bagian 1-128_2

Proceeding of Industrial Engineering and Service Science, 2011, September 20-21

Copyright © 2011 IESS.

Reliability, Maintenance and Its Management: The

Current State of Play

Kym Fraser School of Advanced Manufacturing and Mechanical Engineering, University of South Australia, Adelaide, Australia

[email protected]

ABSTRACT

Maintenance and its management is now of strategic importance for most organisations around the world. Problems

which surround the current maintenance literature include the identification of maintenance management models and

use of these models in real world applications. It has been argued that the gap between theory and practice is wider in

the maintenance field than any other research discipline. In this study 37 different maintenance management models

are identified and from these models three were found to clearly dominate the literature; Total Productive Maintenance

(TPM), Condition-Based Maintenance (CBM), and Reliability-Centred Maintenance (RCM). A comprehensive review

of these three models was undertaken to establish links to empirical ‗real world‘ applications, determining model

popularity and details of study methods, sector, industries, author and country. Further investigation of three leading

journals in maintenance found that 401 published articles on these popular models produced 48 articles with links to

practice, giving an empirical evidence rate of 12% when compared to the overall number of papers published. While

this paper, importantly, examines links between maintenance theory and practice, a clear picture emerges on the lack of

empirical research undertaken by academics in the area of maintenance and its management.

Keywords:Maintenance, management models, literature review, empirical evidence

1.Introduction

According to a US National Research Council Report in 1990, one of the research priorities of US manufacturing is

equipment reliability and maintainability [1]. Historically, maintenance activities have been regarded as a „nece s-

sary evil‟ by the various management functions in an organisation [2,3]. However, over the past 15 to 20 years, this

attitude has increasingly been replaced by one which recognises maintenance as a strategic issue in the organisation.

In 2006 Carnero [4] summed up the situation by stating “the setting up of a predictive maintenance programme is a

strategic decision that until now has lacked analysis of questions related to its setting up, management and control”

(p.945). The role of maintenance in maintaining and improving the availability of plant and equipmen t, product

quality, safety requirement and plant cost-effectiveness levels, constitute a significant part of the operating budget

of manufacturing firms [5].

According to [6] between 15 to 40 percent (average 28 percent) of the total production cost is att ributed to

maintenance activity in the factory. Ten years later [7] goes further by suggesting that maintenance department

costs represent from 15 to 70 percent of total production costs. [8] explained that next to energy costs, maintenance

spending can be the largest part of the operational budget. [9] discussed how the cost of maintenance for a selected

group of companies increased from US$200 billion in 1979 to US$600 billion in 1989, three -fold in just 10 years.

With the advent of more automation, robotics and computer-aided devices, maintenance costs are likely to be even

higher in the future [10].

Therefore, the effective integration of the maintenance function with engineering and other manufacturing

functions in the organisation can help to save huge amounts of time, money and other resources in dealing with

reliability, availability, maintainability and performance issues [11]. For most organisations it is now imperative

they take opportunities via maintenance management programs to optimise their productivity, while maximising the

overall equipment effectiveness. With increasing focus on just-in-time, quality and lean manufacturing, the reliabil-

ity and availability of plant are vitally crucial. Poor machine performance, downtime, and ineffective plan t mainte-

nance lead to the loss of production, loss of market opportunities, increased costs and decreasing profit [12]. This

has provided the impetus to many organisations worldwide to seek and adopt effective and efficient maintenance

strategies over the traditional firefighting reactive maintenance approaches [13,14].

Reliability, Maintenance and Its Management: The Current State of Play

2

The problem which currently exists involves firstly, the limited number of maintenance papers providing r e-

views of maintenance management strategies, and secondly, papers exploring and combining the various mainte-

nance strategies and their links to real world applications (empirical evidence) is non-existent. It would seem that

the gap between theory and practice in regards to maintenance is greater than in other research fields. In 1996,

Dekker [8] argued that mathematical analysis and techniques, rather than solutions to real problems, have been ce n-

tral in many papers on maintenance models. He goes on to say “It is astonishing how little attention is paid either to

make results worthwhile or understandable to practitioners, or to justify models on real problems” (p.235). In 1998

Rausand [15] supported Dekker by claiming “there is more isolation between practitioners of maintenance and the

researchers than in any other professional activity” (p.130). In 2002, [16] stated “Since the late seventies, examples

of models assessing corrective and preventive maintenance policies over an equipment life cycle exist in the liter a-

ture. However, there are not too many contributions regarding real implementation of these models in industry”

(p.367). In a recent discussion on the problems and challenges of reliability engineering, [17] states that the main-

tenance literature is strongly biased towards new computational developments.

Therefore, the key objective of this paper is to provide links between literature and practice by firstly, revie w-

ing the maintenance literature and determining the various maintenance management models/strategies discussed

within it. Secondly, while the number of maintenance related papers in the literature is high (numbering in the

thousands), only papers providing empirical evidence will be further analysed to determine popular maintenance

management strategies in practice today, identifying the country, sector and industry that these models are being

employed in around the world. Articles of a purely mathematical nature, theoretically derived, or of a conceptual

basis were not analysed. The outcomes will provide practitioner and researchers with a practical insight of a bus i-

ness process which now holds significant strategic implications for nearly every organisation.

2. Identification of Maintenance Management Models in Literature

This section will establish the various maintenance management models found within the literature. The review in-

volved all peer reviewed journals and textbooks available on the University of South Australia library databases. This

source included well respected databases such as Business Source Complete (EbscoHost), Emerald fulltext, ScienceDi-

rect, Wiley InterScience, SAGE full-text collection and Compendex. These databases represent the major publishers in

the maintenance field such as Elsevier, Emerald and Taylor & Francis. To keep findings as contemporary as possible

the search for empirical evidence linking popular maintenance models to practice was restricted to articles published

within the last 15 years (1995 – 2009).

Table 1. Model Description and Categorisation

Model Main Focus Benefits/Requirements

Practical

application:

Holistic/

Singular

Literature

evidence:

Empirical/

Theoretical

Advanced tero-

technological model

Moves focus from Life-cycle-cost (LCC) to

Life-cycle-profit (LCP)

Integrates TQM/ terotechnology/ LCP.

Requires integrated IT system Holistic Theoretical

Age-based Maintenance

An extension of RCM Allows better management of items that fail due to wear and/or related to age.

Holistic Theoretical

Availability-based

Maintenance

An extension to both RCM and TPM Needs to be integrated with manufacturing

resource planning (MRP) system Holistic Theoretical

Basic terotechnology model

Focus on maintaining system‟s life cycle Establishes information feedbacks to maintain system‟s life cycle


Breakdown

Maintenance

Action is taken once the item/equipment has

failed

Applied quickly with limited resources

and information. A high risk and commer-

cially expensive strategy

Holistic or singular

Theoretical

Campaign

Maintenance

Simular to shutdown maintenance. Used

when non-maintenance restraints take prior-

ity e.g. military operations

Replaces regular maintenance program but

completion time-frames are limited Holistic or

singular Empirical

Computerised Maintenance

Management

System

Provides capabilities to store, retrieve and analyse information

Deals with computer-aided integration of maintenance in an enterprise. Used in

conjunction with a maintenance manage-

ment system e.g. TPM

Holistic

Empirical

and theoretical

Condition-based

Maintenance

(CBM)

Based on the monitoring and detection of

equipment to determine vital warnings of

impending failure

CBM allows a reliable, accurate assess-

ment of service life while reducing reli-

ance on maintenance personnel

Holistic

Empirical

and

theoretical

Condition

Monitoring (CM)

Similar to CBM where condition monitoring

of selected equipment is undertaken to detect

potential failures

CM is commonly applied to individually

selected equipment. Should be integrated

with other maintenance programs

Singular

Empirical

and

theoretical

Reliability, Maintenance and Its Management: The current State of Play

Copyright © 2011 IESS. 3


Practical

application:

Holistic/

Singular

Literature

evidence:

Empirical/

Theoretical

Corrective

Maintenance

Unplanned activities undertaken to return the

equipment to its operating condition

Requires management processes to iden-

tify defects and eliminate root causes Holistic Empirical

Effectiveness-

centred

Maintenance

Based on “doing the right things” instead of

“doing things right”

Encompasses the concepts of TQMain and

features of TPM and RCM to provides a

more effective maintenance system

Holistic Empirical

E-maintenance Integrates existing telemaintenance princi-ples with Web services and modern

e-collaboration principles

Used in conjunction with CBM. Ideal for military and commercial aircraft operators

to reduce aircraft downtime


Equipment Asset Management

An optimum combination of best practice, technology, organisation, and administration

Maximise lifetime value from process, production, and manufacturing equipment


Kelly‟s philosophy Control of reliability through the physical

control of engineering systems

Develops links between quality and main-

tenance. Mixture of elements from TPM,

RCM and terotechnology


Maintenance

Management Metric

Maintenance management is the allocation

of value added resources

Systematically improves overall equip-

ment effectiveness, while optimizing the

cost of per unit production


Operating mainte-nance training and

administration

An organisational-wide approach which considers all aspects of the supporting infra-

structure

Operations, maintenance, training, and administration are integral parts of the

whole system


Outsourcing Transfer to outsiders with the goal of getting higher quality maintenance at faster, safer

and lower costs

While firm can concentrate on core com-petencies, the maintenance service con-

tract still requires management

Holistic or selected

areas/items

Theoretical

Planned Maintenance

Maintenance functions performed on a pre-planned basis

Firms able to determine optimal intervals for various machines and failure types


Empirical

Preactive

Maintenance

Defines equipment maintenance require-

ments before the process, line or individual

machine commences operation or before major expansion

Provides early evaluation of maintenance

costs and man hours Holistic or

singular Theoretical

Predictive

Condition Monitoring

The application of multiple technologies to

monitor the condition of machines for pend-ing failure

Technology is combined with various

analysis techniques through computerised applications

Singular Theoretical

Predictive Mainte-

nance

Consists in deciding whether or not to main-

tain a system according to its state

Recommended to be use in conjunction

with traditional periodic preventive main-

tenance programs


Theoretical

Pre-planned

Maintenance (PPM)

Divides the working calendar into discrete

separate elements and assigns PPM jobs to

the various elements

Able to determine optimal intervals for

various machines and failure types. PPM

can attract criticism for over-servicing


Empirical

Preventive Maintenance (PM)

A series of tasks performed at a frequency dictated by time, amount of production and

machine condition

PM can either extend the life of an asset or detect that an asset has critical wear and is

going to fail or break down

Holistic or

singular Empirical

Proactive Maintenance

Advanced maintenance approach that fo-cuses on reducing total maintenance required

and maximizing life of machinery

Individual maintenance activities are re-engineered to enable preven-

tive/predictive maintenance practices


Productive

Reliability

Based on TPM with the purpose of reducing

costs and improving capacity through con-tinuous maintenance improvement

Needs to utilise failure mode and effect

analysis techniques Holistic Theoretical

Profit Center

Maintenance

The maintenance of machinery, equipment

of fixed asset is considered a profit activity.

Assets are optimised for maximum value

rather than the least cost Holistic Theoretical

Reliability-centred

Maintenance

(RCM)

An asset maintenance management system

oriented towards maintenance critical indus-

tries such as airlines, power plants

Analyses each physical asset in its operat-

ing context and assesses what must be

done to ensure it fulfils its function

Holistic

Empirical

and

theoretical

Risk Based Maintenance

Focus is on the dual objectives of minimisa-tion of hazards caused by unexpected failure

of equipment and a cost effective strategy

While minimising the probability of sys-tem failure, risk analysis also evaluates

other consequences such as; safety, eco-

nomic and environment


Run-to-destruction Reactive approach. Equipment is used nor-

mally until it fails, then discarded or re-

placed

Normally confined to carefully selected

equipment and the consequences of failure

known and accepted in advance


Run-to-failure Reactive approach. Equipment is used nor-mally until it fails, then discarded or re-

placed

Requires very little ongoing and routine maintenance. Suitable for small,

non-critical, low cost equipment



4


Practical

application:

Holistic/

Singular

Literature

evidence:

Empirical/

Theoretical

Scheduled

Maintenance

Periodic replacement of parts based on their

age

Firms able to determine optimal timing of

maintenance

Holistic or

singular Theoretical

Strategic

Maintenance

Management

An overall business perspective which fur-

ther builds on TPM and RCM

Integrates technical, commercial and op-

erational aspects of business with mainte-

nance program

Holistic

Empirical

and

theoretical

Time-based Maintenance

Maintenance activity based on a time period Economically beneficial when dispersion of the item lifetime is small


Theoretical

The Eindhoven

University of Technology(EUT)

Developed to fill gaps in terotechnology

models.

Lists 14 sub-functions of maintenance but

no links to IT system Holistic Theoretical

Total Productive

Maintenance

(TPM)

An asset maintenance methodology that

combines the effort of plant operators,

safety, energy, materials, and quality with the planning and maintenance efforts

Designed to be integrated with JIT, TQM,

employee involvement and environmental/

organisational factors Holistic

Empirical and

theoretical

Total Quality

Mainte-nance(TQMain)

Converts a singular platform (CM) into a

holistic model

Recommends production schedules should

incorporate time for maintenance Holistic Theoretical

Value-driven Plan

Maintenance

Enhancement of RCM with company, plant

and maintenance objectives being integrated

Relies on the utilisation of knowledge and

expertise within plant Holistic Theoretical

While 37 differently named models were identified (Table 1) an analysis of these models indicates a number of

similarities. Approximately half (18 models) share either a similar focus and/or the benefits/requirements are homoge-

neous. Models identified to offer only minor and subtle variations included: Basic / Advanced terotechnology;

Age-based / Time-based / Scheduled maintenance; Availability-based / Campaign maintenance; Breakdown / Correc-

tive maintenance; Condition Monitoring / Predictive Condition Monitoring; Effectiveness-centred / Total Quality main-

tenance; Planned / Pre-planned / Preactive / Scheduled maintenance; and Run-to-destruction / Run-to-failure. In regards

to similarities it could be argued that the model name, Preventive Maintenance (PM), has broad generic meanings for

maintenance. [18] described preventive maintenance as being a practice which encompasses all planned, scheduled and

corrective actions before the equipment fails. Another point of similarity is the fact that many models are a direct exten-

sion or based on the platform of the three most popular models found in the maintenance literature, being TPM, RCM

and CBM. A common theme to emerge from a majority of models was the need for the maintenance system to be inte-

grated with the organisations information and data systems.

When analysing the 37 indentified models an important consideration, especially for this paper, is the level of em-

pirical evidence found in the literature. While theoretical examples and descriptions can be found for most of the mod-

els, documented practical (real world examples) evidence was found for only 12 models (32%). When the four popular

models (TPM, RCM, CBM, and CM) are removed less than a ¼ of the remaining 33 models have any empirical evi-

dence on which the model can be practically evaluated. Adding to the empirical limitations of these remaining models is

the fact that only 1or 2 papers exist on each of the models and a number of the models are based on the four popular

models. In the case of practitioners the point on empirical evidence is important because it allows the model to be

evaluated in a real world environment. For them, developing an understanding of issues surrounding implementation

and success of the maintenance system are key points. Having limited practical evidence on the various models is prob-

lematic and not desirable.

3. Empirical examples of popular models analysed

A final list of 76 articles (the 3 models were represented 87 times) were extracted from the many hundreds of papers

reviewed and these were examined to establish model type, empirical evidence, author origin, study country, field of

study, and the research industry. The empirical evidence of each article involved methods such as surveys, interviews,

case studies and anecdotal experience. To clarify „anecdotal experience‟ papers classified as „anecdotal‟ were personal

accounts of the author/s experiences working and researching in the field. These articles, while providing empirical

evidence, must be viewed with caution as no empirical data was presented, only a personal view, therefore the use of

the term „anecdotal‟. With the removal of hundreds of papers, due to the conceptual/theoretical nature of these papers, a

clear picture emerged of the „real world‟ examples for the three popular models in practice today.

On the surface it would seem that the rate of empirical research output over the 15 year period of the reviewed lit-

erature (average of 5.07 per year) has remained reasonably consistent, with peak years occurring in 2000 (12 publica-

tions), 2002 (7) and 2006 (8) . A closer analysis of the figures tend to indicate that there has been a decline in empirical

research in the three most popular maintenance models. The overall output in five of the last six years has been below

Reliability, Maintenance and Its Management: The current State of Play


the yearly average of 5.07. Between 1995 to 2000 (6 years) there were 8 articles on CBM but since 2000 (last 9 years)

only 3 empirical studies have been published, with two out the three being in 2006. In regards to RCM, 23 articles were

published between 1995 to 2004 (10 years), and only 3 articles have been published in the five years since 2004.

Analysis of study sector and study industries shows that the Manufacturing sector (55%) clearly dominate, fol-

lowed by a General classification (18%) and Energy (13%). When narrowing the fields into specific industries, power

plants were clearly identified as being popular for maintenance research with eight papers, followed by steel mills and

the semiconductor industries with four each, part suppliers with three papers, and the automotive industry with two.

Interestingly, out of 76 empirical papers only two have direct practical links to the automotive industry. This industry is

a massive global influence on manufacturing around the world and has provided researchers with many practical exam-

ples of modern improvement philosophies such as just-in-time (JIT), total quality management (TQM), lean manufac-

turing (LM), flexible manufacturing systems (FMS), and world class manufacturing (WCM). While authors from Hong

Kong and Taiwan produced five and one publications respectively, Asian powerhouses such as Japan and mainland

China produced only three combined. With TPM being developed in Japan and also being the dominate maintenance

model in the literature (66%) it is therefore interesting that only two studies were conducted in Japan.

Model Popularity Study Sector Study Industry

TPM – 66% [50] Manufacturing – 55% (42) Power plants – 8

RCM – 34% [26] General classification –18% (14) Semiconductor – 4 CBM – 15% [11] Energy – 13% (10) Steel mills – 4

[ ] No. of studies Construction – 3 (4%) Part suppliers – 3 ( ) No. of papers Automotive – 2

Author Origin Study Country Study Methods UK – 24% (18) India – 21% (12) Case study –50% (38)

India – 17% (13) UK – 19% (11) Anecdotal –24% (18)

USA – 12% (9) USA – 9% (5) Survey – 14% (11) Sweden – 9% (7) Sweden – 7% (4) Descriptive – 6% (5)

HK/Taiwan – 8% (6) Japan/China – 7% (4) Comparison – 3% (2)

Canada – 5% (4) Canada – 5% (3) Pilot study – 3% (2) Spain – 4% (3) Spain – 5% (3)

Italy – 4% (3) Italy – 5% (3)

Japan/China – 4% (3) HK/Taiwan – 5% (3)

In summary, it is worth pointing out that caution should be taken when trying to make comparisons between these

three popular maintenance models. It is clear that the applicability of TPM, RCM, and CBM are situation specific.

