carried

out as a co-operation between Chalmers University

of Technology, Linköping University, and IVF

Industrial Research Corporation. The authors express

their gratitude to the Swedish Agency for Innovation

Systems (VINNOVA) for generously sponsoring the

project. We also want to thank the operators that

participated in the field studies, as well as Professor

Erik Hollnagel and associate Professor Johan Stahre

for their valuable contributions.

REFERENCES

Billings, C. E. (1997). Aviation Automation The

Search for a Human-Centered Approach,

Mahwah, New Jersey.

Chapanis, A. (1996). Human Factors in Systems

Engineering, New York.

Daouk, M. (2000). Air Traffic Control

Task Allocation and Decision Processes.

Fitts, R. M. (1951). Human Engineering for an

Effective Air-navigation and Traffic-control

system.

Frohm, J., Karsvall, A., Hassnert, M., Stahre, J. and

Hollnagel, E. (2002). User-driven decision

automation within industrial production systems.

In: Humans in a Complex Environment

Proceedings of the 34th Annual Congress of the

Nordic Ergonomic Society

(Ed)).Vildmarkshotellet, Kolmården, Sweden.

Goom, M. K. (1993). Industrial view of

MANPRINT. In: Proceedings of the 1993 Annual

Conference of the Ergonomics Society, Apr 13-

16 1993 (Ed)).Edinburgh, Scotl.

Gramopadhye, A. K., Drury, C., Sharit, J. and Sudit,

M. (1992). A framework for function allocation

in inspection. In: Proceedings of the 3rd

International Conference on Human Aspects of

Advanced Manufacturing and Hybrid Automation

(Ed)).Gelsenkirchen.

Greenstein, J. S. and Lam, S. T. (1985). An

Experimental Study of Dialogue-Based

Communication for Dynamic Human-Computer

Task Allocation. International Journal of Man-

Machine Studies, 23, 605-621.

Greenstein, J. S. and Revesman, M. E. (1986).

Human-Computer Communication for Dynamic

Task Allocation. In: Human-Computer Dialogue

Design (Williges, R. W. E. a. R. C. (Ed)),

Elsevier, Amsterdam.

Hollnagel, E. (2003). Cognition as control: A

pragmatic approach to the modelling of joint

Cognitive systems. (in print).

IBM Autonomic Computing, (2001). Autonomic

Computing: IBM’s Perspective on the State of

Information Technology. [Online] Available:

http://www.research.ibm.com/autonomic/manifes

to/ [2003, June 25]. In English.

Jordan, N. (1963). Allocation of functions between

man and machines in automated systems. Journal

of Applied Psychology, 47, 161-165.

Karsvall, A., Frohm, J., Hassnert, M. and Stahre, J.

(2003). On Decision-Making: Making Sense of

System Performance. In: In Proceedings of EAM,

the 22th European Conference on Human

Decision Making and Manual Control

(Ed)).University of Linköping.

Meister, D. (1987). Systemsdesign, development and

testing. In: Handbook of Human Factors

(Salvendy, G. (Ed)), John Wiley & Sons, New

York.

Mouloua, M., Parasuraman, R. and Molloy, R.

(1993). Monitoring automation failures: effects of

single and multi-adaptive function allocation. In:

Proceedings of the Human Factors and

Ergonomics Society 37th Annual Meeting

Pacaux-Lemoine, M. P. and Debernard, S. (2002).

Common work space for human-machine

cooperation in air traffic control. Control

Engineering Practice, 10, 571-576.

Parasuraman, R., Sheridan, T. B. and Wickens, C. D.

(2000). A model for types and levels of human

interaction with automation. IEEE transactions on

system, man, and cybernetics - Part A: Systems

and humans, 30, 286-296.

Rouse, W. B. (1988). The Human Role in Advanced

Manufacturing Systems. In: Design and Analysis

of Integrated Manufacturing Systems (Compton,

W. D. (Ed)), National Academy Press,

Washington D.C.

Sheridan, T. B. (1992). Telerobotics, Automation,

and Human Supervisory Control, Cambridge,

MA.

Sheridan, T. B. (1997). Task allocation, task analysis

and supervisory control. In: Handbook of human-

computer interaction (Helander, M., Landauer, T.

K. and Prabhu, P. (Ed)), 87-105. Elsevier Science

B.V., Amsterdam.

Sheridan, T. B. (2002). Humans and Automation:

System Design and Research Issues, Santa

Monica.

PAPER II

Frohm, J., Lindström, V. and Bellgran, M. (2005). A model for parallel levels of automation within manufacturing. In Proceedings of the 18th International Conference on Production Research, Salerno, Italy, July 31 – August 4, 2005.

18th International Conference on Production Research

A MODEL FOR PARALLEL LEVELS OF AUTOMATION WITHIN MANUFACTURING

J. Frohm1, V. Lindström2 and M. Bellgran1,3

1 Division of Production Systems, Chalmers University of Technology, SE- 412 96 Gothenburg, Sweden

2 Department of Industrial Engineering and Management, School of Engineering, Jönköping University, SE-551 11 Jönköping, Sweden

3 Department of Engineering, Physics and Mathematics, Mid Sweden University, SE- 891 18 Örnsköldsvik, Sweden

Abstract A central question for the design of automation has not only been how to design the best possible technical system, but also how to optimize the human-machine interaction. Since manufacturing involves support and replacement of both cognitive as physical working tasks, the view on level of automation (LoA) has to be broadening within the area of manufacturing. This paper focuses on the idea that the process of manufacturing can be separated into two basic categories of functions, I.e. tasks that deal with mechanization (replacement of human muscle power), and computerization (replacement of human sensory and thought processes) in order to control the manufacturing process. The level of automation should, therefore, be separated into two basic scales. By variation within the two scales it may be possible to achieve ”the right automation for the right manufacturing situation”.

Keywords:Levels of Automation, Manufacturing, Flexibility, Task Allocation, Supervisory Control.

1 INTRODUCTION

A clear and present concern for many manufacturing companies is how to compete with low-cost countries. One of the answers has been that manufacturing has to improve its usage of technological resources, i.e. automation. The standard interpretation of “improved usage” has generally been to increase the extent and level of automation (LoA). However, an increased level of automation does not necessary result in increased benefits [1]. In fact, an excessive level of automation may result in poor operator performance caused by too low workload (e.g. loss of vigilance), loss of specific working skills, and loss of situation awareness [2-4]. Modern manufacturing system [5] includes a wide range of activities, spanning from employment of workers and investments in automation, to the processes needed for manufacturing of costumer-ready products. The manufacturing system itself includes both human and technological resources, as well as, procedures, software and facilities. All dependent on each other in a complex combination [6, 7]. Quite in contrast to the early 20th

century ideas of Ford and Taylor [8], modern human operators can rarely be regarded as expendable or replaceable components, but rather as an enhancement of system performance [9]. Consequently, both advanced technical systems and skilled human workers are necessary for successful manufacturing. Therefore, it appears to be a large potential in the area of manufacturing not only to find appropriate levels of automation for each operation, but also to be able to vary LoA for critical operations in order to achieve the right level of automation for the right manufacturing situation as a way of increasing the system robustness.

1.1 Background

It has often been shown that the human element of any production system is by far the most important part in any manufacturing process to achieve a flexible but still robust

manufacturing [10]. However, the human abilities within manufacturing are still not utilized to its full potential.The task of designing and operating manufacturing systems has been approached from many different views. One of those approaches is the technical that mainly focused on the technical parts of the system; often implying a rather high level of automation in physical tasks (mechanization), and possible also within the cognitive tasks (computerization). A central question for the human approach has therefore not only been how to construct the best possible technical system, but how to optimize the human-machine interaction. Sometimes, this approach has implied an emphasis on the human aspects rather than on the integrational aspects. In order to achieve the most optimal manufacturing system in terms of flexibility and robustness, the integrational aspects of human and technology has to be considered. A manufacturing system can be manual, fully automated or most commonly: a combination implying both manual and automated parts, i.e. a hybrid system. Hence, the level of automation can vary on a scale related to the technical system and how the operations are physically and cognitive performed. However, most research on level of automation has been made within the domain of human factors and thereby mainly focused on task allocation. The view on LoA has therefore to be broadening within the area of manufacturing, since manufacturing involves support and/or replacement of both cognitive and physical working tasks.

1.2 Method

From a review of LoA applied on manufacturing which primarily included journal articles, books, conference papers, and internet web pages, it can be concluded that most of the present models of LoA from a research perspective has been focused from a human-centred approach. However, since research within manufacturing often has considered automation in its extremes; fully automated or total manual, we can find an interest in viewing the manufacturing system as integration between

the human and the technical system in terms of level of automation.

1.3 Level of automation applied on manufacturing

The research project called DYNAMO (see chapter 5) presents the scope of varying the level of automation as a way of achieving robust manufacturing systems. The goal of the DYNAMO project is to provide industry with design, measurement, visualization, and management tools fordynamic levels of automation in manufacturing.

In this paper have we choose to define LoA as:

“The relation between human and technology in terms of task and function allocation, which can be expressed as an index between 1 (total manual work) and 10 (total automation) of physical ascognitive tasks.”

By varying LoA in a controllable way it may be possible to cope with both planned and unplanned variations. A robust manufacturing system is depending on the flexibility to beable to handle variations in the process (e.g. introduction of new product, tool-change, product disturbance, etc.).Automation strategies have been applied in order to increase system robustness in areas where risks andinvestment costs are extremely high, such as in aviation, aerospace, military and process industries. General strategies for function or task allocation betweenautomation/technology and people (i.e. LoA) have beensuggested by e.g. Rouse [11] who described three (frequently employed) strategies for allocation:

� Comparative allocation

� Leftover allocation

� Economic allocation

Function or task allocation of man/machine shows that in highly automated systems, manual tasks are mainlyembedded in human supervisory control tasks andmaintenance operations. More recent strategies for designing those semi-automated systems emphasize sharing and trading of control, but also scenario-based, dynamic, and adaptive approaches [12, 13].

2 LOA IN MANUFACTURING

From the manufacturing perspective it can be seen that automation has been of huge importance for the development of the modern society. Automation has been one of the most important parts in manufacturing due to the need of quick information turnaround to be able to hit process targets with high quality, and to sustain the ever-increasing speed of process technology conversions. Automation can also speed up information turns in many ways such as supporting execution and overall work flow control in the plant, automatic control of tools and material handling and to give decision support to help the operators monitor plant performance. However, as mentioned before, an increased level of automation may not necessarily result in a more efficient manufacturing, since a robust manufacturing system is depending on the flexibility to be able to handle variations in the process.Therefore, it is important to understand how to obtain a balanced manufacturing system, meaning a system that has the proper mix of operators and machines to obtain the highest profit possible, and without suffering any losses of product quality.

The basic definition of automation relates to the allocationof tasks and/or functions between the human and her

technological support, i.e. tools, machines, computers.Even if automation fundamentals emerge during the 19th

century and excelled during the 20th, the modern manufacturing still relates to the allocation of physicaltasks but increasingly addresses also control andmonitoring tasks. It is thereby relevant for modern manufacturing systems to combine information andmechanical technology. According to Langmann [14], it has not been until now any clear defined subject content of automation, and it is therefore depending on the different kind of operating range of the automation. Langmann sorts automation into two main fields (table 1); core technologiesand integration/support technologies. The core technologies are basic technologies, which vary dependingon the operation where they are used. The support technologies have a purpose of integration with the outer world:

Table 1: Main fields of automation [14]

Co

re

tech

no

log

ies

- Sensor and acting technology

- Control

- Control technology

- RoboticsIn

teg

rati

on

an

d

su

pp

ort

te

ch

no

log

ies

- Computer technology

- Information technology

- Communication technology

- Man-machine systems

- Structure- and systems technology

- Management technology

A similar division of automation into mechanization andcomputerization were made by Hubka and Eder [15], theydefined:� Mechanization as “the progressively transferring

working functions from humans to technical systems”,� Automation as “the progressively transferring

regulating and controlling functions from humans to technical systems”, and

� Computerization as “the transferring routine decision-making from humans to technical systems, usuallyadding some functions of monitoring systemperformance, and progressing towards expertsystems and artificial intelligence”.

With reference to LoA have we have chosen to defineautomation as the concept that ties mechanization and computerization together (figure 1In the same way as Hubka and Eder we here seemechanization as the transfer of physical tasks betweenhuman and the technical system. Computerization is also defined like Hubka and Eder do, i.e. as the transfer of cognitive tasks between human and the technical system.

Figure 1: Connection between automation, computerizationand mechanization

Automation

Computerization Mechanization


2.1 LoA models of computerization

In order to design a manufacturing system that has theright level of automation in relation to the actual manufacturing situation the system designer should, outgoing from the process flow, consider what operationsthat should performed by technology and what operations that needs to be handled by humans or a combination ofboth. Some operations may be hazardous or fatigue etc for being performed by an operator, and other operations maynot imply a realistic technical solution due to complexityetc. The remaining tasks that both human and technologyare able to handle should be designed in such a way thatthe human can get maximum support by the technicalsystem.

A taxonomy that applies to most human-machine systemsis the sequence of operations showed in figure 2.

Figure 2: Five stages of a human-machine task

The figure above folds together several relevant dimensions according to [16], such as:

� The degree of specificity required by the human for inputting requests to the machine,

� The degree of specificity with which the machine communicates decision alternatives or action recommendations to the human;

� The degree to which the human has responsibility forinitiating implementation of action, and

� The timing and detail of feedback to the human aftermachine action is taken.

It is evident that these many dimensions of human-machine tasks, presented above, can be ordered in variousways on a multidimensional scale as well as one -dimensional scales. The main question according toSheridan [16], is to consider all the options that are more or less ordered and then to decide the degree of automation at each stage.

2.2 Task allocation for LoA

The model in figure 2 implies that different tasks can beordered in various ways, either multidimensional or onone-dimensional scales. However, nothing indicates who

or what should perform the given task. This is not strictlycorrect according to Chapanis [6], because in many cases,such decisions are almost literally forced on the designers of the manufacturing system, by the state-of-the-art capabilities or limitations, cost, or other considerations.To choose or determine an appropriate level of automation largely deals with the question of which tasks should beallocated to the human and which should be handled bythe technology. To give some guidance as to how and what to automate or not, Fitts [17] introduced his rather well-known list where general functions best performed bymachine, or best be left to a human are presented. As mentions in Fitts list, allocation between human andtechnology (e.g. the manufacturing system) can be decided by different reasons, such as:

� Those tasks of information and control that can be“automated” by computers or other “control” devices.

� Those tasks or processes that can be automated or rather mechanized by the “mission fulfilment” equipment.

� Those functions carried out by humans, whether of the control or mission fulfilment class.

Many of the common automation design models acknowledge that the ideal task allocation is only plausiblein theory, since studies of automated systems tend to focus on stable task allocation between human andmachine. It means that we often regard the human and the technical system as two static and individual parts to be able to allocate the tasks between [16, 18, 19]. There are however several difficulties with such a direct allocation method, since it usually address performance differences in short-term or under laboratory settings, with noconsideration for long-term consequences such as whether the manufacturing system is capable of successively perform a given task or not. A number of unknown factors must also be accounted for; such ashidden assumptions, unanticipated situations, and non-independence of tasks [13] since they are a part of the overall context for the manufacturing system.Task allocation models that standardise the relationshipbetween human and machine can, in other words, be criticised for manifesting an artificial distance between theoperator and his/her equipment, as they avoid the complex interaction between system components [20].

3 PARALLEL SCALES FOR MANUFACTURING LOA

Manufacturing systems designed for flexible automation provide means for successful manufacturing of a widevariety of high-quality products primarily due to precisechangeover between products with minimal start-up and running-in periods. These expectations fuelled the rapiddevelopment of Flexible Manufacturing Systems (FMS)and Computer Integrated Manufacturing (CIM) systems,starting in the early 1960’ [5]. The inherent complexitycreated by integrating several sub-systems has since then been successfully addressed by massive use ofinformation technology and control engineering. Althoughvery efficient and productive, automated manufacturing and assembly systems also have drawbacks. They have to deal with tasks that could not be automated, and therebyare they vulnerable to unforeseen situations, wereautomated systems having no predefined solutions. Since each change in a complex technical systemintroduces risks for errors, and the higher the complexitygets, the greater the risk of causing disturbances in the system will be, as well as manufacturing erroneous products. Complex and partly automated systems puts therefore high demands on people to be active parts of the system, dealing with disturbances, unforeseen situations,

Acquire information

Analyze and Display

Decide action

Implement action

Feedback and Inform

non-automated tasks, and maintenance of subsystems.Allocation of tasks between human and technical subsystems in an advanced manufacturing system is therefore a non-trivial problem.

3.1 Automation scales for physical and cognitivetasks

Even if the manufacturing process seems to only involve automation of mechanical or electrical tasks, most of those tasks are controlled by computers for optimalperformance. The process of manufacturing can therefore be separated into two basic classes of automation, mechanization and computerizing (see figure 3). Mechanization is defined as the replacement of human muscle power, such as material and energy transformation. Computerization is in a similar way connected to cognitive tasks, such as replacement of human sensory and thought processes, e.g. collection, storage, analysis and use of information in order to control the manufacturing process.

Figure 3: Separation of functions into mechanization andcomputerization?

Even if most tasks now seem to involve a mix of both technologies (mechanization and computerization), should each task or operation be broken down into mechanization and computerized activities. For example, to operate a cutting machine involves both controlling the cutting tool (computerized activities) as handling the work pieces(material handling, mechanization). By breaking down the activities into mechanized and computerized activities, but still acknowledge co-operation between human and technology, can it be useful to see the allocation as two parallel scales of LoA (figure 4).

Figure 4 outlines one model of making those necessary choices between human implementation and automation.The figure shows that the allocation between the human and the technological components (computers, machines, etc.), should not only be a question of activities or “whodoes what” but rather a question of how the activities can be shared between the human and technologicalcomponents.

Figure 4: Determine the place of the human within the manufacturing system

To make the manufacturing system as robust and flexible as possible, there is a need to assign the appropriate tasks and functions to the right element of the system. Asmentioned earlier and also showed in figure 4, the process of manufacturing can be separated into two basic classes of activities; 1) Information and control and 2) physical activities. Each of those two activities can then be ordered in ten steps, from total manual control (LoA 1) to fully automated control (LoA 10) according to the original LoA scale by Sheridan [21]. However as noted by Price [22], some tasks are not possible to allocate to the human or to the technological system. As seen in figure 5, there is a maximum extent of the human capacity as well as the technological capability, illustrated by the line ofAutomatibility and Humanizability.

Figure 5: Introduction to the extent of human and technical capability

The Automatibility Line shows the absolute extent of pure technologies in their capability to actually automate the tasks and functions. The line is limited by the fact that many tasks and functions require human automation, andas Bainbridge [1] pointed out; “The more advanced a control system is, the more crucial the contribution of the human operator may be”. Thereby can it not be possible to automate all tasks with the available technology. For example, some assembly tasks can be so complex that is impossible for technology to handle it. In the same waycan some tasks be too difficult for the human to attempt.E.g. tasks that demands consistent performance over long periods of time. This is showed by the Humanizability Line

Performed by the technical system

Performed by the human

LoA 10 LoA 10

LoA 1 LoA 1

Physicalactivities

Maximum extent of

human capability

Maximum extent of

technical capability

Humanizability Line

AutomatibilityLine

Information

andcontrol

ctivities

Performed by the technical system

a

Performed by the human

LoA 10

LoA 1

Computerized

tasks

Cognitive

tasks

Mechanized

tasks

Manual

tasks

Information and control activities

Physicalactivities

Tasks that cannot be conducted by human

Tasks that cannot be conducted by technology

LoA 1

LoA 10


in figure 5. This line is limited by human cognitive and physical abilities, such as speed of response, breadth of comprehension, range of vision, physical strength, and so forth. The location of the Line of maximum extent of technical capability is also influenced by economic factors, political aspects, social factors, as well as technological factors.Another aspect of figure 4 and 5 is to standardize the relationship between human and machine by introducing more technology to solve some tasks, leaving the human to deal with the remaining parts. By manifesting such artificial distance between the operator and his/her equipment thereby obstructs the human, from understand the manufacturing situation that he or she is put to interact with.

4 CONCLUSION

Even if many manufacturing processes seems to only involve automation of mechanical or electrical tasks, most of those tasks are controlled or supervised by computers for optimal performance over time. It is, therefore, important to recognize that automation in manufacturing, as well as in other domains, can be seen as two types of technologies; mechanized, which are the basic core technologies, such as drilling, grinding, etc. The other type of technology is the computerized that deals with the control and support of the mechanized technologies. As presented here, it is important to recognize that automation is not only the process of reconfiguring the working part into a product. It is also the part the earlier theories within LoA has put forward, the replacement of human cognitive processes, i.e. the control of the physical activities such as mechanization. Most research on LoA has been focused on the cognitive parts, where automation help to speed up information flow, and give decision support in order to help the operator monitor the situation. It is, however; also important to recognize that from a manufacturing perspective is it the automation of the mechanized tasks that often have been focused.To be able to make the manufacturing system as robust, flexible and adaptable as possible, the system has to be able to handle variations in the process such as introduction of new product, tool-change, product disturbance, etc. Therefore, it is important to understand how to obtain a balanced manufacturing system that has the proper mix of operators and machines in order to obtain the highest profit possible without suffering any losses of product quality. One way to achieve this balanced manufacturing system is by separating the system into the two basic classes of activities, i.e. Information and control and physical activities. Each of those two activities can then be categorized in ten steps, from total manual control (LoA 1) to fully automated control (LoA 10) according to the original LoA scale by Sheridan. By measuring LoA with Value Stream Mapping (VSM), using symbolic figures and collecting data, it would be possible to draw a road map to measure levels of automation in manufacturing. The advantage of using VSM is that it provides mapping of both information and product flows, were the product flow follows the mechanization within manufacturing

5 ACKNOWLEDGMENTS

The work presented is partial results from the project DYNAMO – Dynamic Levels of Automation for Robust Manufacturing System, which is carried out as a co-operation between Chalmers University of Technology, Jönköping School of Engineering, and IVF Industrial Research Corporation. The authors express their gratitude to the Swedish Foundation for Strategic Research (ProViking) for generously sponsoring the project.

6 REFERENCES

[1] L. Bainbridge, "Ironies of Automation," in Analysis, Design, and Evaluation of Man-Machine Systems, G. Johannsen and J. E. Rijnsdorp, Eds. Düsseldorf: International Federation of Automation Control, 1982, pp. 151-157.

[2] M. R. Endsley, "Level of automation: Integrating humans and automated systems," presented at Proceedings of the 1997 41st Annual Meeting of the Human Factors and Ergonomics Society. Part 1 (of 2), Albuquerque, NM, USA, 1997.

[3] M. R. Endsley and E. O. Kiris, "Out-of-the-loop performance problem and level of control in automation," Human Factors, vol. 37, pp. 381-394, 1995.

[4] R. Parasuraman, T. B. Sheridan, and C. D. Wickens, "A model for types and levels of human interaction with automation," IEEE Transactions on System, Man, and Cybernetics - Part A: Systems and humans, vol. 30, pp. 286-296, 2000.

[5] M. E. Merchant, "The Manufacturing System Concept in Production Engineering Research," in Annals of CIRP, vol. 10, 1961, pp. 77-83.

[6] A. Chapanis, Human Factors in Systems Engineering. New York: John Wiley & Sons, Inc., 1996.

[7] G. Chryssolouris, Manufacturing systems: theory and practice, (2., corr. printing) ed. New York; Berlin; Heidelberg; London; Paris; Tokyo; Hong Kong; Barcelona; Budapest: Springer, 1993.

[8] F. W. Taylor, The principles of scientific management. New York: Harper & Brother, 1911.

[9] C. D. Wickens, J. D. Lee, Y. Liu, and S. E. Gordon Becker, An introduction to human factors engineering, 2. ed. ed. Upper Saddle River, N.J.: Prentice Hall International, 2004.

[10] C. D. Wickens, J. D. Lee, Y. Liu, and S. E. Gordon Becker, An Introduction to Human Factors Engineering, 2. ed. ed. Upper Saddle River, N.J.: Prentice Hall International, 2003.

[11] W. B. Rouse, "The Human Role in Advanced Manufacturing Systems," in Design and Analysis of Integrated Manufacturing Systems, W. D. Compton, Ed. Washington D.C: National Academy Press, 1988.

[12] T. Inagaki, "Adaptive Automation: Sharing and Trading of Control," in HANDBOOK OF COGNITIVE TASK DESIGN, E. Hollnagel, Ed. Mahwah, New Jersey: Lawrence Erlbaum Associates, 2003, pp. 147-169.

[13] T. B. Sheridan, Telerobotics, Automation, and Human Supervisory Control. Cambridge, MA: MIT Press, 1992.

[14] R. Langmann, Taschenbuch der Automatisierung: mit 110 Tab. München; Wien: Fachbuchverl. Leipzig im Carl-Hanser-Verl., 2004.

