peer-olaf siebers

Modelling Human Variation in Manual Assembly Line Models

The Impact of Human Performance Variation on theAccuracy of Manufacturing System Simulation Models

Peer-Olaf Siebers

Manufacturing Department – Cranfield University

Research is funded by the Ford Motor Company and EPSRC

Presentation prepared for: EASSS 2004

06/07/2004 Modelling Human Variation In Manual Assembly Line Models 2/15

Content

• Introduction

• Background

• Aim and Method

• Data Collection

• Experimentation

• Results

• Conclusions

• Outlook


Introduction

Intro OutlookConclusionExperimentationData CollectionAim & MethodBackground Results ?

Discrete Event

Simulation

Business

need

Evaluate&

RefineImplement

Operational

Manufacturing

Facility

Concept

design

•Product based measures such as

lead time and volume

•Resource based measures such as

availability and utilisation

•Resources: e.g.

machines, operators

•Part flow and process

times

• Manufacturing system design process:


• An assembly line is set of sequential workstations, typically connected by a continuous material handling system.

• Assembly lines are quite complex constructs due to natural variation in processing times, and breakdowns.

• Causes for breakdowns: machine failure, unavailability of operators or unusual long processing times.

Background (1/3)

Background OutlookConclusionExperimentationData CollectionAim & MethodIntro Results ?


Background (2/3)

Common observations:Common observations:

• A gap exists between the performance prediction of a system model and the performance of the real system. Simulation tends to model the real world too optimistically.

• The magnitude of the gap is bigger when simulating non existing systems compared to existing ones.

• The magnitude of the gap is bigger when simulating manual lines compared to automated ones.

Simulation is the process of constructing a model that describes the behaviour of a real world system. A system model is always a restricted copy of a real system.



Background (3/3)

Initial test has shown that:Initial test has shown that:

• Task completion times (cycle times) vary significantly for the same task between different operators but also for the same operators whenconducting a task several times.

• The time between the last part produced before a break and the first part produced after a break is usually significantly longer then the break defined in the shift pattern and also varies between different operators.



Aim & Approach

Aim:Aim:

• To investigate the importance of incorporating human performancevariation models into manufacturing system simulation models of labour intensive manufacturing systems.

Approach:Approach:

• To assess expert manufacturing system simulation models.

• To enhance the assessed simulation models by using frequency distributions representing operator cycle time variations.

• To enhance the assessed simulation models by using frequency distributions representing break start and break duration variations.

Under examination:Under examination:

• Two discrete event simulation models build by a simulation expert representing two different manual engine assembly flow lines (101/157 operations).

Aim & Method OutlookConclusionExperimentationData CollectionBackground Results ?Intro


Method

Aim & Method

Research question

Preliminary experiment design

Integrating human variation models into system model

Analysis of results

Experiment execution

Recommendations

Human Variation ModelsSystem Model

Refine

Model validation

Model enhancement

Data collectionData collection

Model design

Final experiment design

OutlookConclusionExperimentationData CollectionBackground Results ?Intro


Human Variation Data Collection

Data Collection OutlookConclusionExperimentationAim & MethodBackground Results ?Intro


Section of Running System Model

Experimentation OutlookConclusionData CollectionAim & MethodBackground Results ?Intro


Results (Averages)

Results OutlookConclusionExperimentationData CollectionAim & MethodBackground ?Intro

122%

111% 111%

105%103%

100%

105%

95%

90%

84%

80%

100%

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

110%

120%

130%

Real SimNrm SimBD SimHV SimBT SimHV&BT

Relative average production per shift (100 % = average of real system)

MS1 MS2


Conclusions

Conclusion OutlookExperimentationData CollectionAim & MethodBackground Results ?Intro

What can we do?What can we do?

Findings:Findings:

• Individuals can influence line behaviour and vice versa.

• Adding human performance variation models into manufacturing system simulation models adds to their accuracy.

• The impact depends on the type of variation to be represented as well as on the system to be modelled.

Main Limitations of Current Approach:Main Limitations of Current Approach:

• Due to the automated data collection it is very difficult to separate system and human variation.

• Omission of considering the interdependencies between events as well as between all system components.


Outlook

Possible Solution:Possible Solution:

• Using Computational Organisation Theory as methodological approach and Multi-Agent Based Modelling as the technique.

Issues:Issues:

• Complexity of the task (what level of abstraction can be used)

• Data collection (performance data of individuals required, data format)

• Validation and validity

• Model development is purpose dependent

• Concept of pro-activeness

• Time driven vs. event driven

OutlookConclusionExperimentationData CollectionAim & MethodBackground Results ?Intro

Organisations, which are basically groups of people working together to attain commongoals, can also be characterised as complex adaptive systems composed of intelligent,

task-oriented, boundedly-rational, and socially-situated agents and faced with an environment that also has the potential for change (Carley & Prietula, 1994).


Concept for Multi Agent Based Integrated System

EMM

EMM

EMM

EMM

EMM

Worker n

PMMs

ws En

Worker 1

PMMs

ws E1

Worker 2

PMMs

ws E2

Mediation

rules for P & E

Generic (global)

environment

Support for swopping of

worker profiles

HPM Tool

Worker specific

environment

Mediation

rules for P & E

Mediation

rules for P & E

op1

opn

op2

Set of worker performances

(dependability, activity time,

error rate)

Worker performance

for the tasks at op1

Discrete Event Simulation

System Data

(e.g. no of workstations,

tasks, machine cycle times)

System performance

(e.g. lead time, utilisation, volume)

Model Setup Data

Knowledge Base

Task DB

Pre-defined activity

times for all the tasks

Worker performance

for the tasks at opn

Worker performance

for the tasks at op2

System data,

Process data

Absenteeism, accident rate and staff turnover

based on group composition

Simulation Core

Continuous and

Discrete Event

What labour is available

(no of stereotypes)?

Historical data of

absenteeism, accident

rate and staff turnover

Worker Setup Data

Legend:ws = worker specific

E = environment

P = person

MM = micro model

op = operation

OutlookConclusionExperimentationData CollectionAim & MethodBackground Results ?Intro


Questions?

Email: [email protected]

OutlookConclusionExperimentationData CollectionAim & MethodBackground ResultsIntro ?

peer-olaf siebers

Documents