While very popular in the manufacturing sector, TPM is more suitable as an integrated holistic improvement system for

the organisation as a whole. RCM and CBM are more equipment specific for critical, complex, high tech applications

like gas compression systems in the offshore oil industry, boiler and turbine auxiliaries in the nuclear industry, and ro-

bots in automobile manufacturing. RCM is often used in more safety-focused sectors, such as the nuclear and aircraft

industries, where maintenance management has usually extensive due to safety regulations.

4. The Need for Greater Empirical/Practical Focus

While a total of 76 empirical articles were analysed in Section 3 it would seem that this figure is somewhat small given

the fact it represents 15 years of academic research and considering the growing level of importance maintenance man-

agement is to most organisations around the world. In an attempt to quantify or present an accurate picture of the cur-

rent situation, further analysis was undertaken of maintenance related journals.

Table 2. Published articles on popular maintenance management models: Comparison

betweentotal papers published and papers with empirical evidence (1995-2009)

Leading Maintenance Journals Maintenance Models

Total TPM CBM RCM

Journal of Quality in Maintenance Engineering 71 81 46 198

Reliability Engineering & System Safety 14 74 73 161

International Journal of Quality & Reliability Management 23 9 10 42

Total 108 164 129 401

Published papers with empirical evidence 22 8 18 48

Percentage of papers with empirical evidence 20% 5% 14% 12%

In this study it was found that three journals: Journal of Quality in Maintenance Engineering, Reliability Engi-

neering & System Safety, and the International Journal of Quality & Reliability Management provided over 50% of the

journals referenced. Table 2 provides a comparison between the total papers published and papers with empirical evi-


6

dence. As can be seen a total of 401 articles were publish between 1995 – 2009, and 48 of these articles made links to

real world applications. This provides a rate of 12% of published papers providing empirical evidence.While it would

seem that the percentage of empirical evidence is low, further research would be needed to establish how these figures

compare with other research areas outside of the field of maintenance.

5. Conclusions

A comprehensive review of the maintenance management literature was undertaken with 37 models being identified.

From this group three models were found to dominate the published literature, namely: Total Productive Maintenance

(TPM), Reliability-Centred Maintenance (RCM) and Condition-Based Maintenance (CBM). Of the many hundreds of

articles reviewed for these popular models only 76 papers were found to contain empirical evidence or „real world‟

examples. Of the remaining 34 maintenance management models identified (excluding Condition Monitoring) very

little theoretical or practical support was found in the literature. Also in the last 5 or so years it was shown that overall

publication output of the maintenance models reviewed is trending lower, and this decline is even more pronounced in

regards to CBM and RCM. As maintenance and its management has increasingly become an important and strategic

issue for nearly every organisation in the world it could easily be argued that empirical based publications should be

increasing, not trending lower. The findings of this papers support the view that maintenance theory, in many respects,

is de-coupled from practical applications.

6. References

[1] V. Ebrahimipour and K. Suzuki, “A synergetic approach for assessing and improving equipment performance in offshore in-

dustry based on dependability”, Reliability Engineering and System Safety,Vol. 91, 2006, pp.10-19.

[2] F. L. Cooke, “Plant maintenance strategy: evidence from four British manufacturing firms”, Journal of Quality in Maintenance

Engineering, Vol.9, No.3, 2003, pp.239-249.

[3] S. Apeland and T. Aven, “Risk based maintenance optimization: foundational issues”, Reliability Engineering and System

Safety,Vol. 67, 2000, pp.285-292.

[4] M. Carnero, “An evaluation system of the setting up of predictive maintenance programmes”, Reliability Engineering and Sys-

tem Safety,Vol. 91, 2006, pp.945-963.

[5] B. Al-Najjar and I. Alsyouf, “Selecting the most efficient maintenance approach using fuzzy multiple criteria decision making”,

International Journal of Production Economics, Vol. 84, 2003, pp.85-100.

[6] R. Mobley,“An Introduction to Predictive Maintenance”, Van Nostrand Reinhold, New York, 1990.

[7] M. Bevilacqua and M. Braglia, “The analytic hierarchy process applied to maintenance strategy selection”, Reliability Engi-

neering and System Safety, Vol. 70, 2000, pp.71-83.

[8] R. Dekker, “Applications of maintenance optimization models: a review and analysis”, Reliability Engineering and System

Safety, Vol. 51, 1996, pp.229-240.

[9] T. Wireman, “World Class Maintenance Management”, Industrial Press Inc., New York, 1990.

[10] S. Blanchard, “An enhanced approach for implementing total productive maintenance in the manufacturing environment”,

Journal of Quality in Maintenance Engineering, Vol.3, No.2, 1997, pp.69-80.

[11] J. Moubray, “Twenty-first century maintenance organization: Part 1 – the asset management model”, Maintenance Technology,

Applied Technology Publications, Barrington, IL, 2003.

[12] C. Cholasuke, R. Bhardwa and J. Antong, “The status of maintenance management in UK manufacturing organisations: results

from a pilot survey”, Journal of Quality in Maintenance Engineering, Vol.10, No.1, 2004, pp.5-15.

[13] K. Fraser, “Maintenance management is now of strategic importance: So what strategies are your competitors using?”, Pro-

ceedings of the 6th International Strategic Management Conference, St. Petersburg, Russia, July 8-10, 2010, pp. 139-152.

[14] I. Ahuja and J. Khamba, “An evaluation of TPM implementation initiatives in an Indian manufacturing enterprise”, Journal of

Quality in Maintenance Engineering, Vol. 13 No. 4, 2007, pp.338-352.

[15] M. Rausand, “Reliability centered maintenance”, Reliability Engineering and System Safety,Vol. 60, 1998, pp.121-132.

[16] A. Marquez and A. Heguedas, “Models for maintenance optimization: a study for repairable systems and finite time periods”,

Reliability Engineering and System Safety, Vol. 75, 2002, pp.367-377.

[17] E. Zio, “Reliability engineering: Old problems and new challenges”, Reliability Engineering and System Safety, Vol. 94, 2009,

pp.125-141.

[18] S. I. Mostafa, “Implementation of proactive maintenance in the Egyptian Glass Company”, Journal of Quality in Maintenance

Engineering, Vol.10, No.2, 2004, pp.107-122.



Human Factors in Manufacturing

Kym Fraser School of Advanced Manufacturing and Mechanical Engineering, University of South Australia,Adelaide, Australia

[email protected]

ABSTRACT

In today‘s competitive environment, cellular manufacturing (CM) is a process which offers global manufacturers im-

proved performance and helps them meet their strategic commitments through product and volume flexibility, lower

costs and improved customer response times. CM is a well-known strategy in removing many of the inefficiencies ex-

perienced in functional batch-type manufacturing environments. Evidence now indicates that some organisations have

achieved results that are less than anticipated and that firms which struggle to achieve the full benefit from CM may in

fact be experiencing problems with the human factors associated with manufacturing cells. As a socio-technical proc-

ess, cellular manufacturing requires careful attention to both its technical and human aspects. Academics and practi-

tioners alike have focused on the technical factors such as cell layout, machine order, family part grouping, and work-

flow balancing. The adoption of CM changes the social relationship and interactions among employees and their su-

pervisors. Given the potential impact on employee‘s attitudes, motivation, and retention, these social changes call for

effective management in a number of areas including HRM, employment relations and industrial structure. This study

presents a review of the various human factors involved in manufacturing cells and tests the importance of each. A sur-

vey of managers, team leaders and operators working within CM systems helps to distinguish between technical and

human aspects and identifies the importance of human factors such as training, communication and teamwork.

Keywords: Human factors, cellular manufacturing, empirical study

1.Introduction

In today‟s competitive environment many companies are endeavouring to improve their manufacturing performance. It

is now widely accepted that cellular manufacturing (CM) is one such method that manufacturers can use to help meet

their strategic commitments, through product and volume flexibility, lower costs and improved customer response

times. CM is based on operators processing part families, or collections of similar parts, in cells, or clusters of dedicated

machines that may be dissimilar in function [1]. The benefits of CM include reduction in setup times, material handling,

work-in-process, cycle time, and tooling requirements [2],[3],[4]. Furthermore, the implementation of CM has been

shown to achieve significant improvements in product quality, space utilization, control of operations, scheduling, and

worker productivity [5],[6],[7]. While there is no doubt about the increasing popularity of CM (studies show that cells

are now adopted by between 43 and 53% of firms in the United States and the United Kingdom [8]) there is also evi-

dence that CM has not been successful in some organisations. Companies converting to CM often struggle with imple-

mentation and achieve results that are less than anticipated [8],[9],[10]. Evidence now indicates that firms who struggle

to achieve the full benefit from CM may in fact be experiencing problems with the human factors associated with CM

[7],[11],[12].

While much of the CM research work has focused on technical issues (machine order, family part grouping,

workflow balancing), it is now accepted that the implementation and ongoing success of CM involves the consid-

eration of both technical and human aspects. [13]found that both technical and social changes take place when a

company adopts advanced manufacturing systems such as cellular manufacturing. They point out that, i f an organi-

sation focuses solely on the technical side at the expense of human factors, its performance will be less favourable

than if it pays attention to both sets of issues. Under traditional batch-type functional manufacturing conditions em-

ployees have well-defined responsibilities for a single operation or machine. The very nature of cells requires that a

pool of individually skilled machine operators be grouped together to share work in the cell.

It is now accepted that a number of fundamental social changes do occur when companies convert from func-

tional manufacturing layouts to manufacturing cells. Given the potential impact on employee‟s attitudes, motiv a-

tion, and retention, these social changes call for effective management in a number of areas including supervision,

HRM, employment relations and industrial structure.This study aims to provide answers to two areas which have

not been adequately addressed in the literature. The first part of the study seeks to determine the level of influence


8

that technical and human aspects may play within CM systems, and secondly, a list of human factors associated

with CM are tested to determine which are the most important factors within CM systems.

2. Technical vs Human Aspects of Cellular Manufacturing

There exists a significant and growing body of academic research exploring various technical facets of cell formation

and design [14],[15],[16],[17]. These include areas such as machine sequence, workflow balancing, machine-part fami-

lies, and cell capacity using mathematical or simulation methodologies. [11]explain that most of this technically focused

research adopts a micro-level focus, investigating one or a few issues within this large and complex process, and giving

only a limited attention to the significant human dimensions. This has led to the situation where we know a great deal

about certain steps in the technical design of cells, but lack a well-developed and broadly-focused theory of cell design

and its human consequences.

[18] argue that a major contributing factor why the full benefits of CM have not been achieved is the fact that

the research literature on cellular manufacturing over the last 15 years has to an overwhelming degree focused on

the development of technical procedures to solve the cell formation problem (machine order/layout, family part

grouping, work flow sequence).While many of the decisions inherent in cell system design are technical in nature

(e.g., how work should be scheduled through the cell), there are significant human dimensions to cell design (e.g.,

how cell operators will be selected, trained and rewarded). [11]conclude that many of the problems and failures in

cellular manufacturing systems occur at the interface between the technical and social subsystems.

[19]state that manufacturing companies can establish a strategic competitive advantage by placing a greater

importance on the human elements early in the design and implementation process. The authors explain that the vast

majority of the cell formation literature places primary emphasis on grouping similar parts and machines. Once the

cells are designed, secondary consideration is given to the assignment of workers to the cells. At this stage, the h u-

man element has typically only considered workers in terms of their labour capacity and/or technical skills.

[19]argue that, in this setting, human skills such as communication, problem solving, teamwork, leadership, and

conflict resolution can become just as important as technical skills, such as mechanics, mathematics, machining and

inspection.

In a study of implementation experiences, [6] concluded by making the following point, “the picture that

emerges from this study is clear – restructuring the factory to adopt cellular manufacturing should not be viewed

merely as a technical, engineering-dominated problem but as a change process where the people element domi-

nates”. [20]found that a number of fundamental social changes occur when companies convert from functional

(batch or job shop) manufacturing layout to manufacturing cells. In CM, employees are moved from segregated

work groups (eg all press operators work in the same department, all lathe operators in same department) into cells

that combine jobs and workers from several specialized skill areas. Cell team members have to work together,

though each may have originally been under different pay or reward system, or possess different levels of training,

skills, and experience. In essence, conversion to CM changes the social interactions among employees and their

supervisors. These social changes require careful attention because of their potential impact on employee attitudes,

motivation, and retention.

[21]conducted a comprehensive evaluation of the various human factors associated with CM by reviewing

both the CM and the advanced manufacturing technologies (AMT) literature. Adding support to the lack of research

is this field, [21] states “while cellular manufacturing is a popular research area, there is a singular absence of art i-

cles that deal with the human elements in cellular manufacturing”. The results of their review identify eight broad

areas of human issues in CM: worker assignment strategies, skill identification, training, communication, auto n-

omy, reward/compensation system, teamwork, and conflict management.

Where socio-technical systems such as cellular manufacturing are involved, both aspects need to be considered

to maximise success. The very nature of manufacturing cells dictates that individuals will be required to work t o-

gether to maximise the benefits that cells can provide manufacturers. What is not clear is the level of influence that

either aspect has on CM systems. Is one aspect considered more important or has a greater influence on the

on-going success of manufacturing cells? This study will endeavour to provide a better understanding to this un-

answered questioned.

3. Methodology

The data used for this study was collected via a questionnaire survey designed to provide information about the impor-

tance of human issues in cellular manufacturing. A sample size of 175 participants involved in cellular manufacturing

took part in the survey. Survey participants included three sub-groups: managers, team leaders, and operators. A brief

summary of the four medium to large organisations involved in the study are as follows:



Company 1 (sites 1 & 2) – Electrical accessories manufacturer (Australian) (2300 employees)

Company 2 – Sanitary ware manufacturer (Australian) (2000 employees)

Company 3 – Automotive components manufacturer (Australian) (800 employees)

Company 4 – Electrical accessories manufacturer (Switzerland) (380 employees)

The aim of this study is to provide answers for two research questions. Firstly, in an attempt to determine the level

of influence that technical or human factors may have on CM, participants were asked to distinguish between technical

and human aspects of cellular manufacturing and determine the level of influence either aspect may have on CM sys-

tems. The second objective was to test the importance of a list of human factors which are associated with CM and es-

tablish which factors are considered the most important within a CM system. The list of human factors used in this

study wereidentified by [21] and the list encompasses most of the social issues presented in the literature. The data col-

lected to answer both research questions was independent of the other.

4. Findings

Participants were asked to rate which problem (between technical and human) they had encountered the most often

while working in cells. A list of technical and human problems were provided to help participants understand the dif-

ference between each issue. The scale used to rate this question was as follows; 1 – mostly technical, 2 – more techni-

cal, 3 – same, 4 – more human, 5 – mostly human.

Table 1 – Source of problem: Technical or Human (Company)

Com N Mean SD

Company 1 – site 1

- site 2 Company 2

Company 3

Company 4

42

23 6

40

61

2.05

2.91 2.33

2.15

2.66

0.909

0.793 0.816

1.027

1.124

Total 172* 2.41 1.042

*Of the 175 participants, 3 operators fail to answer this question (N=172)

The results showed that the problem being experienced within manufacturing cells is skewed toward „technical‟

issues (mean 2.41) (see Table 1). The mean for each of the four companies and five sites surveyed fell within 2 (being

more technical) to 3 (experiencing the same amount of problems for both issues). Of the 172 participants, 52% indi-

cated that they had experienced either „more technical‟ or „mostly technical‟ problems involving CM systems. Of the

remaining 48% of participants, 34 % indicated that they had experienced the same amount of problems for both aspects

leaving only 14% to indicated that they had experienced „more human‟ or „mostly human‟ problems. It is worth noting

that the maximum and minimum mean values for the survey occurred between the two sites of the same company, 2.05

to 2.91.

Table 2 – Source of problem: Technical or Human (Position)

Position N Mean SD

Manager

Team Leader Operator

10

23 139

3.00

2.65 2.33

1.054

1.027 1.031

Total 172* 2.41 1.042

*Of the 175 participants, 3 operators fail to answer this question (N=172)

When comparing the data for the different positions held within the companies (see Table 2), operators indicated

that they experienced more technical problem than human issues within manufacturing cells with a mean of 2.33. For

team leaders the mean increases to 2.65 (indicating less technical and more human issues than operators) and for man-

agers the mean value is 3.00. For managers the amount of problems they experience between technical and human is-

sues are the same. The results indicate that people in leadership or management positions within a cellular environment

experience increased human issues as compared to operators of the cells.

In regards to the second research objective, participants were ask to rank the eight (8) human factors from 1=most

important to 8=least important. Each factor was characterised by a short description to help participants understand the

various factors being ranked.


10

The overall rankings of human factors for the various companies/sites (see Tables 3) indicate three sub-groupings

of the eight human issues listed. The three factors ranked most important (means between 2.79 and 3.47) were „com-

munication‟, „teamwork‟, and „training‟. The next sub-group (means between 4.38 and 4.83) was „skill identification‟

and „worker assignment strategies‟. The final group (means between 5.34 and 5.75) were, „conflict management‟,

„autonomy‟, and „reward/compensation‟.

When comparing the individual companies and sites to the overall rankings the following differences are observed.

Of the three top ranked factors, the biggest difference occurred with the European company (Company 4) which ranked

„training‟ as the 7th most important factors while the Australian companies/sites ranked „training‟ either 1st or 2nd .

Another notable difference was Company 2 ranking „teamwork‟ as the 5th most important factors. While „teamwork‟

ranked highly overall, one possible reason for this lower ranking may be due to the fact that the cells in this company

were only 2-3 person cells as compared to the surveys overall average of 6 people per cell. The low number of partici-

pants (N=6) for Company 2 also make it difficult to draw meaningful comparisons on an individual level.