[15] V. Hubka and W. E. Eder, Theory of Technical Systems: A Total Concept of Technical Systems. Berlin: Springer-Verlag, 1988.

[16] T. B. Sheridan, Humans and Automation: System Design and Research Issues. Santa Monica: John Wiley & Sons, Inc., 2002.

[17] R. M. Fitts, "Human Engineering for an Effective Air-navigation and Traffic-control system," National Reseach Council, London 1951.

[18] C. E. Billings, Aviation Automation: The Search for a Human-Centered Approach. Mahwah, NJ: Lawrence Erlbaum Associates, 1997.

[19] T. B. Sheridan, "Task allocation, task analysis and supervisory control," in Handbook of human-computer interaction, M. Helander, T. K. Landauer, and P. Prabhu, Eds., 2nd ed. Amsterdam: Elsevier Science B.V., 1997, pp. 87-105.

[20] J. Frohm, A. Karsvall, M. Hassnert, J. Stahre, and E. Hollnagel, "System perspective on task allocation and automation design," presented at 8th IFAC symposium on automated systems based on human skills and knowledge, Göteborg, Sweden, 2003.

[21] T. B. Sheridan, "Computer Control and Human Alienation," Technology Review, vol. 83, pp. 60-73, 1980.

[22] H. E. Price, "The Allocation of Functions in Systems," Human Factors, vol. 27, pp. 33-45, 1985.

PAPER III

Frohm, J., Granell, V., Winroth, M. and Stahre, J. (2006). The industry’s view on automation in manufacturing. In Proceedings of the 9th symposium IFAC on Automated Systems Based on Human Skills and Knowledge, Nancy, France, May 22-24, 2006.

THE INDUSTRY’S VIEW ON AUTOMATION IN MANUFACTURING

Frohm, J.1 Granell, V.2 Winroth, M.2 and Stahre, J.1

1 Division of Production Systems, Chalmers University of Technology, SE- 412 96 Gothenburg, Sweden

2 Department of Industrial Engineering and Management, School of Engineering, Jönköping University, SE-551 11 Jönköping, Sweden


Abstract: Many manufacturing companies in Europe are presently focusing on automation as a weapon for competition on a global market. This paper focuses on industry’s view of automation. The paper presents data on benefits and disadvantages of automation, based on one pilot study and one Delphi study in two rounds. Copyright � 2006 IFAC

Keywords: Automation, Production

1. INTRODUCTION

Highly automated product realization is an important means for industry to meet competition from low-cost countries, primarily due to relatively high wage costs observed in e.g. the US and Europe. In a time of rapidly changing technologies and shortening product life cycles, many companies are focusing on automation as a means for competing on a more demanding market. From a scientific perspective the view of automation and the humans roll in the system, has been described by e.g. Fitts (1951), Bainbridge (1982) and Sheridan (1992)

Many authors e.g. Merchant (2000) and Saridis (2000) has concluded that the benefits of automation, among others, are that it may led to increases in productivity and decreases in costs. Further, Wei et al. (1998) concluded that automation improves production efficiency and enhances operational safety.

However, an increased usage of automation doesn’t necessary result in increased benefits (Bainbridge 1982). According to Bellgran (1998) the most common reason for running-in problems, are that new product variants increase demands beyond what was planned. Carlsson and Jacobsson

(1995) noted that users must have competence to formulate technical needs. Bellgran and Säfsten (2005) also noted that investment- and development costs must be put in relation to demands for flexibility and shorter change-over times.

There is also a tendency that we focus the automation solutions on the easiest tasks and leave the rest to the operator (Sheridan 1992). This focus may lead to a hodgepodge of partial automation, which will led to a less coherent, meaningful and more complex that they need (Sheridan 1992). Merchant (2000) also noted that technology will only perform at its full potential if the utilization of the system’s human resources are fully engineered. There is little question that the human is seen as an essential element in the system (Sheridan 1995). Inagaki (1993) noted that the human operator is often kept within the system, to handle situations appropriately even if unanticipated events happen.

The purpose of this paper is to present a preliminary review of industry’s view of automation in manufacturing. In earlier reviews of literature on “level of automation” and “automation strategies” (Frohm et al. 2005; Lindström et al. 2005; Säfsten et al. 2005), the following questions have been stated:

o What is the foremost benefit of automation?

o What types of problems are linked to automation?

The question is how the previous scientific work can be put into consideration from a manufacturing perspective.

1.1. Levels of automation

In 1961, a pioneer in manufacturing and production, Eugene Merchant, suggested to the production engineering research community that previous focus on automation of processes and standalone automation should be expanded to cover what he called “the manufacturing system” (Merchant 1961). He claimed that such a system should include the whole chain of events starting with the product designer’s activities and ending with the delivery of finished parts. Automation efforts should then be applied to the design, programming, and fabrication as well as the machine and its control system, thus covering what we today call the product realization chain. Chryssolouris (1993) and Chapanis (1996) suggested that an “equipment system” should include both human and technological resources as well as procedures, software, and facilities, all intrinsically linked to form an efficient system.

Consequently, both advanced technical systems and skilled human workers are necessary for successful manufacturing. Therefore, it appears to be a large potential in the area of manufacturing in finding appropriate levels of automation. Thereby, the correct level of automation can be achieved for the right manufacturing situation as a way of increasing the system robustness. In this paper have we chosen to define the Levels of Automation (LoA) as:

“The relation between human and technology, in terms of task and function allocation, which can be expressed as an index between 1 (total manual work) and 9 (total automation) of physical as cognitive tasks.”

2. METHOD

The main research method for this study is based on two types of surveys.

2.1. Preliminary survey

In a preliminary survey, 16 respondents from seven medium-sized to large manufacturing companies were interviewed. All of them had extensive experience from automation and held positions where they could influence automation strategies in their companies. The preliminary survey data was collected through semi-structured interviews that included three main sections. The first part focused

on the participants’ background, experience, and expertise. In particular, the interviewees were asked about the type of manufacturing practices they had most expertise in. Then, he or she was asked to concentrate on these practices for the rest of the interview. The second part of the interview was based upon a questionnaire focusing their view of automation, and on the allocation of tasks between technology and humans.

Topics covered in the second part of the interview were e.g.:

o How much of the manufacturing tasks, such as assembly, machining, material handling, etc. are done manually or automated?

o What is the benefit of using automation?

o When is automation or manual work inappropriate?

In the third and final part of the interview the topics were e.g.:

o Is the level of automation (LoA) of interest and how can LoA be measured?

o Could a flexible or dynamic LoA be of interest and usable?

2.2. Delphi survey

Based on results from the preliminary survey, a second survey was conducted, focusing on a selection of questions from the preliminary survey. The purpose of the survey was to partly capture the view of automation and the perception of terms like “levels of automation” and “automation strategies” among production managers and production technicians working with automation in industry.

Since the respondents needed for the second study could not be obtained for a face-to-face group meeting a Delphi methodology were chosen.

The Delphi technique (Williamsson 2002) is a methodology for systematic attempt to produce a consensus of opinion, as well as identify opinion divergences.

The initial procedure in a Delphi is to prepare, distribute and synthesize a series of problem statements for evaluation. Participants will then receive feedback in the form of their previous responses as well as data describing the responses from the entire group in each preceding round. This enables the participants to reaffirm original options, modify some, and “collectively brainstorm” by adding new items to the list.

The survey tool used for this paper was a web-based questionnaire which included either open

answer questions or statements requiring grading on a four-grade scale (“don’t know” was also an option).

3. RESULTS

3.1. Preliminary survey

The preliminary survey involved 16 respondents from seven medium- to large-sized companies. The results showed that usage of automation differs in activities (figure 1). The interviewed persons where asked “To what extent are different activities automated in your specific production system?” Available answers were; 1) to a very low extent, 2) to a low extent, 3) to a high extent and 4) to a very high extent.

0% 20% 40% 60% 80% 100%

Assembly

Machining

Manufacturing

Material handling

Controlling

Information

Material supply

Supervision

Packaging

Maintenance

Change-over

Very low extent Low extent High extentVery high extent Not relevant

Figure 1. In your production system, to what extent is

activity X automated?

0% 20% 40% 60% 80% 100%

Assembly

Machining

Manufacturing

Material handling

Controlling

Information

Material supply

Supervision

Packaging

Maintenance

Change-over

Very low extent Low extent High extent

Very high extent Not relevant

Figure 2. In your production system, to what extent is activity X performed manually?

Figure 1 confirms that machining and manu-facturing are automated to a larger extent than assembly, which is more difficult to automate due to the complexity of the product and to the

investment cost in more advanced automation. Assembly, packaging, and maintenance are on the other hand tasks that are conducted manually to a large extent (figure 2). If e.g. material handling is automated to a high extent, the manual part is low according to the study. The questions on what is suitable and not suitable to automate are thereby linked to the decisions regarding automation strategies.

In the preliminary survey, the respondents were asked what tasks they considered not suitable for automation, but also which tasks that are not suitable to carry out manually. It was also confirmed here that automation of introduction and ramp-up of new product, production of occasional products, or making of products with short life cycle are not suitable for automation. Further, several companies acknowledge that too many products or variants in production can be a problem when automating. On the other hand, tasks that involve bad ergonomic conditions and great production volumes are not suitable to be conducted manually.

3.2. Delphi survey

The sample for the Delphi survey was derived from respondents, mainly production and industrial managers, from 108 industrial companies in Sweden (figure 3).

Figure 3. Role distribution of respondents in survey

The Delphi survey was based on 62 respondents out of 86, which answered the questionnaire in the first round and 45 in the second and final round. The respondents represented 43 medium- to large-sized companies, primarily from the manufacturing and automotive industry (figure 4).

Industrial management

34%

Production management

38%

Production development

15%

Production technician

5%

Other 8%

Figure 4. Sector distribution of companies in survey

Example questions from the Delphi survey are:

o What is the main benefit of automation?

o What are the most difficult problems when automating?

o How can LoA be defined from an industrial perspective?

Based on the results from the first round of the ongoing Delphi survey and data from the preliminary study, the major benefits of using automation according to the respondents (figure 5), were that automation will give cost savings within production (100%), give possibilities for higher efficiency (98%), increased competitiveness (97%), and productivity (94%). Automation will also give possibilities to improve working environment (93%), by eliminate monotonous and physically demanding work-situations. Many respondents also acknowledged that the main benefit of automation is that it enables production with a minimum of employees (85%).

0%

20%

40%

60%

80%

100%

Incr

ease

d ef

ficie

ncy

Impr

oved

qua

lity

Incr

ease

d co

mpa

rativ

enes

s

Cos

t-cu

ts

Impr

ovem

ent

of p

rodu

ctiv

ity

Pro

duci

ng w

ith a

min

imum

of

empl

oyee

s

Bet

ter

prod

uct

dist

urba

nce

hand

ling

Pos

sibi

litie

s fo

r in

crea

sed

volu

me

capa

city

Impr

oved

wor

king

env

ironm

ent

High Very high

Figure 5. Automation benefits

Results from the first round showed that the main benefits of automation are cost reductions, improvement of efficiency, and enhanced competitiveness and productivity.

The result also indicated that automation did not necessarily lead to better production disturbance handling (PDH). The respondents in the second round were asked whether they anticipated that automation could cause production disturbances (PD), and if so, how automation would affect the production.

Almost all respondents in round 2 acknowledged that automation could lead to production disturbances. The major opinion was that automation includes more complex production systems, which are more difficult to handle.

Many of the respondents also acknowledge having too many products or variants in production (figure 6), which can be a problem when automating (73%). Almost 70% of the respondents also said that they have some difficulties in adapting the products for automated production. 50% of the respondents said that they did not have sufficient time for planning the usage of automation or for training operators on the new investment. It is notable that around 60% of the respondents said that it may be difficult to get payback on investments in automation.

0%

10%

20%

30%

40%

50%

60%

70%

80%

Ada

pt th

e pr

oduc

t to

auto

mat

icm

anuf

actu

ring

Suf

ficie

nt k

now

ledg

e fo

r or

derin

gau

tom

atio

n

Suf

ficie

nt ti

me

for

plan

ning

the

auto

mat

ion

Inve

stm

ents

cos

ts

Uni

on

Kno

wle

dge

and

know

-how

from

the

user

s

Tra

in a

nd e

duca

te th

e us

er o

n th

e ne

wau

tom

atio

n

Diff

icul

ties

to r

amp/

up th

e ne

w p

rodu

ct

To

man

y pr

oduc

ts o

r va

riant

s

High Very high

Figure 6. Automation disadvantages

Since none of the response alternatives in round 1 reached a response rate of over 75%, the results were individually re-connected to each respondent. The respondents were then asked what they thought was the primary problem with automation. It can be concluded from the answers in the second round that almost 30% of the respondent thought that the competence was the primary problem when automating.

Manufacturing industry 51%

Automotive industry31%

Electronic industry8%

Process industry 2%

Other 8%

Many of the respondents also acknowledge that they have too many products or variants in production (11%), and that they are experiencing difficulties in adapting the produced products for automated production (11%).

Based on the Delphi survey, many of the companies also acknowledge (figure 7) that tasks, such as production of occasional products or small batches (72%), or occasional products over a limited time (73%), would not be suitable to automate.

0%

20%

40%

60%

80%

Intr

oduc

tion

of n

ew p

rodu

ct

Pro

duct

ion

of o

ccas

iona

lpr

oduc

ts

Occ

asio

nal p

rodu

cts

with

sho

rtle

ngth

of l

ife

Offi

cial

insp

ectio

n of

fini

shed

prod

ucts

Tas

ks th

at d

eman

ds “

feel

ing”

Tas

ks th

at d

eman

ds h

igh

flexi

bilit

yHigh Very high

Figure 7. What tasks are not suitable to be automated?

Since none of the alternatives in round 1 exceeded 75%, the results were also here individually re-connected to the respondents and the questions were re-asked. There were few changes of the results. Production of occasional products or small batches, and production of occasional products over a limited time still resulted in around 70% answering rate.

The respondents were also asked what they thought was the primary reason for not automating production of occasional products or small batches. The two main answers were that automation of such products would not be cost-efficient and that the change-over time would be too high.

0%

20%

40%

60%

80%

100%

Hea

vy li

fting

or

mon

oton

ous

wor

king

ope

ratio

ns

Haz

ardo

us ta

sks

for

the

hum

an Big

vol

umes

Dem

ands

for

chec

k of

qual

ity

Tas

ks th

at d

eman

ds h

igh

accu

racy

Tas

ks th

at d

eman

ds h

igh

repe

atne

ss

High Very high

Figure 8. What tasks are not suitable to be handled manually?

In the same way as some tasks are not suitable for automation; there are some tasks that are not suitable to be handled manually (figure 8). Both from the preliminary study and the conducted Delphi study it could be concluded that work in hazardous environments (97%) or monotonous and physically demanding working situations (94%) are not suitable to be conducted manually. Also tasks that demand high precision over time (89%) are not suitable to be automated according to the survey.

4. DISCUSSION

The general purpose of automated systems is to perform functions more efficiently, more reliably, and more accurately than human operators. Also, expectations are that automated systems can perform functions and tasks at lower cost than human operators. There are few arguments that can be put against the efficiency, reliability, and accuracy of automated systems. With higher reliability, it could be argued that a system would be a safer system as well.

It could also be argued that keeping the operator out of the control of the system will in one sense protect the human from his or her actions, thereby improving overall safety of the system. Although financial measures have been a main driving force in the development and increase of automation in recent years it could be argued that an automated system is a more economic as well as safer system.

According to results from the studies in relation to earlier research by Merchant (2000) and Wei et al. (1998), the primary driving force for automation is to provide possibilities for increased efficiency and productivity. Automation will also enable cost reductions that, together with increased efficiency and productivity, will lead to increased competitiveness on a more demanding market. Many companies acknowledge that introduction of automation will provide possibilities for improving the work-environment by eliminating monotonous and physical demanding work situations. This is in line with Wei et al. (1998) for example.

Many respondents also acknowledge that a primary problem with automation is too many products or variants in production. Also, that there are difficulties in adapting products for automated production. These findings are in line with Bellgran and Säfsten (2005) who concluded that investment- and development costs must be put in relation to demands for flexibility and shorter change-over times. Almost one third of the respondents also acknowledged in round 2 that the knowledge of the users (e.g. operators) was one of the main difficulties when automating.

5. CONCLUSION Our studies and earlier scientific notation suggest that the foremost benefit of automation, are to provide possibilities for increased competitiveness by enabling cost reductions as well as increased efficiency and productivity.

A number of respondents acknowledge that benefits and problems related to automation are closely linked to the balance between, on one side, investment and development costs and, on the other side, demands on flexibility and shorten change-over time.

The results in relation to previous scientific work suggest that some of the main problems with automation occur when new product variants increase demands on the production system, both in terms of the user knowledge and know-how.

6. ACKNOWLEDGEMENTS

The work presented interim results from the project DYNAMO – Dynamic Levels of Automation for Robust Manufacturing System, which is carried out as collaboration between Chalmers University of Technology, Jönköping School of Engineering, and IVF Industrial Research Corporation. The authors express their gratitude to the Swedish Foundation for Strategic Research (ProViking) for generously sponsoring the project.

REFERENCES�

Bainbridge, L. (1982). Ironies of Automation. Automatica, 19, 775-779.

Bellgran, M. and Säfsten, K. (2005). Produktionsutveckling - Utveckling och drift av produktionssystem. Studentlitteratur, Lund, Sweden.

Carlsson, B. and Jacobsson, S. (1995). What makes the automation industry strategic? In: Technological Systems and Economic Performance: The Case of Factory Automation (Carlsson, Bo. (Ed)), Kluwer Academic Publishers, Cleveland, Ohio.

Chapanis, A. (1996). Human Factors in Systems Engineering. John Wiley & Sons, Inc., New York.

Chryssolouris, G. (1993). Manufacturing systems: theory and practice. Springer, New York; Berlin; Heidelberg; London; Paris; Tokyo; Hong Kong; Barcelona; Budapest.

Frohm, J., Lindström, V. and Bellgran, M. (2005). A model for parallel levels of automation within manufacturing. Pasquino, Raimondo (Eds). In proc. of the 18th Int. Conf. on Production Research. Fisciano, Italy.

Inagaki, T. (1993). Situation-adaptive degree of automation for system safety. (Eds). In proc. of 2nd IEEE Int. Workshop on Robot and Human Communication. Tokyo, Japan.

Lindström, V., Winroth, M. and Frohm, J. (2005). Choosing levels of automation in production systems - Finding critical and supportive factors. Demeter, K. (Eds). In proc. of the Int. Euroma Conf. Budapest, Hungary.

Merchant, M. E. (1961). The Manufacturing System Concept in Production Engineering Research. In: Annals of CIRP (Ed)), 77-83.

Merchant, M. E. (2000). The Future of Manufacturing. In: Handbook of Industrial Automation (Shell, Richard L. and Hall, Ernest L. (Ed)), Marcel Dekker, Inc., Cincinnati.

Saridis, G. N. (2000). Intelligent Manufacturing in Industrial Automation. In: Handbook of Industrial Automation (Shell, Richard L. and Hall, Ernest L. (Ed)), Marcel Dekker, Inc., Cincinnati.

Sheridan, T. B. (1992). Telerobotics, Automation, and Human Supervisory Control. MIT Press, Cambridge, MA.

Sheridan, T. B. (1995). Human centered automation: oxymoron or common sense? (Eds).In proc. of IEEE Int. Conf. on Systems, Man and Cybernetics. Vancouver, Canada.

Säfsten, K., Winroth, M. and Stahre, J. (2005). The content and process of automation strategies. Pasquino, Raimondo (Eds). In proc. of the 18th Int. Conf. on Production Research. Fisciano, Italy.

Wei, Z.-G., Macwan, A. P. and Wieringa, P. A. (1998). A Quantitative Measure for Degree of Automation and Its Relation to System Performance and Mental Load. Human Factors, 40, 277-295.

Williamsson, K. (2002). The Delphi method, Research methods for students and professionals. Wagga wagga, NSW.

PAPER IV

Frohm, J., Karsvall, A., Thunberg, A. and Stahre, J. (submitted). Possibilities for Automated Decision-making in Industrial Systems. (Submitted for publication in International Journal of Industrial Ergonomics).

Possibilities for Automated Decision-making in Industrial Systems

J. Frohm1,*, A. Karsvall2, A. Thunberg1 and J. Stahre1

1 Department of Product and Production Development, Chalmers University of Technology, Göteborg, Sweden 2 Department of Computer and Information Science, Linköping University, Linköping, Sweden

Abstract The human role in advanced production systems has historically been debated. However, axiomatic

belief in automation as the primary means to achieve productivity has gradually been replaced by joint cognitive system approaches. This paper elaborates on the possibility of automating decision-making in production systems. Theory and results from three industrial case studies suggest that automating operative decision-making may be questionable, while tactic and strategic decisions could be supported by automated decision support systems. Relevance to industry

As manufacturing environments increasingly rely on computerized and automated systems for control and human operator support, it is necessary to understand how human performance is supported, or replaced by sophisticated information and control technology. This paper elaborates on whether it is possible to automate decision-making, in the light of modern industrial requirements for stable and robust, yet flexible production. © 2007 Elsevier B.V. All rights reserved.

Keywords: Automation; Task allocation; Decision-making; Production; User-driven Introduction

* Corresponding author. Department of Product and Production Development, Division of Production System, Chalmers University

of Technology, 412 96 Göteborg, Sweden

E-mail: [email protected] (J. Frohm)

1 Introduction

As a general concept, production covers all activities from the employment of workers or investment in machinery, to the process of production and delivery of consumer-ready products (Merchant, 1961). Thus, the production system can be described as a socio-technical system that consists of both human and technological resources (Eijnatten, et al., 1993), which depend on each other. However, since humans often are seen as an unreliable system component (Riley, 1996), one of the main concerns in engineering has been to increase overall system performance by use of automation (Parasuraman, et al., 2000). In some cases, the ambition was to create so-called lights-out factories with complete automation in every production unit (Mital, 1997).

The automation movement peaked in the 1980s throughout early 1990s, a wave of automation efforts swept through several sections of industry under names such as “computer-integrated manufacturing”,”intelligent manufacturing systems” and “artificial intelligence”. Consequences were that the human workers were regarded as more or less expendable or replaceable components (Lennerlöf, 1993).

However, while automation-centred approaches for information and control automation has contributed to both innovations and overall increased industrial productivity, it has also been very problematic (Satchell, 1998). Therefore, during the last two decades, scientific management of production systems has been challenged by an increasing insight on the importance of the human contribution to system performance (Satchell, 1998), especially in order to achieve maximum flexibility of operations. This does not mean that industry has stopped treating computerisation as a primary way to enhance production. Applications for technology are as noted by e.g. Parasuraman and Riley (1997) and Sheridan (2002), rather seen in a wider context, where human performance is supported, or at least substituted, by sophisticated information and control technology. One of its central applications is to automate or support decision-making, since inadequate decision-making or problem solving may have damaging (even fatal) effects on both productivity and workers health. Hence, as noted by Beach and Lipshitz (1993) intelligent decision-making applications also promote efficiency and security in production by reducing human deficiencies.

The question is, how can such applications be implemented without repeating historical mistakes of over-automation?

The purpose of this paper is to elaborate on whether it is possible to automate decision-making, in the light of modern industrial requirements for stable and robust, yet flexible production.

Initially, the paper focuses on previous findings in the field of decision-making in real-time practice and Task Allocation (TA). Then the paper will relate previous findings to three empirical examples: One from the petrochemical industry, one within the area of Advanced Manufacturing Systems (AMS), and one from the area of nuclear power, to discuss the possibilities of automating decision-making.

2 Decision-making and Task Allocation within Production

A system may be defined in many different ways. From a classical definition of an equipment system, Chapanis (1996) defines a system as ”an interacting combination, that at any level of complexity, of people, materials, tools, machines, software, facilities, and procedures are designed to work together for some common purpose.” The purpose of the system is thus relative to the system in the general definition, but classical ergonomics and human factors usually begin with a clear distinction between the human operator and the machinery in use (Karwowski, et al., 1997). This seemingly natural decomposition means that the interaction between the two, and their isolated functionality, becomes the focus of attention, and that the overall perspective is easily neglected. The discipline of Cognitive Systems

Engineering (CSE) overcomes that by shifting the focus from human and machine as two separate units to the Joint Cognitive System (JCS), which view the system as a single unit (Hollnagel, et al., 1988). It follows from this that what should be studied is neither the internal functions of either human or machine nor the interaction between them according to Hollnagel and Woods (2005), but rather the external functions of the JCS as based on human-machine cogency.

2.1 A Joint Cognitive System

Within CSE, the system boundary can be defined by considering two ways of describing sets of objects (i.e. system components) or functions (Hollnagel and Woods, 2005). Firstly (1), whether the objects or factors may or may not be affected by the artefact; secondly (2), whether the objects or factors that are changed by artefact use or not (Hollnagel, 2003). By using these criteria, it is possible to propose a pragmatic definition of how to determine the boundary of a JCS, as indicated in Table 1.