Table 3 – Human Factor Ranking: Companies

Rank Respondent Category

Company 1 -

Site 1

Company 1 -

Site 2

Company 2

Company 3 Company 4 All Companies

1 Training Training Training Communication Communication Communication

2 Teamwork Teamwork Communication Training Teamwork Teamwork

3 Communication Communication Skill Identification Teamwork Autonomy Training

4 Skill Identification Skill Identification Worker Assign-

mentStrategies

Skill Identification Worker Assign-

ment Strategies

Skill Identification

5 Reward/ Compensation

Conflict Management

Teamwork Worker Assign-mentStrategies

Conflict Management

Worker Assign-mentStrategies

6 Worker Assign-

mentStrategies

Worker Assign-

mentStrategies

Reward/

Compensation

Reward/

Compensation

Skill Identification Conflict

Management

7 Conflict Management

Autonomy Autonomy Conflict Management

Training Autonomy

8 Autonomy Reward/

Compensation

Conflict

Management

Autonomy Reward/

Compensation

Reward/

Compensation

The overall ranking of „skill identification‟ and „worker assignment strategies‟ at 4th

and 5th

respectively seem a true

reflection as all individual companies/sites ranked both between 3rd

and 6th

. When comparing differences within the

three least important factors the notable difference again occurred in the European company (Company 4) when it

ranked „autonomy‟ as the 3rd

most important while the other companies/sites (Australian) ranked „autonomy‟ either 7th

or 8th

most important.

When comparing the three respondent categories; managers, team leader, operators (see Tables 5), the overall re-

sults is strongly influence by the high number of operators (81%) of the total participants. The rankings given to the

eight human factors by both „team leaders‟ and „operators‟ was the same. While the rankings were the same, the mean

value for each factors for „team leaders‟ was higher (indicating greater importance) except for „reward/compensation

system‟. The biggest difference between these means occurred for „communication‟ (2.13 to 2.85) and „skill identifica-

tion‟ (3.70 to 4.51). The notable difference in the rankings occurred between „managers‟ and the other two positions.

Manager considered „teamwork‟, „training‟ and „communication‟ as the top three factors while team leaders and opera-

tors ranked „communication‟, „teamwork‟ and „training‟ as their top three.

Table 4 – Human Factor Ranking: Job Position

Rank Respondent Category

Managers Team Leaders

Operators All Groups

1 Teamwork Communication Communication Communication

2 Training Teamwork Teamwork Teamwork

3 Communication Training Training Training

4 Skill Identification Skill Identification Skill Identification Skill Identification

5 Autonomy Worker Assignment

Strategies

Worker Assignment

Strategies

Worker Assignment

Strategies

6 Worker Assignment Strategies

Conflict Management Conflict Management Conflict Management

7 Conflict Management Autonomy Autonomy Autonomy

8 Reward/

Compensation

Reward/

Compensation

Reward/

Compensation

Reward/

Compensation



5. Discussion

In regards to the first researchquestion, the biggest difference occurred between the two sites of the same company

(Company 1). This significant difference may in some way be explained by each sites attitude to training. At Site 1 the

training records (both past and future needs) of each operator were displayed in a strategic position within the plant for

all employees to observe. While such a method was not evident at Site 2, many employees at Site 2 openly complained

about the lack of coordinated training within the plant. Support for this issue was also evident in Question 2 when par-

ticipants at Site 2 ranked training (mean value of 2.04) higher than any other company or site. When questioned about

employee training at Site 2, management stressed that adequate training had been provided. Another notable difference

between the two sites of the same company occurred with the ranking of „reward/compensation system‟. While Site 2

operators were unhappy about the lack of training, it was not the case in regards to reward/payment. Site 2 participants

clearly ranked the issue of payment as the least important (mean 7.09). At site 1 it was ranked the 5th most important

with a much higher mean of 4.58.

When comparing the three respondent categories (job positions), the two notable differences was that managers

ranked „autonomy‟ the 5th most important while the other two positions put it at 7th. Secondly, while all three positions

ranked „reward/compensation system‟ the least important factor, the mean of 7.40 by managers was the lowest mean

recorded in the survey, indicating the very low importance given to reward/payment issues by managers.

When looking for notable differences between countries the two factors to stand out was the low ranking for

„training‟ (ranked 7th) and the high ranking of „autonomy‟ (ranked 3rd) from the European results. Without further re-

search it would be difficult to state the reasons for these differences but the following point will be made. The surveys

distributed to participants in the Swiss company were converted to the common language of the factory being

Swiss-German. It was observed that many operators were experiencing some problems understanding the

Swiss-German language being used, even with the use of a senior and long time employee of the company to help ex-

plain the survey questions. While many of these workers were not native to Switzerland, it was interesting to note that

so many workers (including younger people) would experience such a problem with the native/common language. It

would seem that some literacy training would be beneficial the company, considering that operators needed to work

within a team environment in manufacturing cells. A second point which may have some influence on the high

„autonomy‟ ranking is that operators may feel more comfortable working in smaller groups or even alone due to this

literacy issue. It could also be argued that cultural differences may affect someof the outcomes,both within the Swiss

company and between the two countries.

The results of this survey clearly indicate that human factors play a significant role in the overall success of cellu-

lar manufacturing. When analysing which individual human factors are important to CM it is shown that communica-

tion, teamwork and training rank the highest while reward/compensation ranked lowest. In determining the importance

of human factors such as skill identification, worker assignment, training, reward etc. in different companies and coun-

tries, it must be noted that these issues are rarely „neutral‟ in nature, and their interpretation will be shaped by the indus-

trial context of the firm and/or country. It is therefore acknowledged that some of the differences in the results between

the four companies may well be shaped by the industrial context in which they operate. The non-testing of this issue in

this research provides the opportunity for further research in this area and in human factors in general.

6. Conclusions

Cellular manufacturing has a lot to offer global manufacturers by reducing both costs and inefficiencies within their

manufacturing processes. While the focus of research has been on the technical side of this socio-technical process, it is

now clear that greater effort must be placed on the human aspects to improve the benefits and success of this form of

manufacturing. This study found that while technical issues still play a major role in the on-going problems experienced

in cellular manufacturing, human issues account for a significant proportion of problems within cells. The study goes on

to identified the various human factors associated with CM and tests the importance of each. While communication,

teamwork and training were ranked as the most important factors, it is hoped that these findings will better inform prac-

titioners on the human aspects of CM and provide future direction in areas such as employment and industrial relations.

7. References

[1] H. Harris and K. Fraser, “Towards virtual manufacturing: An implementation framework from feasibility to product develop-

ment,” International Journal of Product Development, Vol. 11, No. 1/2, 2010, pp. 136-162.

[2] V. L. Huber and N. Hyer, “The human factor in cellular manufacturing”, Journal of Operations Management, Vol.5, No.2,

1985, pp.213-228.

[3] F. Olorunniwo, “A framework for measuring success of cellular manufacturing implementation”, International Journal of


12

Production Research, Vol.35, No.11, 1997, pp.3043-3061.

[4] A. Gunasekaran, R. McNeil, R. McGaughey andT. Ajasa, “Experiences of a small to medium size enterprise in the design and

implementation of manufacturing cells”, International Journal of Computer Integrated Manufacturing, Vol.14, No.2, 2001,

pp.212-223.

[5] V. L. Huber, and K.A. Brown, “Human resource issues in cellular manufacturing: A sociotechnical analysis”, Journal of Op-

erations Management, Vol. 10, No.1, 1991, pp.138-159.

[6] U. Wemmerlov and D. J. Johnson, “Cellular manufacturing at 46 user plants: implementation experiences and performance

improvements”, International Journal of Production Research, Vol. 35, No.1, 1997, pp.29-49.

[7] K. S. Park and S. W. Han, “Performance Obstacles in Cellular Manufacturing Implementation – Empirical Investigation”, Hu-

man Factors and Ergonomics in Manufacturing, Vol.12, No.1, 2002, pp.17-29.

[8] D. J. Johnson and U. Wemmerlov, “Why does cell implementation stop? Factors influencing cell penetration in manufacturing

plants”, Production and Operations Management, Vol.13, No.3, 2004, pp. 272-289.

[9] C. A. Yauch, “Moving towards cellular manufacturing: The impact of organisational culture for small businesses”, PhD Thesis,

University of Wisconsin, USA, 2000.

[10] K. Fraser, H. Harris and L. Luong, “Improving the implementation effectiveness of cellular manufacturing: A comprehensive

framework for practitioners”, International Journal of Production Research, Vol. 45, No 24, 2007, pp. 5835-5856.

[11] N. L. Hyer, K. A. Brown and S. Zimmerman, “A socio-technical systems approach to cell design: case study and analysis”,

Journal of Operations Management, Vol.17, 1999, pp.179-203.

[12] K. Fraser, “Labour flexibility: Impact of functional and localised strategies on team-based product manufacturing”, CoDesign,

Vol. 5, No. 3, 2009, pp. 143-158.

[13] G. G. Udo and A. Ebiefung, “Human Factors affecting the Success of Advanced Manufacturing Systems”, Computers & In-

dustrial Engineering, Vol.37, 1999, pp.297-300.

[14] N. Singh, “Design of Cellular Manufacturing Systems: An Invited Review”, European Journal of Operational Research, Vol.

69, No.3, 1993, pp. 248-291.

[15] M. Kazerooni, “An integrated methodology for cellular manufacturing system design”, PhD Thesis, University of South Aus-

tralia, Adelaide, 1997.

[16] G. Shambu and N. C. Suresh, “Performance of hybrid cellular manufacturing systems: A computer simulation investigation”,

European Journal of Operational Research, Vol.120, 2000, pp.436-458.

[17] Z. Albadawi, H. Bashir and M. Chen, “A mathematical approach for the formation of manufacturing cells”, Computers & In-

dustrial Engineering, Vol. 48, 2005, pp.3-21.

[18] U. Wemmerlov and D. J. Johnson, “Empirical findings on manufacturing cell design”, International Journal of Production

Research, Vol. 38, No.3, 2000, pp.481-507.

[19] B. A. Norman, W. Tharmmaphornphilas, K. L. Needy, B. Bidanda and R. C. Warner, “Worker assignment in cellular manu-

facturing considering technical and human skills”, International Journal of Production Research, Vol.40, No.6, 2002,

pp.1479-1492.

[20] F. Olorunniwo and G. Udo, “The impact of management and employees on cellular manufacturing implementation”, Interna-

tional Journal of Production Economics, Vol.76, 2002, pp.27-38.

[21] B. Bidanda, P. Ariyawongrat, K. L. Needy, B. Norman and W. Tharmmaphornphilas, “Human related issues in manufacturing

cell design, implementation, and operation: a review and survey”, Computers & Industrial Engineering, Vol.48, 2005,

pp.507-523.

Proceeding of Industrial Engineering and Service Science , 2011, September 20-21


Developing A Model For Measuring

Organizational Knowledge: A Case Study of

PT.Telekomunikasi Indonesia, Tbk.

Fransiscus Rian Pratikto Department of Industrial Engineering, Parahyangan Catholic University, Bandung, Indonesia

[email protected], [email protected]

ABSTRACT

Knowledge management (KM) has become the main attention of today business organizations due to its important role

in developing and sustaining their competitive advantage. Many organizations have benefited from KM initiatives. As

an old proverb in management goes, ―you cannot manage what you cannot measure‖, knowledge measurement is one

of the essences of KM. The importance of knowledge measurement is not supported by the availability of an appropriate

framework to do such effective measurement. Reperence [2] proposed a framework for measuring knowledge in an or-

ganization that consists of three component, i.e. stock, flow, and enabler. Stock is existing level of knowledge at a point

in time; flow is movement of knowledge between entity, including individuals, organizations or organization levels; and

enablers are investment, processes, structures, and activities established by organizations aimed at changing or main-

taining knowledge stocks, or influencing knowledge flows. This research aims to contribute in designing a measurement

model based on Boudreau‘s framework by exploring and operationalizing this framework, and finally empirically test-

ing the model in a case study. In this research, a model of knowledge measurement based on Boudreau‘s framework is

developed. This model identifies some constructs that theoretically affect (either directly or indirectly) the flow of

knowledge in an organization, i.e.: network attributes, external sources and knowledge complementarity, potential ab-

sorptive capacity, realized absorptive capacity, nature of knowledge (tacitness), knowledge integration, and social

mechanism integration. An empirical study in the largest telecommunication company in Indonesia is conducted. This

survey resulted in the significance and magnitude (in standardized value) of the total effect of each of the above con-

struct on the flow of knowledge as follows: network attributes (.51), external sources and knowledge complementarity

(.15), potential absorptive capacity (.22), realized absorptive capacity (.62), nature of knowledge (tacitness) (.25),

knowledge integration (.30), and social mechanism integration (.22).

Keywords: knowledge management, stock, flow, enabler.

1. The Importance of Knowledge Measurement

Knowledge management (KM) has become the main attention of today business organizations due to its important role

in developing and sustaining their competitive advantage. Many organizations have benefited from KM initiatives. Best

practices from many organization showed that the return on investment (ROI) of KM initiative implementation varied

between 2,5:1 to 10:1 [1].

Based on research by Davenport, et.al. [2] which studied 31 KM projects in 23 companies, it is found that KM im-

plementation project that effectively increase organization‟s efficiency and effectiveness are those that focused on (i)

creating knowledge repositories, (ii) improving access to knowledge, (iii) enhancing culture that support knowledge

usage in an organization, and (iv) managing knowledge as asset. Managing knowledge as asset requires a good and

proper knowledge measurement.

2. Boudreau’s Framework of Knowledge Measurement

Boudreau [3] offered a quite comprehensive framework for measuring knowledge of an organization. Boudreau‟s

framework consists of 3 components, i.e.: stock, flow, and enabler. Stocks are the level of knowledge at a point in

time; flows are movement of knowledge between entities, including individuals, organizations, and organization levels;

and enablers are investment, processes, structures, and activities established by organizations aimed at changing or

maintaining knowledge stocks, or influencing knowledge flows.

According to Boudreau, knowledge measures that are categorized as stock includes accounting of intangibles, fi-

mailto:[email protected]


Developing A Model For Measuring Organizational Knowledge: A Case Study of PT.Telekomunikasi Indonesia, Tbk.

14

nancial statement augmentation, patents, publications and citations, organization experience and rivalry patterns, learn-

ing curves, and unit-level competencies, education and experience. Measures of knowledge flow are categorized in 2

groups, one group of measures focuses on business units and alliance partners, and another group focuses on groups and

teams. Enabler measures comprise geographical and political proximity, international and domestic organizational and

alliance design, research & development (R&D) expenditures, absorptive capacity, network attributes, and tacitness.

This research aims to contribute in designing knowledge measurement model based on Boudreau‟s framework.

This measurement model can be used to portray the condition of knowledge in an organization, and focuses on knowl-

edge measures that are measurable at individual level (see Table 1).

Table 1: Knowledge measures of Boudreau’s framework and corresponding level of measurement

Knowledge measures Level of measurement

Individual Organization

Stocks

Accounting for intangibles assets √

Financial statement augmentation √

Patents or publications and their citation patterns √

Organization experience and competitive rivalry √

Learning curves √ Unit-level education, experience, and job requirements √ √

Flows Measures that focus on flow of knowledge between business units and

alliance partners √

Measures that focus on flow of knowledge between colleagues and team

√

Enablers

Geographycal and physical proximity √

International and domestic organizational and alliance design √

Research and development expenditures √

Absorptive capacity √ √

Networks attributes √ √

Tacitness √ √

3. The Knowledge Measurement Model

The following enabler measures are chosen as the model‟s components:

Absorptive capacity

Reference [4] defines absorptive capacity as dynamic capability embedded in a firm‟s routines and processes mak-

ing it possible to analyze the stocks and flows of a firm‟s knowledge and relate this variables to the creation and

sustainability of competitive advantage.

Absorptive capacity has 2 components, i.e.: potential absorptive capacity (PACAP) and realized absorptive capac-

ity (RACAP). PACAP comprises the ability to acquire and assimilate knowledge, whereas RACAP comprises the

ability to convert and exploit knowledge. Several factors significantly affect these two capabilities [4], e.g. knowl-

edge integration, sources of external knowledge, knowledge complementarity, and social integration mechanism.

Network attributes

Individual and organizational network attributes are key enablers that determine flow of knowledge. This network

can be network between individuals, organization and suppliers, buyers, financial institutions, etc. [3].

Tacitness

Tacitness reflects the efforts required to move the knowledge [5]. Tacitness is an enabler because it determines the

ease of knowledge transfer process. Tacitness can be harmful when it restricts desired knowledge flow between

groups, but also valuable in making knowledge difficult for competitors to copy (Teece, Pisano, & Shuen, 1997;

Barney, 1991 in [3]).

The complete model consists of 8 constructs, i.e.:

Network attributes; consists of 3 variables: size, coverage, and strength of network.

Sources and complementarity of external knowledge; consists of 4 variables: coverage and strength or relationship

with external paty in term of acquisition, purchasing through licensing and contract, inter-organization relationship

including research & development, alliance, and joint ventures, and knowledge complementarity between

individual in the organization and contacts in their networks [6].

Social integration mechanism consists of variables that indicate the form and level of social mechanism in the or-

Developing A Model For Measuring Organizational Knowledge: A Case Study of PT. Telekomunikasi Indonesia, Tbk.


ganization, either formal or informal [3].

Tacitness consists of 5 variables: type of knowledge, codifiability, teachability, complexity, and system dependence

[6].

Knowledge integration, consists of variables that indicate the availability and level of integration of knowledge in

the organization either in formal or informal form [7].

Potential absorptive capacity consists of variables that indicate the ability to acquire and assimilate external

knowledge [7], [4].

Realized absorptive capacity, consists of variables that indicate the ability to convert and exploit knowledge [7],

[4].

Knowledge flow, consists of variables that indicate the flow of procedures, tools and ideas including patents [3];

The relationships between the above constructs result in the following directional hypotheses:

Hypotheses 1: The better the network attributes of an organization, the more flow of knowledge into the organiza-

tion.

Hypotheses 2: The better the network attributes of an organization, the higher external knowledge sources and

complementarity of the organization.

Hypotheses 3: The higher external knowledge sources and complementarity of an organization, the higher potential

absorptive capacity of the organization.

Hypotheses 4: The better the social mechanism integration of an organization, the higher the realized absorptive

capacity of the organization.

Hypotheses 5: The higher the level of knowledge integration in an organization, the higher the potential absorptive

capacity in the organization.

Hypotheses 6: The higher the level of knowledge integration in an organization, the higher the realized absorptive

capacity in the organization.

Hypotheses 7: The higher the level of knowledge tacitness in an organization, the more difficult to attain a degree

of knowledge integration in the organization.

Hypotheses 8: The higher the level of potential absorptive capacity in an organization, the higher the level of real-

ized absorptive capacity in the organization.

Hypotheses 9: The higher the level of realized absorptive capacity in an organization, the higher the knowledge

flow in the organization.

Hypotheses 10: The higher the level of knowledge tacitness in an organization, the more difficult to attain a degree

of knowledge flow in the organization.

Hypotheses 11: The higher the social integration mechanism in an organization, the higher the level of knowledge

integration in the organization.

4. Case Study

The model was implemented in PT. Telekomunikasi Indonesia, Tbk. (TELKOM), Indonesia‟s largest state-owned

telecommunication company. TELKOM is a full service and network provider telecommunication company. TELKOM

provides fixed wireline services, fixed wireless services, mobile/cellular services, data & internet services, and

interconnection & network services. As of 31 December 2006, TELKOM has about 48.5 million customers, consists of

8.7 million fixed wireline service customer, 4.2 million fixed wireless customers, and 35.6 million mobile service

customers [8]. In line with its vision “To become a leading InfoComm player in the region”, TELKOM has been

making a continual effort to stay the top position among telco operators in Indonesia.

Knowledge management initiative has ben formally implemented in TELKOM which is managed by an AVP

(Assistant Vice President). In 2007 TELKOM also awarded Top-5 of Most Admired Knowledge Enterprise (MAKE)

2007 in Indonesia.

Online survey through TELKOM‟s intranet was conducted in April 2007, 1269 sampel of TELKOM‟s employee

are collected from which 133 completed questionnaire were considered invalid, hence 1136 samples are used for further

analysis. The reliability of the questionnaire was tested using Cronbach Alpha and resulted in alpha coefficient of

0.922, leading to the conclusion that the instrument is reliable.

Majority of respondent are 41 – 45 years old (39,96%) and 46 – 50 years old (22,36%) with a mean of 43,88 years

old. Most of them have Sarjana degree (44,63%) dan Diploma degree (37,15%). About 56,16% of respondents have

been working in Telkom for more than 20 years, and 20,95% of them for 16 – 20 years. About 53,17% of the

respondents have just been no more than 1 year in their current position, and on average all of the respondents have

been in their position for 24,58 months.


16

5. Model Parameterization and Analysis

The model was parameterized using Structural Equation Modeling (SEM) which employed Maximum Likelihood esti-

mation method. Two trimming process were conducted to ensure that all constructs, variables, and relationships are

significant.

The parameterization‟s result in standardized value is depicted in Figure 1.

The final model was then tested for fitness and the result is depicted in Table 2, in which 5 of 9 criteria indicate

that the model was still in the acceptable level of fitness.

Table 2: Final model fitness

Model Fit Criteria Value Acceptable Level [9] Interpretation

Chi-square (c2) 8005.58 < 1022,82 Not fit

Goodness-of-Fit Index (GFI) 0.69 0 (not fit) s.d. 1 (perfect fit) Partially fit

Adjusted GFI (AGFI) 0.65 0 (not fit) s.d. 1 (perfect fit) Partially fit

Root-mean-square residual (RMR) 0.16 0,05 Not fit

Root-mean-square error of approximation (RMSEA) 0.10 < 0,05 Not fit

Tucker-Lewis Index (TLI) 0.69 0 (not fit) s.d. 1 (perfect fit) Partially fit

Normed fit index (NFI) 0.69 0 (not fit) s.d. 1 (perfect fit) Partially fit

Normed chi-square (c2/DF) 11.59 1,00 c

2/DF 5,00 Not fit

Parsimonious fit index (PGFI) 0.61 0 (not fit) s.d. 1 (perfect fit) Partially fit

The impact of each construct on flow of knowledge is depicted in Table 3.

Table 3: Impact of each construct on flow of knowledge

Constructs Direct Effect Indirect Effect Total Effect

Realized Absorptive Capacity 0.62 0.00 0.62

Network Attributes 0.39 0.12 0.51

Knowledge Integration 0.00 0.30 0.30

Tacitness 0.10 0.15 0.25

Potential Absorptive Capacity 0.00 0.22 0.22

Social Integration Mechanism 0.00 0.22 0.22

External Sources and Knowledge Complementarity 0.00 0.15 0.15

Compared to the initial model, the final model contains the same construct, but some variables were eliminated

due to insignificance, i.e. variable J23, J24, J33, J34, J35, J36, and J45. Variable J23, J24, J33, J34, J35, and J36 were

trimmed due to insignificant factor loading (<0.30).

Variable J23 measures the complexity of knowledge that is required in a certain job such that an individual in-

volved in the job needs to be a specialist. The insignificant factor loading maybe due to the lack of respondent‟s com-

petence in judging whether or not a job requires a specialist. The question might be more appropriate for an expert on

job analysis and competency.

Variable J24 measures the system dependence characteristic of knowledge and its depedence on knowledge of

other individual in the organization. The insignificant factor loading may be caused by insufficient respondent‟s

knowledge to make a proper judgement about whether their knowledge is dependent on other‟s.

Variable J33 measures the inidividual ability to recognize changes that occur in market, especially changes on

regulations, competition, and technology. Variable J34 measures individual ability to analyze and interpret those

changes. Those 2 variables measures second aspects of Potential Absorptive Capacity (PACAP), i.e. assimilation abil-

ity.

The insignificant factor loading of those 2 variables is due to the fact thet not all respondent‟s jobs require them to

recognize, analyze, and interpret changes on regulation, competition, and technology. The questions are more appropri-

ate for employee whose job is related to market intelligence, business intelligence, atau technology watch.

Variable J35 measures individual ability to recognize the relationship between new external knowledge and cur-

rent internal knowledge, while variable J36 measures individual ability to capture an opportunity from new knowledge.

These 2 variables actually measure the first aspect of Realized Absorptive Capacity (RACAP), i.e. conversion ability.

The insignificant factor loading of these 2 variables may be caused by the fact that not all respondent‟s jobs require

them to recognize the relationship between external knowledge and current internal knowledge, and to recognize op-

portunities regarding the new knowledge. These questions are more appropriate for empoyees whose jobs are related to

product development, business development, or one who act as a gatekeeper in knowledge management initiative im-



plementation.

While variable J45 was trimmed because its regression coefficient was not significant. This variable measures the

extent to which an individual knows about knowledge outside of the organization. Eliminating this variable does not

bring any significant impact on the construct because it is still represented in 6 other variable.

Network

Attributes

Sources and

Complement

arity of

External

Knowledge

Potential

Absorptive

Capacity

Realized

Absorptive

Capacity

Flow of

Knowledge

Social

Integration

Mechanism

Knowledge

IntegrationTacitness

J4 J5 J6 J7J1 J2 J3J11 J12 J13J8 J9 J10

J14

J15

J16

J17

J21 J22J18 J19 J20J28J25 J26 J27 J29

J30

J31

J32

J37 J38 J39J40

J41

J42

J43

J44

J46

e1 e2 e3 e4 e5 e6 e7e8 e9 e10 e11 e12 e13

e37 e38 e39

e29

e30

e31

e32

e14

e15

e16

e17

e40

e41

e42

e43

e44

e46

e25 e26 e27 e28 e18 e19 e20 e21 e22

ePE

ePAC

eRAC

eAP

eIP

0,81

0,390,70

0,39

0,35

0,36

0,13

0,10

0,62

0,49

0,590,690,810,780,56

0,41

0,71

0,64

0,70

0,47

0,44

0,59 0,78 0,83

0,59

0,76

0,72

0,760,700,670,64 0,76

0,44

0,44

0,81

0,82

0,63 0,81 0,72 0,52 0,47 0,72 0,670,65 0,74 0,66 0,83 0,79 0,83

0,44

Figure 1: Final model after second trimming in standardized value.

6. Conclusions

Based on the parameterization and analysis, the following conclusions are drawn:

The flow of knowledge in Telkom is affected by some enablers, i.e. network attributes, sources and complementar-

ity of external knowledge, potential absorptive capacity, realized absorptive capacity, tacitness, knowledge integra-


18

tion, and social integration mechanism of the organization.

The impacts between enablers and of enablers on the flow of knowledge in Telkom that are significant (in stan-

dardized value) are: the positive impact of network attributes on flow of knowledge (0.39); the positive impact of

network attributes on sources and complementarity of external knowledge (0.81); the positive impact of sources

and complementarity of external knowledge on potential absorptive capacity (0.70); the positive impact of potential

absorptive capacity on realized absorptive capacity (0.35); the positive impact of realized absorptive capacity on

flow of knowledge (0.62); the positive impact of knowledge explicitness (or negative impact of knowledge

tacitness) on flow of knowledge (0.10); the positive impact of knowledge explicitness (or negative impact of

knowledge tacitness) on knowledge integration in the organization (0.49); the positive impact of knowledge

integration in the organization on potential absorptive capacity (0.39); the positive impact of knowledge integration

in the organization on realized absorptive capacity (0.36); the positive impact of social integration mechanism on

realized absorptive capacity (0.13); and the positive impact of social integration mechanism on knowledge

integration in the organization (0.44).

The overall measures of organizational knowledge in Telkom consist of the following constructs and variables:

- Network attributes, which has 6 measures: number of internal contacts; number of external contacts; strength of

internal network; strength of external network; scope of internal network; and scope of external network.

- Sources and complementarity of external knowledge, which has 6 measures: the extent to which individuals in

the organization have opportunities to be involved in relationships with external parties in the form of

acquisition, purchasing through licensing & contract, research & development, alliance, and joint venture;

individual access to external sources of knowledge; the extent to which individuals in the organization access and

use knowledge from external sources; the extent to which external contacts are willing to share their knowledge;

and complementarity of knowledge from external sources with internal knowledge.

- Tacitness, which has 4 measures: knowledge codifiability; knowledge teachability; knowledge complexity; and

knowledge system dependence.

- Social Integration Mechanism, which has 4 measures: intensity of problem solving through formal teams,

intensity of problem solving through informal mechanism, effectivity of problem solving through formal teams,

and effectivity of problem solving through informal mechanism.

- Knowledge integration, which has 5 measures: knowledge usefulness; ease of access to knowledge repository;

knowledge repository usefulness for knowledge sharing; ease of communication between individuals in the

organization; and effectivity of communication for knowledge sharing.

- Potential absorptive capacity, which has 3 measures: frequency of interaction between individuals in the

organization to acquire new knowledge; frequency of interaction between individuals in the organization and

customer or individuals outside the organization to acquire new knowledge; and intensity of knowledge

acquisition through informal mechanisms.

- Realized absorptive capacity, which has 3 measures: the extent to which individuals record new knowledge for

further reference; the extent to which individuals understand how to do their jobs and responsibilities; and the

extent to which individuals looking for better ways in doing their jobs.

- Flow of knowledge, which has 6 measures: flow of procedures and tools between individuals in the organization;

flow of ideas between individuals in the organization; flow of ideas between individuals in the organization and

individuals outside the organization; individual‟s awareness about knowledge in other individuals or units;

intensity of knowledge exchange between individuals in the organization; and intensity of knowledge exchange

between individuals in the organization and those outside the organization.

7. References

[1] W. Vestal, “Measuring Knowledge Management”, American Productivity and Quality Center, 2002.

[2] M. R. Trent, “Assessing Organization Culture Readiness for Knowledge Management Implementation: The Case of Aeronauti-

cal Systems Center Directorate of Contracting”, Thesis, Air-Force Institute of Technology, Wright-Patterson Air Force Base,

Ohio, USA, 2003.

[3] J. W. Boudreau, “Strategic Knowledge Measurement and Management”, Working Paper, Center for Advanced Human Re-

sources Studies, School or Industrial and Labor Relations, Cornell University, 2002.

[4] S. A. Zahra and G. George, “Absorptive Capacity: A Review, Reconceptualization, and Extension”, Academy of Management

Review 2002, 27 (2), 2002, pp. 185 - 203.

[5] P. Almeida and B. Kogut, “Localization of Knowledge and the Mobility of Engineers in Regional Networks”, Management

Science, 45 (7), 1999, pp. 905 – 917.

[6] U. Zander and B. Kogut, “Knowledge and the Speed of the Transfer and Imitation of Organization Capabilities: An Empirical



Test”, Organization Science, 6 (1), 1995, pp. 76 – 92.

[7] J. J. P. Jansen, F. A. Bosch, and H. W. Volberda, “Managing Potential and Realized Absorptive Capacity: How Do Organiza-

tional Antecedents Matter?”, Accepted to be published in The Academy of Management Journal, 2005.

[8] PT. Telekomunikasi Indonesia, Tbk., “Menjadi Model Korporasi Terbaik Indonesia”, Annual Report 2006, 2007.

[9] R. E. Schumacker, and R. G. Lomax, “A Beginner‟s Guide to Structural Equation Modeling”, Second Edition, Lawrence Erl-

baum Associates, Publishers, 2004.


20



Large Scale Optimization Based on Self-Directed

Local Search

Eman Hasan; Daryl Essam; Ruhul Sarker University of New South Wales at Australian Defense Force Academy, Canberra, Australia

[email protected], [email protected], [email protected]

ABSTRACT

In this paper we propose, a Memetic Algorithm (MA) that is based on a self-directed Local Search (MA-sd-LS), for

solving large scale problems using a random grouping decomposition technique. To overcome the large dimensionality

difficulties, the large scale problem is decomposed into smaller subproblems. As the correct subproblem size is hard to

determine, the motivation of this work is to investigate the effect that the subproblem size will have on the optimization

process for problems of different types of structure. Also, this work considers the role that self-directed Local Search

has in guiding the search to the most promising solutions MA is applied in the large scale domain. MA-sd-LS has

achieved significantly higher performance than one of the Evolutionary Algorithms (EA) in the literature

(DECC-CG[1]), in most of the benchmark problems defined in the Special Session on Large Scale Optimization in the

IEEE Congress on Evolutionary Computation in 2010.

Keywords: Memetic Algorithms, Evolutionary Algorithms, Local Search, Large Scale Optimization, Problem Decomposition.

1. Introduction

Over the last few decades, solving large scale optimization problems has become a challenging area in the fields of

computer science, operations research, and industry. This is mainly due to both the increased need for high quality

decision making for large scale problems in practice, and also the availability of increased computational power. In

general terms, solving small scale nonlinear optimization problems is not easy. The high dimensionality of the large

scale problems, along with their complex structure and interdependencies of variables, make them even more diff i-

cult to solve when using the same approaches which are used for small scale problems. The optimization of a large

scale problem usually consumes a massive amount of computational effort that might go beyond the capabilities of

the available computing resources. This has lead to the development of optimization algorithms for large scale

problems which use a decomposition technique to divide the total computation tasks of the large scale problems into

several smaller subproblems. Moreover, it is difficult to develop a mechanism to keep the same value for an inte r-

dependent variable which has more than one instance in the decomposed subproblems, and to also later merge back

the solutions of the subproblems to generate the complete solution of the problem is a demand.

One of the attempts to improve the performance of the optimization algorithms for large scale optimization

problems, is the divide and conquer strategy which was first introduced by Potter and DeJong [2], which has since

then been referred to as Cooperative Coevolution (CC). In this approach, a large optimization problem is deco m-

posed into smaller scale subproblems which can then be optimized separately using any optimization algorithm.

Although CC seems a promising framework for large scale optimization problems, its performance varies depen d-

ing on the separability of the considered optimization problem [3]. For example, CC is inefficient in solving non-

separable optimization problems [4]. This is due to the fact that CC does not have a systematic way to group the

interdependent variables of a nonseparable problem. When these interdependent variables are optimized in different

subproblems, there will be a major decline in the overall performance of the optimization algorithm [5]. This indi-

cates the importance of grouping interdependent variables in one subproblem.

To overcome the CC limitation, it is necessary to use an appropriate technique to decompose large scale opt i-

mization problems, and to also have an efficient mechanism for information exchange for the interdependent var i-

ables when they are optimized in one subproblem and have instances in other subproblems. The early efforts for

decomposing large scale problems used a decomposition approach that divided the problem into one variable for

each subproblem (one-dimension based), or into two equal size subproblems (splitting-in-half strategy) [6-7]. Later,

some other approaches were introduced to decompose the large scale problem into many subproblems of a certain

size [8-10]. However, specifying a subproblem size is a compromise between complexity and the algorithm‟s pe r-




Large Scale Optimization Based on Self-Directed Local Search

22

formance [5, 11]. The smaller the subproblem sizes in the separable problems, the easier the optimization, but the

larger the number of subproblems that must be evaluated. Also, for nonseparable problems, the larger the subprob-

lem‟s size, then the better the algorithm performance, but the more complex the optimization [11].

An optimization algorithm can compensate for the complexity of specifying the appropriate subproblem size if

it is powerful and efficient. Over the last few decades Evolutionary algorithms (EAs) have successfully proven

themselves as efficient optimization techniques[12]. EAs are a set of techniques which have the common feature of

being inspired by the natural evolution of species. They have achieved great success on many numerical and co m-

binatorial optimization problems [13], and they can deal with multimodality, discontinuities, and noisy functions

[3]. Moreover, EAs have the advantages of being widely applicable as a simple and flexible approach, and of ha v-

ing a robust response to dynamic optimization problems [14]. In spite of these advantages, EAs‟ performance de-

clines when dealing with large scale optimization problems. However because it is a flexible algorithm, EAs can be

hybridized with Local Search (LS) techniques. It is known that hybridizing EAs with other techniques can improve

the performance of the optimization process[15]. EAs that have been hybridized with LS are often called Memetic

Algorithms (MA). LS is a technique that iteratively improves its estimate of better solutions by searching in the

local neighborhood of the current solution [16]. Combining EAs with LS to form a MA, has been proven to refine

the search mechanism of optimization algorithms when they are applied to large scale problems [17].

In this work, we present a MA algorithm with a self-directed LS to optimize large scale problems. To investi-

gate its performance, we have applied this algorithm to the specific test suite proposed in the Special Session on

Large Scale Continuous Global Optimization in the 2010 IEEE Congress on Evolutionary Computation. The results

of the MA were analyzed against DECC-CG [1], and has been proven to be comparable in most of the benchmark

problems. This shows the role of the self-directed LS in achieving better performance. When applying the model

with variant subproblem sizes, the algorithm achieves high performance with a large subproblem size in the pro b-

lems that contain many interdependent variables and with a small subproblem size for other problems. This reveals

the relationship between the subproblem size and the interdependencies among variables.

The rest of this paper is organized as follows: section 2 presents our proposed methodology. The experiments

are presented in section 3. Section 4 views the results and analysis. Finally section 5 concludes this paper.