Table 1. Pragmatic definition of the JCS boundary. Derived from (Hollnagel, 2003)

1

Objects or factors that may affect artefact use Objects or factors that do not affect artefact

use

Objects or factors that

are changed by

artefact use

Objects or functions that must be considered

by JCS

Objects or functions that should be considered

by JCS

2

Objects or factors that

are not changed by

artefact use

Objects or functions that should be considered

by JCS

Objects or functions that not need be considered by JCS

Thus, from a JCS point of view, the environment provides inputs to the system and reacts to outputs from the system, while the system boundary depends on the purpose of the analysis. This means in line with Hollnagel and Woods (2005), that the operator-control system ensemble constitutes a JCS, which exists within the larger production system. Alternatively, control-room operators can also in line with Hollnagel and Woods (2005) be considered as a JCS, if studied separately. The environment of the operator-control system or operator staff ensemble is, however, not just the surrounding buildings, or the market, but also as noted by Hollnagel and Woods (2005) the plant management and other systems that might be considered (fig. 1).

Fig. 1. Levels of Joint cognitive systems. Derived from (Hollnagel, 2003)

So, provided that operator work is seen as a JCS, is it possible to automate decision-making. If so, how can automated decision-making be achieved?

2.2 Operators in Production Systems

Even if information and control technology historically was incorporated into the production systems in order to remove the operator from direct control (Lennerlöf, 1993), or at least support the human; this introduction has not only been positive (Wickens and Hollands, 2000). In the area of process control it has been shown that an increase in automation and computerization, may lead to an increase in mental workload (Zuboff, 1988). Additional information provided by information technology increases the requirements and needs for reasoning (Brehmer, 1997); which may replace or even add to earlier routines and skills for the operator (Brehmer, 1997). It can also be seen that increasing usage of computer technology (e.g. information and control) not necessarily has lead to a decrease in manual contribution; even if the requirements for cognitive skills in supervision of the computerised system have increased (Lennerlöf, 1993). Also Inagaki (1993) noted that since the human locus of control is required, the role of the human has become more important than before automation was introduced.

One reason for this is that even if automation relieves the user, he or she still has to have full awareness over the present and future situations (Endsley and Garland, 2000). This shift from manual and distributed activities to relatively static or passive monitoring of computer screens has placed the human operators into a role that they are ill-suited to handle (Bainbridge, 1982). Ironically, the technology that aimed to give conceptual simplicity also seems to decrease the sense of direct control in the actual work situation. It is also noted by Wickens and Hollands (2000), that an increase in the level of technological system complexity, associated with the incorporation of more automation, will also correspond to an increased probability for catastrophic failures.

The out-of-the-loop performance problem (Endsley and Kiris, 1995) is therefore a major type of risk associated with human interaction and automation. Hence, modern operator work still calls for a significant amount of manual labour, which makes it necessary (and possible) for the operator to gain detailed knowledge about the work place in general (Endsley, 1995) and the production processes in particular. This is especially apparent in cases where the human operators have to ensure that the automated system is performing properly and to detect situations that diverge from the expected performance (Sheridan, 1992).

The importance of the operators in critical system events has therefore put a focus on user-centred, or user-driven, approaches on system design (Billings, 1997). Such focus as Human-Machine Interaction (HMI) or Human-Computer Interaction (HCI), which acknowledged that also the most advanced systems need human supervision, decision-making, and control (Sheridan, 2002) – given that no automation can be fail-proof in every aspect or all the time.

A central question for the design of automation and information technology has therefore not only been how to construct the best possible automatic system, or to optimise the human-machine interaction, but also how to optimise the task allocation (TA) between humans and technology.

2.3 Task Allocation

Initial contributions to the field of task allocation were made by Fitts (1951), who put together a list of general physical and cognitive tasks for both humans and machines; illustrating that the performance of one category of either physical or cognitive advantages exceeds that of the other (table 2). However, this paper will only focus on the cognitive side.

Table 2. Fitts’ List, reprinted with permission in (Hoffman, et al., 2002)


� Ability to detecting small amounts of visual or

acoustic energy

� Ability to perceiving patterns of light or sound

� Ability to improvise and use flexible procedures

� Ability to store very large amounts of information

for long periods and to recall relevant facts at the

appropriate time

� Ability to reason inductively

� Ability to exercise judgment

� Ability to respond quickly to control signals, and to

apply great force smoothly and precisely


� Ability to store information briefly and then to

erase it completely

� Ability to reason deductively, including

computational ability

� Ability to handle highly complex operations, i.e.,

to do many different things at once.

Over time, there has been a range of criticism regarding the applicability of Fitts’ list in light of system engineering. For instance, Jordan (1963) stated that the attempt to compare abilities of humans and machines is inappropriate since only machines can be designed for a specific pre-defined task. According to Jordan, the idea of comparing the human and technology should therefore be discarded, yet system designers should still keep in mind what humans and machines do best, and embrace that the two are complimentary rather than conflicting entities. Although many of the common automation design models (Jordan, 1963), acknowledge that, the ideal task allocation is only plausible in theory. Instead, studies of automated systems tend to focus on stable task allocation between human and machine, and thereby emphasising static and individual attributes of the system components (Billings, 1997). Task allocation models that standardise the relationship between human and machine can, in other words, be criticised for manifesting an artificial distance between the operator and his/her equipment, as they avoid the complex interaction between system components (Frohm, et al., 2003).

From the user-driven perspective, operators can be set to solve unanticipated problems that the present technology cannot handle (Billings, 1997) (e.g. maintaining the machinery, or to keep track on problematic events in the production process). As a result, the machine operators, and other employees, sometimes only get passive roles in the production system, in which they stand for external credibility to, or responsibility for, the automatic system (Sheridan, 2002).

These variations on the human side of the system also mean that the roles of the automation are not given. To use computer aid in order to give more time for the process operators, will probably result in an entirely different time-frame, and generating a new kind of decision-making process; a process where the operators are given strategic capabilities, and responsibilities, rather than tactical. Instead, there needs to be different design solutions at different levels of operator activity, and that the purpose of the technical system may vary in these tasks (Karsvall, et al., 2003).

2.4 Operator Decision-making in Production Systems

The common interactive performance goal for humans and machines in the production system is the overall production criteria; such as producing reliable and cost-efficient products (Kilbo, 2001). Humans, however, may have alternative personal or social motives, which will influence the overall performance. From an actor’s point of view, a decision can therefore be referred to as the personal choice between two or more alternatives in situations that call for some change (Beach and Lipshitz, 1993). If this process is performed correctly, the theoretical outcome is the path that leads to a desired goal, or at least to a positive outcome or a way to avoid an unpleasant one (Tversky and Kahneman, 1986). However, there are several competing notions about decision-making, which make it hard to generalise the concept. Psychobiological theory for example, tells us that decisions are governed by implicit physical and psychological

needs, which steam from a continuous adjustment to the human habitat (Hastie and Dawes, 2001). This is said to explain why humans are skilled at seemingly complicated decision-making tasks, associated to pre-historic conditions, but less capable in situations that are related to the modern life of informational stress (Karsvall, et al., 2003). Decision theories within traditional engineering, on the other hand, often propose rational or prescriptive definitions, and consequently emphasise the normative nature of decision-making (Beach and Lipshitz, 1993). Decision-making is thereby treated as a distinctive phenomenon, which is close to the essence of human ability, yet explained as a high-level cognitive process that involves thorough analysis of available data and explicit evaluation of possible outcomes (Tversky and Kahneman, 1986).

In real-time practice, however, decision alternatives are not always in explicit conflict with each other during real-time settings (Hedelin and Allwood, 2002), which is why decision-makers often have to rely on communication with other stakeholders, rather than on objective calculations of pros and cons for each alternative (Hedelin and Allwood, 2002). To compare each identified alternative in complex settings is, if possible, also very time-consuming and thereby counter-productive in decision situations with strict time-limits (Hastie and Dawes, 2001). Hence, a major shortcoming in the normative and individualistic theories is that it requires that decision-making is considered as a strictly logical process with little, if any, relation to the time-pressure or similar environmental constrains (Hollnagel, 1984). It is thereby obvious and in line with Klein (1993), that in many situations where rapid decisions are necessary, the decision-maker has to act in less analytical and instinctively manner. As evolutionary survivors, we as humans have brought with us mental decision tools based on very little information and simple rules or decision heuristics (Beach and Mitchell, 1978). Empirically driven researchers such as Klein (1993) and Zsambok (1997) therefore propose that decisions should firstly be studied as the decision-makers perform it, in their natural environment, rather than as it is supposed to be happening in controlled settings. This descriptive (or behavioural) take on decision-making is often referred to as the Naturalistic Decision-Making (NDM) paradigm, a framework for research that describes how people make decisions in action, and how they use their domain-specific experience (i.e. expertise) in order to solve complex real-world problems (Zsambok, 1997).

In the NDM sense, decision-making is an on-going and often routinised activity that is performed during familiar settings (Klein, 1998). Still, the NDM-tradition also emphasises decision-making as a spontaneous activity taken in conjunction with the demands of the present situation (Klein, 1989). According to Brehmer (1993), operator work should be summarised as dynamic decision-making; an ongoing dialogue between the technology, its users, the operative task, and the system environment.

Decision levels and problem structures

From the literature of production and human factors it is possible to single out three basic conditions of the internal nature of a problem (derived from e.g. Michon (1989) and Harlin (2000)), which has to be solved by different kinds of decisions. Firstly, strategic decisions (decisions with long-term effect), represents explicit, conscious, non-routinised, or pro-active activity. Secondly, tactic decisions (mid-term effect) of conscious but semi-routinised, and pro-active but spontaneous activity. Thirdly, operative decisions or actions (short-term effect) are characterised as unconscious, routinised, re-active or spontaneous activities.

If it also can be expected that problem in itself can be categorized as structured, unstructured or semi-structured; the structured problems are those that can be defined by explicit procedures. Unstructured problems are problems where the decision-maker has to rely on his or her judgment and evaluating capacity.

Not apparent is the general tendency to find the majority of structured problems in the operational level. Further, most semi-structured problems are to be most common at the strategic level. Structured decisions seldom involve managers and can hence be made by

operators or even by computers. Semi-structured decisions are by their nature appropriate for managers with computer support. Unstructured problems are not able to be formalized in a technical sense and hence are impossible to feed into a computer. The nature of the problem, the volume of data, or lack of an appropriate method makes any decision entirely dependent upon human experience and intuition.

Thus, the core question of this paper is then: given that operator work situation is seen as a JCS:

Is it possible to automate decision-making from an NDM-perspective, on operative, tactical and strategic levels; and if so, how can it be achieved?

3 Examples from Industry

Based on the previous findings in the theoretical field of decision-making in real time practise and in conjunction with task allocation; three empirical studies were conducted to confirm the previous findings, during real time situations.

3.1 Process control room example

The first study was carried out in an automated process control system in a petroleum refinery during two nightshifts. The refinery consisted of plants for processing crude oil into end-user products such as light propane gases (butane and propane), motor spirits, aviation fuel, and diesel-, and heating- or heavy fuel oils. The crude oil was supplied by ships and imported via pipelines from a nearby-situated harbour. The finished products where then sent from the refinery in pipes to an oil terminal, and then shipped. Monitoring of the production processes was centralised in a control room where there was one console for each subgroup A, B and C (fig. 2). Each console was staffed with two operators. The ensemble could be considered as one JCS (see fig. 1). Subgroup A and B were both divided into two workstations (fig. 3), where each workstation together with its process operator can also be seen as a JCS.

Fig. 2. Overview of the refinery control room work setting

Fig. 3. Overview of one of the workstations (subgroup A) in the refinery control-room. Label 1-3 refer to separate

sets of control panels, label 4 refers to an ordinary PC connected to a local network

In each subgroup there were two consol operators (30-40 years old), both with 10-20 years of experience as process control operators. In the control room there was also one process supervisor and on the plant-floor additionally 3-4 factory-floor operators. The console operators’ tasks were to observe the conditions of the process (e.g. the variable ranges and their tendencies during different periods) and to detect and handle deviations of the process. Two operators assigned to the workstations in subgroup A where observed during the first nightshift. On the following nightshift, the other operator in subgroup A and one of the operators in subgroup B, were studied. The activities of the operators where noted in observation protocols by the researchers. The observations where also recorded by a digital video camera, for the purpose of deeper analyses in a later stage. The observation protocols contained three main fields for recording data and at the bottom, four fields for keyword to support the observer; (the design of the protocol had been evaluated for usability in an experimental pre-study). By the end of the work shift, a semi-structured interview with the participant was conducted to further understand the activities that had taken place, as well as on individual background data.

Results

� During the study, the observed operators were involved in handling three major tasks.

� Adjusting for outdoor temperature changes due to e.g. sunrise.

� Handling unsatisfactory laboratory results based on routine analyses of the produced product. Deviations were due to the varying amount of salt in the sample.

� Handling temperature changes, resulting from a malfunctioning cooling fan.

In the matter of handling adjustments to outdoor temperature which was a tactical decision, the operators did not have rules related to regulations; although this type of task was something that appeared every summer and was observed during both nightshifts. This indicated that procedures had to be built on previous experience and knowledge.

In a similar way, it was observed that the operators acted as assistants to colleagues when they had to handle the unsatisfactory laboratory results of the produced product. The laboratory results from the previous shift deviated from the control values. Since the process responded so slowly, many of the changes made during the eight-hour shift had to be handled by the next shift. This time-delay made it difficult for the operators to learn what the actual effects of their actions would be.

In this case, the operators thought that the reason for value deviations was that the work floor-operators could not manage to take proper samples. So, the operator dealt with the problem by ordering the operator on the workshop floor to take a new sample for analysis. In

the case of the malfunction or broken cooling fan, the operator in charge discussed different options and opinions with his colleges.

It was also observed that the operative and tactical tasks at hand were performed very much in a nested and parallel ways, where single activities related to tasks that were extended over the entire work shift; both in terms of time and in terms of system actors. The problem with the cooling fan, resulted in continuous flow of alarms that where connected to temperature problems, but those problems should not be seen as consequences of the raise of outdoor temperature per se. In order to further enhance the coding of the data, the operator activities were therefore divided in a set of low-level action categories: as various ways to look at screens; using keyboards; or talking with colleagues. These were later processed and summarised into four more high-level activitycategories:

� Dealing with system alerts/alarms

� Looking at screens

� Using touch-screens or keyboards

� Communicating with colleagues

It was observed that the operators spent most of their time monitoring the system by looking at the different screens. The operators’ observations where made in order to access relevant information for a better understanding of the present situation, and to be able to predict further states of the system. However, out of the authors’ observations and additional interviews with the two nightshifts, approximately one third of the time was spent on activities not related to predefined and ordinary work tasks. The operators also had conversations with colleagues in a more personal manner, i.e. smaller breaks for coffee and/or meals. Operators also spent time informing the researchers regarding details about their work.

Discussion

In the following interview, the operators concluded that there were seldom any clear guidelines how problems that where observed during the study should be dealt with. It was more or less up to the different shift-groups to solve the problems in their own way; since the results were more important than the working methods.

Thus, from the researchers’ point of view, the operators seldom performed clear-cut decisions, and the observations indicated that there were no particular routine solutions for handling process adjustments. From the interviews that followed with the operators, it was clear that most of the interactions where directed by personal operator experience and common pre-defined rules. This notation is in line with Zsambok (1997) which also conclude that people often use their domain-specific experience when solving complex real-world problems.

Therefore, the process operators did actually not see them selves as decision-makers. Instead, it was the task of the researchers to identify the decisions made by the operators, based on a detailed and retrospective analysis of the operators’ activities.

3.2 Manufacturing industry example

The second study was conducted in a manufacturing system that processes complex machined products. The study was conducted during 4-shift operations, and the selected manufacturing system was a highly automated manufacturing line. Its manufacturing was distributed along a line of 5-6 machines for processing raw material — from casting to a large, machined part that was passed on for further assembly. Each workstation together with its dedicated operator can be seen as a JCS. The production line in the example consisted of three legs (fig. 4); each of them staffed by 1-2 operators (25-35 years old), and the whole production line with one production supervisor, (around 45 years old).

Fig. 4. The observed production line

Some of the operators had only worked for 4-8 months on this particular line, and they where therefore still learning to handle each individual machine. The production supervisor on the other hand was a well-experienced operator. Since the computer network was integrated virtually into all manufacturing equipment; each operator could thereby manage and control several machines in a particular part of the line. Thereby one operator could be followed and tracked by one researcher focused on operator tasks that could be connected to maintenance(involving personal from the maintenance department), disturbance (smaller problems in production) and conversion (switching fixture and/or tools for a new type of product). The observations were recorded through observation protocols and followed up by semi-structured interviews to get a deeper understanding for the working procedures.

Results

The observed operators were involved in three major tasks connected to maintenance, disturbances, and conversion.

� Disturbance related to grinding machine, which lead to that the personnel from maintenance department had to be involved.

� Malfunction of the tool-head at one of the industrial robots.

� Problem with the automatic identification of the machined product at one of the milling machines.

Since the operators were not expert operators, limited rules built on earlier experience existed; this could be seen from the observation, where the operators in most cases turned to each other for help. When a problem occurred, which the operator could not handle, he or she contacted the production supervisor for help and support. In the matter of handling the disturbance with the grinding machine, the operator had to contact the maintenance department; since the failure code on the machine indicated that there was an electrical problem connected to one of the sensors. The ordinary procedure to handle such problems with sensors was to clean the sensor. The problem with the cleaning in this case was that the failure code told very briefly, what the problem was all about; but not where to find the sensor in the machine. So, one of the main reasons to take an operative decision to contact the maintenance department, was that the operators did not know where the problem was located in the machine. The operator then placed a work-order into the local IT-system, with information about the malfunction; this vague problem was then converted into a short description that included characteristics about the machinery, and information about the operator that placed the order.

The electrician who got this vague description from the IT-system tried to recall if there had been any earlier problems with that particular machine. If so, the electrician had a model of an appropriate solution; if not, the electrician had to contact the operator at sight for more information of the problem before dealing with it. In this case, the problem was new for both the operator and the electrician. The electrician was accordingly forced to go back to the basic documentation for the present machine to see if he could get any contextual cues. Later on, after that the problem was solved, the electrician noted what he had done to solve the problem into a personal notebook. The reason for using a personal notebook was that he could not be sure when he would have the possibility to register the work into the common IT-system right after the disturbance handling.

The second type of problem that occurred was a malfunction with one of the industrial robots. As above, the operator made an operative decision and placed a work-order, for what he thought in this case was an electrical problem in the automatic control system of the robot. After a more accurate investigation from the electrician, it was however concluded that there was not an electrical problem, but a mechanical. Therefore, the electrician returned with a specialised mechanic to deal with the problem, and then remained on-site, as the work on the robot also needed to be handled by an electrician. During the whole, disassemble operation, the tool-head of the robot, the operator, electrician, and the mechanic talked with each other in order to solve the problem together.

In contrast to the two previous disturbances, during the problem with the automatic identification of the product, the machine itself did not recognise the problem and consequently did not give any failure-code or stopped the production process. Since no failure could be identified by the machine, the testing could not either be conducted. The operator therefore contacted a process supervisor for assistance in order to locate the problem. But since there were no failure codes to go on, he could not proceed. Thus, during the following shift, different electricians and other personnel with expert skills appeared to see if they could be of any assistance. However, since the solution for the problem could not be found during the study, no observation for any solution to this malfunction was observed.

Discussion

Out of the observation, the informal discussion and the semi-structured interviews with the operators, it could be concluded that there were many different types of supervision and monitoring tasks. The operators moved between 3-4 machines along the production line that were included in to the JCS, to get a clear picture of the production situation that he/she was responsible of. Moreover, as in the case with the process-control room, the observed operators did not consider themselves to be decision-makers; it was up to the researchers to identify operative decisions. It was also difficult to do so, since no explicit routine solutions for handling adjustments in the production seemed to exist. Even if most of the interactions where built on knowledge and experience, the most important part of the decision-making process was in the communicative interaction between the operators and their production supervisor.

3.3 Nuclear control room example

The third study was carried out in a full-scope simulator of a central nuclear power control room. The study was performed during a yearly training session for a shift team.

The nuclear power plant was a Boiling Water Reactor (BWR). The shift team in the study consisted of a shift supervisor, a reactor operator, an assistant reactor operator, a turbine operator, and a plant operator, which together with the technical equipment in the control room constitute a JCS. The crew works together to monitor the plant process and to manage incidents. All members of the team had 8-20 years of experience.

The reactor operator controls equipment that affects the effect of the power of the reactor, e.g. the control rods. The turbine operator monitors and operates the turbines, the condensers,

feed water pumps, and has responsibility of the electricity supply in the nuclear station. The control room reflects the division of work between the operators where each operator has a separate work desk with controls and monitoring equipment (Fig 5).

Fig. 5. Layout of the central nuclear power control room

Observations where performed to investigate how the shift team dealt with incidents. The working tasks for the team were to identify incidents and take corrective or mitigating actions. Examples of incidents are low oil level in tank, loss of feed water pump and bypass of heaters. In total, five different incidents were studied.

The observations where performed from the instructors’ room. The instructors’ room has a one-way mirror and video display unit equipment that reflect all the screens the team operators work on. Furthermore, the communication between the shift members is listened in to via microphones installed in the simulator control room. Observations protocols where used to record the communication and actions. Furthermore, attendance at the de-briefing session was also included in the study.

Results

� The disturbance management tasks studied in the simulator were:

� Management of insufficient vapour lock

� Low oil level in tank

� Loss of feed water pump

� Bypass of heaters

� Isolation of containment

The turbine operator was mostly affected by the situation with insufficient vapour lock.Therefore, the turbine operator had the main responsibility to manage the situation whereas the reactor operator and shift supervisor assisted and tried to maintain a good overall picture of the plant process by monitoring key parameters. When the disturbance had subsided, the operators together tried to identify the cause of the incident. The operators identified different possible causes that might have had triggered the incident and evaluated the probabilities. The evaluation was based on training, documentation and experience. Thereafter, investigations where taken to evaluate the different possibilities from a tactical perspective. The operators were working as a team to solve this problem. The problem-solving phase was lead by the shift supervisor. The management of the situation was supported by instructions whereas the problem solving required more creative thinking and relied on the operators’ previous experience.

When handling the situation with low oil level in a tank the operators did not have to do much. When the alarm was generated, the turbine operator made an operative decision to close a valve and then to contact the plant operator who had to check which problem that had occurred.

Reactoroperator

Turbine operator Shift supervisor

When the plant operator reported back, he had identified leakage as the cause. The operators and the shift supervisor then discussed the possible mitigating or corrective actions on a tactical and strategic level.

The management of the situation with loss of a feed water pump was facilitated by the fact that the turbine operator already had the pumps controlled by their number of revolutions and not by the effect. This was a result of an action performed in an earlier disturbance. However, the turbine operator needed to monitor the feed water system very carefully during the switch of pumps. Instead, the shift supervisor made a tactical decision and told the assistant reactor operator to monitor the turbine section for a while. The distribution of workload and work tasks is only possible if the operators have double competencies, which were the fact in this case. When the operators have training and competence for more than one role in the shift team, the team is more flexible and can better adapt to different situations.

The two last disturbances, bypass of heaters and isolation of containment, where performed with guidance of operative decisions from instructions and checklists. The operators therefore considered both of these tasks easy to handle. It could also clearly be seen that the operators knew what to do in these situations. They never hesitated, but immediately used the correct procedures and performed the required checks of equipment.

Discussion

A general conclusion about disturbance management in nuclear control rooms is that large disturbances are easier to handle than unanticipated minor incidents, since emergency operating procedures and instructions are available to the operators for the large disturbances. Minor incidents put higher demands on the operators’ abilities for problem solving, identification and correction of the cause. The availability and use of emergency operating procedures and instructions contrasted to the findings from the process and manufacturing industry examples, where the operators did not have routine solutions for the management of several different situations.

A common denominator between the operators in the process industry and the operators in nuclear power was that the operators did not see themselves as decision-makers. The operators in process control stated that they were directed by pre-defined rules or their experience, and the nuclear operators declared that they usually “just followed written procedures”.

The disturbance management work in nuclear power was lead by the shift supervisor who kept an overall view of the plant process and who also distributed the workload and work tasks between the shift members in high workload situations. Since the operators have different responsibilities in the control room, the workload can vary between the two roles very much in different situations. The supervisor can then distribute the task more evenly between the operators, leading to a situation with better overall control of the incident. However, this is only possible with good communication strategies within the team. Further, the communication is very important even if the operators follow their respective roles. Since the different parts of the process are dependent on each other, so are the operators. Therefore, communication is a key to successful management of disturbances. This finding corresponds to the results from the two other examples. This indicates that both the technical system and the social system need attention when modelling the overall performance.

Further, communication and experience are very important to manage unanticipated events where procedures are not applicable. This corresponds to the findings in the process control study. In addition, to improve the performance of the operators in nuclear power and to facilitate the management of incidents, the operators are trained regularly. The operators who participated in the study were also very experienced. The training of operators to improve the performance of disturbance management was not found in the other two examples. In those industries, the management of incidents was solely dependent on the operators’ knowledge and experience.