2. Proposed Methodology

In this work we propose an MA algorithm with a self-directed LS for solving Large Scale problems (MA-sd-LS). The

proposed model has been applied on 20 benchmark problems [18], where the large problems are decomposed into

smaller subproblems so as to investigate the effect of the subproblem size on the algorithm‟s performance. MA-sd-LS

achieves higher and comparable results to other algorithms in literature. The used decomposition technique in

MA-sd-LS is Random Grouping (RG). RG is one of the techniques that showed significant improvement over the

original CC for large scale optimization problems [11]. RG decomposes the large problem by grouping the variables

randomly into smaller size subproblems. In this approach, RG increases the probability of grouping dependent variables

in the same subproblem, which is recommended for the nonseparable problems. Each subproblem is optimized sepa-

rately, where any EA can be used. In this proposed methodology we are using a Genetic Algorithm (GA). In GA, Simu-

lated Binary Crossover (SBX) is used to generate offspring y1 and y2 of two parents x1, x2 as in (1) and (2). Where β is

generated from (3), and is a constant value ( =2 is used by most practitioners).

𝑦𝑖1 =

1

2 1 + 𝛽 × 𝑥𝑖

1 + 1 − 𝛽 × 𝑥𝑖2 (1) 𝑦

𝑖2 =

1

2 1 + 𝛽 × 𝑥𝑖

2 + 1 − 𝛽 × 𝑥𝑖1 (2)

𝛽 =

(2𝑢)

1

1+𝜂 ,𝑢 ≤ 0.5

(1

2(1−𝑢))

1

1+𝜂 , 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒

(3)

The mutation operator of the used GA is nonuniform mutation. As in non-uniform mutation, the step size decreases as

the generations increase, the search is uniform in the initial space and gets smaller as the algorithm proceeds [19]. Off-

spring xi′ t = xi,1

′ , xi,2′ ,… , xi,n

′ are created according to (4).

𝑥𝑖 ,𝑗′ = 𝑥𝑖 ,1

′ + 𝛿𝑖 ,𝑗 𝑡 (4)

𝛿𝑖 ,𝑗 𝑡 = 𝑥𝑖 ,𝑢𝑝𝑝𝑒𝑟 − 𝑥𝑖 ,𝑗 × 1 − [ 𝑢 𝑡 1−

𝑡

𝑇 𝑏

,𝑤𝑖𝑡ℎ 𝑝𝑟𝑜𝑏𝑎𝑏𝑖𝑙𝑖𝑡𝑦 0.5

𝑥𝑖 ,𝑙𝑜𝑤𝑒𝑟 + 𝑥𝑖 ,𝑗 × (1 − [ 𝑢 𝑡 (1−𝑡

𝑇)𝑏 ),𝑤𝑖𝑡ℎ 𝑝𝑟𝑜𝑏𝑎𝑏𝑖𝑙𝑖𝑡𝑦 0.5

(5)



In MAs, LS is the component which is most directly affected by dimensionality. LS is used to explore the neighborhood

around the current solution, and so high dimensionality increases that region and the overall domain. As LS is a com-

putationally expensive tool, for it to be scalable in a large search domain, the number of search iterations made by LS

should be minimized, and each executed iteration should achieve effective improvements to the solution. This is what

motivates the development of the self-directed LS in our proposed algorithm.

In this paper, MA-sd-LS uses an adaptive search step (d) in the LS which is directed towards the good solutions.

This makes the LS scalable so that it can be applied in large search domains, as it only steps forward if it enhances the

solution. The d value is expanded as in (6) after each successful search iteration. If the enlarged d fails, this is an indica-

tion of a local optimum, or that the global optimum is located in the previously searched domain. In this case, the d

value is doubled a few times to push the solution out of the local optimum and to thus expand the search area so that it

might contain the global optimum. If these new d values fail to move the solution forward, then the new d should take a

small step from the last successful d as in (7), to thus extensively concentrate the search in the current search space.

Regardless of the added cost of the adaptation of d, it helps to overcome the difficulty that is imposed by the large di-

mensionality.

dedd /1 (6) )/1( dedd (7)

The present decomposition techniques cannot detect the variables‟ interdependencies. This means that the decom-

posed subproblems may include variables which are interdependent with other variables in a different subproblem.

Hence, because the subproblems are optimized separately, the different instances of all variables should be maintained

during the optimization process to represent the latest value of the optimized interdependent variable. The migration of

the interdependent variables to the other subproblems is controlled and is taken into consideration in our proposed algo-

rithm. All the variables are collected and the complete solution is upgraded throughout the optimization process.

Based on the previous discussion, the detailed steps of the proposed methodology are summarized as follows:

1. Generate the initial population with size NP for the dimension D variables.

2. Decompose the large scale problem randomly into subproblems subk where k=[1,m] and D=k*m

3. Optimize each subk

4. Apply LS to one random variable of the few best xi,j of subk:

a. Create d that changes with the LS_iter

b. add, and also subtract d and select the direction that enhances fitness

c. repeat until l=LS_iter

i. if fitness increases, then use (6) to enlarge d

ii. else, use (7) to decrease d

5. copy the value of the optimized variables in all other subk

6. While k<=m go to step 3

7. If FE < max_FE go to step 2

3. Experiments

MA-sd-LS has been tested on 20 benchmark large scale optimization problems [18]. These problems are designed into

4 categories where f1 to f3 are separable, f4 to f8 are partially-separable such that a small number of variables (m=50)

are dependent while all the remaining ones are independent. Functions f9 to f18 are partially-separable functions that

consist of multiple independent groups, each of which is m-nonseparable. The experiments are implemented with sub-

problem sizes of 5, 50, and 100 and the results are analyzed to investigate the effect of subproblem size on the algo-

rithm‟s performance. In this experiment, subproblems are optimized sequentially for a certain number of generations.

An interdependent variable is optimized in a subproblem and its final value is copied into the other subproblems which

contain the dependent variables. After finishing the optimization of all subproblems - which will be referred to as a new

cycle – the algorithm starts again with a different random grouping. The subproblems are repeatedly optimized until

reaching the stopping criteria, which is the maximum number of fitness evaluations (max_FE) of 3e+6 in these experi-

ments. Parameter setting: Tournament selection of size two, Simulated Binary Crossover (SBX), and Nonuniform

mutation are used. The mutation probability changes adaptively through generations from 0.15 to 0.1. The LS is

self-directed by the performance as in (6) and (7).


24

4. Results

To analyze the differences directly, we view the results obtained by each algorithm in Table 1. The mean and median

values are remarked as bold for the best algorithm.

We can conclude the following from Error! Reference source not found.:

MA-sd-LS achieved the best results in 16 functions out of 20 where 12 of them is partially-separable, two separable,

and 2 fully nonseparable.

MA-sd-LS achieved the best results in all the Rosenbrocks functions (such as f8, f13, f18, and f20)

MA-sd-LS achieved higher and relatively close results in Rastrigins (f10, f15) and the best results in Schwefels (f12,

and f17) only when the variables are partially-separable and consist of multiple independent groups, each of which is

m-nonseparable.

For the separable functions and fully-nonseparable categories, MA-sd-LS obtains the best results for the multimodal

functions (f2, f3, and f20).

The results of most of the partially-separable problems and one of the nonseparable problems in [18] with subprob-

lem size 100, achieved higher performance to the results obtained from the DECC-CG [1].

As the differences between mean and median are small in most of the functions, we can conclude that the proposed

MA-sd-LS algorithm is robust and stable.

Table 1: Results of MA-sd-LS and DECC-CG, subproblem size=100 and FE=3e+06

f1 f2 f3 f4 f5 f6 f7 f8 f9 f10

Mean 3.42e+08 1.85e+02 2.85e-05 2.79e-03 5.61e+11 1.67e+06 4.88+08 4.15e+07 5.18e-04 7. 54e+03

Median 3.38e+08 1.86 e+02 2.92 e-05 2.87e-03 5.85 e+11 1.67e+06 4.87e+08 3.00 e+07 5.05e-04 7.61e+03

MA-sd-LS Std. 1.26e+07 4.39e+00 2.40e-06 2.13e-04 1.39 e+11 4.72-02 3.18e+8 3.12e+07 7.73e-05 1.81e+03

Best 3.28e+08 1.81e+02 2.45 e-05 2.49e-03 3.66 e+11 1.67e+06 1.61e+08 2.21e+06 4.15e-04 7.25e+03

Worst 3.61e+08 1.91e+02 3.04 e-05 3.03e-03 6.92 e+11 1.67e+06 8.21e+8 7.41e+07 6.23e-04 7.71e+03

Mean 2.93e-07 1.31e+03 1.39e+00 1.70e+13 2.63e+08 4.96e+06 1.63e+08 6.44e+07 3.21e+08 1.06e+04

Median 2.86e-07 1.31e+03 1.39e+00 1.51e+13 2.38e+08 4.80e+06 1.07e+08 6.70e+07 3.18e+08 1.07e+04

DECC-CG Std. 8.62e-08 3.26e+01 9.73e-02 5.37e+12 8.44e+07 8.02e+05 1.37e+08 2.89e+07 3.38e+07 2.95e+02

Best 1.63e-07 1.25e+03 1.20e+00 7.78e+12 1.50e+08 3.89e+06 4.26e+07 6.37e+06 2.66e+08 1.03e+04

Worst 4.84e-07 1.40e+03 1.68e+00 2.65e+13 4.12e+08 7.73e+06 6.23e+08 9.22e+07 3.87e+08 1.17e+04

f11 f12 f13 f14 f15 f16 f17 f18 f19 f20

Mean 1.77e+01 8.01e+03 7.27e+02 2.54e-03 1.52e+04 3.33e+01 1.39e+04 5.10e+03 4.24e+5 1.83e+01

Median 1.77e+01 7.38e+03 6.27 e+02 2.58e-03 1.52 e+04 3.33e+01 1.32e+04 4.66e+03 4.34e+05 9.44e+00

MA-sd-LS Std. 6.49e-06 1.38e+03 4.96e+02 2.68e-04 3.59e+02 5.47e-06 3.32e+03 2.77e+03 5.60e+04 1.81e+01 Best 1.77 e+01 6.76 e+03 3.06 e+02 2.13e-03 1.47 e+04 3.33e+01 1.11e+04 2.49e+03 3.29e+05 1.85e+00

Worst 1.77 e+01 1.02 e+04 1.58 e+03 2.86e-03 1.61 e+04 3.33e+01 1.95e+04 9.15e+03 4.71e+05 3.82e+01

Mean 2.34e+01 8.93e+04 5.12e+03 8.08e+08 1.22e+04 7.66e+01 2.87e+05 2.46e+04 1.11e+06 4.06e+03

Median 2.33e+01 8.87e+04 3.00e+03 8.07e+08 1.18e+04 7.51e+01 2.89e+05 2.30e+04 1.11e+06 3.98e+03 DECC-CG Std. 1.78e+00 6.87e+03 3.95e+03 6.07e+07 8.97e+02 8.14e+00 1.98e+04 1.05e+04 5.15e+04 3.66e+02

Best 2.06e+01 7.78e+04 1.78e+03 6.96e+08 1.09e+04 5.97e+01 2.50e+05 5.61e+03 1.02e+06 3.59e+03

Worst 2.79e+01 1.07e+05 1.66e+04 9.06e+08 1.39e+04 9.24e+01 3.26e+05 4.71e+04 1.20e+06 5.32e+03

After analyzing the results and the concluded observations from Error! Reference source not found., it is obvious that

our algorithm is successful in solving different types of functions whether separable or nonseparable, and all the par-

tially-separable functions except for the Rastrigins (which is originally separable) and Schwefels (which is originally

nonseparable) when a small number of variables are dependent and all the remaining ones are independent (f5, f7, and

f15). However, MA-sd-LS has been proven to be successful in solving Rastrigins and Schwefels in their original struc-

ture (f2, and f19). Although DECC-CG is better in the Ackleys‟ function [20], our proposed algorithm achieved higher

performance in all of them (f3, f6, f11, and f16) and is comparable to DECC-CG.

The mean results obtained by MA-sd-LS are compared with the others obtained by DECC-CG in Table 2 using the

Wilcoxon‟s test which is described in [21]. This test shows that although there is no significance difference, MA-sd-LS

achieves a higher rank than DECC-CG.

Table 2: MA-sd-LS versus DECC-CG (Wilcoxon's test with p-value=0.05)

Algorithm R+ R- Sig. difference?

DECC-CG 134 56 No

We have provided the convergence curves of the average values of problems f2 (shifted-Rasrigins), f3

(shifted-Ackelys), f5 (rotated-Rastrigins), f9 (rotated-Elliptic), f12 (Schwefel), and f18 (rotated-Rosenbrocks) as represen-



tative samples for subproblem size of 100 in figures Figure1 - Figure 6. We observe that f2, f9 convergence quickly at

the first fitness evaluations and the convergence continue but slower. From the convergence curve of f3, we observe that

it achieves fast improvement at the beginning, and then the fitness value is almost stabilized after 1e+06 evaluations to

the end. We observe that the curve of f5 has regions of different convergences, and from evaluation 2e+06 until 3e+06

the improvement is too slow. The convergences curve of f12, and f18 have a steep slope and continue in a horizontal

convergence after 6e+05 evaluations.

The results in Table 3 show the relationship between the performance and the subproblem size. It is clear that the

separable problems (such as f1, f2, and f3) achieve higher results when they are decomposed into smaller subproblems.

Even at the partially-separable problems where the separability is represented by Rastrigins separable function (such as

f5, f10, and f15) we can notice that the smaller the subproblem the higher the performance. For the partially-separable and

fully-nonseparable functions (such as f4, f5, f7, f8, f9, f12, f13, f14, f17, f18, f19 and f20); they achieve higher performance when

they are decomposed into larger subproblems. We can observe from Table 3 that the partially-separable Ackelys f6, f11

and f16, obtain the same results when decomposed into different subproblem sizes. These remarks from Table 3 indicate

how the large scale problem decomposition in highly related to the problem separability.

Table 3: Different subproblem size using MA-sd-LS with Dimension of 1000

Size f1 f2 f3 f4 f5 f6 f7 f8 f9 f10

Mean 3.42e+08 1.85e+02 2.85e-05 2.79e-03 5.61e+11 1.67e+06 4.88+08 4.15e+07 5.18e-04 7. 54e+03

Median 3.38e+08 1.86 e+02 2.92 e-05 2.87e-03 5.85 e+11 1.67e+06 4.87e+08 3.00 e+07 5.05e-04 7.61e+03

S=100 Std. 1.26e+07 4.39e+00 2.40e-06 2.13e-04 1.39 e+11 4.72-02 3.18e+8 3.12e+07 7.73e-05 1.81e+03

Best 3.28e+08 1.81e+02 2.45 e-05 2.49e-03 3.66 e+11 1.67e+06 1.61e+08 2.21e+06 4.15e-04 7.25e+03

Worst 3.61e+08 1.91e+02 3.04 e-05 3.03e-03 6.92 e+11 1.67e+06 8.21e+8 7.41e+07 6.23e-04 7.71e+03

Mean 4.40e+07 1.99e+02 2.09e-05 1.02e-02 5.89e+11 1.67e+06 6.89e+09 5.11e+07 6.07e-04 7.34e+03

Median 4.67e+07 1.95e+02 2.06e-05 1.02e-02 5.32e+11 1.67e+06 7.02e+09 3.78e+07 6.04e-04 7.39e+03

S=50 Std. 6.91e+06 1.94e+01 5.33e-07 1.02e-03 1.40+10 1.63e-01 1.58e+09 2.42e+07 6.87e-05 3.58e+02

Best 3.22e+07 1.73e+02 2.04e-05 9.20e-03 4.92e+11 1.67e+06 4.97e+09 2.70e+07 5.44e-04 6.78e+03

Figure 1: Mean results for f2







26

Size f1 f2 f3 f4 f5 f6 f7 f8 f9 f10

Worst 4.92e+07 2.21e+02 2.16e-05 1.16e-02 8.37e+11 1.67e+06 9.06e+09 7.89e+07 7.20e-04 7.70e+03

Mean 3.08e+07 5.38e+01 8.46e-06 4.14e-01 5.23e+11 1.67e+06 6.23e+10 6.92e+07 1.08e-02 7.22e+03

Median 2.74e+07 5.50e+01 8.49e-06 2.50e-01 5.22e+11 1.67e+06 5.76e+10 6.22e+07 9.31e-03 7.15e+03

S=5 Std. 2.54e+07 3.93e+00 3.41e-07 3.06e-01 3.17e+10 0.00e+00 1.75e+10 1.28e+07 4.19e-03 2.49e+2

Best 3.03e+05 4.76e+01 7.96e-06 1.39e-01 4.76e+11 1.67e+06 4.23e+10 5.59e+07 5.72e-03 6.96e+03

Worst 6.92e+07 5.76e+01 8.85e-06 8.31e-01 5.63e+11 1.67e+06 8.95e+10 8.32e+07 1.56e-02 7.59e+03

f11 f12 f13 f14 f15 f16 f17 f18 f19 f20

Mean 1.77e+01 8.01e+03 7.27e+02 2.54e-03 1.52e+04 3.33e+01 1.39e+04 5.10e+03 4.24e+5 1.83e+01

Median 1.77e+01 7.38e+03 6.27 e+02 2.58e-03 1.52 e+04 3.33e+01 1.32e+04 4.66e+03 4.34e+05 9.44e+00

S=100 Std. 6.49e-06 1.38e+03 4.96e+02 2.68e-04 3.59e+02 5.47e-06 3.32e+03 2.77e+03 5.60e+04 1.81e+01

Best 1.77 e+01 6.76 e+03 3.06 e+02 2.13e-03 1.47 e+04 3.33e+01 1.11e+04 2.49e+03 3.29e+05 1.85e+00

Worst 1.77 e+01 1.02 e+04 1.58 e+03 2.86e-03 1.61 e+04 3.33e+01 1.95e+04 9.15e+03 4.71e+05 3.82e+01

Mean 1.77e+01 1.32e+05 1.36e+03 1.37e-02 1.53e+04 3.33e+01 2.64e+05 2.18e+04 9.97e+05 1.51e+02

Median 1.77e+01 1.35e+05 6.88e+02 1.36e-02 1.55e+04 3.33e+01 2.67e+05 2.32e+04 9.92e+05 1.45e+02

S=50 Std. 9.03e-06 1.49e+04 1.49e+03 1.68e-02 3.70e+02 8.68e-05 1.77e+04 1.02e+04 5.10e+04 1.54e+01

Best 1.77e+01 1.13e+05 6.28e+2 1.20e-02 1.48e+04 3.33e+01 2.56e+05 1.15e+04 9.26e+05 1.33e+02

Worst 1.77e+01 1.52e+05 4.03e+3 1.56e-02 1.58e+04 3.33e+01 2.90e+05 3.56e+04 1.06e+06 1.71e+02

Mean 1.77e+01 6.24e+05 8.09e+02 5.68e-01 1.45e+04 3.33e+01 1.17e+06 2.38e+04 2.57e+06 2.20e+02

Median 1.77e+01 6.16e+05 8.52e+02 5.95e-01 1.44e+04 3.33e+01 1.15e+06 2.21e+04 2.51e+06 2.24e+02

S=5 Std. 5.57e-06 2.86e+04 9.63e+01 9.57e-02 2.09e+02 1.14e-05 4.07e+04 7.86e+03 2.10e+05 1.22e+01

Best 1.77e+01 5.91e+05 6.96e+02 4.25e-01 1.42e+04 3.33e+01 1.14e+06 1.66e+04 2.36e+06 1.99e+02

Worst 1.77e+01 6.63e+05 9.06e+02 6.76e-01 1.48e+04 3.33e+01 1.23e+06 3.38e+04 2.80e+06 2.30e+02

5. Conclusion

In this paper, we have proposed a Memetic Algorithm MA-sd-LS that is based on self-directed Local Search. In it, large

scale problems are decomposed into smaller subproblems which are optimized separately. MA-sd-LS has obtained good

results comparable to the DECC-CG algorithm in most of the separable, partially-separable, and all nonseparable prob-

lems proposed by the organizers of the Special Session of Large Scale Global Optimization, in the IEEE Congress on

Evolutionary Computation 2010 [18]. This emphasises the advantage of using the self-directed LS to guide the search to

the most promising solutions. We have carried out empirical studies to analyze how the subproblem size affects the

performance of the optimization algorithm, following the benchmark problems [18]. Experiments have shown that there

is a relationship between the performance and the subproblem size. The separable problems achieve high results when

they are decomposed into smaller subproblems, even at the partially-separable problems. But for the partially-separable

problems in which the nonseparable problems are used and the fully-nonseparable functions, they achieve higher per-

formance when they are decomposed into larger subproblems. To get better results, the problem separability should be

known a head before starting the optimisation. This conclusion indicates the importance of a systematic approach that

can determine the problem structure that suits a certain subproblem size. Our future work will focus more on the prob-

lem identification before decomposition, in order to make more accurate groupings of the interdependent variables and

to specify the most appropriate subproblem size. As the decomposition of subproblems is applicable for optimization in

a parallel computing environment, this will also be implemented in future research.

6. References

[1] Yang, Z., K. Tang, and X. Yao, Large scale evolutionary optimization using cooperative coevolution. Information Sciences,

2008. 178: p. 2986-2999.

[2] M. A. Potter and K.A.D. Jong. A Cooperative Coevolutionary Approach to Function Optimization. in The Third Prallel Prob-

lem Solving From Nature. 1994. Berline, Germany: Springer-Verlag.

[3] T. Ray and X. Yao. A Cooperative Coevolutionary Algorithm with Correlation Based Adaptive Variable Partitioning. in IEEE

Congress on Evolutionary Computation. 2009.

[4] M. N. Omidvar, X. Li, Z. Yang, and X. Yao. Cooperative co-evolution for large scale optimization through more frequent

random grouping. 2010.

[5] M. Omidvar, X. Li, and X. Yao. Cooperative Co-evolution with Delta Grouping for Large Scale Non-separable Function Op-

timization. in 2010 IEEE World Congress on Computational Intelligence. 2010. Barcelona, Spain.

[6] Potter, M. and K. De Jong, A cooperative coevolutionary approach to function optimization, in Parallel Problem Solving from

Nature — PPSN III, Y. Davidor, H.-P. Schwefel, and R. Männer, Editors. 1994, Springer Berlin / Heidelberg. p. 249-257.



[7] Potter, M.A. and K.A.D. Jone, Cooperative Coevolution: An Architecture for Evolving Coadapted Subcomponents. Evolution-

ary Computation, 2000. 8(1): p. 1-29.

[8] Y. Liu, X. Yao, Q. Zaho, and T. Higuchi. Scaling Up Fast Evolutionary Programming with Cooperative Coevolution. in Com-

gress on Evolutionary computation. 2001.

[9] X. Li and X. Yao. Tackling high dimensional nonseparable optimization problems by cooperatively coevolving particle

swarms. in IEEE Congress on Evolutionary Computation (CEC) 2009. 2009. Trondheim, Norway IEEE.

[10] Z. Yang, K. Tang, and X. Yao. Differential Evolution for High-Dimensional Function Optimization. in 2007 IEEE Congress on

Evolutionary Computation. 2007.

[11] Z. Yang, J. Zhang, K. Tang, X. Yao, and A. Sanderson. An adaptive coevolutionary differential evolution algorithm for

large-scale optimization. in Proceedings of the Eleventh conference on Congress on Evolutionary Computation. 2009. Trond-

heim, Norway: IEEE Press.

[12] Sareni, B., L. Krahenbuhl, and A. Nicolas, Efficient genetic algorithms for solving hard constrained optimization problems.

Magnetics, IEEE Transactions on, 2000. 36(4): p. 1027-1030.

[13] R. Sarker, M. Mohammadian, and X. Yao, Evolutionary Optimization. 2002, MA, USA: Kluwer Academic Publishers Nor-

well.

[14] Lozano, M. and C. García-Martínez, Hybrid metaheuristics with evolutionary algorithms specializing in intensification and

diversification: Overview and progress report. Computers & Operations Research, 2010. 37(3): p. 481-497.

[15] Davis., L., Handbook of Genetic Algorithms. 1991, New York: Van Nostrand Reinhold.

[16] Hart, W.E., Adaptive Global Optimization with Local Search. 1994, University of California at San Diego, La Jolla, CA, USA.

[17] Zhao, S.Z., J.J. Liang, P.N. Suganthan, and M.F. Tasgetiren. Dynamic multi-swarm particle swarm optimizer with local search

for Large Scale Global Optimization. in Evolutionary Computation, 2008. CEC 2008. (IEEE World Congress on Computa-

tional Intelligence). IEEE Congress on. 2008.

[18] Tang, K., X. Li, P.N. Suganthan, Z. Yang, and T. Weise, Benchmark Functions for the CEC 2010 Special Session and Compe-

tition on Large Scale Global Optimization. Technical report, in Nature Inspired Computation and Applications Laboratory.

2009: USTC, China.

[19] Z. Michalewicz, Genetic Algorithms + Data Structures = Evolution Programs. 1992, New York: Springer-Verlag.

[20] Molina, D., M. Lozano, A. Sánchez, and F. Herrera, Memetic algorithms based on local search chains for large scale continu-

ous optimisation problems: MA-SSW-Chains. Soft Computing - A Fusion of Foundations, Methodologies and Applications,

2010: p. 1-20.

[21] S. Garc´ıa, D. Molina, M. Lozano, and F. Herrera., A study on the use of non-parametric tests for analyzing the evolutionary

algorithms‟ behaviour: A case study on the CEC‟2005 special session on real parameter optimization. Journal of Heuristics,

2009. 15: p. 617–644.


28



A Recovery Model for a Production-Inventory

System with Transportation Disruption

Hawa Hishamuddin, Ruhul Sarker, Daryl Essam School of Engineering and IT, University of New South Wales, ADFA Campus, Northcott Drive, Canberra 2600, Australia


ABSTRACT

Supply chains (SC) are becoming increasingly competitive and complex in order to effectively meet customer demands.

This nature and complexity of the SCs make them vulnerable to various risks, including disruptions due to interruptions

in supply, transportation and many other sources. In the presence of a disruption, managers are required to make quick

and reliable decisions to recover from the unexpected event with as minimal costs as possible. In this study, a recovery

model is proposed for a two stage production and inventory system that experiences a transportation disruption. The

model is capable of determining the optimal ordering and production quantities during the recovery window such that

the total relevant costs are minimized, while seeking to recover the original schedule. Such tools are useful to assist

managers in effective decision making in response to disruptions, in particular when determining the optimal recovery

strategy for the longevity and sustainability of their businesses.

Keywords: Transportation disruption, recovery model, two stage inventory-production system, supply chain

1. Introduction

SC disruption is defined as an event that interrupts the normal course of operations of the effected SC entities. Disrup-

tions can be caused by internal or external sources to the SC, including machine breakdowns, transportation failure,

natural disasters, labor dispute, terrorism, war, and political instability. In recent years, we have come to see many dis-

ruption occurrences that have severely affected SCs [1]. Transportation disruption is slightly different from other forms

of SC disruptions, in that it only stops the flow of goods, whereas other disruptions may stop the production of goods as

well. It is distinctive in that the goods in transit have halted, even though the other operations of the SC are intact [2].

It is crucial that managers take appropriate preparatory measures of response, such as mitigation or contingency

strategies, to reduce the negative effects of these disruptions [3]. One of the goals of Disruption Management is to im-

plement the correct strategies that will enable a SC to quickly return to its original state, while minimizing the relevant

costs associated with recovery from the disruption [4].

In the literature on supply uncertainty or supply-disruption, where the supplier is not always available, numerous

studies have been performed for inventory models under the continuous review [5], [6] and the periodic review frame-

works [7], [8]. Although SC disruption in general has recently gained the interest of many researchers, the study on

transportation disruption in particular has received much less attention. Wilson [2] investigates the effect of transporta-

tion disruption on SC performance using system dynamics. The work concluded that the most severe impact is experi-

enced when transportation disruption exists between the tier 1 supplier and the warehouse. Studies of transportation

disruption can also be found in the literature on Emergency Logistics Scheduling, which is the integration of machine

scheduling and job distribution to customers with the consideration of disruption events [9]. Another study has been

conducted in the area of the Vehicle Routing Problem (VRP) by Sun et al.[10], who presented a hybrid knowledge rep-

resentation framework for disruption management problems in urban distribution decisions.

The model proposed in this paper studies a real-time rescheduling mechanism for an economic lot sizing problem

of a two stage SC system subject to transportation disruption. The recovery model is different from the works men-

tioned earlier in a number of ways. Our problem differs from Xia et al.'s model [11], in that to make our model more

realistic, disruption is in the form of a transportation disruption, which is not known a priori. Additionally, we have

considered penalty costs, as well as stock-out costs consisting of both backorder and lost sales cost. We are reluctant to

make assumptions such as that the inter-arrival time of supply disruption and the duration of the supply disruptions are

exponentially distributed [6], [7]. Rather the two parameters are assumed to be a random variable in our model. As the

key contribution, we introduce a novel approach that determines the optimal recovery plan for a two stage produc-

tion-inventory system, subject to the system‟s costs and constraints.

A Recovery Model for Production-Inventory System with Transportation Disruption

30

The contents of the paper are organized as follows. Section 2 discusses the model development. This section in-

cludes derivation of the cost functions. Section 3 deals with the solution approach for the model. Section 4 addresses the

related computational results and analysis. Lastly, section 5 summarizes our research findings and offers potential di-

rections for future research.

2. Model Formulation

2.1. System Description

In this paper, we consider a two stage production and inventory system consisting of a manufacturer and a retailer. The

manufacturer has production and inventory, and thus follows the economic production quantity model, while the retailer

only has inventory and follows the economic order quantity model. The notations used in developing the cost function

are as follows:

A1 setup cost for the first stage ($/setup)

A2 ordering cost for the second stage ($/order)

D demand rate for the system (units/year)

H1, H2 annual inventory cost for stage 1 and 2 ($/unit/year)

P production rate (units/year)

Q1 production lot size for stage 1 in the original schedule (units)

Q2 ordering lot size for stage 2 in the original schedule (units)

Xi production lot size of cycle i in the recovery schedule for stage 1 (units)

Si order lot size of cycle i in the recovery schedule for stage 2 (units)

Bq back order quantity for stage 2

Lq lost sales quantity for stage 2

Td disruption period

ρ production up time for a normal cycle (Q/P)

u production down time for a normal cycle

te start of recovery time window

tf end of recovery time window

T production cycle time for a normal cycle (Q/D)

B1, B2 unit back order cost per unit time for stage 1 and 2 ($/unit/time)

L1, L2 unit lost sales cost for stage 1 and 2 ($/unit)

CT unit transportation cost for each delivery ($/shipment)

W warehouse capacity for stage 2 (units)

T1i production time for cycle i in the recovery window for stage 1

T2i production time for cycle j in the recovery window for stage 2

n number of cycles in the recovery window

m number of lots in the recovery window

z number of optimal production lots in the recovery window

Ii inventory level at the end of cycle i in the recovery window

f1 the penalty function for the delay in recovering the original schedule in the first stage

f2 the penalty function for the delay in recovering the original schedule of the second stage handled by stage 1

f3 the penalty function for the delay in recovering the original schedule in stage 2

It is assumed that the demand rate is less than the production rate, i.e. D<P. As a preliminary study, we have cho-

sen the lot-for-lot policy to be applied to the model. For this particular type of shipment policy, the manufacturing lot

size for the first stage is equal to the ordering lot size of the second stage (Q1= Q2=Q) under ideal conditions, due to

coordination of the two stage system. The current production-inventory system is a modified version of the model pro-

posed by Banerjee [12], where the optimal production lot size (Q) is:

PHDH

AAPDQ

21

212

(1)

However, our model assumes that a truck experiences a disruption that prevents the goods from being dispatched to the

retailer as scheduled. The disruption may be caused by a major accident involving the truck or it can be due to natural

disasters, such as floods, earthquakes, or snow blizzards, which disrupt the truck from operating normally. In addition,

the disruption may or may not cause damage to the finished goods being transported. This paper investigates the dam-

aged lot case.



In the case where the goods are damaged, the manufacturer has to reproduce the damaged lot in order to satisfy

demand. Therefore, there will be changes in the original production schedule at the manufacturer‟s side. The retailer

only receives the goods after the production of the damaged lot is completed and this delay will result in shortages to

the retailer. Furthermore, no shortages are allowed during the subsequent cycles following the disruption. Moreover, our

model assumes that the transportation has unlimited capacity. It is assumed that the retailer owns limited amount of

warehouse capacity, (W). In addition, it is assumed that the second stage follows the zero-order inventory policy, where

an order is made only when the on-hand inventory reaches zero. Therefore, there is no inventory at the end of each cy-

cle. The first stage, on the other hand, can have left over inventory at the end of the cycles in the recovery window.

The objective of the problem is to determine the new recovery plan, consisting of the optimal order quantities for

the retailer and the production quantities for the manufacturer, so as to minimize the total recovery cost for the system.

Additionally, the aim is to return back to the original manufacturing and ordering schedule as soon as possible. This has

been the common purpose of various models in the disruption management literature and our model has the same aim.

Note that recovery is achieved when both stages are back to their original schedule. The duration in which the schedule

is allowed to have changes to achieve recovery is defined as the recovery time window [11], [13], which will be n cycle

times from the start of disruption. Extra costs are incurred in order to recover the system from the disruption, including

backorder (B1, B2), lost sales (L1, L2) and penalty costs (f1, f2, f3) for both the manufacturer and the retailer. Similar to

our previous work [13], our model assumes that the pre-disruption period in the disruptive cycle is zero.

2.2. Mathematical Representation

Let z be the optimal number of production lots in the recovery window, n be the number of cycles in the recovery win-

dow, m be the number of lots i.e. the demand to be satisfied and y be a binary parameter to represent the state of goods.

The relationship between n, m and y can be stated as follows:

m = n +y where

undamagedislotif

damagedislotify

0

1

(2)

Figure 1 depicts the inventory lines for stage 1 (manufacturer) and stage 2 (retailer). The dotted lines represent the

original non-disruption schedule, whereas the solid lines represent the new recovery schedule with the presence of dis-

ruption. The stripe-shaded triangle shows the amount of shortages, consisting of backorders (Bq) and lost sales (Lq),

incur- ed by the retailer during the disruption period, Td. In this figure, we have n = 3 recovery cycles and z = 4 produc-

tion lots for the recovery time window (te – tf). We define the decision variable Xi as the production quantity for cycle i

in the re-

Figure 1: Production Inventory Curve for a two stage SC for the damaged lot case

covery time window for the first stage (manufacturer) and T1i as its respective production time, where i = 1, 2, …, n.

The second decision variable, Si, is the ordering quantity for cycle i in the recovery window for the second stage (re-

tailer) and T2i is its respective consumption time. After a disruption of Td occurs, recovery takes place by utilizing the

S1-Bq

S2

S3

Tdte tf

T21 T22 T23Time

X1 X2

X3

te tf

T11 T12 T13

Time

S2

S31 2

3

4

5S4

Q

Q

STAGE 2

STAGE 1

Bq

Lq

T14

X4

6

S1

S1-Bq

S2

S3

Tdte tf

T21 T22 T23Time

X1 X2

X3

te tf

T11 T12 T13

Time

S2

S31 2

3

4

5S4

Q

Q

STAGE 2

STAGE 1

Bq

Lq

T14

X4

6

S1


32

production idle times, δ, in the original schedule. The time horizon is finite, such that only the costs in the recovery

window are considered. The total cost function considered is the sum of the average setup, inventory, transportation and

penalty costs per unit time, plus the total costs for shortages (backorders and lost sales). Given that shortages only occur

during the first cycle, we find that it is better to record it as a total and not as a time average.

2.3. Damaged Goods

For this particular case, we assume that the goods being transported are damaged during the disruption (y = 1). The total

costs for the second stage will first be formulated as this will ease in determining the costs for the first stage later, since

the production schedule for the first stage is dependent on the order schedule of the second stage. The setup cost equa-

tion for the first stage is rather straight forward and can be obtained by: A2(z-1) (3)

The inventory holding cost is derived as the unit inventory holding cost, H, multiplied by the total inventory dur-

ing the recovery time, which is equivalent as the area under the curve. This is calculated as:

...2

12332222112 TSTSTBqSH

1

2

22

12

2

z

i

iSBqSD

H (4)

The backorder cost formulation for the second stage can be derived as: LqDTTB

d

d

2

2

(5)

Finally, the lost sales cost is obtained as:

1

1

2

z

i

iSnQL

(6)

The penalty function derived in this model is based on the assumption that the longer it takes to recover the original

schedule, the higher the associated penalties. These penalties represent the extra costs incurred by the system when

there are changes in the original plan. Here we have derived it as a function of the number of recovery cycles, as indi-

cated below: f3(n2)

(7)

The sum of all the cost components above gives the total relevant costs of the recovery plan for the second stage, as

presented below:

1

1

2

22

3

1

2

22

12

222

)(2

11

),(z

i

id

dz

i

ii SnQLLqDTTB

nfSBqSD

HzA

nTzSTC

(8)

Next, the total relevant costs for recovery for the first stage will be calculated. The setup cost is given by: A1(z)

(9)

Let us define Ii as the inventory level at the end of cycle i in the recovery window, where

Ii = Ii-1+ Xi - Si for i = 1, 2, ... , z (10)

The inventory cost for the first stage is:

...