There were many similarities especially between nuclear power and process control. In both industry sectors, the operators perform tasks in a nested and parallel way and the time lags can be long. The operators also tried to maintain a good understanding of the present situation and predict future states by monitoring the system via the process control screens.

Short 5-minute-meetings where held during the management of disturbances to update the overall information about the plant and to facilitate for the operators to get a holistic view of the disturbance, its probable causes and mitigating actions. To facilitate problem solving and to ensure that no possible causes might be forgotten, the operators use a whiteboard for information about the disturbance. Further, to facilitate the decision-making the operators really focused on the problem. For example, once the reactor operator said to the turbine operator “No,do not answer the telephone; we need to solve this situation first”. In this situation, a logic process chart could be useful to the operators to identify possible causes and their likelihood. The prioritisation of tasks also mirrors the importance of diagnosing the cause to be able to proceed with the start-up phase. If equipment is not working as supposed, safety cannot be guaranteed and then the station is not allowed to take into operation. Safety and efficiency must therefore be considered as very dependent on the operators’ ability to solve problems and make decision.

Decision-making and problem solving are not the result of an individual operators cognitive processes but rather a result of the organisation of the team and work, the communication within the shift team and the individual operator’s cognition and attitudes. Therefore, decision-making needs to be studied from an overall perspective and as a distributed task.

In conclusion, the nuclear power operators were much more directed by procedures and instructions than the operators in the other examples. Both the identification of incidents, identification of causes, problem solving and decision-making, and direction of actions to take were supported by established instructions. Furthermore, the nuclear operators have an advantage since they regularly train to manage different situations.

4 Summarizing Discussion

Operator decision-making may take many different forms, when manufacturing industry is concerned with very complex products with high quality requirements. In process and nuclear power industry, the difficult tasks consist in line with Brehmer (1993) of interpreting a complex event with many variables, circular flows, or unclear interactions. Decision-making in this environment is thereby governed in line with Klein (1993) by user-experience; the operator must know how the plant is physically built, know the petrochemical or nuclear power process, its control mechanisms, and the dynamics of the overall system. From elaboration of existing models of decision-support systems (e.g. Beach and Lipshitz (1993)), and in conjunction with the observations from the field studies; it can certainly be discussed if operator work really can be defined as some sort of leftover activity; meant only to serve the automation.

As seen in the field studies, the decision-making activities are not only contributing to the general system performance, they are also in line with Brehmer (1993) one of the most crucial events for the overall efficiency and safety. This shift in focus from system automation to system autonomy has therefore the potential to enhance the ecological validity of research by emphasising that system models should not only explain the internal functionalities. Autonomy, in contrast to traditional task allocation, also considers more open and flexible work settings that involve interpersonal interaction as well as human-machine interaction and system automation. When addressing the distinctions between human and machine elements in the JCS (Joint Cognitive System), it is therefore essential to recognise in line with Brehmer (1993) that the most prominent division of labour/functionality within the system; is the division of decision-making power between actor (human) and agent (machine).

It can further be argued in line with Hastie and Dawes (2001) that if the identified decision-maker is not in absolute conjunction with the identified decision. The individual decision-maker

rather acts as promoter of a common point of view, which is in line with Beach and Lipshitz (1993), or serves as an important, but nevertheless interdependent, system component. As seen in both theory and practise, decision-making in actual work-settings should in other words not be seen as clear-cut cases, but as nested or parallel events during which the decision-makers are able to proceed with several projects, or to change agenda in accordance with the demands of the current situation. Thereby, it is even hard to observe any decisions at all from the observer point of view. The occurring activities were rather anticipated as part of the system design (pro-actively or strategic) or just operative reactions to real-time events. However, by relating decision-making to the present real-world event; research in the NDM-tradition and related prescriptive theories, risk to treat decision-making as something that is distinguishable from other forms of observed activity. That is, stressing that decisions are natural occurrences, which can be studied or identified in the present; as seen in more prescriptive theories.

So, as a non-routine event, operative decision-making is, per definition, not possible to automate in the classical sense. As long as we are not talking about (future) AI-technology, machines rather implement readymade operations in reaction to pre- or user-defined decision situations and, consequently, fully automatic systems will not make decisions in action. The machine operator, in contrast, has the ability to make tactical decisions in action. In fact, as Billings (1997) argue, this is one of the main reasons why we employ humans in control systems at all. If a JCS risk encountering non-anticipated situations, (a situation that has a high probability), human decision-makers need to be an active part of that system. When the system organisation and behaviour become conceptually complex, however, the human cognitive capabilities/skills are as noted by Beach and Lipshitz (1993) not enough to assess all the information or practical skills necessary for reliable or efficient decision-making. It is in these cases that the operator needs Decision-support from automated information systems. The quality of the interaction process between human and machine is therefore essential to the JCS, and the critical situated decisions are often made explicit during the interface interaction.

Table 3. Decision levels and problem structures

Kind of

problem Operative level Tactical level Strategic level

Structured � Operational quality control in

manufacturing

� Disturbance handling with

support of check-list in nuclear

power industry

� Production planning


support of check-list in nuclear

power industry

E.g. the handling of the loss of feed water pump in case 3

� Investment in a new production

line

Semi-

structured


limited information

E.g. the malfunction fan in case 1 and grinding machine in case 2

� Rescheduling of production


limited information

E.g. the handling of low oil level in case 3

� Choice of sub-supplier

Unstructured � Ad hoc emergency handling

E.g. the malfunction of identification of machined parts in case 2

� Marketing new products

Yet, while the prescriptive and traditional decision-making theories such as Tversky and Kahneman (1986) and Hastie and Dawes (2001) are focused on the long-term effects, descriptive (and in particular NDM) are more often concerned with the short-term decision-making on the operative or tactical levels. One explanation for the emphasis on long-term decision-making is, of course, the difficulties involved when analysing each and every detail of

a complex event, but it can also be a consequence from the fact that researchers/designers and operators do not share the same view at lower levels. One can normally not ask the practitioner about operative events, and seldom even about tactical considerations. That is, it is only possible to get a mutual understanding between the observed activity and the experienced event at the strategic level. As seen, decision-making is not a routinised work process, but something that it determined in real-time (practice) and in interaction with the system environment.

5 Conclusions

This paper has elaborated on decision-making and task allocation to study whether it is possible to automate decision-making. Theoretical findings have been further related to the authors’ empirical observations in three industrial case studies in three different industry sectors.

It can be concluded that the primary function of technology aimed to support short-time, operative decision-making should be to mediate and amplify operator skills, rather than present static lists of decision alternatives. Tactical and strategic decision-making could be supported by automated information systems. However, it is questionable if operative decision-making can be fully automated.

To make decisions meaningful for both external observers and involved decision-makers, decision-making should be studied as an instance of meaningful system performance from an overall system perspective. Also, from the users’ point of view, decision-making could be studied as a high-level, but often distributed, cognitive task.

Finally, when it comes to understanding the purpose of automated systems, we propose that less time should be spent on allocation tasks between human and machine attributes, and more time should initially be spent on describing the requirements of the overall system.

Acknowledgements

The work presented is in parts results from the project “User-Driven Decision Automation”, which was carried out as a co-operation between Chalmers University of Technology, Linköping University, and IVF Industrial Research Corporation. Other parts of the paper are results from the research project "Alarm system design and presentation", a collaborative project with Chalmers University of Technology, the Swedish Nuclear Power Inspectorate, Swedish nuclear power plants and the Institute for Energy Technology.

The authors express their gratitude to the Swedish Agency for Innovation Systems (VINNOVA), the Swedish Nuclear Power Inspectorate and, the Nuclear power plants; Forsmark, Oskarshamn and Ringhals for generously sponsoring the projects. The authors also thank Professor Eric Hollnagel, Associate Professor Anna-Lisa Osvalder, and psychologist Mathias Hassnert for their contributions. We also want to thank the companies and operators that participated in the field studies.

References

Bainbridge, L., (1982). Ironies of Automation. Automatica Vol. 19 pp. 775-779.

Beach, L. R. and Lipshitz, R., (1993). Why classical decision theory is an inappropriate standard for evaluating and aiding most decision making. In: G. K. R. Calderwood (Eds.), Decision Making in Action: Models and Methods. Ablex, New York, USA.

Beach, L. R. and Mitchell, T. R., (1978). A Contingency Model for the Selection of Decision Strategies. The Academy of Management Review 3 (3), pp. 439-449.

Billings, C. E., (1997). Aviation Automation: The Search for a Human-Centered Approach, ed. Lawrence Erlbaum Associates, Mahwah, New Jersey, USA.

Brehmer, B., (1993). Processoperatörernas arbete i moderna kontrollrum. In: L. Lennerlöf (Eds.), MÄNNISKOR - DATORTEKNIK - ARBETSLIV. C. E. Fritzes, Stockholm, Sweden.

Brehmer, B., (1997). Styrning av system: kognitionspsykologiska aspekter på modernt arbete. In: L. Lennerlöf (Eds.), Människan i arbetslivet. Stockholm, Sweden.

Chapanis, A., (1996). Human Factors in Systems Engineering, ed. John Wiley & Sons, Inc., New York, USA.

Eijnatten, F. M. V., Sitter, U. D., Gustavsen, B., Emery, F. and Beinum, H. V., (1993). The paradigm that changed the work place, ed. Arbetslivscentrum, Stockholm, Sweden.

Endsley, M. R., 1995. Toward a theory of situation awareness in dynamic systems. Human Factors Vol. 37 (1).

Endsley, M. R. and Garland, D. J., (2000). Situation Awareness Analysis and Measurement, ed. Lawrence Erlbaum Associates, London, England.

Endsley, M. R. and Kiris, E. O., (1995). Out-of-the-loop performance problem and level of control in automation. Human Factors 37 (2), pp. 381-394.

Fitts, R. M., (1951). Human Engineering for an Effective Air-navigation and Traffic-control system, ed. National Research Council, London, England.

Frohm, J., Karsvall, A., Hassnert, M., Stahre, J. and Hollnagel, E., (2003). System perspective on task allocation and automation design. In: Proceedings of the 8th symposium on automated systems based on human skills and knowledge, Göteborg, Sweden.

Harlin, U., (2000). Towards Strategic Disturbance Management in Advanced Manufacturing Systems, Licentiate of Engineering, Chalmers University of Technology, Göteborg, Sweden.

Hastie, R. and Dawes, R. M., (2001). Rational choice in an uncertain world: the psychology of judgment and decision making, ed. Sage Publications, Thousand Oaks, Calif., London, England.

Hedelin, L. and Allwood, C. M., (2002). IT and strategic decision-making. Industrial Management and Data Systems Vol. 102 (3-4), pp. 125-139.

Hoffman, R. R., Feltovich, P. J., Ford, K. M. and Woods, D. D., (2002). A rose by any other name...would probably be given an acronym [cognitive systems engineering]. Intelligent Systems, IEEE [see also IEEE Intelligent Systems and Their Applications] Vol. 17 (4).

Hollnagel, E., (1984). Inductive and deductive approaches to modelling of human decision making. Psyke & Logos Vol. 5 (2), pp. 288-301.

Hollnagel, E., (2003). Prolegomenon to Cognitive Task Design. In: E. Hollnagel (Eds.), Handbook to Cognitive Task Design. Lawrence Erlbaum Associates, London, England.

Hollnagel, E., Mancini, G. and Woods, D. D., (1988). Cognitive engineering in complex dynamic worlds, ed. Academic Press, London, England.

Hollnagel, E. and Woods, D. D., (2005). Joint cognitive systems: Foundations of cognitive systems engineering, ed. CRC Press / Taylor & Francis, Boca Raton, FL., USA.

Inagaki, T., (1993). Situation-adaptive degree of automation for system safety. In: Proceedings of the 2nd IEEE International Workshop on Robot and Human Communication, Tokyo, Japan.

Jordan, N., (1963). Allocation of functions between man and machines in automated systems. Journal of Applied Psychology Vol. 47 (3), pp. 161-165.

Karsvall, A., Frohm, J., Hassnert, M. and Stahre, J., (2003). On Decision-Making: Making Sense of System Performance. In: Proceedings of the In Proceedings of EAM, the 22th European Conference on Human Decision Making and Manual Control, University of Linköping, Linköping, Sweden.

Karwowski, W., Warneke, H. J., Hueser, M. and Salvendy, G., (1997). Human Factors in Manufacturing. In: G. Salvendy (Eds.), Handbook of Human Factors and Ergonomics. John Wiley & Sons, Inc., New York, USA.

Klein, G., (1993). Decision making in action: models and methods, ed. Ablex, Norwood, NJ, USA.

Klein, G., (1989). Recognition-primed decisions. In: W. B. Rouse (Eds.), Advances in man-machine systems research. Greenwich, England.

Klein, G., (1998). Sources of Power : How People Make Decisions, ed. MIT Press, New York, USA.

Lennerlöf, L., (1993). Människor och datateknik i arbetslivet. In: L. Lennerlöf (Eds.), MÄNNISKOR - DATORTEKNIK - ARBETSLIV. C.E. Fritzes AB, Stockholm, Sweden.

Merchant, M. E., (1961). The Manufacturing System Concept in Production Engineering Research. Annals of CIRP Vol. 10 (2), pp. 77-83.

Michon, J. A., (1989). Explanatory pitfalls and rule-based driver models. Accident Analysis & Prevention Vol. 21 (4).

Mital, A., (1997). What role for humans in computer integrated manufacturing? International Journal of Computer Integrated Manufacturing Vol. 10 (1-4).

Parasuraman, R. and Riley, V., (1997). Humans and automation: Use, misuse, disuse, and abuse. Human Factors Vol. 39, pp. 230-253.

Parasuraman, R., Sheridan, T. B. and Wickens, C. D., (2000). A model for types and levels of human interaction with automation. IEEE transactions on system, man, and cybernetics - Part A: Systems and humans Vol. 30 (3), pp. 286-296.

Riley, V., (1996). Operator Reliance on Automation: Theory and Data. In: R. Parasuraman and M. Mouloua (Eds.), Automation and Human Performance: Theory and Applications. Lawrence Erlbaum Associates, Mahwah, NJ, USA.

Satchell, P., (1998). Innovation and Automation, ed. Ashgate, USA.

Sheridan, T. B., (2002). Humans and Automation: System Design and Research Issues, ed. John Wiley & Sons, Inc., Santa Monica, USA.

Sheridan, T. B., (1992). Telerobotics, Automation, and Human Supervisory Control, ed. MIT Press, Cambridge, MA., USA

Tversky, A. and Kahneman, D., (1986). Rational Choice and the Framing of Decisions. Journal of Business Vol. 59 (4), pp. 251-278.

Wickens, C. D. and Hollands, J. G., (2000). Engineering psychology and human performance, 3rd ed. Prentice Hall, Upper Saddle River, NJ., USA

Zsambok, C. E., (1997). Naturalistic Decision Making: Where Are We Now? In: C. E. Zsambok and G. Klein (Eds.), Naturalistic decision making. Lawrence Erlbaum, Mahwah, NJ., USA.

Zuboff, S., (1988). In the age of the smart machine : the future of work and power, ed. Basic Books, New York, USA.

PAPER V

Frohm, J., Lindström, V., Stahre, J. and Winroth, M. (submitted). Levels of Automation in Manufacturing. (Submitted for publication in Ergonomia).

REVIEWED MANUSCRIPT – FOR RESUBMISSION

Levels of Automation in Manufacturing Communicating Author:

Jörgen Frohm, PhD-student1, Division of Production Systems, Chalmers University of Technology, Sweden

Veronica Lindström, PhD-student, Department of Industrial Engineering and Management, School of Engineering, Jönköping University, Sweden

Johan Stahre, Associate Professor, Head of division, Division of Production Systems, Chalmers University of Technology, Sweden

Mats Winroth, Assistant Professor, Head of department, Department of Industrial Engineering and Management, School of Engineering, Jönköping University, Sweden

ABSTRACT

The objective of this paper is to explore the concept “Levels of Automation” (LoA). The paper presents a literature review on definitions and taxonomies for LoA across multiple scientific and industrial domains. A synthesizing concept is suggested, including a LoA definition and taxonomy aimed at application in the manufacturing domain. Results suggest that the level of automation should be divided in two separate variables, i.e. physical/mechanical LoA and cognitive/information-related LoA. Further, that LoA in a manufacturing context can be described and assessed using 7-step reference scales for the physical and cognitive LoA respectively.

Key words: Manufacturing, automation, allocation, taxonomy

1 Communicating author:

Jörgen Frohm

Division of Production Systems

Department of Product and Production Development

Chalmers University of Technology

SE-412 96 Göteborg, Sweden


1(28)


1 INTRODUCTION

Highly automated product realization has been an important mean for industry to meet the competition from low-cost countries, primarily due to relative high wage cost observed in e.g. the US and in Europe (Kilbo 2001; Teknisk framsyn 2000). Throughout the 20th century, extensive efforts to develop automated production processes where used by manufacturing companies to radically increase efficiency and sustain a high quality of production. Increased automation was not only concerned with actual manufacturing processes, but also focused on supporting tasks (e.g. material handling, transport and storage) (Reveliotis 1999). However, even if there were ambitions during the 1980’s even to create so-called “lights-out factories” with full automation in each production unit (Mital 1997), most automated systems within manufacturing are still semi-automatic with manufacturing systems consisting of combinations of automated and manual tasks. This is especially apparent within assembly operations, which have generally been more difficult to automate at a justifiable cost (Boothroyd 2005).

In addition, product customization demands and increased product complexity have resulted in increasingly complex manufacturing systems as well as increased levels and extent of automation (Satchell 1998). However, neither automation to fulfill efficiency requirements, nor automation to achieve flexibility does necessarily lead to expected results (Youtie et al. 2004). In fact, excessive levels of automation (LoA) may result in poor system performance (Endsley 1997; Endsley and Kiris 1995; Parasuraman et al. 2000). Further, complex manufacturing systems are generally vulnerable to disturbances, which might lead to possible overall equipment efficiency (OEE) degradation (Ylipää 2000). In parallel, degraded operator performance may be caused by e.g. lack of knowledge, gradual loss of special working skills, degradation of situation awareness (Endsley 1997; Endsley and Kiris 1995; Parasuraman et al. 2000), or unexpected increase in cognitive workload (Connors 1998)

Consequently, the human actor is usually a component in the manufacturing system, and as such, she has to be involved in the technical advancements and needs to be able to handle machines and equipment. In other words, both advanced technical system and skilled human workers are necessary for achieving flexible and efficient manufacturing. Thus, automation decisions are not trivial; indicating a need for a deeper understanding of automation as well as the way automation is approached by manufacturing industry as well as manufacturing research.

The objective of this paper is to increase understanding of task allocation within semi-automated systems and to provide a systematic approach for changing the level of automation. Our approach towards this goal is to:

� Review Level of Automation (LoA) definitions and taxonomies

� Suggest a LoA definition and taxonomy for use in manufacturing

Research presented is part of a Swedish research project named DYNAMO - Dynamic Levels of Automation. This was a three-year project, ending in 2006, initiated to address industrial needs for adaptive solutions and dynamic change in automation levels in manufacturing systems during different phases of the system life cycle. The project aimed to provide industry with design, measurement, visualization, and management tools for dynamic levels of automation in manufacturing.

2 RESEARCH APPROACH

The paper focuses on level of automation taxonomies at an operative level, i.e. allocation and distribution of tasks between human operators and technical subsystems. The results presented in this paper are based on key publications in the areas advanced manufacturing technologies (AMT), human factors, and teleoperations. The review concludes with a definition and taxonomy of LoA to be used in manufacturing. The taxonomy was developed through case studies in manufacturing environments in relation to the literature review. Finally, the proposed LoA taxonomy is visualized through a generalized

2(28)


example in a manufacturing context, based on the empirical findings from the case studies. The literature search was carried out by means of Science Direct, Compendex, Inspec, IEEE Xplore, Ergonomics Abstract, and Google Scholar. Keywords used were e.g. levels, automation, degree of, task, allocation, control, information process, production, and manufacturing.

3 WHAT IS AUTOMATION?

One difficulty in defining automation lies in the multitude of context-specific definitions available. The Oxfords English Dictionary (2006) defines automation as:

Automatic control of the manufacture of a product through a number of successive stages; the application of automatic control to any branch of industry or science; by extension, the use of electronic or mechanical devices to replace human labor

As seen in this definition, automation often refers to the mechanization and integration of the sensing of environmental variables. With the emergence and rapid growth of information technology, the relevance of systems, which integrate information technology and mechanical technology, increases (Satchell 1998). Today, the term automation has grown beyond manufacturing, which was the popular context of historical automation implementation. However, with emerging concepts like ‘knowledge workers’ (Drucker 1999), human work practices have gone from physical labor, to also covering cognitive labor. Today we use computers to interpret and record data, make decisions, and visualize information. Such tasks are regarded as automation, including the sensors that go with them. Therefore, in industrial sectors where human capability is important to safety, e.g. airplanes, user-centered approaches to automation have evolved and thus different degrees of human cognitive process that are in touch with machines (Billings 1997). Recently, other definitions of human-machine integrations have emerged (Satchell 1998), focusing on sharing of tasks, control, and authority between human and machines, regarding the two as being mutually complimentary. A central question for designing automation within manufacturing systems has therefore not only been how to design the best system, but also how to optimize the Task Allocation (TA) between human and automation.

3.1 Task allocation Initial contributions to the field of task allocation were made by Fitts (1951), who presented a list of general tasks covering both humans and machines, illustrating where the performance of one category exceeds that of the other (table 1).

Table 1 Fitts’ List, in (Hoffman et al. 2002)


� Ability to detecting small amounts of visual or acoustic

energy

� Ability to perceiving patterns of light or sound

� Ability to improvise and use flexible procedures

� Ability to store very large amounts of information for long

periods and to recall relevant facts at the appropriate

time

� Ability to reason inductively

� Ability to exercise judgment

� Ability to respond quickly to control signals, and to apply

great force smoothly and precisely


� Ability to store information briefly and then to erase it

completely

� Ability to reason deductively, including computational

ability

� Ability to handle highly complex operations, i.e., to do

many different things at once.

3(28)


However, since its introduction, a range of criticism has arisen regarding the applicability of Fitts’ list in systems engineering. Jordan (1963), for instance, stated that the attempt to compare the abilities of human and machines is inappropriate since only machines can be designed for a specific pre-defined task. According to Jordan, the idea of comparing human and technology should there for be discarded, yet system designers also should keep in mind what humans and machines do best, and embrace that the two are complimentary rather than conflicting entities when designing a human-machine system. In order to make use of ideas of Jordan, Price (1985) represents the allocation of tasks in a decision space, in which the x-axis represents the ‘goodness of humans’, scaled from unsatisfactory to excellent, and the same for the machine on the y-axis (figure 1).

Fig. 1 Decision matrix for allocation of functions (Price 1985)

In line with Fitts’ notation, as can be seen in figure 1, some tasks have to be allocated to the machine (Uh) or to the human (Ua) for reasons such as physical strength or demands for problem solving. There are also tasks that both human and automation are qualified to handle (Pa, Ph and Pah in figure 1). It can be noted that tasks that do not need to be solely allocated to human or to the automation, are tasks where the human or the automated system need support from each other.

As mentioned in Fitts’ List, decisions on task allocation between human and technology (e.g. within the manufacturing system) can be based on different factors and criteria (Frohm et al. 2003). For example, according to the Purdue Enterprise Integration Reference Architecture suggested by Williams (1999), task allocation can be based on e.g. financial, technical, or social factors. These factors can in turn be organized into three categories of implemented tasks or business processes according to Williams:

� All direct mission enabling elements of the enterprise (that is, all of the equipment providing the product and/or service functions which comprise the mission to the customers of the enterprise),

� All control and information function enabling elements (equipment again), and

� All humans involved in the enterprise (humans may serve along with the equipment in either the direct mission enabling functions and/or the control and information functions, which monitor and control the mission elements).

According to Williams (1999), only two of these classes of tasks have to be considered when the functionality is in focus (operating the ‘processes’ and ‘control’ the mission in an ‘optimal’ manner), but there are three classes of implementation, since humans may be asked to implement any of the tasks from either functional class (figure 2).

4(28)


Mission fulfillment tasks Information and Control tasks

Fig. 2 Human implemented tasks may be classified either as mission fulfillment task or as information and control task, based on the PURDUE Architecture for Manufacturing Systems (PERA). (Williams 1999)

To make a manufacturing system as robust and flexible as possible is consequently in line with Billings (1997) and Sheridan (2002) not only a question on how to assign the right task and function to the right element of the system. It is also a question on how the human and automated elements can support eachother during different levels of automation, to make the manufacturing system as robust and flexible as possible.

4 THE LEVEL OF AUTOMATION CONCEPT It is often the case that progression from manual operations to full automation is said to be made in one single step, i.e. when the operators are replaced by robots or advanced machinery. This is however not completely true. According to Encyclopedia Britannica (2006), manual operations are defined as “work by hand and not by machine”, and machines are in the same way defined as “instruments (e.g. a lever) designed to transmit or to modify the application of power, force, or motion”. The term manual can thereby be defined as work performed without any tool or support. Thus, giving the user tools or other support to achieve the task can be seen as increasing the level of automation and approaching full automation. Another example is a bolt that can be fitted into a construction by hand, which may be seen as the lowest level of manual work. However, by supplying the operator with manual or automated hand tools (e.g. spanner or hydraulic bolt machine) the level of technological support is raised. By further replacing the electrical or hydraulic hand tool with a machine or robot on the workshop floor, we reach almost full automation.

Even if the example may seem trivial, it can be noted from literature that the concept of “level of automation” (LoA) has been considered by many authors (see table 2) from areas such as aviation, telerobotics, and process industry. From the review in this paper, it can be seen that the simplest form of automation often operates in two modes, manual or automatic. However, within more complex systems, e.g. aviation, control rooms within nuclear power plants and within manufacturing, it is not uncommon to find multiple automation modes of both physical as control and information support.

Mechanizedtasks

Automated

tasks

Human implementedtasks

5(28)


Table 2 A selection of level of automation definitions

Author Levels of Automation definition

Amber and Amber (1962) The extent to which human energy and control over the production process are replaced

by machines.

Sheridan (1980) The level of automation incorporates the issue of feedback, as well as relative sharing of

functions in ten stages.

Kern and Schumann (1985) Degree of mechanization is defined as the technical level in five different dimensions or

work functions.

Billings (1997) The level of automation goes from direct manual control to largely autonomous

operation where the human role is minimal.

Endsley (1997) The level of automation in the context of expert systems is most applicable to cognitive

tasks such as ability to respond to, and make decisions based on, system information.

Satchell (1998) The level of automation is defined as the sharing between the human and machines with

different degrees of human involvement.

Parasuraman et al. (2000) Level of automation is a continuum from manual to fully automatic operations.

Groover (2001)

The level of automation can be defined as an amount of the manning level with focus

around the machines, which can be either manually operated, semi-automated, or fully

automated.

As indicated in table 2, levels of automation are quite well covered within area of human factors. There are, however, only a few publications of LoA within the manufacturing area. From the manufacturing perspective, LoA has often been seen in line with Groover (2001) as the manning level, i.e. a comparison between the actual numbers of operators on the workshop floor in relation to the number of machines.

Based on our review, it can be noted that automation is not all or nothing, but should rather be seen as a continuum of automation levels, from the lowest level of fully manual performance (based on the capabilities of the human) to the highest level of full automation (without any human involvement).

Also, based on the literature review, it can be argued that automation can vary across different levels, and exists as a continuum of full, partial, or no replacement of a function previously carried out by the human operator.

4.1 Taxonomies for the Level of Automation concept Based on the review in this paper, can it be argued that the level of automation concept should be described in more depth. In previous case studies (Frohm et al. 2005; Lindström et al. 2005), it was observed that most automated processes within manufacturing in the beginning seem to only involve automation of mechanical tasks. However, those tasks are mostly controlled by computers for optimal performance (Frohm et al. 2005).

6(28)


Fig. 3 Separation of functions into mechanization and computerization (Frohm et al. 2005)

It is therefore important to recognize that automation in manufacturing can be seen in line with Chiantella (1982) and Williams (1999) as two basic classes of automation, mechanization and computerization (figure 3). Computerization is defined in this paper as the replacement of cognitive tasks, such as human sensory processes and mental activity. This would include e.g. collection, storage, analysis, and use of information in order to control the manufacturing process. Mechanization is in a similar way connected to computerization and is defined as the replacement of human muscle power, such as material and energy transformation (Frohm et al. 2005).

4.1.1 Mechanization - automation of physical tasks

Mechanization as such is often associated with individual manufacturing machines and robots (Allum 1998; Bright 1958; Duncheon 2002; Groover 2001; Kern and Schumann 1985). Further, the concept of automated systems can also in line with ISO, be applied to various levels of factory operations (Gullander 1999), since manufacturing machines are made up of subsystems, which can each be automated individually. One example of this is the Computerized Numerical Control (CNC) machine, which is composed of numerous integrated and automated subsystems. The CNC-machine may further be integrated with automated transportation systems, thus forming an automated group of manufacturing equipment (a manufacturing system), which in itself may be part of a highly automated manufacturing plant.

For example Kern and Schumann (1985) suggested that the ‘level of mechanization’ can be defined as the technical level of the manufacturing system. As seen in table 3, the ‘level of mechanization’ can be classified into pre-mechanization, mechanization, and automation. These three categories can be further divided into nine steps, ranging from manual to automated manufacturing.

7(28)


Table 3 Three levels of mechanization, adopted from (Kern and Schumann 1985)

ManualPre-mechanization

Line flow

Single units with manual work

Single units with mechanical control

Multi-functional units without manual control Mechanization

Systems of units

Partly automated single units

Partly automated systems Automation

Automated manufacturing

In similarity to Kern & Schumann, Groover (2001) suggests that the term ‘level of mechanization’ can be defined as the manning level, with focus on operating of the machines. According to Groover, the level of automation can be either operated manually, semi-automatically, or fully automatically.

Also Duncheon (2002) suggests, in the context of fiber optic component assembly, that the ‘level of automation’ can be visualized in three main automation levels (table 4).

Table 4 Three levels of mechanization (Duncheon 2002)

Manual

Automated alignment

Automated process Semi-automatic

Automated cassettes

Robotic material handling Automatic

Automated inter-cell transfer

As indicated in table 4, Duncheon (2002) defines in a similar way as Kern & Schumann (1985); ‘manual’ as tasks were the humans are responsible for conducting the task (e.g. application of epoxy). ‘Semi-automatic’ is a higher level of automation and involves, according to Duncheon, automated alignment and application of epoxy by a robot. Material handling, on the other hand, is still conducted by humans unlike ‘automatic’, where also material handling is automated.

Based on the three previous taxonomies, it can be presumed that the level of automation or mechanization can be seen as three levels, from manual to fully automated manufacturing. However, most manufacturing systems consist of both humans and automation in connection. This means that most tasks are falling between manual and full automation.

From a generic manufacturing strategy perspective, Chiantella in Kotha and Orne (1989) presented a model of process complexity, where the ‘level of automation’ consists of mechanization (physical tasks) and systemization (control of the physical tasks). Here in chapter 4.1.1, only the physical (mechanization) part of LoA will be addressed. As seen in table 5, Chiantella (1989) classifies mechanization in four levels, and emphasizes similar to Duncheon (2002) that ‘manual’ tasks can involve operations with a minimum of tools. In line with previous taxonomies, it can be seen that the ‘level of automation’ increases as more tasks are being carried out automatically.

8(28)


Table 5 Four levels of mechanization (Kotha and Orne 1989)

Manual A human operator performs an operation manually with a minimum of tools. Component assembly

using simple fixtures and hard tools would be an example

Machine The operator employs mechanical assistance in performing an operation, as in the fabrication of

parts using milling machines, lathes or presses

Fixed program A fixed program machine may employ pneumatic logic, mechanical sequencing or numerical control

to execute a sequence of operations. No provision are made for exceptions to the normal process

Programmable

control

Under programmable control a machine may execute a sequence of operations and compensate for

exceptions that may occur. A machine may be programmed to perform different tasks as well.

As previously stated, automation often relates to allocation of physical tasks, i.e. those functions that according to Williams (1999) are involved in operating the processes which results in making the ‘product’. However, for some time focus of automation efforts has been extended to address control tasks such as supervision, problem solving etc. Therefore, it seems to be appropriate to separate the level of automation into a) physical tasks, such as manufacturing technologies, and b) control tasks such as supervision and problem solving.

4.1.2 Computerization – automation of control and information handling

Based on previous case studies (Frohm et al. 2005; Lindström et al. 2005) can it be argued that computers are often implemented into many modern manufacturing operations for optimal performance control. Furthermore, recent literature, e.g. Mital and Pennathur (2004) concerning research related to humans in advanced, dynamic and automated systems (e.g. remote controlled machines and robotics), has presented a number of taxonomies of level of automation. In the same way as regarding mechanized task performance, most taxonomies on the computerized and information/control-oriented side also offer intermediary levels of automation between fully manual control and full automation. These levels are, according to Kaber et al. (2000), intended to maintain both human and computer involved in active system control, and thereby improving the operators understanding and awareness of the present and future situations (Endsley 1997). From literature, it can be seen that there are mainly two ways of looking at levels of computerization. The first is the classical ten-level taxonomy suggested by Sheridan and Verplanck (Sheridan 1980) focused on human-computer decision making in the context of undersea teleoperation systems (table 6).

9(28)


Table 6 The Sheridan-Verplanck 10 levels of automation (Sheridan 1980)

1. Human considers alternatives, makes and implements decision.

2. Computer offers a set of alternatives which human may ignore in making decision.

3. Computer offers a restricted set of alternatives, and human decides which to implement.

4. Computer offers a restricted set of alternatives and suggests one, but human still makes and implements final

decision.

5. Computer offers a restricted set of alternatives and suggests one, which it will implement if human approve.

6. Computer makes decision but gives human option to veto prior to implementation.

7. Computer makes and implements decision, but must inform human after the fact.

8. Computer makes and implements decision, and informs human only if asked to.

9. Computer makes and implements decision, and informs human only if it feels this is warranted.

10. Computer makes and implements decision if it feels it should, and informs human only if it feels this is

warranted.

As seen in table 6, Sheridan and Verplanck’s taxonomy incorporates issues of feedback as well as the relative sharing of functions, determining options, selecting options, and implementing tasks between the human and the technical system. This taxonomy is also one of the most descriptive taxonomies that can be found in literature in terms of identification of ‘what’ the operator (human) and the technical system (computer for information and control) are to do under different LoAs, and ‘how’ they should cooperate. Also Satchell (1998) stresses that sharing between humans and machines can occur in many forms, and these forms can be stratified into different levels depending on the degree of human involvement. Inagaki (2003) further develops task allocation methods, introducing not only sharing but also trading of control.

The other way of viewing levels of computerization is from a perspective of human information processing (Wickens and Hollands 2000), which determines that human information processing consists of the following four steps: acquire the information, analyze and display the information, decide action based on the analysis, and finally implement the action based on the decision (Parasuraman, 2000). As mentioned in the previous chapter, Chiantella model of process complexity also includes systemization in Kotha and Orne (1989), which is employed to control a process. Chiantella describes the level of systemization in six levels ranging from collecting data from the process, which involves acquiring information, to closing the control loop by feedback the information and using the collected data directly into the manufacturing process.

Table 7 Levels of systemization according to Chiantella in (Kotha and Orne 1989)

Data collecting Recording past occurrences – documents (reports) produced at some later time

Event reporting Capturing information as event occur – documents are produced when and where required

Tracking A continuing profile of event information for a series of operations or movements

Monitoring Dynamically comparing actual events to those planned. Alert messages are produced

Guide Providing action alternatives, and capturing the course of action taken

Control Executing a control action when predefined event conditions occur.

As shown by e.g. Parasuraman (2000), the tasks at each of these four stages can be automated to different degrees or levels. By combining the Sheridan-Verplank taxonomy with the model of human

10(28)


information processing, Parasuraman showed that a particular system could involve automation in all four dimensions at different levels. For example, a given system A (black line in figure 4) could be designed to have moderate to high level of acquisition (level 5-7), low level in analysis and decision selection (level 2-3), but high in implementation (level 6-8). Another system B (gray line) on the other hand, might have high levels of automation across all four stages.

Fig. 4 Examples of different levels of automation at different task stages (Parasuraman, et al. 2000)

Endsley and Kaber (1999) presented a taxonomy that could give support to the human by means of expert systems, which where developed for e.g. air traffic control, advanced manufacturing, and teleoperations. Endsley (1997) had previously proposed that there are five functions in a human-machine system that either human or an expert system can perform: suggest, concur, veto, decide, and act. By structuring those five functions to serve the full range between manual and full automation, Endsley and Kaber (1999) suggested a ten level taxonomy for wider applicability to a range of cognitive and psychomotor tasks requiring real-time control, see Table 8.

Table 8 Five levels of automation (Endsley and Kaber 1999)

Roles

Level of automation

Monitoring Generating Selecting Implementing

(1) Manual control Human Human Human Human

(2) Action support Human/computer Human Human Human/computer

(3) Batch processing Human/computer Human Human Computer

(4) Shared control Human/computer Human/computer Human Human/computer

(5) Decision support Human/computer Human/computer Human Computer

(6) Blended decision-making Human/computer Human/computer Human/computer Computer

(7) Rigid system Human/computer Computer Human Computer

(8) Automated decision-making Human/computer Human/computer Computer Computer

(9) Supervisory Control Human/computer Computer Computer Computer

(10) Full Automation Computer Computer Computer Computer

In comparison to the models of Endsley and Kaber (1999) and Parasuraman (2000), the Lorenz et al. (2001) model concluded that automated systems do not only differ as a function depending on LoA, but also on different types of automation, corresponding to which of the four information-processing stages

11(28)


(monitoring, generating, selecting, and implementing) that are supported by automation. Lorenz et al. (2001) instead proposed that LoA could be seen through four stages of automation (table 9).

Table 9 Four levels of automation (Lorenz et al. 2001)

Absence of automation No support is given

Notification The system provided information about what happened, and assists the operator in his or her

decision-making

Suggestion The system notified and suggested the appropriate action to take, but leave the operator to

decide and to act

Action The system resolved the problem by itself and informs the human to allow him or her to veto

Another way of describing LoA, according to Billings (1997), is to see LoA not as a single progression from total manual control to fully automated control, but as a parallel control—management continuum. Applied on air-traffic controllers, this would mean ranging from unassisted control to autonomous operations in six levels (table 10).

Table 10 Continuums of control and management for air-traffic controllers (Billings 1997)

AUTOMATION FUNCTION MANAGEMENT

MODEHUMAN FUNCTION

Fully autonomous operation; Controller not

usually informed. System may or may not be

capable of being by passed

Autonomous

operations

Controller has no active role in operation.

Monitoring is limited to fault detection. Goals

are self-defined; pilot normally has no reason to

intervene.

Essentially autonomous operation. Automatic

decision selection. System informs controller

and monitors responses.

Management by

exception

Controller is informed of system intent; May

intervene by reverting to lower level.

Decisions are made by automation. Controller

must assent to decisions before implementation.

Management by

consent

Controller must consent to decisions. Controller

may select alternative decision option

Automation takes action only as directed by

controller. Level of assistance is selectable.

Management by

delegation

Controller specifies strategy and may specify

level of computer authority.

Control automation is not available. Processed

radar imagery is available. Backup computer

data is available.

Assisted control

Direct authority over decisions; Voice control

and coordination.

Complete computer failure; No assistance is

available

Unassisted

control

Procedural control of all traffic. Unaided

decision-making; Voice communication.

Even if Billings (1997) indicated that the levels of automation are a division of tasks between the human and the automated system, the important point is that the role of the operator can vary. It can e.g. go from direct authority over the entire manufacturing process to a relatively passive surveillance function, in which the operator control task is handled by the technical system. From areas such as aviation, nuclear power plants, and advanced manufacturing systems, it can be noted that none of them today can be operated entirely at either end of this spectrum of control and management. Indeed, complex systems such as airplanes or advanced manufacturing systems can be under direct manual control, while still incorporating several kinds of control automation. Even remotely controlled systems (e.g. NASA’s Mars explorer) are not fully autonomous in the sense that the locus of control of such systems simply has

12(28)


been relocated. Similar to Billings (1997), Ruff et al. (2002) describe a context-specific LoA for remotely operated vehicles, or unmanned air vehicles (table 11), which is based on the taxonomy by Rouse and Rouse (1983) and corresponds to Sheridan’s level 1, 5 and 6.

Table 11 Three levels of automation (Ruff et al. 2002)

Manual control � Automation is dormant until initiated by operator

� Manual control is the 3 situation where automation is dormant unless initiated

Management-

by-consent

� Automation proposes action, but cannot act without explicit operator consent

� Management-by-consent is the situation where automation proposes action, but cannot act

without explicit operator consent.

Management-

by-exception

� Automation acts without explicit operator consent, requiring specific commands from the

operator to cancel automation

� Management-by-exception is the situation where automation acts without explicit operator

consent and only fails to act when explicitly commanded by the operator

A further development of LoA in the context of Telerobotics control was developed by Milgram et al. (1995), see table 12, who defined LoA by considering the different roles a human operator can play in controlling telerobotics, including being a decision-maker and direct controller. As Milgram et al. point out; one of the key aspects that have to be considered is the role of the human operator in relation to other elements of the control system. This means that, for anything other than completely automated systems, it will resolve into a spectrum of different tasks for the human operator, ranging from manual teleportation to autonomous robotics.

Table 12 Five levels of automation (Milgram et al. 1995)

Manual Teleoperation The most basic operating mode, defines all situations in which the human operator is

constrained to remain continuously in the control loop.

Telepresence Typically, this involves some form of master-slave control system, where the human operator

initiates all actions of the master arm.

Director / Agent Control Director / Agent (D/A) control can be considered as a basic form of supervisory control, where

the human operator acts as a director and the limited intelligence robot acts as her or his

agent.

Supervisory Control Supervisory control describes a wide range of options were the human operator can take on a

variety of supervisory roles.

Autonomous Robotics Fully autonomous teleoperations. The system work without remote control and the human has

no part in controlling the system.

Also Draper (1995) discusses the level of control and how to combine human operators with machine control in the context of teleoperations, see Table 13. Similar to Endsley (1997) Draper identified five different functions (programming, teaching, controlling, commanding and monitoring), that has to be carried out by the human. In relation to the human function, Draper also identified four functions (controlling, modifying, communication, and displaying) that have to be allocated to the machine.

13(28)


Table 13 Five levels of automation (Draper 1995)

Human control Total human control

Manual Control with

Intelligent Assistance

Human control and teaching with machine modification of control inputs.

Shared Control Human control and monitoring and machine control of (routine) subtasks.

Traded Control This level involves consecutive assignment of subtask control to the human and machine

depending on the characteristics of the task and server capability.

Strategic control Involving human long-term operations planning accompanied by machine performance of tasks.

Based on that allocation of tasks, Draper proposed five levels of automation, ranging from human control to strategic control involving human long-term operations planning, accompanied by machine performance of tasks (table 13). Also Anderson (1996) presents a similar context-specific LoA approach as Draper (1995) and Milgram et al. (1995), for telerobots in three levels (table 14).

Table 14 Three levels of automation (Anderson 1996)

Autonomous Control An operator programs a series of points that a robot is to move to in order to perform

manipulative functions.

Direct Teleoperation An operator is required to directly command all activities of the robot in real-time, using a hand-

pilot instead of programming.

Shared Control The robotic system involves a blend of the characteristics of these two modes including

superimposing inputs of the operator and computer control on each other.

Similar to Draper, Schwartz et al. (1996) identified that there are five different tasks that users have to carry out to some extent in the context of teleoperations of satellites. From the assignment of those five tasks, Schwartz identifies six levels of automation (table 15), ranging from no automation to fully automated operation.

14(28)


Table 15 Six levels of automation (Schwartz et al. 1996)

No Automation Satellite downlinks raw telemetry data to ground. A human controller then performs the function

without any filtering or processing.

Data Filtering Satellite downlinks raw telemetry data to ground. Ground segment then filters and processes the data

and a human controller performs the function.

Cueing Satellite performs function and suggests possible solutions. Ground segment must verify all solutions

(using filtered data) before they can be implemented.

Supervision Satellite performs function but ground segment supervises recovery activities. Ground is aware of the

satellite's progress (through filtered data from the satellite) and can override any actions, but is not

required to intervene unless stall occurs.

Paging Satellite performs function but ground segment notified if processor stalls. In such a case, ground

controller then resolves problem-using data filtered from the satellite.

Fully Automated Satellite performs function with no communication to ground. No ability to recover if satellite processor

stalls.

Based on the presented taxonomies for automation of control and information, it can be argued that many of the presented taxonomies are designed for specific predefined tasks, and thereby might have limited applicability in other systems such as manufacturing. It has also been noted by Kaber et al. (2000) that taxonomies of LoA that are designed for specific tasks might result in a reduction of the human operators’ understanding of the automated systems during an automation failure. Bainbridge (1982) already foresaw this conclusion in her classical statements on the ironies of automation.

4.1.3 LoA in Manufacturing

In previous case studies (Frohm et al. 2005; Lindström et al. 2005) the authors of this paper have observed that most tasks within modern manufacturing seem to involve a mix of both mechanization and computerization. Further, that observed tasks and operations could be broken down into mechanized and computerized activities. For example, to operate a cutting machine, tasks such as controlling the cutting tool (computerized activity) as well as handling the work pieces and performing the cutting (mechanized activity) are involved.

We now propose that a selection of taxonomies for mechanization and computerization reviewed by this paper can be assembled in a way that makes them relevant for the manufacturing field.

The idea to combine automation of physical tasks with control tasks is not new. In 1958, when automation introduced a new manufacturing era, Bright (1958) described and presented a concept of mechanization profile, consisting of 17 levels of mechanization, see Table 16.

15(28)


Table 16 Levels of mechanization (Bright 1958) In

itia

ting

Co

ntr

ol

So

urc

e

Typ

e o

f M

ach

ine

Re

spo

nse

Pow

er

So

urc

e

LEVEL OF MECHANIZATION

17 Anticipates action required and adjusts to provide it.

16 Corrects performance while operating M

od

ifie

s o

wn

a

ctio

n o

ver

a

wid

e r

an

ge

o

f va

ria

tion

15 Corrects performance after operating

14 Identifies and selects appropriate set of actions

13 Segregates or rejects according to measurement

Re

spo

nd

s w

ith a

ctio

n

Se

lect

s fr

om

a

lim

ited

ra

nge o

f p

oss

ible

pre

fixe

da

ctio

ns

12 Changes speed, position, direction according to measurement signal

Records performance 11

10Signals pre-selected values of measurement (Includes error detection)F

rom

a v

ari

ab

le in

th

e e

nvi

ron

me

nt

Re

spo

nd

s w

ith

sig

na

l

Measures characteristic of work 9

Actuated by introduction of work piece or material 8

Power Tool System, Remote Controlled 7

6 Power Tool, Program Control (sequence of fixed functions)

Fro

m a

co

ntr

ol

me

cha

nis

m t

ha

t d

ire

cts

a

pre

de

term

ine

d

pa

tte

rn o

f a

ctio

n

Fix

ed

with

in t

he

m

ach

ine

Power Tool, Fixed Cycle (single function) 5

4 Power Tool, Hand Control

Me

cha

nic

al (

No

-ma

nu

al)

3Powered Hand Tool

2 Hand Tool

Fro

m m

an

Va

ria

ble

Ma

nu

al

Hand1

Each of the 17 mechanization levels in table 16 is classified depending on the power source (either the human or mechanical source), type of machine response, and initiating control source (the origin of control information). As can be seen in Table 16, Bright (1958) lists the different levels into three categories depending on who is initiating the control, the human (1-4), the human together with automation (5-8) or the automation (9-17). In each of those tasks, Bright lists what type of machine response that can be expected.

Similar to Bright, Marsh and Mannari (1981) define what they call “automaticity” in six levels from conducting the tasks manual, without any physical support, to full automated cognition with computer control (table 17).

16(28)


Table 17 Levels of automaticity (Marsh and Mannari 1981)

Hand tools and manual

machines

E.g. Pliers, hammer, files, etc.

Powered machines and

tools

Muscles are replaced for the basic machine function, but machine action and control are

completely dependent on the operator. Uses mechanical power, but man positions work and

machine for desired action. E.g. electric tools

Single-cycle automatics

and self-feeding

machines

Completes an action when initiated by an operator. Operator must set up, load, initiate actions,

adjust, and unload. E.g. Production machines without accessory automatic control system,

grinder, planer, lathe, etc.

Automatic repeating

cycle

At this level, all energy is mechanized. The automation carries out routine instructions without

aid by the human. Starts cycle and repeat actions automatically – self-feeding. Loads, goes

through sequence of operations, and unloads to next station or machine. No self-correction but

obeys internal program such as cams, tapes, or cards. E.g. engine production lines, self-feed

press lines, automatic copying lathe, etc.

Feedback by self-

measuring and self-

adjusting

Measures and compares results to desired state and adjusts to minimize error. Although

feedback control of the actual surface of the Workpiece is preferable, positional control of the

machine table or tools is of great value. E.g. feed-back from product, automatic sizing grinders,

size-controlled honing machines, process controllers, etc.

Computer control and

automatic cognition

Computer monitors multiply factors on which machine or process performance is predicated –

evaluates and reconciles them to determine proper control action.

It can thereby be noted in both Bright’s and in Marsh and Mannari’s taxonomies that both taxonomies put focus on the support that the technological tools can give. However, both taxonomies combine the physical support with the cognitive in one single taxonomy.

4.2 Summary of the LoA concept Williams (1999) stated:

“there must be a simple way of showing where and how the human fits in the enterprise and how the distribution of functions between humans and machines is accomplished”.