2

1

2

1

2

1

2

11441431331321221211111101 TXTITXTITXTITXTIH

z

i

i

iiP

XXIH

1

112

1 (11)

For this model, it is assumed that the manufacturer incurs a penalty for backorders and lost sales of the retailer‟s.

In other words, the manufacturer incurs a cost whenever a customer is unable to purchase the manufacturer‟s product

from the retailer. The backorder cost and the lost sales cost for the manufacturer follows the concept by Cachon and

Zipkin [14] and is given by equations (12) and (13) respectively.

LqDTTB

d

d

2

1

(12)

1

1

1

z

i

iSnQyQL

(13)

Instead of having a parameter that constitutes the fraction of shortages that are backordered or lost, like most mod-

els do [15], our model determines this by way of optimization to ensure the overall cost of the system is minimized.

Notice that for the lost sales cost formulation, we have considered the damaged lot as lost sales, which is given by Q(y).

The transportation cost for each delivery can be formulated as: CT(z-1)

(14) Lastly, the penalty for delay in recovery is given as f1(n

2) + f2(n

2). Thus, the first stage‟s total relevant cost for the

recovery plan is represented as follows:



1

1

1

12

2

2

1

1

11112

)()(12

11),(

z

i

i

d

d

T

z

i

i

iii

SnQyQL

LqDTTB

nfnfzCP

XXIHzA

nTzXTC

(15)

The optimal recovery plan for the damaged lot case is obtained by solving the following mathematical problem,

which is minimizing the total cost of recovery for the two-stage system:

Min[TC1(Xi, i = 1, …, z) + TC2(Si, i = 1, …, z)]

(16)

subject to the following constraints (17) – (22):

Si ≤ W for i = 1, 2, …, z-1 (17)

z

i

i nPTX1

(18)

z

i

i LqmTDX1

(19)

i

j

qi

j

jjD

BS

DX

P 2

1

1

11 for i = 2, ..., z (20)

Io = Iz = 0 (21)

Sz = Q (22)

The objective function (16) comprises of the two total cost components of the first stage (8) and the second stage

(15). Equation (17) ensures that the inventory storage at the retailer side does not exceed its warehouse capacity. Equa-

tion (18) represents the production capacity constraint; whereas (19) ensures that the total demand during the recovery

period is accounted for. Equation (20) ensures that the retailer receives each of its shipments on time and never runs out

of stock. Equation (21) states that there is zero inventory at the start and end of the recovery window and (22) guaran-

tees recovery of the original schedule after m cycles. The above model can be categorized as a constrained integer

nonlinear programming model.

By solving the above model (16) for Xi, Si and z subject to the constraints (17) - (22), one can obtain the optimal

recovery plan for the two stage SC system under disruption. Without disruption, this model will reduce to the original

model as in (1) that was presented earlier.

3. Solution Approach

This section presents several numerical examples to demonstrate the applicability of the proposed model in practice,

particularly on determining the new ordering and production schedule in the presence of a transportation disruption.

Two optimization methods have been used to solve the model, namely LINGO 10.0 and evolution strategy (λES

[16] with stochastic ranking [17]. The mathematical model presented in this paper was used to solve five different test

problems. The test problems were generated by arbitrarily changing the cost parameters as well as the disruption dura-

tion. Under the ES method, 30 independent runs were performed based on a (30, 200)-ES with a total of 1750 genera-

tions before termination. The value of Pf used was 0.45. As used in [17], Pf is the probability of using only the objective

function for comparisons in ranking. The solution procedure was coded in MATLAB and executed on an Intel Core

Duo processor with 1.99 GB RAM and a 2.66 GHz CPU.

The same test problems were solved using LINGO 10.0 to judge the quality of the solutions. Table 1 summarizes

the results of the experiment, which shows the optimal ordering quantities for the retailer, the optimal production quan-

tities for the manufacturer, the optimal number of recovery cycles, and the best objective values found by the two ap-

proaches. It can be seen that the ES method gives near optimal solutions as compared to the optimal solutions given by

LINGO. Moreover, the computed error of the ES method is exceptionally low (0.0% to 0.578%).

Analysis of the results shows that the solution for the model is highly dependent on the relationship between the

shortage cost parameters. When the backorder cost is lower than the lost sales cost and the rest of the parameters are

fixed, it can be observed that having backorders is more attractive. On the contrary, when the lost sales cost is lower

than the backorder cost, it is more optimal to have lost sales in the recovery schedule. It is worth highlighting that the

number of recovery cycles for the latter case will be shorter than for the former case (see test instances 1 and 2). The

extent of the disruption has an effect on the production quantities and z as well. For a large Td, X1 is found to be larger,

which in turn yields a lower z value. This finding can be seen when comparing test instances 1 and 3.


34

Table 1. Parameters for 5 Test Problems.

5. Conclusion

In this study, a disruption recovery model for a two stage production and inventory system subject to transportation

disruption was analyzed. The cost structure for the above model was developed for the case where the goods are dam-

aged while being transported. The objective of the study was to determine the optimal ordering and production quanti-

ties for the recovery schedule that yields the minimum relevant costs of the system. To implement the solution proce-

dure, two methods, namely, LINGO and ES with stochastic ranking, were used to obtain optimal solutions for the pro-

posed model. Numerical examples were provided to demonstrate the applicability of the model to real life problems.

Analysis of the results shows that the optimal recovery schedule is highly dependent on the cost parameters and the

length of the disruption. The presented model can assist decision makers who take a pro-active approach in maintaining

business continuity in the event of a transportation disruption in their SC system. Future work will focus on developing

a heuristic as an alternative method to solve the presented model.

6. References

[1] Y. Sheffi, "The resilient enterprise: Overcoming vulnerability for competitive advantage", The MIT Press, Cambridge,

Massachusetts, 2005.

[2] M. C. Wilson, "The impact of transportation disruptions on supply chain performance," Transportation Research Part E:

Logistics and Transportation Review, Vol. 43, No. 4, 2007, pp. 295-320.

[3] B. Tomlin, "On the value of mitigation and contingency strategies for managing supply chain disruption risks," Management

Science, Vol. 52, No. 5, 2006, pp. 639-657.

[4] X. Qi, J. F. Bard, and G. Yu, "Supply chain coordination with demand disruptions," Omega, Vol. 32, No. 4, 2004, pp. 301-312.

[5] M. Parlar and D. Berkin, "Future supply uncertainty in eoq models," Naval Research Logistics, Vol. 38, 1991, pp. 107-121.

[6] M. Parlar and D. Perry, "Analysis of a (q, r, t) inventory policy with deterministic and random yields when future supply is

uncertain," European Journal of Operational Research, Vol. 84, 1995, pp. 431-443.

[7] A. Arreola-Risa and G. A. DeCroix, "Inventory management under random supply disruptions and partial backorders," Naval

Research Logistics, Vol. 45, 1998, pp. 687-703.

[8] S. Chopra, G. Reinhardt, and U. Mohan, "The importance of decoupling recurrent and disruption risks in a supply chain,"

Naval Research Logistics, Vol. 54, No. 5, 2007, pp. 544-555.

[9] F. Ke-Jun, H. Xiang-Pei, and W. Xu-Ping, "Research on emergency logistics scheduling model based on disruptions,"

Proceedings of the International Conference on Management Science and Engineering, 2006.

[10] L. Sun, X. Hu, and Y. Fang, "Knowledge representation for disruption management problems in urban distribution decisions,"

Proceedings of the The 3rd International Conference on Innovative Computing Information and Control, 2008.

[11] Y. Xia, M.-H. Yang, B. Golany, S. M. Gilbert, and G. Yu, "Real-time disruption management in a two-stage production and

inventory system," IIE Transactions, Vol. 36, 2004, pp. 111-125.

[12] A. Banerjee, "A joint economic-lot-size model for purchaser and vendor," Decision Sciences, Vol. 17, 1986, pp. 292-311.

[13] H. Hishamuddin, R. A. Sarker, and D. Essam, "A recovery model for an economic production quantity problem with

disruption," Proceedings of the International Conference of Industrial Engineering and Engineering Management, Macau,

2010.

[14] G. P. Cachon and P. H. Zipkin, "Competitive and cooperative inventory policies in a two-stage supply chain," Management

Science, Vol. 45, No. 7, 1999, pp. 936-53.

[15] K. S. Park, "Inventory model with partial backorders," International Journal of Systems Science, Vol. 13, No. 12, 1982, pp.

1313-1317.

[16] H.-P. Schwefel, "Evolution and optimum seeking", Wiley, New York, 1995.

[17] T. P. Runarsson and X. Yao, "Stochastic ranking for constrained evolutionary optimization," IEEE Transactions on

Evolutionary Computation, Vol. 4, No. 3, 2000.

LINGO ES

1 200 20 1.2 1.8 1 1 15 15 0.003 4 4 531144.3 534212.3 0.578%

2 200 20 1.2 1.8 1000 1000 1 1 0.003 2 2 175017.0 175013.9 0.002%

3 200 20 1.2 1.8 1 1 15 15 0.03 5 2 569185.1 569185.1 0.000%

4 400 25 4 5 2 2 20 20 0.03 6 2 958554.5 958554.5 0.000%

5 400 25 4 5 1 1 2 2 0.008 2 2 338239.4 338240.6 0.000%

B1Test Instance A1 A2 H1 H2 B2 L1 L2 T d znTC

Error



A Measurement Framework and Obstacles to

Align Educational System Output with

Employment Demand in Indonesia

Effi Latiffianti*; YudhaPrasetyawan** Department of Industrial Engineering, InstitutTeknologiSepuluhNopember (ITS),Surabaya, Indonesia

[email protected] *, [email protected] **

ABSTRACT

Currently the Ministry of National Education, Republic of Indonesia, is working on several important programs related

to educational system, including the program of alignment between educational system and employment market. The

program aims to match the output of educational system to those required in the employment market. In this research,

the relationship between these two can be described as the same principle works between demand and supply where

educational system acts as the supply side and the employment market as the demand side.This paper proposed a

measurement framework so called Alignment Index (AI) model as an index to measure the level of alignment between

educational system and employment market in certain region. This index includes four dimensions suggested in the

program: quantity, quality, time, and location. Furthermore, challenges to achieve a perfectly aligned system are also

discussed.

Keywords: Alignment Index,Educational System,Employment, Indonesia

1.Introduction

Rapid changes almost in all aspects of life have been under way across the globe. Higher demand for better quality of

life has forced the emergence of wide variety of improvement methods. To cope up with market changes, million

manufacturing and service systems across nations changed themselves by applying selected improvement alternatives.

Moving forward with the changes certainly needs to be balanced with the improvement of human resources. In the past,

a secondary education might be sufficient to guarantee the economic success, but today the economic health of devel-

oped and developing nations has increasingly come to depend on higher level of education and more specialized voca-

tional training [1]. However, this is not always the case. For instance, in Indonesia we found that more than 29% of total

unemployment in 2009 was graduated from general and vocational secondary education, while almost 27% was

post-secondary graduates [2]. This fact shows us that higher level of education may not be a sufficient solution in all

cases and we believe that there must be an alignment between outputs resulted by the educational system with the hu-

man resources requirement in an employment market.

Most countries in the world are continuously reforming its educational system to better capitalize on its natural,

social and economic resources [3], and so does Indonesia. Unemployment rate in a certain education level indicates a

mismatch between educational system and employment market. This relationship can be described as the same principle

works between supply and demand. Higher availability in one side will make its value to the other side less, and with

the same cost higher quality products will be more preferable than the less ones. This paper only reports a part of bigger

research scope and critically examines the alignment level between educational system as a supply side and employ-

ment market as a demand side of a region that later on may be associated as city/town, province, or a country. We aim

to provide a general framework to measure alignment of the region based on four aspects: quality, quantity, time, and

location. The result of this paper is then used for further research involving all captured variables and their interaction in

the system, as well as uncertain behavior, which in turn will affect the average performance of the examined system.

The research is expected to provide important points for policy maker as considerations in the policy making.

2. Alignment Index (AI) Model

In the context of Indonesian educational system, alignment is defined as efforts to match the educational system as hu-

man resource suppliers with the employment market that requires human resources with a certain gradeof competencies

and its variance. This requirement is continuously changing so that educational system must respond it as needed. When



A Measurement Framework and Obstacles to Align Educational System Output with Employmt Demand

36

a mismatch exists, problems may arise, such as increasing number of unemployment and decreasing productivity due to

delegation of improper men for certain job positions. Mismatch in this case may occur in terms of quality, quantity,

time, and location [4]. Thus, alignment index for each education level and field of study should include those four as-

pects in the measurement as follow:

AI = a (AIQt) + b (AIQ ) + c (AIT ) + d (AIL) (1)

With a + b + c + d = 1

Where a, b, c, and d aredesired weights for alignment index in term of quantity (AIQt), quality (AIQ), time (AIT), and

location (AIL) respectively. Because Indonesia has been implementing “9 years basic education” program, or 12 years

for specific towns/cities due to the local policy, the education levels and fields of study to be measured should start from

secondary school levels and others that can be considered as equal with secondary education. It includes but is not lim-

ited to the following mentioned in Table 1. The alignmentindex of a region can be obtained by aggregating the align-

ment indexes of all level and field of study in the related region. The same way, we can also calculate the alignment

index for the country.

Table 1. Education level and fields of study in Indonesia

Levels Fields of study Levels Fields of Study General Secondary

School General

Post-secondary (professions) Medical school and health sciences,

Pharmacy, special education for teacher, and others Special school for

disabled Post graduate non professions

Vocational Secondary

School

Engineering and technology, Information and communication

technology, Healthcare services, Arts, crafts and tourism, Agribusiness and

agro-industry, Business and

management Source: [5]

Vocational post-secondary

(Diploma)

Engineering, healthcare services, special

education for teacher (Keguruan),

language study, social science, business school, economy, language studies

Other vocational

programs (non school) Language and technical skills.

Post-secondary non professions Engineering, business and management,

law and social science, science, arts,

language studies, and others Post-graduate non professions

2.1.The Quantity Alignment Index (AIQT)

The quantity alignment index (AIQt) describes level of alignment between educational system and employment in term

of quantity. Ideally educational system should produce an equal amount of human resources to those required in the

employment market. The closer the AIQt value to 100%, the better the system performance. The value above 100% is

possible to be obtained, but that is a very rare case and usually only happens in a very specific or narrow area of exper-

tise.

Figure 1. The quantity alignment index measurement model

Basically, the value of AIQt can be obtained by simply calculating the ratio between the total number of educated

human resources available in a certain region in year (i) and the number of available employments in the same year and

region. This index should be measured for each level and field of education, which all then to be aggregated to obtain

the AIQt. Number of available employment should be identified in all possible sectors, including public sector (govern-

ment employees), manufacturing, farming, service, entrepreneur, and others. Figure 1 describes how AIQt can be meas-

ured.



Having the information of available employment in the near future, it allows the supply side to arrange how many

students to produce in a certain semester or year by adjusting number of intake and considering the proportion of stu-

dents graduated in each intake. The index interpretation, whether it is good, average, or bad, can be set according to the

achievement target of the measured system.

2.2.The Quality Alignment Index (AIQ)

The quality alignment index (AIQ) measures how good the educational system satisfies the market requirement in term

of the human resources quality. Quality in this case is associated with competencies of human resources. Good AIQ is

expected to reduce unbalanced supply-demand condition, for example when nursing school capacity is way higher than

the actual number of nurses required, the number of unemployment (nursing school graduates) will increase.

In this model, competencies should be assessed in two aspects: hard skills and soft skills. While soft skills score

(SS) may be simply measured as a percentage of individual‟s soft skills in comparison with the total soft skills required

in therelated employment, the hard skills score (HS) needs to be measured based on level of study appropriateness,field

of study appropriateness, andthe required hard skills that successfully fulfilled by the assessed individual (equation 2).

HS = C .B . HP (2)

AIQ = x(HS) + y(SS), where x + y = 1 (3)

In equation 2, C notates level of study appropriateness (C = 1 if individual‟s level of study equal to the required

criteria, C=0.71 for other condition), B is field of study appropriateness (B =1 if individual‟s education back-

ground/field of study is equal to the required criteria, B=0.71 for other condition), and HP is the average proportion of

all individual‟s hard skills to the total required hard skills. It can be calculated as follow:

Table 2. Hard skills assessment

Required competencies Individual assessment (0-100%)

Competency 1 _______ %

Competency 2 _______ %

.

.

.

.

.

.

Competency n _______ %

Individual HP = average value of assessed competencies

The quality alignment index AIQcan be obtained using equation (3) where the score of hard skills and soft skills may be

weighted by x and y respectively.

2.3.The Location Alignment Index (AIL)

In term of location, alignment efforts are made in purpose to maintain the fulfillment of human resources requirement in

a specific region. A region of 100% aligned system should be able to produce graduates that would fill out 100% avail-

able employments. To measure the location alignment index we use scores for demand and supply sides as shown in

Table 3.

Table 3. Graduates and Employments Scoring

Score Graduates in the assessed region (city/town) Score Employment in the assessed region (city/town)

1 Working in the assessed region (G1) 1 Filled out by graduates from the city(E1)

1.25 Working outside the region, but in the same

state/province(G2) 0.75 Filled out by graduates from other region, but in the same

state/province (E2) 1.5 Working outside the state/province (G3) 0.5 Filled out by graduates from other state/province (E3)

1.75 Working outside the country (G4) 0.25 Filled out by graduates from other country(E4)

Score should be calculated in both demand (equation 5) and supply sides (equation 6) at which the maximum

scores are 100%. The location alignment index is the score ratio between demand and supply as shown in equation (7).

(5)


38

(6)

(7)

whereAISL(i), AIDL(i), and AIL(i) are alignment index of supply side, alignment index of demand side, and the location

alignment index respectively,Gi and Ei for i = 1, 2, 3, 4 are the number of each assessed criteria explained in Table 3. In

this model, the numerator maximum score is 1, while the denominator will be equal or greater than 1. Thus, the location

alignment index model will not produce any value larger than 1. It should be underlined that the number of individuals

assessed in the supply side and the demand side may differ because this index does not attempt to measure alignment in

quantity.