Based on the reviewed taxonomies and the division into mechanization and computerization, it can be presumed that there are a number of different taxonomies that could be fit into the model (in figure 2) by Williams. From a mechanized perspective, the level of automation or mechanization can be expressed in three levels, in line with Kern & Schumann (1985) and Duncheon (2002). The lowest LoA is manual, where the tasks are achieved without any support of the automation, tools or other form of technology. The intermediate LoA is semi-automated, and consists of collaboration between the human and the technological support to achieve the task. The highest level of physical automation is automated, which exclude the human from conducting the physical tasks. From the review of computerized taxonomies, it can be seen that there are in the mainly two ways of viewing levels of computerized automation. The first is similar to the physical support approach. Focus is on the interaction between the human and the technology, thereby identifying what the human and the technical system do under different LoA conditions and how they should co-operate. This approach, promoted e.g. by Sheridan (1980) has often be used as a starting point to optimize the task allocation of decision-making and information processing. The second way of viewing levels of computerized automation is to focus on how humans acquire information, process information and makes decisions according to psychological theory (Card et al. 1983). By combining those two views the level of computerized automation can differ, depending

17(28)


on what step of the information processing is supported. Parasuraman, et al. (2000) and Kaber, et al. (2000) presents a taxonomy based on the support the technical system could give the human at different stages by means of expert systems for more efficient information processing. As noted by Billings (1997), the level of computerized automation is a division of tasks between the human and the technical system. It is important to note in line with e.g. Sheridan (1980), that the role of the operator can vary depending on the level of computerized support. As seen in the review it can be an advantage to separate the physical and cognitive tasks when assessing the level of automation. Even so, in a manufacturing context consisting of a mix of both mechanized and computerized support there is a need of assessing both perspectives. Bright (1958) and March & Mannari (1981) did that by defining the level of automation depending on who was initiating the control task. By combining the two perspectives, the levels of physical support are connected to an equal level of cognitive support. There may be an advantage in assessing the level of automation using two connected reference scales, were the level of physical and cognitive support may vary depending on the need from the human.

5 Suggestion of a LoA definition and taxonomy to be used in manufacturing

Based on this review of different LoA taxonomies and in relation to the forthcoming empirical example (chapter 6) from industry, it can be argued that the basic definition of LoAs relates to the allocation of tasks of all types between the human and her technological support. On the other hand, automation within manufacturing is generally focused on mechanical components and physical task allocation strategies, such as the task of assembling the plinth or the roll front in the forthcoming empirical example. Even if many of those manufacturing processes seem only to involve automation of mechanical tasks, most of those tasks are controlled or supervised by computers for optimal performance over time, e.g. the hydraulic screwdrivers that are programmed to tighten the screw to a pre-defined torque. Therefore it is suggested in this paper, in line with Williams (1999), that it is important to recognize that automation in manufacturing, as well as in other domains, should be seen as an interaction between two types of tasks: physical tasks and cognitive tasks. The physical tasks are the basic core technologies, such as drilling, grinding, etc, and the cognitive tasks deal with the control and support of the physical tasks.

Based on the previous review, the level of automation within manufacturing can be defined as:

“The allocation of physical and cognitive tasks between humans and technology, described as a continuum ranging from totally manual to totally automatic”

Focus in the definition should be focused on how the task can be achieved in the best possible way.

As seen earlier in table 2, most LoA taxonomies are either focused on the allocation of strictly cognitive tasks, e.g. Billings (1997), Endsley (1997) and Sheridan (1980) or on the manning level around the machines, e.g. Groover (2001).

By breaking down activities into physical and cognitive tasks, while still acknowledging the co-operation between human and technology, it can be useful to consider the task allocation as two independent reference scales relating to the two kinds of level of automation. Each of those tasks can then be ordered in seven steps, from totally manual control to fully automatic (table 18).

18(28)


Table 18 LoA-scales for computerized and mechanized tasks within manufacturing

LoA Mechanical and Equipment Information and Control

1 Totally manual - Totally manual work, no tools are used, only the users own muscle power. E.g. The users own muscle power

Totally manual - The user creates his/her own understanding for the situation, and develops his/her course of action based on his/her earlier experience and knowledge. E.g. The users earlier experience and knowledge

2 Static hand tool - Manual work with support of static tool. E.g. Screwdriver

Decision giving - The user gets information on what to do, or proposal on how the task can be achieved. E.g. Work order

3 Flexible hand tool - Manual work with support of flexible tool. E.g. Adjustable spanner

Teaching - The user gets instruction on how the task can be achieved. E.g. Checklists, manuals

4 Automated hand tool - Manual work with support of automated tool. E.g. Hydraulic bolt driver

Questioning - The technology question the execution, if the execution deviate from what the technology consider being suitable. E.g. Verification before action

5 Static machine/workstation - Automatic work by machine that is designed for a specific task. E.g. Lathe

Supervision - The technology calls for the users’ attention, and direct it to the present task. E.g. Alarms

6 Flexible machine/workstation - Automatic work by machine that can be reconfigured for different tasks. E.g. CNC-machine

Intervene - The technology takes over and corrects the action, if the executions deviate from what the technology consider being suitable. E.g. Thermostat

7 Totally automatic - Totally automatic work, the machine solve all deviations or problems that occur by it self. E.g. Autonomous systems

Totally automatic - All information and control is handled by the technology. The user is never involved. E.g. Autonomous systems

The advantage, in relation to previous taxonomies, of using the two proposed reference scales in table 18 is that both the level of physical and cognitive support can be assessed in the same taxonomy. Since the two reference scales for assessment of physical and cognitive support are independent, the level of automation for physical and cognitive support can vary depending on the need from the human. Further, by using the taxonomy presented in table 18 as a base for a measurements methodology might it be possible to estimate the potential of technology and automation in different types of human-manufacturing system.

To explain the taxonomy (table 18) an empirical example representative for the manufacturing industry will be described in the next section.

6 An industrial example of a LoA assessment The following description of a LoA assessment procedure (figure 5) is a generalized and simplified example based on our empirical experiences. It is presented to visualize the levels of automation during a defined part of office cabinet assembly in the Swedish furniture industry.

Work--station

1

Work-station

2

Work-station

3

Work-station

4

Work-station

5

Work-station

6

Officecabinetframe

Work-station

7

Plinth Rollfront

Rollfront

Shelves,doors, cases,

etc.strips

19(28)


Fig. 5 An empirical example of office cabinet assembly

6.1 Context and list of operations in the example When the cabinet enters the equipment-paced assembly line, the frame (consisting of the cabinet’s sides, top, and bottom) has already been assembled. The cabinet is also positioned upside-down to simplify the task of assembling the plinth on to the bottom of the office cabinet.

- Workstation 1: Application of glue on the base of the office cabinet

Before the plinth can be assembled, the bottom side has to be prepared by applying glue on the bottom

of the cabinet. The application of glue is done by an industrial robot, which means that the cabinet has to be positioned so that the robot can conduct the gluing operation.

- Workstation 2: Application of the plinth on base of the office cabinet

In the next step, the plinth is attached to the bottom of the cabinet by a robot in the plinth-assembly workstation, where the robot picks the right plinth from the plinth stock. Even if the assembly of the plinths works most of the time, the operator at the following workstation has to inspect the assembly. If the plinth has not been assembled correctly, the operator corrects the assembly by using a rubber hammer to force the plinth and bottom of the cabinet together. In some cases, the operator also has to correct and patch the gluing operation before the cabinet is sent to the next station.

- Workstation 3: Assembly of cabinet roll front

In the station where the application of glue and the assembly of the plinth are being checked, the roll front of the cabinet is assembled by hand. The operator picks the roll front with the right width from the roll front stock, and than inserts it into the roll front magazine on the topside of the cabinet.

- Workstation 4: Turning office cabinet from upside-down

The cabinet has to be turned for the remaining assembly. This is done by a robot, since the task is considered too heavy to be permanently allocated to humans. However, in case the robot is malfunctioning, the task is still performed by humans.

- Work station 5: Assembly of shelves

After the assembly of the roll front, the cabinet is turned and shelves are assembled into the cabinet. The shelves are picked from the shelf stock and assembled into the office cabinet with support of hydraulic screwdrivers.

- Workstation 6: Assembly of cabinet roll front strips

After the shelves have been assembled, the next step is to assemble the strips to guide the roll front. This is done by a robot that picks the strips from the strip magazine and nails it on to the cabinet.

- Workstation 7: Equipping the office cabinet

The final step of assembly is to equip the office cabinet according to the costumers’ demands. The operators at the workstation get their instructions from the attached order form on how the cabinet will

20(28)


be equipped. Based on the order form, the operator equips the cabinet with shelves, doors, cases, etc. The assembly of the different equipments is done by hydraulic screwdrivers and, since the torque is of importance, the hydraulic screwdrivers have been programmed to tighten the screws to a pre-defined torque.

The manufacturing processes described represent the basis for a LoA assessment described in the next section.

6.2 Assessment of LoA Based on the task descriptions in the empirical example in 6.1 and the LoA taxonomy in table 18, the LoA for Mechanical/Equipment and Information/Control can be estimated for each of each workstation. This is presented in table 19.

Table 19 Estimated LoA from the empirical example

Workstation Task description LoA –

Mech. & Equip.

LoA –

Info. & Contr.

1 Application of glue on base of the office cabinet 6 5

2 Application of plinth on base of the office cabinet 4 5

3 Assembly of cabinet roll front 1 2

4 Turning office cabinet from upside-down 5 5

5 Assembly of shelves 4 3

6 Assembly of cabinet roll front strips 6 5

7 Equipping the office cabinet 4 4

Further, by identifying the sub-tasks based on the main task for each of the seven workstations the allocation of tasks between human and automation can be visualized. E.g. workstation 2 that has the main task of assemble the plinth on to the cabinet has a number of physical and cognitive sub-tasks that has to be allocated between the human and the technology, as seen in table 20.

Table 20 The allocation of physical and cognitive tasks between human and automation in workstation 2

Workstation 2 Physical task Cognitive task

Human Correct and/or conduct the assembly if the plinth has not been assembled correctly

Supervise the industrial robot

Inspect the assembly

Automation Automatic positioning of the cabinet before conducting the automated task

Pick the plinth from the plinth stock

Automatic assembly of the plinth on the bottom of the cabinet

Decide which plinth that is going to be picked for assembly based on the control program

Decide where to apply the plinth on cabinet based on the control program

21(28)


The physical task for the automation is in the first step is to position the cabinet before the plinth can be assembled. According can to the reference scale in table 18 for the physical task can the LoA be assesses as LoAM&E =5, since the workstation is designed for the specific task of positioning the cabinet. The same task can accord the cognitive reference scale in table 18 be assessed as LoAI&C =6, since automation intervene and corrects the positioning of the cabinet if it is necessary. Next task for the automated workstation is to pick the plinth from the plinth stock. According can to the reference scale can the physical task be assessed as LoAM&E =6, since this part of the automated system is design to both pick the shelf and assemble it on to the cabinet. However, the robotic assembly of the plinth is not always successful which means that humans, with or without support of a rubber hammer, sometimes do the assembly task. This means that the task is sometimes conducted at LoAM&E =1 (by hand without the rubber hammer) and at LoAM&E =2 (with support of the rubber hammer). By using the cognitive reference scale on the same example of picking the plinth from the stock and assembly, it can be determined that the cognitive LoA corresponds to LoAI&C=5, since the robot solution calls for the operators’ attention in case of malfunctioning. On the other hand, if the task is conducted manually (with or without the rubber hammer), the cognitive LoA assessment would be LoAI&C =1, since no support is given. The understanding for achieving the task is based on the operator’s own experience and knowledge.

As noted by Price (1985), some tasks are neither possible to allocate to the humans, nor to the technological system. As seen in figure 1 by Price (1985), there is a maximum extent of the human capability to handle some tasks and in the same way has the technological system its limitations. LoA is also not static over time, but can be modified and adapt to the changing conditions for the manufacturing context. It would thereby be suitable to view the level of automation in terms of what are the limits of automation, or a relevant maximum and minimum for a particular machine or cell. Based on the empirical example in figure 5, the maximum and minimum LoA for each workstation can be estimated (table 21). Williams (1999) came to a similar conclusion, stating that there are identifiable sets of limits in manufacturing to what is humanly accomplishable. He also suggested that there are definable limits to automation in mission fulfillment tasks as well as in information and control tasks (figure 2).

Our empirical example in chapter 6.1, based on earlier case studies show that for the practical manufacturing situation a maximum or minimal level of automation in many cases may prove unrealistic in terms of investment cost etc. In the empirical case presented in chapter 6.1, an assessment of LoA based on e.g. interviews of the personnel will not reveal the “real” maximum of the LoA, instead a “relevant” LoA will result from the assessment. In our empirical example, the relevant LoA-levels are indicated (table 21).

22(28)


Table 21 Relevant maximum and minimum of LoA for a particular machine or cell

Mechanical and Equipment

LoA Workstn. 1 Workstn. 2 Workstn. 3 Workstn. 4 Workstn. 5 Workstn. 6 Workstn. 7

7

6 Relevant max Relevant max Relevant max Relevant max

5 Relevant min

4 Relevant max Relevant min Relevant max

3

2 Relevant min Relevant max Relevant min

1 Relevant min Relevant min Relevant min

Information and Control

LoA Workstn. 1 Workstn. 2 Workstn. 3 Workstn. 4 Workstn. 5 Workstn. 6 Workstn. 7

7

6

5 Relevant max Relevant max Relevant max Relevant max Relevant max Relevant max

4 Relevant min Relevant min

3 Relevant max Relevant min



In the two reference scales presented by table 21, two assessed levels of automation are marked for each workstation:

The relevant maximum LoA describes a possible technical situation. It is considered when the investment cost not exceeds the value of the advantages achieved through automation. The consequences of a possible stop of manufacturing or breakdown also need to be taken into account. It is important not to automate more than what is necessary or justified, since investment costs rise rapidly. Another disadvantage of high LoA is that the system may become inflexible and rigid.

A relevant minimum of LoA can be favorable when work can be carried out at a suitable pace at an acceptable cost without jeopardizing work environment for people involved in the process.

The assessment of LoA makes it possible to choose a suitable LoA for each manufacturing process, and moreover, to identify a suitable span of LoA.

The existing and potential level of automation for each workstation is located somewhere in the range between maximum and minimum LoA, indicated by the grey area in Table 21. Thus, it is now possible to analyze the present situation and to judge possible future automation potentials.

7 DISCUSSION We have presented a review of recent research on “level of automation” concept across several fields of research and industries. Our analysis suggests a possibility to apply two independent reference scales for assessment of automation, one for physical and one for cognitive tasks.

Further, as noted by Billings (1997), Endsley (1997), Endsley et al. (1997) and Satchell (1998), much of the previous research approaches in areas such as aviation and telerobotics have been primarily

23(28)


concerned with cognitive automation. The aim of the cognitive automation has often been aiming to speed up information flow and to provide decision support, thus supporting the operator in monitoring the situation. However, from a manufacturing perspective it is important to recognize in line with Williams (1999) that manufacturing processes in general consist of both physical tasks, e.g. replacement or support of human muscle power as well as cognitive tasks e.g. carrying out control and information tasks. Further, this approach is also in line with Bright (1958) and Marsh and Mannari (1981), who concluded that the manufacturing context consists of a mix of both mechanized (physical) and computerized (cognitive) tasks.

This indicates a potential in combining results from multiple fields of research to arrive at a relevant description of the actual automation situation in manufacturing contexts.

The review shows profound support for treating automation as a continuum of automation levels, from the lowest level of fully manual performance (based on the capabilities of the human), to the highest level of full automation (without any human involvement). Thereby it can be argued with support of earlier research by Endsley (1997), Inagaki (1993), Parasuraman et al. (2000) and Sheridan (1992), that automation is not all or nothing.

By breaking down the activities in line with Williams (1999) into physical and cognitive tasks, while still acknowledging the co-operation between the human and the technical system, it can be relevant to consider the allocation of tasks as two independent reference scales of LoA, relating to the cognitive and physical tasks. Each of those can then be ordered in seven steps, ranging from totally manual to fully automatic.

The advantage of using the two proposed reference scales, in relation to previous taxonomies of LoA by e.g. Billings (1997), Endsley (1997) and Sheridan (1992), is that both the level of physical and cognitive support can be assessed in the same taxonomy. Since the two reference scales for the assessment of physical and cognitive support are independent, the level of automation of physical and cognitive tasks can vary depending on the individual human needs.

However, as noted by Price (1985) and Williams (1999), some tasks are not possible to allocate to the human or to the technological system. As showed by Williams (1999), there is an maximum and minimum extent of the human as well as the technological capability to handle different tasks. Based on the model of limitations by Williams (1999) and the cases studies within this thesis, it can be noted that in term of e.g. investment cost, it is unrealistic for practical industrial situations to consider the entire span of each of the two reference scales. Therefore, and in line with Williams (1999) it is not relevant to discuss the assessment of LoA in terms of absolute maximum and minimum levels, but instead in terms of relevant levels of automation for each of the two reference scales.

8 CONCLUSION The objective of this paper was to take a systematic approach towards changing levels of automation in semi-automated manufacturing systems. From a literature review performed across disciplines and industrial sectors it was concluded that levels of automation within manufacturing generally does not constitute a single step from manual to fully automated tasks. Instead, we have identified two independent “continuums”, one for physical tasks and one for cognitive tasks. To enable assessment of the two levels of automation, two seven-step references scales where suggested and delimited by lower and upper boundaries. By assessing a) the relevant maximum, b) the relevant minimum, and c) the actual levels of automation for each work-task or workstation, the potential residing in increased or decreased automation can be identified. The resulting assessment can be used to increase total manufacturing system performance.

24(28)


ACKNOWLEDGEMENTS

This paper contains results from the DYNAMO project. The authors are deeply grateful to the Swedish Fund for Strategic Research, specifically its ProViking programme, for funding the project. We would also like to thank colleagues and industries involved in the multiple parts of the DYNAMO project.

REFERENCES

Allum, A, (1998), Automation requirement matching through TMC. Assembly Automation, Vol.18, No. 2, pp. 107-111.

Anderson, RJ, (1996), Autonomous, Teleoperated, and Shared Control of Robot Systems. [in] Proceedings of International Conference on Robotics and Automation, April 1996, Minneapolis, Minnesota, USA.

Bainbridge, L, (1982), Ironies of Automation. Automatica, Vol.19, pp.775-779.

Billings, CE, (1997), Aviation Automation: The Search for a Human-Centered Approach. Lawrence Erlbaum Associates, Mahwah, New Jersey, USA.

Boothroyd, G, (2005), Assembly Automation and Product Design. CRC Press, New York, USA.

Bright, JR, (1958), Automation and management. Maxwell Reprint Company, Boston, USA.

Card, SK, Moran, TP, and Newell, A, (1983), The Psychology of Human-computer Interaction Lawrence Erlbaum Associates London, England.

Chiantella, N, (1982), Achieving Integrated Automation Through Computer Networks. SMA/CASA Computer Integrated Manufacturing Series, Vol.1, No. 2, pp. 2-21.

Connors, MM, (1998), Teaming Humans and Automated Systems in Safely Engineered Environments. Life Support & Biosphere Science, Vol.5, pp.453-460.

Draper, JV, (1995), Teleoperators for Advanced Manufacturing: Applications and human Factors Challenges. The International Journal of Human Factors in Manufacturing, Vol. 5, No. 1, pp53-85.

Drucker, PF, (1999), Management Challenges for the 21st Century. Butterworth Heinemann, New York, USA.

Duncheon, C, (2002), Product miniaturization requires automation - but with a strategy. Assembly Automation, Vol. 22, No. 1, pp. 16-20.

Encyclopedia Britannica. (2006), Encyclopedia Britannica. Retrieved December 28, 2007, from http://search.eb.com/eb/article-9109393

Endsley, MR, (1997), Level of automation: Integrating humans and automated systems. [in] Proceedings of Proceedings of the 1997 41st Annual Meeting of the Human Factors and Ergonomics Society. Part 1 (of 2), Albuquerque, NM, USA. Human Factors and Ergonomics Society, Inc., Santa Monica, CA, USA, 200-204.

Endsley, MR, and Kaber, DB, (1999), Level of automation effects on performance, situation awareness and workload in a dynamic control task. Ergonomics, Vol. 42, No. 3, pp. 462-492.

Endsley, MR, and Kiris, EO, (1995), Out-of-the-loop performance problem and level of control in automation. Human Factors, Vol.37, No. 2, pp. 381-394.

25(28)


Endsley, MR, Onal, E, and Kaber, DB, (1997), The Impact of intermediate levels of automation on situation awareness and performance in dynamic control systems. [in] Proceedings of Proceedings of the 1997 IEEE 6th Conference on Human Factors and Power Plants, Jun 8-13 1997, Orlando, FL, USA.

Fitts, PM, (1951), Human Engineering for an Effective Air-navigation and Traffic-control system. London: National Reseach Council.

Frohm, J, Karsvall, A, Hassnert, M, Stahre, J, and Hollnagel, E, (2003), System perspective on task allocation and automation design. [in] Stahre, J (eds.). Proceedings of 8th symposium on automated systems based on human skills and knowledge, Göteborg, Sweden.

Frohm, J, Lindström, V, and Bellgran, M, (2005), A model for parallel levels of automation within manufacturing. [in] Pasquino, R (eds.). Proceedings of the 18th International Conference on Production Research, Fisciano, Italy.

Groover, MP, (2001), Automation, production systems, and computer-integrated manufacturing. Prentice Hall, Upper Saddle River, N.J., USA.

Gullander, P, (1999), On Reference Architectures for Development of Flexible Cell Control Systems. Doctoral thesis, Chalmers tekniska högskola, Göteborg, Sweden.

Hoffman, RR, Feltovich, PJ, Ford, KM, Woods, DD, Klein, G, and Feltovich, A, (2002), A Rose by Any Other Name... Would Probably Be Given an Acronym. IEEE Intelligent Systems and Their Applications, Vol. 17, No. 4, pp. 72-80.

Inagaki, T, (1993), Situation-adaptive degree of automation for system safety. [in] Proceedings of 2nd IEEE International Workshop on Robot and Human Communication, Tokyo, Japan.

Jordan, N, (1963), Allocation of functions between man and machines in automated systems. Journal of Applied Psychology, Vol. 47, No. 3, pp. 161-165.

Kaber, DB, Onal, E, and Endsley, MR, (2000), Design of automation for telerobots and the effect on performance, operator situation awareness, and subjective workload. Human Factors and Ergonomics In Manufacturing, Vol. 10, No. 4, pp. 409-430.

Kern, H, and Schumann, M, (1985), Industriearbeit und Arbeiterbewusstsein : eine empirische Untersuchung über den Einfluss der aktuellen technischen Entwicklung auf die industrielle Arbeit und das Arbeiterbewusstsein. Suhrkamp, Frankfurt am Main, Germany.

Kilbo, P, (2001), Produktionslogik för företag efter en tredje industriell revolution (IVF-report No. 01003). Mölndal: IVF Industrial Research and Development Corporation, Swden.

Kotha, S, and Orne, DL, (1989), Generic Manufacturing Strategies: A Conceptual Synthesis. Strategic Management Journal, Vol. 10, No. 3, pp. 211-231.

Lindström, V, Frohm, J, and Bellgran, M, (2005), Developing a methodology based on Value Stream Mapping for the Measurment of Automation Levels in Production Systems. [in] Proceedings of the 3rd International Conference on Reconfigurable Manufacturing, May 10-12, Ann Arbor, Michigan, USA.

Lorenz, B, Nocera, FD, and Parasuraman, R, (2001), HUMAN PERFORMANCE DURING SIMULATED SPACE OPERATIONS UNDER VARIED LEVELS OF SYSTEM AUTONOMY. [in] Proceedings of Bioastronautics Investigators' Workshop, January 17–19, 2001, Galveston, Texas, USA.

26(28)


Marsh, RM, and Mannari, H, (1981), Technology and Size as Determinants of the Organizational Structure of Japanese Factories. Administrative Science Quarterly, Vol. 26, No. 1, pp. 33-57.

Milgram, P, Rastogi, A, and Grodski, J, (1995), Telerobotic Control Using Augmented Reality. [in] Proceedings of 4th IEEE International Workshop on Robot and Human Communication (ROMAN '95), 5-7 July, 1995, Tokyo, Japan.

Mital, A, (1997), What role for humans in computer integrated manufacturing? International Journal of Computer Integrated Manufacturing, Vol. 10, No. 1-4, pp. 190-198.

Mital, A, and Pennathur, A, (2004), Advanced technologies and humans in manufacturing workplaces: an interdependent relationship. International Journal of Industrial Ergonomics, Vol. 33, No. 4, pp. 295-312.

Parasuraman, R, (2000), Designing automation for human use: empirical studies and quantitative models. Ergonomics, Vol. 43, No. 7, pp. 931-991.

Parasuraman, R, Sheridan, TB, and Wickens, CD, (2000), A model for types and levels of human interaction with automation. IEEE transactions on system, man, and cybernetics - Part A: Systems and humans, Vol 30, No. 3, pp. 286-296.

Price, HE, (1985), The Allocation of Functions in Systems. Human Factors, Vol. 27, No. 1, pp. 33-45.

Reveliotis, SA, (1999), Production planning and control in flexibly automated manufacturing systems: current status and future requirements. [in] Proceedings of the IEEE International Conference on Robotics and Automation, Marriott Hotel, Renaissance Center, Detroit, Michigan, USA.

Rouse, WB, and Rouse, SH, (1983), Analysis and classification of human error IEEE Transactions on Systems, Man, and Cybernetics, SMC-13, pp. 539-549.

Ruff, HA, Narayanan, S, and Draper, MH, (2002), Human Interaction with Levels of Automation and Decision-Aid Fidelity in the Supervisory Control of Multiple Simulated Unmanned Air Vehicle. Presence, Vol. 11, No. 4, pp. 335-351.

Satchell, P, (1998), Innovation and Automation. Ashgate, London, England.

Schwartz, R, Kuchar, J, Hastings, D, Deyst, J, and Kolitz, S, (1996), A probabilistic model for the determination of the effects of automation of satellite operations on life cycle costs. [in] (Proceedings of SpaceOps '96, Munich, Germany.

Sheridan, TB, (1980), Computer Control and Human Alienation. Technology Review, Vol. 83, No. 1, pp. 60-73.

Sheridan, TB, (1992), Telerobotics, Automation, and Human Supervisory Control. MIT Press, Cambridge, MA, USA.

Sheridan, TB, (2002), Humans and Automation: System Design and Research Issues. John Wiley & Sons, Inc., Santa Monica, USA.

Teknisk framsyn, (2000), Rapport från Teknisk framsyn - panel:4 Produktionssystem. Stockholm: The Royal Swedish Academy of Engineering Sciences (IAV): Teknisk framsyn. Stockholm, Sweden.

The Oxfords English Dictionary. (2006), The Oxfords English Dictionary. Retrieved October 8, 2007, from http://dictionary.oed.com/

27(28)


Wickens, CD, and Hollands, JG, (2000), Engineering psychology and human performance. Prentice Hall, Upper Saddle River, NJ., USA.

Williams, TJ, (1999), PERA AND GERAM:. ESTABLISHMENT OF THE PLACE OF THE. HUMAN IN ENTERPRISE INTEGRATION. [in] Proceedings of IFAC Congress, Beijing, China.

Ylipää, T, (2000), High-Reliability Manufacturing Systems. Chalmers University of Technology, Göteborg, Sweden.

Youtie, J, Shapira, P, Urmanbetova, A, and Wang, J, (2004), Advanced Technology and the Future of U.S. Manufacturing. [in] Shapira, P, Youtie, J and Urmanbetova, A (eds.). Proceedings of Georgia Tech research and policy workshop, Georgia Tech Economic Development Institute, Atlanta, GA, USA.

28(28)

PAPER VI

Granell, V., Frohm, J. Dencker, K. and Bruch, J. (2007). Validation of the Dynamo Methodology for Measuring and Assessing Levels of Automation. In Proceedings of the Swedish Production Symposium, Göteborg, Sweden, August 28-30, 2007.

VALIDATION OF THE DYNAMO METHODOLOGY FOR MEASURING AND ASSESSING LEVELS OF AUTOMATION

Granell, V. 1, 2 *, Frohm, J. 2, Bruch, J. 1, 2, Dencker, K. 3, 4

1 Department of Industrial Engineering and Management, School of Engineering, Jönköping University, Sweden 2 Department of Product and Production Development, Chalmers University in Gothenburg, Sweden

3 IVF Industrial Research and Development Corporation, Stockholm, Sweden 4 Department of Production Engineering, Royal Institute of Technology, Sweden

* Corresponding author


Abstract: The purpose of measuring Levels of Automation (LoA) is to find an appropriate level for balancing the system for achieving high effectiveness. Results from an interview study with manufacturing experts working in the Swedish industry showed that measurement of automation levels is not common, mainly because of unfulfilled prerequisites and barriers to be overcome. This paper forms a platform for fulfilling some of the expressed prerequisites and overcoming some of the barriers to measure and assess LoA in manufacturing value flows. The paper provides a deeper knowledge on how to measure and assess the automation used in an industrial setting.

Keywords: Levels of Automation, measurement, validation, methodology

1. INTRODUCTION AND PROBLEM AREA

Results from an interview study with manufacturing experts working in the Swedish industry in 2004 showed that Swedish manufacturing firms are not acquainted with the term ‘level of automation’ (LoA). Moreover, measurement of automation levels is not common, mainly because of unfulfilled prerequisites and barriers to be overcome (Granell, 2007). The purpose of measuring and assessing LoA is to find an appropriate and right level for balancing the system for achieving high effectiveness of it. For the industry, the benefits would be maintained productivity through diminishing disruptions. Disruptions in the industry appear because of too complex systems, i.e. too high LoA, or because of sick leaves where the cause could be a too low level of automation. Historically, researchers have defined a similar concept to LoA, called levels of mechanization (Bright, 1958; Kern and Schumann, 1985; Kotha and Orne, 1989). However, the approach presented by these researchers has been focusing on the machines rather on the support given to the human. On the other hand, levels of automation may be seen as a different approach, namely as the allocation of a task and thereby as an interaction between human and technology (Sheridan, 1980; Satchell, 1998; Parasuraman et al,

2000). The concept Levels of Automation is defined in this paper as (Frohm, et al. 2007):

‘The allocation of physical and cognitive tasks between humans and technology, described as a continuum ranging from totally manual to totally automatic’

This means that when measuring LoA, the focus is on the tasks performed by the human, the technology, or both. Measurement of LoA should be understood as judging the level of interaction between the human and two types of technology; mechanization and computerization (e.g. information systems). The tasks are judged by the type of interaction between the human and technology. In forming a methodology for judging (here: measuring) and assessing LoA, it is possible to analyse if the current level of automation is too low, too high, too static etc. depending on the requirements of the company where the measurement takes place. This paper forms a platform for fulfilling some of the expressed prerequisites in earlier case studies and overcoming some of the barriers to measuring LoA in manufacturing value flows by presenting a validated methodology. Purpose of the paper is to describe the Dynamo methodology, which comprises measurement, assessment, and analysis of LoA, and the validation of the methodology. The use of the

expression methodology should in this paper be understood as ‘a set of principles of method’ (Checkland, 1999). A methodology contains elements of both ‘what’ and ‘how’ and this description corresponds with the LoA measurement methodology presented here. Further, the development of the methodology has been done in the DYNAMO project (Dynamic Levels of Automation) and the short name for the methodology is therefore the Dynamo methodology.

The following research questions are answered within this paper:

• What steps does the Dynamo methodology consist of?

• How was the validation conducted?

• Which problems and deviations were discovered by the validation of the Dynamo methodology?

The validation of the methodology presented in this paper was carried out in collaboration between researchers from the two research projects DYNAMO and ProAct. In conducting the validation as collaboration between these two projects, a transformation of knowledge could be possible between the projects with the purpose to develop the Dynamo methodology further.

2. BACKGROUND: DEVELOPMENT OF THE DYNAMO METHODOLOGY

The implementation and development of the LoA measurement methodology was done during the years 2004 to 2007 in seven case studies (Case study A to G in table 1) using the single case study method in sequence (Yin, 2003). Based on the finding from the first six case studies, the method was rewritten into the final Dynamo methodology, which was tested in case study G, with the purpose to validate the Dynamo methodology. The methods used within the case studies comprise of interviews, observation and participant observation. Moreover, a review of academic literature within the fields of human factors and manufacturing was done to find out how levels of automation have been measured earlier. From the review it could be concluded that the usefulness of automation is very much dependent on finding appropriate distribution of tasks between the human and the technical system (Frohm, et al., 2007). Since production systems also can be described as flows of materials and information (Sheridan, 2002), the lean production tool Value Stream Mapping (VSM) was used as an starting point for measuring LoA in those production flows (Lindström et al., 2005) together with reference scales, which are used for judgement of observed LoA, see example table 2.

Table 1. Case studies for development of the Dynamo methodology

Case study

Study period Unit of analysis

Methods

AOctober-December 2004

Manuf. process

Interviews Observation

BJanuary-May 2005

Manuf. process

Interviews Participant observation

C May-July 2005 Manuf. process

Observation

DMarch 2005-October 2006

Manuf. process


ENovember-December 2005

Manuf. process

Interviews

FOctober 2005-June 2006

Manuf. process and information process


GJanuary-March 2007

Validation of the Dynamo methodology

Observation

However, as levels of automation focus on the task in the working situation between the human operator and the technology, the methodology separates the description of the material and information flows, and the task description. Based on the results from the first six case studies (A-F), the Dynamo methodology was presented (see figure 1), containing methods for measuring, assessing and analysing LoA.

Table 2. The Dynamo reference scales for LoA (Frohm et al., 2007)

LoA Mechanical Information

1 Totally manual Totally manual

2 Static hand tool Decision giving

3 Flexible hand tool Teaching

4 Automated hand tool Questioning

5 Static machine/workstation

Supervision

6 Flexible machine/workstation

Intervene

7 Totally automatic Totally automatic

3. THE DYNAMO METHODOLOGY

The Dynamo methodology, as described before the validation, consists of eight steps (figure 1). Each of the sections below describes step 1 to 8 in the Dynamo methodology.

1. Plan ahead before the measurement2. On site, start with a pre-study to identify the process3. Visualize and document the production flow4. Identify the main task for each section/cell5. Identify sub-tasks for each section/cell6. Measure LoA7. Assess LoA, set relevant max. and min. levels8. Analysis of results

1. Plan ahead before the measurement2. On site, start with a pre-study to identify the process3. Visualize and document the production flow4. Identify the main task for each section/cell5. Identify sub-tasks for each section/cell6. Measure LoA7. Assess LoA, set relevant max. and min. levels8. Analysis of results

Fig. 1. The Dynamo methodology in eight steps

3.1 Step 1

The first step in the Dynamo methodology, which is conducted off-site, is to discuss the goal and purpose of the measurement, as well as delimitations of the production flow with the company. For example, the goal of the LoA measurement can be to maximize or minimize the level of automation within the flow.

3.2 Step 2

The second step, which is conducted on-site, is to carry out a pre-study to identify and document the purpose of the production flow and where it starts and ends, documented in form no. 1. Also the number of products and variants produced within the production flow are identified and documented, as well as work organization and the purpose of the machines and humans.

3.3 Step 3

After the basic data for the production flow has been documented and understood, the next step is to visualize the production flow. This is done by “walk the process” and defining which sections/cells the production flow consists of. Data such as the number of products and variants that pass through the section/cell or buffer are documented in form no. 2, as well as the physical and cognitive tasks that have been allocated to the technology or to the human. Also the number of operators that are allocated to the section/cell or buffer is documented, and if the operator is responsible for more then one section/cell or buffer.

3.4 Step 4

After the production flow has been understood, visualized and documented, the identification of the main task is done. Based on the data collected in step 2 and 3, the main task of each section/cell or buffer is identified and documented (e.g. assembly of X electronic components on a circuit card).

3.5 Step 5

The identification of the sub-tasks is done once the main task is identified. The identification of sub-tasks is done by observing how the main task is achieved, which is done by breaking down the task until it reaches a level of operations, where only the human or the technology can be responsible for achieving the task. In support of the observation, the operation instructions are used as a starting point and explanation of what is going to be observed. By using the documentation of the task from the company, the measurement crew can easily identify the sub-tasks and deviations from how the task is intend to be done. To simplify and structure the down breaking of the main tasks in to sub-tasks, Hierarchical Task Analysis (HTA) is used. The HTA is a method for description of activities under analysis in terms of a hierarchy of goals, sub-goals, operations, and plans (Stanton et al, 2005).

3.6 Step 6

After the tasks have been broken down and identified, the LoA is judged based on the two reference scales for mechanical and information LoA (see table 2) and an observation of the sub-tasks performed and described in the HTA in step 5. The judged LoA for each task is than based on how the task is conducted, and what type of interaction that is observed for fulfilling the task. The type of interaction for each sub-task is mapped against the reference scale. The measurement data from the observation is documented in form no. 3. By observing more than one operator it is possible to increase the strength of the observations, and also to identify if tasks are conducted with different LoA:s depending on which operator that conducts the task. If the case is conducted under different LoA:s, the task can then be said to be a dynamic LoA.

3.7 Step 7

After the measurement of the observed LoA-values have been judged, the observer together with the operator or/and production technician on-site estimates the relevant maximum and minimum LoA for each measured task. By using respondents that has an understanding on how the tasks that has been observed is conducted, a good estimation on the relevant maximum and minimum can be assessed during the discussion. The date is documented in form no. 4.

3.8 Step 8

The final step of the Dynamo measurement methodology is to analyse the collected data from the LoA-measurement on-site (form no. 3), with the assessed data on relevant maximum and minimum of LoA (form no. 4). The analysis starts with placing the the LoA value from the observed LoA value as a black dot in the Mechanical-Information-LoA diagram for all documented sub-tasks. The LoA-

measurement value shows the actual flexibility/dynamic of the automated task. By drawing the boarders for the relevant maximum and minimum of each LoA, a potential area of automation of the task is given (see figure 2). Depending on the purpose of the LoA-measurement, an analysis of the Mechanical-Information-LoA diagram indicates how to take advantage of the automation potential. For example, if the purpose is to maximize the automation of different reasons by the company, the task is then to move the actual LoA to the upper right corner of the max-min LoA square in figure 2.

Fig. 2: Analysis of LoA in the Mechanical-Information-LoA diagram

4. METHOD: VALIDATION OF THE DYNAMO METHODOLOGY

The case study for validation took place at a company not participating in any of the previous described case studies. The purpose with the validation was to perform the whole Dynamo methodology at an industrial company which had not been participating in the development of the methodology. By stepwise performing the instructions of the Dynamo methodology and carry out each activity the research group intended to identify:

• Problems with understanding the instruction of the methodology

• Problems with logic sequences of the methodology

• Efficiency potential within the methodology

A reliable methodology is one that, when used to measure LoA of a work task, repeatedly and consistently gives the same findings. Methods for determining methodology and scale reliability and validity often rely on statistic procedures requiring large sample size (Yin, 2003). This has not yet been fulfilled but we can establish the face and construct validity of the methodology by the following case study. Further construct validity will be established

for the Dynamo methodology during the ProAct project (Dencker et al, 2007), with the collection of a large sample of work task ratings.

Three researchers including one of the researchers from the methodology development phase set up the methodology for validation. The validation was conducted in the following sequence, see figure 3:

1. Studying the Dynamo methodology.

2. Introducing the methodology to the participating company where the validation took place.

3. Conducting the Dynamo methodology, step 1 to 7 (the analysis was not realized in this case)

4. Evaluation of the measurement and the results.

5. Developing an improved methodology

6. Documentation and reporting

The above described sequence forms the validation process described next. The validation process of the Dynamo methodology contains both planning, testing the methodology, evaluation, development of an improved methodology, and finally documentation.

4.1 Studying the Dynamo methodology

At a first meeting the Dynamo methodology was introduced to two of the researchers in the ProAct project by a third researcher in the Dynamo project with the purpose to study the methodology. The researcher in the Dynamo project had participated in the development of the methodology. The meeting was also used to conduct step one of the Dynamo methodology, ‘Plan ahead before the measurement’. Further, the first visit to the company was planned and arrangements were done.

4.2 Introducing the methodology

When visiting the company for the first time, another researcher from the methodology development phase joined the group with the purpose to observe the realization of the methodology and to document the validation. Moreover, the Dynamo methodology was introduced to two company representatives.

4.3 Conducting the methodology

After the introduction of the methodology the researchers started with conducting the methodology in a first visit at the company. The researchers continued with the second step of the Dynamo methodology ‘On site, start with a pre-study to identify the process’ by using a pre-designed form (form no. 1). The research group also used pre-designed forms for step 3 ‘Visualize and document the production flow’ (form no. 2) and step 4 ‘Identify the main task for each section/cell’ (form no. 2). Step 5 ‘Identify sub-tasks for each section/cell’ could not be completed due to time restrictions at this point. The researchers had to leave the assembly hall without having observed the assembly of the product

7

1

1 7

Mech. LoA

Info. LoA

Rel. min

Rel. max

Rel. min

Rel. max

once from the start to the end. Yet the researchers decided to conduct step 6 ‘Measure LoA’ with the collected material. The group left therefore the assembly hall to improve communication possibilities. However, the observation notes were incomplete and therefore the researcher had to return to the assembly station and ask complementary questions. A following discussion summarizing the day resulted in a first deviation of the validation. Deviations, problems and comments during the validation process resulted in a change of the instructions in the Dynamo methodology. The change decided was to measure LoA preliminary off-site (step 1) by conducting the judgement of LoA based on the operation instruction. The advantage of doing the preliminary LoA measurement off-site is that the measurement crew can get an understanding of what tasks that are going to be observed. Observations on-site may then be easier to do and there may be more time to detect deviations from the instructions. To be able to complete the validation and include the changes in the preparation the researcher decided to return to the company for a second visit.

At the second visit the company was dealing with rearrangement of the assembly layout of the studied product. Therefore, it was impossible for the researchers to continue with the Dynamo methodology. However, the time was used to discuss the Dynamo methodology.

When returning to the company the third time the work places had been rearranged. The rearrangement included both movements in the manufacturing hall as well as a new layout of the two workplaces. Rearrangement of the assembly stations resulted in a U shaped layout, instead of the former arrangement of the two workplaces arranged in a line. Moreover, the operator responsible for preparing the right kit needed for product assembly was now separated from the operator who assembled the product. Since the operator was still doing the same assembly tasks the LoA measurement data was not changed and/ or influenced in spite of changes in the assembly layout. The researchers received the operation instruction for the cell and the instruction had not been affected by the rearrangement. The researchers could therefore continue to observe the subtasks (step 5) with the assistance of the operation instruction. Further, step 6 ‘Measure LoA’ and 7 ‘Asses LoA, set relevant maximum and minimum levels’ were performed.

4.4 Evaluation and improvement of the methodology

Evaluation of the different steps in the methodology was done by discussing deviations and possible improvements from the original methodology. Different alternatives were discussed and an improved methodology was presented, following the evaluation. However, the overall methodology presented in figure 1 was not changed.

Studying the methodology

Introducing the methodologyto the company

Visit 1

Performing the methodology, validation

Visit 2 Visit 3

Evaluation of the methodology

Documentation and reporting

Improved methodology

Studying the methodology

Introducing the methodologyto the company

Visit 1

Performing the methodology, validation

Visit 2 Visit 3

Evaluation of the methodology

Documentation and reporting

Improved methodology

Fig. 3. Validation process

4.5 Documentation and reporting

The documentation of the validation was primarily done by the researcher observing the work done by the other three researchers, who were performing the LoA measurement. Deviations, problems, and comments were captured for activities in each step of the Dynamo methodology and those are presented in the result chapter, see below. Reporting of results was done in internal project reports and conference papers.

5. RESULT OF THE VALIDATION

A discussion of deviations, problems and the comments given during the validation resulted in changes of the Dynamo methodology and its instructions. For example, more activities were moved to step 1, the off-site planning of the measurement. Another major change was also to measure LoA (step 6) of the activities described in the assembly instruction, and to capture LoA both of the operational instruction and observed tasks in the workplace. Moreover, additional forms were added to support the accomplishment of the LoA measurement, steps 1 to 8. The final step 8 in the Dynamo methodology was not validated because the company where the validation took place did not have any purpose regarding LoA strategy. The purpose of the LoA measurement is a basis for analysing observed LoA, relevant maximum and

minimum levels. The company may have one of four different scenarios. The LoA scenario chosen by the company set prerequisites for analyzing consequences in step 8 of the methodology. For deviations and problems detected during the validation, see table 3. Some of the comments are case-dependent issues, e.g. difficulty to follow operation instruction, which did not resulted in any change of the methodology.

Table 3 Deviations found in the validation

Step in the measurement methodology

Deviations, problems, comment

Loose-leaf system should be avoided A run-through should be used when new or untrained people that make the measurement

1. Off-site planning

Operation instruction should be collected for the chosen flow before on site visits Complex flows with long lead times take long time; an estimation of the time for realization should be done Data can vary depending on the respondent. Some data needs to be answered of operative people, like a production engineer

2. Identification of process

Operators should be informed about the measurement and documentation method. Video documentation is an advantage Discussion about which sort of symbols and method to use. Visualize if operation instruction exist or not, form 2 It is easy to forget the cognitive support tools, form 2

3. Flow visualization anddocumentation

It is easy to forget the cognitive tasks, form 2

4. Identification of main task

According to the methodology, both purpose and task should be documented Time constraints did that the observed flow was shortened. The documentation is made both by observing the tasks done and by looking in the operation instructions by the workplace. To use the operation instructions to identify the sub-tasks could be helpful

5. Identify sub-tasks

It would be beneficial to use video

Discussion about which figure to measure; either real (‘used’) information support or the level suggested by the operation instruction (‘available’)

Difficulty to follow operation instruction by observation

6. Measure LoA

Tasks in operation instruction are not accomplished Go through measurement method and inform the person about the taxonomy

7. Assess LoA Conduct a dialogue with the support of the taxonomy and observation

8. Analysis No validation

The experiences from validating the LoA measurement and assessment showed that by moving more activities to the preparation and planning step, step 1, the methodology would benefit in efficiency. Moreover, by informing operative personnel about the measurement in step 2 will increase motivation and may be advantageous by the actual LoA measurement in step 6. Informing about the measurement method will also make the dialogue easier in step 7 when assessing LoA. Documentation of sub-tasks is better when done with a video recorder. One experience from the validation is that the observation and LoA measurement could be very time consuming, especially if there is bad conformity with operation instruction. Both estimating LoA based on the operation instruction off-site as well as measuring LoA by observation in step 6 will make the data broader in the analysis in step 8. Setting relevant maximum and minimum levels is done by using dialogue and the person involved in this step should be informed about the LoA measurement methodology before assessing LoA.

6. CONCLUSION

Through the combination of methods for measuring, assessing and analysing LoA, the Dynamo methodology was developed. The implementation and development of the presented Dynamo methodology is based on six case studies using the single case study method in sequence. To increase the reliability of the methodology, the first seven steps of the methodology were tested and validated in an additional, seventh, case study. The tools used in the Dynamo methodology are various forms, Hierarchical Task Analysis for identifying the subtasks, a LoA-taxonomy for measuring the sub-tasks, and a tool for assessment of relevant minimum and maximum levels of automation.

The case study for validation took place and the purpose with the validation was to perform the whole Dynamo methodology at an industrial company which had not been participating in the development

of the methodology. By stepwise performing the instructions of the Dynamo methodology and carry out each activity the research group identified problems and deviations. By understanding the instruction of the methodology, problems with logic sequences and efficiency potential within the methodology were discovered.

The validation process of the Dynamo methodology contains both planning, conducting the methodology, evaluation, development of an improved methodology, and finally documentation. Some deviations from the validated methodology were that operation instructions should be collected before the LoA measurement. By judging LoA off-site based on the operation instruction for the chosen flow and mapping LoA against the reference scales, the efficiency of the on site work can be improved. For the same reason, time estimation for the chosen flow should be done before starting measurement on site. Moreover, data can vary depending on the respondent by the identification of the process and it could be difficult to follow the operation instruction by observation of tasks. It is therefore beneficial to use video recorder by the observation of work done. Based on the deviations and problems detected during the validation, an improved methodology was developed. However, the overall steps of the methodology were kept, but instructions were changed. Finally, the results of this paper are relevant to researchers exploring and interested in effects of automation and possible ways of designing production systems balancing human and technical equipment. For industrialists, the paper provides a deeper knowledge on how to measure and assess the automation used.

ACKNOWLEDGEMENTS

The authors wish to thank the participating company and our colleagues in the DYNAMO and ProAct projects. This research is financially supported by the Swedish Foundation for Strategic Research (ProViking) and the Swedish Agency for Innovation Systems (VINNOVA). The DYNAMO project is part of the production research collaboration ROMUS, www.romus.se.

REFERENCES

Bright, J. R. (1958) Automation and Management,Maxwell reprint company, Elmsford, New York

Checkland, P. (1999) Systems Thinking, Systems Practice, Wiley

Dencker, K., Stahre, J., Gröndahl, P., Mårtensson, L. and Lundholm, T. (2007) Proactive assembly systems – higher productivity with collaboration between human and machine,In Proc. of the Swedish Production Symposium 2007, Gothenburg, Sweden.

Frohm, J., Granell, V., Stahre, J. and Winroth, M. (2007). Levels of automation in manufacturing, Submitted to Human Factors and Ergonomics in Manufacturing.

Granell, V. (2007). Levels of automation in manufacturing systems: aligning strategic and tactical decisions by means of operational measurement. Research Series from Chalmers University of Technology,Report no. 18, Göteborg, Sweden

Kern, H., Schumann, M. (1985) Industriearbeit und Arbeiterbewusstsein – eine empirische Untersuchung über den Einfluss der aktuellen technischen Entwicklung auf die industrielle Arbeit und das Arbeiterbewusstsein, Suhrcamp Taschenbuch Verlag, Frankfurt am Main

Kotha, S., Orne, D. (1989) Generic manufacturing strategies: a conceptual synthesis, Strategic Management Journal, Vol. 10, pp.211-231

Lindström, V., Frohm, J. and Bellgran, M. (2005) Developing a methodology based on value stream mapping for the measurement of automation levels in production systems, In Proc. of CIRP 3rd International Conference on Reconfigurable Manufacturing, Ann Arbor, USA

Parasuraman, R., Sheridan, T. B. and Wickens, C. D. (2000). A model for types and levels of human interaction with automation, IEEE transactions on system, man, and cybernetics - Part A: Systems and humans, 30, 286-296.

Satchell, P. (1998) Innovation and Automation,Ashgate Publishing Company, USA

Sheridan, T. B. (1980) Computer Control and Human Alienation, Technology Review. 83(1) pp. 60-73.

Sheridan, T. B. (2002). Humans and Automation: System Design and Research Issues, JohnWiley & Sons, Inc., Santa Monica.

Stanton, N. A., Salmon, P. M., Walker, G. H., Baber, C. and Jenkins, D. P. (2005) Human Factors Methods - A Practical Guide for Engineering and Design, Ashgate

Yin, R. K. (2003). Case study research: design and methods, Sage Publications Inc.

PAPER VII

Fasth, Å., Stahre, J. and Frohm, J. (2007). Relations between parameters/performers and level of automation. In Proceedings of the IFAC Workshop on Manufacturing Modelling, Management and Control, Budapest, Hungary, November 14-16, 2007.

��

RELATIONS BETWEEN PARAMETERS/PERFORMERS AND LEVELS OF AUTOMATION

��

Åsa Fasth1, Johan Stahre and Jörgen Frohm

Department of Product and Production Development Division of Production system

Chalmers University of Technology S-412 96 Gothenburg, Sweden

Abstract: Manufacturing companies struggles with the fact that costumers are becoming more and more aggressive in demanding new products and services within a short period of time. A result of this is that companies have to find the most effective manufacturing for their products. This means finding new tactics to decrease the time- and cost-parameters in a system i.e. cycle-time, ramp-up time, investments etc. It is a challenge to find the best solution for this when it comes to flexibility, robustness and efficiency. This paper presents some of the parameters that exist in a manufacturing system and the relationship between them. This is done to be able to identify the parameters that is the most important when it comes to simulate and visualise different Levels of Automation (LoA). Copyright © 2007 IFAC�Keywords: Levels of Automation, assembly systems, flexibility, Parameters

1Corresponding author. Tel +46317723686

e-mail: [email protected]

1. INTRODUCTION �Automation is used to increase productivity and reduce the cost of man-hours. By definition, automation is a technology by which a process or procedure is accomplished without human assistance (Groover 2001). Unfortunately, automation is not always fully fulfilling expectations, as the need for human intervention in cases of disturbances and system failures is still high. However, there is a tendency in industry to consider automation investments as” black or white” decisions. This may be suboptimal, since there is not always a need to distinctly choose between humans or machines. The interaction and task division between the human and the machine should instead be viewed as a changeable factor which can be called “the level of automation” (Parasuraman, Sheridan et al. 2000). Finding and implementing the right level of automation in a controlled way may be a way to maintain the effectiveness of a system (Bellgran and Säfsten, 2005).

Findings by Frohm, et al. (2007), indicate that there is a methodology to measure or assess different levels of automation or LoA;

“The relation between human and technology in terms of tasks and function allocation, which can be expressed as an index between 1 (total manual work) and 7 (total automation) of physical and cognitive support.”

By physical support, Frohm, et al. (2007), mean the level of automation for mechanical activities, mechanical LoA while the level of cognitive activities is called information LoA (see table 1)

Table 1 Levels of Automation (Frohm et al., 2007)

Levels Mechanical Information 1 Totally manual Totally manual 2 Static Hand tool Decision giving 3 Flexible hand tool Teaching 4 Automated hand tool Questioning 5 Static work station Supervising 6 Flexible workstation Intervene 7 Totally automatic Totally automatic

Companies may want to employ multiple LoAs in different parts of their system, depending on what they want to achieve. According to Seliger et al. (1987), the manufacturing system could be described from a hierarchical perspective, where every system can be divided into elements or stations. These can further be divided into part elements or tasks and every task can have its own LoA (see figure 1).

Fig 1 Relationship between tasks-stations-system (Modified from Seliger et al., 1987)

This means that both mechanical and information levels of a certain work task, which is integrated in automated systems, can be measured and analyzed at the same time. This is the rationale behind the DYNAMO measurement methodology. The methodology can be used to analyze current as well as new manufacturing systems. An important step of the DYNAMO method is setting goals for the analysis:

� Current: The Company has old line/stations but wants to increase effectiveness in this line by increasing the LoA.

� New: a company has a new product or set of products planned and has to produce in a new manufacturing system.

1.1 parameters/Performers in a system

In the next step of the LoA measurement methodology, LoA is measured in the current version of the target system. An information LoA value and a mechanical LoA value are determined. After the measurements, a workshop with personal from different levels and areas within the company is organized. This group discusses the different tasks and sets a relevant minimum and a maximum LoA value for each task. These values form a Square of Potential Improvements (SoPI). An example of this square is shown in figure 2

Figure 2 Relevant Min-and Max boundaries for LoA (Modified from Frohm, et al., 2007)

To navigate within the square, companies need to consider a number of parameters and there relationships with LoA. For example, the mechanical LoA is depending on the products’ weight and tasks performed at the different stations. Some simple examples will demonstrate this; if a product is too heavy or a product needs some kind of tool to be assembled, level 1 (see table 1 for definitions) is not recommended. In the same way, if there are more than 7 ± 2 different tasks (Miller, 1956) performed at one station, then level 1 of information LoA is not recommended. Furthermore if a company want to have a LoA = (5; 4) instead of (3; 3) this might indicate the need for investments in terms of machines, layout changes etc. In this paper a range of parameters that companies might consider have been reviewed (table 2).

Table 2 Performers in a system

Parameters

(Bjö

rkm

an, 1

990)

(Joh

nson

, 199

2)

(Gro

over

, 200

1)

(B

ellg

ran

and

Säf

sten

, 20

05)

(Lik

er, 2

004)

(Sla

ck, 2

005)

(Stå

hl, 2

006)

� Flexibility� Complexity � Robustness � Efficiency � Proactivity � Capacity/Resources

o Humans o Machines o Vehicles o Space (layout)

� Overall Equipment Efficiency – OEE

o Availability o Performance o Quality

� Re-work � Cassation � Delivery

� Cost parameters o Investment o Pay-back o Running cost o Set-up costs o Ramp-up costs

� Disturbances o Bottle-necks o System losses o MTTR o MTBF o Waste

� Time parameters o Lead-time o Cycle-time o Set up time o Ramp up time

� Production Flow o Assembly

stations

o Line o Transport o Buffers o Push or pull

MaxMin

Measured value (3; 3)

LoAmech

LoAInfo

Max

Min

Square of potentialimprovements

TasksStations

System

There are numerous parameters and performance indicators to consider if the company wants to have an optimized and effective system. Due to the complex interaction between parameters, there are strong reasons to consider simulation and visualization tools to analyse the parameters before. Such tools may aid the decision-maker in determining the kind of active relationships between different parameters. Further, what impact a change of LoA may have on these parameters? Figure 3 shows a general illustration of proposed, complex relations between parameters in table 2.

Figure 3 Relations between different main areas and parameters/performance

The main areas Efficiency, Flexibility, Complexity, Robustness, and Proactivity have different relation-ships between themselves. Efficiency, for example, should be valued in conjunction with flexibility and robustness (Bellgran, et al., 2005). Another interesting relation is found between the efficiency of automation and the flexibility of manual work.

The bubbles placed “on top” of the main areas are parameters that could be measured or visualised in some way. These have relationships with each other, but also with the main parameters. For example, if the company wants to increase the efficiency they could increase the capacity in the system by changing resources and decrease the timeparameters (i.e. cycle-time). Figure 3 indicates how quickly complexity increases when trying to describe relations between all the main areas and parameters simultaneously.

2. FLOW IN ASSEMBLY SYSTEMS

This paper focuses on delimitations of an internal flow in an assembly system. The target production situation is limited to:

� Two or more kinds of products in the system at the same time.

� Products are assembled in batches with 50 products or less.

� Two or more stations in the system

� New products in the production “pipeline”

To reduce the complexity caused by multiple relations between different parameters and areas, the discussion will be limited to one main area and a selection of parameters. Thus, it can be investigated how and if they have an impact when changing LoA.

Figure 4 The chosen main area and parameters; Flexibility, flow and time

Figure 4 shows the main area that has been chosen which is flexibility, and the parameters are flow- and time parameters. These factors were determined because they are considered the most important factors in the four points listed above. The remaining main areas and parameters listed in chapter 1 are considered unlimited or non-existing.

3. DIFFERENT APPROACHES OF FLEXIBILITY

An early definition of flexibility was provided by Stigler (1939): who defined it as;

“Those attributes of a manufacturing technology which can accommodate grater output variations”.

Since then, numerous definitions of flexibility have been suggested. Sethi and Sethi (1990) demonstrated the use of over 50 separate terms describing flexibility. The question is what kind of impact these types of flexibility have on the type of flow that is in the system and the time parameters that are listed in chapter 1.1. Below, six different types of flexibility are presented in terms of flow and time:

� Process/Product flexibility – The flow has to have workstations and tools that can assemble different kind of products at the same station. The set up-time between different products and the ramp-up time for new products has to be short.

� Usage flexibility – The Company has to have capacity flexibility but at the same time they can’t have too low degree of utilisation of the resources during normal production.

Table 3 Definitions of flexibility

(Gerwin, 1983) Product flexibility Mix Flexibility Volume Flexibility Design Flexibility The ability to add and

remove details from the mix over time.

The ability in arbitrary time to be able to treat a mix of different details, inadequa-tely related to each other.

The ability to handle a change in volume for a specific unit.

The ability to quickly introduce changes in a product design.

Routing flexibility Material flexibility The ability to reroute a

product’s path if a unit doesn’t work.

The ability to handle unexpected variations in dimensions or quality in part material.

(Browne, et al., 1984) Product flexibility Production Flexibility Volume Flexibility Machine Flexibility The ability to manufacture

a product in an economical way.

The ability to manufacture a mix of products.

The ability to manufacture profitable in spite of a shifting manufacture volume.

Easiness to change batches.

Process Flexibility Operation Flexibility Routing Flexibility Expansion Flexibility The ability to manufacture

a product in different ways. The ability to change the operation order of a product.

The ability to continue manufactures a product in spite of a tool breakdown.

The capacity to expand if needed.

(Björkman, 1990) Recycling flexibility Usage flexibility Adjust flexibility Set up flexibility A measurement of the

equipment and system ability to recycle.

A measurement of the equipment and system technical usability/ versatility.

A function of the cost to adjust the equipment and system to different circumstantialities.

A sum of usage and adjustment flexibility.

(Slack, 2005) Product flexibility Mix flexibility Volume flexibility Delivery flexibility The ability to introduce

novel products, or to modify existing ones.

The ability to change the range of products made within a given time period.

The ability to change the level of aggregated output.

The ability to change planned or assumed delivery dates.

(Ståhl, 2006) Product flexibility Production flexibility Capacity flexibility Resource flexibility The ability to develop, buy

and produce new products and to modify the product and the assembly system for normal and nominal production.

The ability to produce a multitude of products and handle changes in the production planning.

The ability to change production volume regarding retaining the effectiveness and the moving costs tied to the production tact.

The ability to add new products or modules to an existing system and adapt the system to these changes.

� Capacity/Volume flexibility – The main flow has to be well planned, with alternative flows and resources that can be used for increasing the capacity when the company has a peak in the production volume. If this could not be done the lead-time for the products increases. The usage flexibility has to be considered as well.

� Routing flexibility - The Company has to have well planned production flows, with alternative flows and resources that could be used for many different products if their normal rout is unusable. The company has to consider the usage flexibility as well. The set up-time for different resources has to be short, e.g. to set-up programs or tools (robot or CNC-machines) if needed.

� Mix/Production flexibility – The production flow has to be adaptable for new products and be able to change many times during its life-time. This means that the resources in this flow have to have a short set-up time and ramp-up time between products and for new products. The cycle-time can not be too long if the company wants to assemble many different products in a short period of time.

� Resource Flexibility - The flow has to have resources that are easy and fast to set up for new products and easy to change/ move in the flow. This results in a need for short set-up as well as ramp-up times between products.

4. DISCUSSION

What happens with these types of flexibility when the company changes the LoA for different tasks? Further, what is the optimal LoA to achieve the most flexible system? Below, some hypotheses on how the different types of flexibilities changes when changing LoA are described.

Process/Product flexibility –The flow might need a lot of recourses if the mechanical LoA is below 3. The cycle-time increases if there are too many totally manual tasks. If the information LoA is below 2 it depends on the number of tasks that are performed at each station. Thus, the ramp-up time for new products could increase.

Assumption: The process/product flexibility has its most optimal LoA-square between 3-6 in for the mechanical LoA and 2-6 for the information LoA.

Capacity flexibility – The flexibility demands capacity to be able to move around (humans) or be able to be set-up for many different products (machines). The information on what and how the products are to be assembled has to be very clear because there are different people and machines that are assembling the products.

Assumption: The optimal LoA-square for the capacity flexibility is 1-6, mechanical LoA, and 2-6 information LoA

Routing flexibility – The flow has to be able to produce the same product at different stations, even if it means to lower the LoA at one side flow that could produce many different products, and have a higher LoA on the main flow. At the side flow the information LoA has to be high if many different products arrive during a varying time period.

Assumption: The optimal LoA-square for the Routing flexibility is 2-6, mechanical LoA, and 2-6 information LoA

Mix/Production flexibility – a flow with this kind of flexibility needs to have resources that could assemble new products fast. This means that the information LoA has to be high, 3 and over, the mechanical LoA has to be adoptable for new products, either with a side flow for testing new products before they enter the main flow or machines and humans that could adopt in the main flow.

Assumption: The optimal LoA-square for the Mix/production flexibility is 2-6, mechanical LoA, and 2-6 information LoA

Resource flexibility – the flow has to be adaptable to new products, both machines and humans. This requires a high Information LoA so that the ramp-up time for the new products becomes short.

Assumption: The optimal LoA-square for the Resource flexibility is 1-6, mechanical LoA, and 2-6 information LoA

The optimal square of potential improvements when it comes to flexibility could be where all the different flexibilities are represented, which figure 5 shows are: LoaMech= 2-6 and LoAinfo=3-6.

Fig. 5 Square of potential improvements for different flexibilities

Assumption: The optimal square for flexibility are: LoaMech= 2-6 and Loainfo=3-6.

5. DISCUSSION AND CONCLUSION

In this paper the complexity of different main areas and parameters or performers in a system has been discussed. Companies need to be increasingly aware of the parameters affecting their production systems. It might be better to optimise one main area and some parameters first. In this paper, assumptions have been made that refer to flexibility, time and flow parameters due to the Levels of automation. To validate these assumptions beyond theoretical analysis, empirical data is being collected and analyzed as a part of a research project called Proactive Assembly Systems (ProAct). Deeper knowledge will be gained and used to improve modelling and simulation tools for different levels of automation. These activities are a part of the research project SIMTER.

6. FUTURE RESEARCH

Future research aims at simulating and visualising manufacturing systems with varying LoA in the system’s stations and tasks. Improvement of flexibility and the flow-and time parameters will be measured. We will be able to simulate and visualise the same system with different LoA in the system’s different stations and tasks. Further investigations will be made to measure changes in selected variables as a consequence of changes in flexibility-, flow-, and time parameters and validate the hypotheses.

The visualising/simulation tool could then be used for: A help/ presentation tool when discussing different investments in a system.

1. Planning the system when it comes to different flows, capacity etc.

2. Analyzing the “square of potential improvements” in the LoA measurement method in a visualised way.

3. Simulate the impacts that changes in LoA have on the flexibility, flow and time parameters.

The tool could be used in a long period of time and the system could be updated or optimized in fast and easy way when different changes is to be tested. This gives the companies great advantages to create and maintain a flexible system.

ACKNOWLEDGEMENT

The authors would like to express their gratitude to the DYNAMO project for providing background knowledge related to levels of automation. Further, we would like to thank the participants in the ProAct project. The ProAct project is funded by the Swedish Agency for Innovation Systems (VINNOVA) and participating industries.

REFERENCES

Bellgran, M. and Säfsten, K. (2005). Produktionsutveckling - Utveckling och drift av produktionssystem. Studentlitteratur, Lund, Sweden.

Björkman, M. (1990). Flexibel automatisk montering - en analys av motiv och förutsättningar. University of Linköping, Linköping, No. 233, 13-17

Browne, J., Dubios, D., Rathmill, K., Sethy, S. P. and Stecke, K. E. (1984). Classification of Flexible manufacturing systems. The FMS Magazine, April

Comstock, M. (2001). Mass Customizing - Perspectives, Applications and Implications for a 'New Manufactoring Paradigm'. Linköping, No. 913, 45-47

Frohm, J., Lindström, V., Winroth, M. and Stahre, J. (2007). Levels of Automation in Manufacturing. To be submitted.

Gerwin, D. (1983).A framework for Analyzing the Flexibility of Manufacturing Process, School of Business Administration, University of Wisconsin, Wisconsin

Groover, M. P. (2001). Automation, production systems, and computer-integrated manufacturing.Prentice Hall, Upper Saddle River, N.J.

Johnson, T. H. (1992). Relevance Regained. The free press, Macmillan Inc,

Liker, J. K. (2004). The Toyota Way: 14 Management Principles from the World's Greatest Manufacturer. McGraw-Hill, USA.

Miller, G. A. (1956). The Magical Number Seven, Plus or Minus Two: Some Limits on Our Capacity for Processing Information. Physical review, 63, pp 81-97,

Parasuraman, R., Sheridan, T. B. and Wickens, C. D. (2000). A model for types and levels of human interaction with automation. IEEE transactions on system, man, and cybernetics - Part A: Systems and humans, No. 30, 286-296.

Seliger, G., Viehweger, B. and Wieneke, B. (1987). Decision Support in design and optimization of flexible automated manufacturing and assembly. Robotics and computer-integrated manufacturing,Vol. 3, No. 2, 221-227

Slack, N. (2005). The flexibility of manufacturing Systems. International Journal of Operations & Production Management Vol. 25 No. 12,

Ståhl, J.-E. (2006). Industriella tillverkningssystem - länken mellan teknik och ekonomi. Lunds tekniska högskola, Lunds Universitet, Vol. 1 No. 2

APPENDIX

DYNAMO form no. 1-5

For

m n

o. 1

– D

ata

colle

ctio

n of

pro

duct

ion

flow

Pa

ge 1

of

2

Dat

e: 2

0__-

__-_

_ P

lace

(C

ompa

ny):

___

____

____

____

____

Mea

sure

men

t co

nduc

ted

by (

Nam

e):

____

____

____

____

___

Pro

duct

ion

flow

incl

udin

g w

here

it s

tart

s an

d en

ds

Pur

pose

wit

h th

e pr

oduc

tion

flow

Pro

duct

s w

ithi

n th

e pr

oduc

tion

flow

The

pro

duct

mix

wit

hin

the

prod

ucti

on fl

ow

Exp

lana

tion

s:

Pro

duct

ion

flow

incl

udin

g w

here

it s

tart

s an

d en

ds: T

he m

ain

func

tion

of th

e pr

oduc

tion

flo

w (

e.g.

ass

embl

y, p

roce

ssin

g) a

nd w

here

the

prod

ucti

on

flow

sta

rts

and

ends

(e.

g. th

e pr

oduc

tion

flow

sta

rts

in s

tati

on X

and

end

s in

sta

tion

Y)

Pur

pose

wit

h th

e pr

oduc

tion

flo

w:

E.g

. Ass

embl

y of

pro

duct

X f

or c

ostu

mer

A a

nd B

, wit

h de

live

ring

pre

cisi

on Y

/mon

th

Pro

duct

s w

ithi

n th

e pr

oduc

tion

flo

w:

The

num

ber

of p

rodu

ct li

nes

to b

e pr

oduc

ed

The

pro

duct

mix

wit

hin

the

prod

ucti

on f

low

: D

escr

iptio

ns o

f th

e nu

mbe

r of

pro

duct

var

iant

s an

d an

app

roxi

mat

ely

num

ber

of c

ompo

nent

s to

be

hand

led

with

in th

e pr

oduc

tion

flow

.

For

m n

o. 1

– D

ata

colle

ctio

n of

pro

duct

ion

flow

Pa

ge 2

of

2

Dat

e: 2

0__-

__-_

_ P

lace

(C

ompa

ny):

___

____

____

____

____

Mea

sure

men

t co

nduc

ted

by (

Nam

e):

____

____

____

____

___

Wor

king

org

aniz

atio

n w

ithi

n th

e pr

oduc

tion

flow

Tas

ks to

be

achi

eved

by

the

mac

hine

s w

ithi

n th

e pr

oduc

tion

flow

Sect

ions

/cel

ls w

ithi

n pr

oduc

tion

flow

Exp

lana

tion

s:

Wor

king

org

aniz

atio

n w

ithi

n th

e pr

oduc

tion

flo

w:

The

num

ber

of s

hift

, des

crip

tion

of

the

wor

k cr

ew, t

he n

umbe

r of

ope

rato

rs a

nd th

eir

task

s

Tas

ks t

o be

ach

ieve

d by

the

mac

hine

s w

ithi

n th

e pr

oduc

tion

flo

w: T

he a

ge o

f th

e m

achi

nes,

type

of

mac

hine

s an

d th

eir

mai

nten

ance

Sect

ions

/cel

ls w

ithi

n pr

oduc

tion

flow

: T

he n

umbe

r of

sta

tions

with

in th

e pr

oduc

tion

flo

w, d

escr

iptio

n of

con

duct

ed a

nd p

lane

d ch

ange

s of

the

layo

ut.

For

m n

o. 2

– V

isua

lisat

ion

of p

rodu

ctio

n fl

ow

Page

1 o

f 2

Dat

e: 2

0__-

__-_

_ P

lace

(C

ompa

ny):

___

____

____

____

____

Mea

sure

men

t co

nduc

ted

by (

Nam

e):

____

____

____

____

___

Vis

uali

zati

on o

f the

pro

duct

ion

flow

Vis

ualiz

atio

n of

the

pro

duct

ion

flow

:

= W

orks

tati

on/-

cell

wit

h op

erat

ion

inst

ruct

ions

= B

uffe

r

= W

orks

tati

on/-

cell

wit

h op

erat

ion

inst

ruct

ions

Tas

k co

nduc

ted

by

the

oper

ator

(s)

Num

ber

of

oper

ator

s

Tas

k co

nduc

ted

by t

he

tech

nica

l equ

ipm

ent

Typ

e of

tec

hnic

al

equi

pmen

t

Pro

duct

var

iant

(s)

Stat

ion

nota

tion

For

m n

o. 2

– V

isua

lisat

ion

of p

rodu

ctio

n fl

ow

Page

2 o

f 2

Dat

e: 2

0__-

__-_

_ P

lace

(C

ompa

ny):

___

____

____

____

____

Mea

sure

men

t co

nduc

ted

by (

Nam

e):

____

____

____

____

___

Stat

ion

nr.

�

� � � � �

��

� � � �

��

� � � �

��

�

��

� � � �

��

� � � �

��

� � � �

��

��

��

� � � �

��

�

Dat

e: 2

0__-

__-_

_ P

lace

(C

ompa

ny):

___

____

____

____

____

Mea

sure

men

t co

nduc

ted

by (

Nam

e): _

____

____

____

____

__�

� ��

� � � �

Com

men

ts

Com

men

ts

Com

men

ts

Com

men

ts

Relevant Max Relevant Min Relevant Max Relevant Min

Est

imat

ed

LoA

Est

imat

ed

LoA

Mechanical LoA Information LoA

For

m n

o. 4

– D

efin

ing

the

rele

vant

max

. and

min

. of

LoA

in th

e pr

oduc

tion

flo

w

Dat

e: 2

0__-

__-_

_ P

lace

(C

ompa

ny):

___

____

____

____

____

Mea

sure

men

t co

nduc

ted

by (

Nam

e):

____

____

____

____

___

Tas

k nr

Tas

k de

scri

ptio

n

For

m n

o. 5

– A

naly

sis

of L

oA m

easu

rem

ent

wit

hin

the

prod

ucti

on f

low

Dat

e: 2

0__-

__-_

_ P

lace

(C

ompa

ny):

___

____

____

____

____

Mea

sure

men

t co

nduc

ted

by (

Nam

e):

____

____

____

____

___

LoA

-inf

o.

LoA

-mec

h.

1 2 3 4 5 6 7

1 2

3 4

5 6

7

mokbil

Documents