2.4.The Time Alignment Index (AIT)

Alignment index in term of time can be measured using several indicators. Average waiting time before a graduate finds

a job is probably the closest indicator to give information about the level of alignment in term of time. If a system were

well aligned, graduates would normally get hired soon after graduation time. Shorter waiting time should indicate that

graduates are produced in the (nearly) right time when they are needed. While companies may perform recruitment

process when there are job positions need to be filled out, each educational system has standard lengths of period where

the schedule may only commence at specific times during the year, for example spring and fall. Thus, time gap will

always exist between the point where students are graduated and the point when they find employment. In the case with

such limitation, it may be better to measure the system using equation (4).

(4)

In equation 4, AIT is the time alignment index, pijis the proportion of graduates get hired in less than the sub-period

of educational system in level iand field of study j, m and n are total numbers of education level and field of studies

observed (see Table 1 as an example). This proportion should be measured for each education sub-period (quarter, se-

mester, or year). For example, in an educational system where graduation occurs each semester or twice a year, meas-

urement should be made every 6 months period and aggregated for the assessed year, and piis the proportion of gradu-

ates that successfully find job in less than 6 months after graduation time in the same year.

3. Discussion

In the model development, we put more stress on the educational system side rather than the employment side. The

main reason is that from our perspectives, the educational system is more controllable. In practice, government in-

volvement in form of policies is easier to implement in the educational system than in industries.

Ideally, a perfectly aligned system should be able to produce an exact number of educated human resources with

the required quality of competencies in a time and place they are needed. However this condition may be difficult to

achieve due to the nature of population behavior, which normally found to be higher than the number of available em-

ployments. Furthermore,there are several reasons that would potentially obstruct the realization of 100% aligned sys-

tem, including:

1. Economic of scale. When a number of human resources in a specific field needed are small, there is no reason to

establish an institution only for single purpose. For example, 5 fresh graduates of shipbuilding engineering per

year are needed in a certain region in which there is no university or institution of engineering to accommodate the

need. A perfectly aligned system should provide an educational process to produce the required engineers. How-

ever, in this case, it may not feasible in term of financial aspect to establish a shipbuilding engineering school with

capacity of 5 per year.

2. Centralized industry. In Indonesia and probably in most countries in the world, it is often found that regulation

prevents the establishment of industrial facilities in just any places. For environmental, safety, and other reasons,

sometimes the government only allows the industries to operate in a specifically dedicated industrial area, and

sometimes not all regions (towns or cities) have this kind of industrial park. In this case, it is difficult to prevent

people migration for employment purpose from one place to another. Also, it may be not wise to approve the ab-

sence of school in a certain region just because there is no available employment there.

3. Errors in forecast. Each education level of specific or general field of study requires a certain length of period to

produce graduates. Thus, forecasting on the number of required employment in the future should be available in



order to prepare the human resources. For example, decision making regarding how many students should be taken

this year in an intake of 4 years education level should be made based on the number of graduates required 4 years

from now. Of course it is not easy to identify this number, especially because it is very closely related to strategic

planning of firms. In fact, only small number of industries (firms) really got involved, established and maintained

direct connection with educational system (institutions), especially in Indonesia.

4. Human migration.Migration has been a popular issue in discussion over the past decades, not only because the

phenomenon has been occurring in most part of the world but also due to the interesting facts of various causes

and impacts related to migration. While employment and economic reasons have been reported as major factors

causing migration as those found in China [6] and United States [7,8,9], other factors such as educational opportu-

nities, culture, and family also play an important but secondary role [9]. So is probably the case in Indonesia. Ad-

ditionally, disparities in the development level among regions in Indonesia are suspected to be the root cause of

most suggested reasons [10]. For example, the fact that universities in Indonesia are commonly found in relatively

more developed regions would somehow make migration unavoidable case for education reasons. Thus, a

well-aligned system is only expected to reduce migration for employment, education, and economic reasons, but

not those for other reasons. The occurrence of migrations for whatever reasons would make it difficult to achieve a

perfect location alignment index (AIL).

5. Culture and Local Value.Changes in all aspects of life have been affecting society and its culture. However, some

of cultural values remain. Tensions between “culture/religion” in one side and “women‟s individual human rights”

in the other side were identified recently [11] and it proves that this value somehow continue to exist. Although

RadenAjengKartini, one of Indonesian national heroine, had pioneered emancipation of women in Indonesia long

time ago, some of cultural and religion values related to women has not changed significantly in several cases.

This is especially true for less developed regions where the society tends to be conservative. For instance, in some

parts in Indonesia, women are expected to stay within the home and family. In such society, woman would proba-

bly choose to stay at home after completing her study rather than being employed and that would cause the align-

ment index to be lower. Differently, members of very rich family may choose not to work because they find it not

necessary to do. To this problem, it may have nothing to do with educational system but somehow it affects the

index value for the assessed system.

6. Close collaboration requirement and organizational purpose. Aligning educational system and the employment

market requires close collaboration of both sides, which is sometimes hard to maintain due to differences of or-

ganizations‟ purpose. In order to be able to consistently produce human resources needed by the employment

market, industries are expected to actively inform any changes they expect from educational system. In the other

hand, educational system members should also put efforts to keep the system informed about any changes may

occur in the market. Furthermore, the harder and more important part is how institutions within the educational

system share and play their roles. For example, in a specific region 100 fresh graduates of electrical engineer are

needed every year while there are three universities offering this subject. To keep the system well aligned, those

three universities are expected to take in total 100 students. This can be very difficult to implement especially

when the universities (or other educational institutions) play roles as profit-oriented organizations, as those found

in most cases. Hence, government policies should take part to make it possible.

7. Knowledge and technology limitation. Knowledge and technology advancementoften come from other part of the

world, especially in the developing countries. Knowledge can be transferred by moving people, specific tools, and

technologies as well as networks that combine people, tools, and routines [12]. Thus, the presence of people as

knowledge owners is sometimes unavoidable when it comes to knowledge transfer efforts. It may not be a first

priority to be avoided, but in the context of alignment, importing people from other regions will significantly re-

duce the location alignment index value.

8. Globalization and the spread of industries across nations. Globalization has made the parent company, facilities

locations, and market borderless. In order to deliver products more quickly and cheaply, agile supply networks are

required. Determining the optimum number and location of factories and distribution centers is very crucial to

successfully set up the network. In doing so, many firms have opened new facilities across nations. To closely su-

pervised the subsidiary companies and transferring knowledge, as well as copying the success of the parent com-

panies, employees from parents companies are often placed overseas. The impact of this phenomenon is lower lo-

cation alignment index value for regions where the new facilities are located.

While alignment is thought as a solution in educational and unemployment problems, some parties would probably

disagree with the concept of alignment. Practically, when a system is well aligned, several undesired phenomenon

might emerge, such as limited human resources alternatives. In a well-aligned system, the number of required human

resources for employment with the available number will be nearly the same. It means employers will have smaller

number of candidates to select from. For most-wanted firms as employment place, it may not be a problem, but for less


40

and least wanted firms they may have to be satisfied with even smaller number of available candidates to select from.

This condition will also made decision-making process of hiring-firing more difficult. For example because the number

of available human resources is small, a firm might rather keep an underperformed employee than to go once again

through the selection process where only very few number of candidates will participate.

4. Conclusion

The proposed alignment index model can be implemented to measure the alignment level of educational system and

employment market in certain regions. This index can be used as an indicator to compare the performance of a system

in a certain region with others. The model also gives information on what aspects within the four dimensions assessed:

quantity, quality, time, and location, which require further improvement. However, measurement may not be easy in

this case due to data availability. Thus, a well-structured data gathering method is required to implement the model.

Although the index value suggests that 100% alignment is an ideal condition, it may not be the final purpose of the

alignment program. In fact, how to increase the index is more important concern in our case. As we all know, abundant

supply in this case is unavoidable and it involves many variables that interact each other, and some probably is not di-

rectly related to the educational system and employment. Further analysis on what impacts may be resulted in an ideal

condition has not been performed yet. In addition, achieving 100% index is a difficult and challenging task for many

reasons, including: economic of scale, centralized industry, errors in employment forecast, human migration, culture

and local value, technology limitation, different organizational purpose, globalization and the spread of industries across

nations, as well as close collaboration requirement among parties involved in the system.

Finally, this alignment model can only be seen as a soft approach in purpose to help evaluating the performance of

national educational system in its role in the supply-demand relationship with employment rather than to reach 100%

alignment. The evaluation is then expected to help the government narrowing the gap between demand and supply.

6. References

[1] Organization for Economic Cooperation and Development (OECD), “Education Policy Analysis 2006-2005,” OECD, Paris,

2006.

[2] BadanPusatStatistik, 2009.

[3] Metzger et al, “A Comparative Perspective on the Secondary and Post-Secondary Education Systems in Six Nations: Hong

Kong, Japan, Switzerland, South Korea, Thailand and the United States”, Procedia Social and Behavioral Sciences, Vol. 2,

2010, pp. 1511-1519.

[4] Tim PenyelarasanKementrianPendidikanNasional, “KerangkaKerjaPenyelarasanPendidikandenganDuniaKerja”, 2010.

[5] Directorate of Vocational Secondary School Development – Ministry of Indonesia National Education, “Data Pokok SMK”,

2009, http://datapokok.ditpsmk.net/index.php?prop=&kab=&status=&kk=&bk=&pk=

[6] Z. Liu, “Human capital externalities and rural-urban migration: Evidence from rural China”, China Economic Review, Vol. 19,

2008, pp. 521-535.

[7] M. J. Greenwood, “Research on Internal Migration in the United States: A Survey”, Journal of Economic Literature, Vol. 13,

1975, pp. 397-433.

[8] C. C. Roseman, “Labor Force Migration, Non-Labor Force Migration, and Non-Employment Reasons for Migration”,

Socio-Econ Plan Sci, 1983, Vol. 17, No. 5-6, pp 303-312.

[9] T. Kontuly, K. R. Smith, and T. B. Heaton, “Culture as a Determinant of Reasons for Migration”, The Social Science Journal,

1995, Vol. 32, No. 2, pp. 179-193.

[10] Fathurrohman, “KerjasamaAntar Daerah dalamPenangananMigrasidanPersebaranPenduduk”, Dialogue JIAKP, Vol. 2, No. 2,

May 2005, pp. 726-734.

[11] B. Winter, “Religion, culture and women‟s human rights: Some general political and theoretical considerations”, Women‘s

Studies International Forum, 2006, Vol. 29, pp. 381-393.

[12] A. C. Inkpen, “Knowledge Transfer and International Joint Ventures: the case of NUMMI and General Motors”, Strategic

Management Journal, John Wiley & Sons Ltd.,2008, Vol. 29, pp. 447-45.



Government Intervention and Performance:

Evidence from Indonesian State-Owned

Enterprises

Bin Nahadi/Graduate School of Asia Pacific Studies Doctoral Program,

Ritsumeikan Asia Pacific University, Oita, Japan.

[email protected].

ABSTRACT

In this study, the impact of government interventions toward performance of Indonesian state-owned enterprises is in-

vestigated, using 114 of total 141 enterprises from the year 2006 to 2009 (456 observations) as a sample. The study is

cross-sectional to estimate how issues of intellectual property, soft budget constraint and political embeddedness affect

the economic performance of enterprises. Form of SOEs, number of state ownership, government loan, capital injec-

tion, number of government officers/politicians seat in the board of commissioners, as well as government assignment

are assigned as government intervention proxies. On the other hand the firm performance is represented by ROA and

ROE values. The result shows that the government ownership, government loan and government assignment have an

adverse impact to SOEs performance, meanwhile number of government officerson the supervisory board is the only

variable with favorable impact to SOEs. The impact from the rest of government actions are unclear and need to be

investigated further. Finally, possible explanations of each empirical findingare elaborated.

Keywords: Government Intervention, Performance, Indonesia, State-Owned Enterprise

1.Introduction

The role of government in transition economies is undeniably critical, which is one of the common ways is through

state owned enterprises (SOEs).It is widely known that SOEs have been suspected as ill-governed business entities sig-

nified by high level of corruption, lack of transparency, as well as severe efficiency. Many market based economist be-

lieve that the main reason of such weaknesses is overwhelming government interventions. Therefore, they actively

promote liberalization trough privatization of SOEs. However, it may be not true for all cases.

This paper aims toexamine the relationship between the level of government interventions and the performance of

state owned enterprises. The paper unfolds as follows. In Section 2, the theoretical review is described. Variables and

hypothesis is developed in Section 3; meanwhile, section 4depicts data and methodology. Section 5presents result and

findings, and then discussion is developed in section 6. Final section concludes the paper.

2. Literature Review in Government Intervention

There are three main issues of government intervention whichare elaborated in this paper. They are intellectual property

aspect through control and ownership, budget constraint aspect, and political embeddedness issue. Each aspect is de-

scribed in the following paragraphs.

2.1.Intellectual Property Aspect

SOEs are a business institution whichbelongs to a society as a whole at the proxy of state. The problem is if everyone

owns itthat means no one actually ownsit, as a result, no one has an incentive to utilize the resources effectively and

efficiently. Therefore, many economists suggest assigning property rights by lowering the government control and

ownership [1].

The problem believed to be related to ownership is the principle-agent problem that arises when managers act not

in shareholder‟s best interest. The deviating management goal often hinders the shareholders goal in maximizing their

share value. Previous study reveals that efficient information and structure of incentive as a result of the existence of

private ownership is believed to be able to reduce agency problems[2]. Also,it is argued in [3] that another reason why

Government Intervention and Performance: Evidence from Indonesian State-Owned Enterprises

42

full or partial privatized SOEsare said to have less agency problem is because those firms have better both external and

internal governance mechanisms.Furthermore, the agency problem in the SOEs sector is worse than their peers in the

private sector since there are two layer of agency problem: owners-to-politicians and politicians-to-managers[4].

2.2.Budget Constraint

SOEs are frequently exploited by governments in emerging economies to produce public necessities, which in turn all

costs occurred will be shouldered by government via loan policy or subsidies [5]. It will lead to the situation of

so-called soft budget constraint. Comprehensive illustration is as described in [6]:

‗‗The softening of the budget constraint appears when the strict relationship between expenditure and earnings has

been relaxed, because excess expenditures over earnings will be paid by some other institution, typically by the state. A

further condition of softening is that the decision maker expects such external financial assistance with high probability

and this probability is built into his behavior‟‟.

From several previous studies, causes of soft budget constraint can be categorized into some causes, such as de-

centralized [7], paternalism [5], and public ownership in socialist economies [8], monopolistic market [9], and policy

burden [10]. In the context of Indonesian SOEs, two latter causes are relevant. Some particular industries, such as sea-

port, airport, and defense industry, have been still monopolized by SOEs. It is not because of competitiveness of SOEs-

but because either the government has not liberalized the market yet or otherwise those industries are not lucrative

enough to attract private firms.

It is said in [11] soft budget constraint will cause the firm become less responsive to price, technological changes,

and unfavorable external condition that lead to arise of organizational slack. In addition, SOEs may not be efficient in

utilizing their finance resources since capital market cannot discipline SOEs.

2.3 Political Embeddedness

The role of the state as the regulator as well as the owner of SOEs at the same time causes the situation, so-called po-

litical embeddednessthat refers to „technical, bureaucratic, or emotional ties to the state and its actors. It includes

wide-ranging and intricate association; official and unofficial, personal and organizational ties to the state [12].Given

the existence of the principle-agent problem mentioned earlier, one way utilized by the shareholder to ensure the man-

agement work toward owner-based interests is through a supervisory board. However, it has been quite common that

the members of supervisory board of most SOEshave been selected among bureaucrats from any associated departments

or politicians from any political parties. As a result,SOEs might be an ideal place of rent seeking activities from the

member of the board of commissioner.

In addition, as mentioned in the issue of soft budget constraint above, SOEs are often utilized as a vehicle for exe-

cuting governmental agenda such as for delivering some government assignments. As a result, SOEs will be charged

with multitasks, not only as a business entity but also as a government body at the same time.

3. Variables and Hypothesis

To address the issue of intellectual property assignment/ownership control, two variables are employed. They are form

of SOEs (FORM) and number of government ownership (OWNERSHIP). Some economistsargue that a source of inef-

ficiencies in SOEsis high control of state over the firms. It is said that the government is more likely to distract the re-

sources of the firm to attain its own political or socioeconomic goals [13]. In addition, government control over enter-

prises is also suspected to have an association with the absence of incentive and lack of monitoring for managers to bet-

ter perform[14]. Moreover, different forms of state ownership are also associated with the level of government officials‟

involvement in the process of corporate governance and it is likely to have different performance [15]. Form transfor-

mation and privatization can be regarded as one ways of defining property rights. Property right theory suggests that the

clearer the property rights are defined, the better the utilization of the assets (governance) will be [16]. According to

those arguments the following hypothesis is proposed:

H1: the higher government control toward SOEs represented by more-bureaucratic form of firm will providene-

gative impact toward SOEs performance;

H2: the higher government control toward SOEs represented by higher percentage of state-ownership will provi-

denegative impact toward SOEs performance;

With regard to the soft budget constraint aspect, this study employs two independent variables, namely capital in-

jection (CAPITAL) and government debt (GOVLOAN). In most cases, if SOEs are facing severe financial hardship the

state will interfere either by providing loan or capital injection as a last resort sources. In contrast to the case of com-

mercial bank loan that requires some rigid requirements in obtaining credit and of course with market rate, the govern-

Government Intervention and Performance: Evidence from Indonesian State-Owned Enterprise


ment frequently releases many requirements so that the SOEs will more easily get a loan at subsidized interest rate. This

government loan present financial benefit to SOEs, mainly because of lower interest rate, no collateral required and

lower transaction cost. In case of capital injection the advantages enjoyed by SOEs are even bigger than government

loan. Nonetheless, both types of government actions can create disincentive for managers to govern the firm properly

and efficiently including finding needed financial resources. This may also hinder sound development of capital or fi-

nancial market. Therefore the following hypotheses are set:

H3 :Government loan will givenegative impact toward SOEs performance;

H4 :Capital injection will providenegative impact toward SOEs performance;

The issue of political embeddedness is examined by employing two variables; government assignment through

public service obligation (PSO) and number of government officers/politicians seating in the board of commissioner

(OFFICERS). PSO is a government program to avail the basic need of people such as electricity, food, medicine, fuel,

transportation and soon. Doing so will provide SOEs both benefit as well as cost. The appointed SOEs will financially

benefit from captive revenue plus certain a percentage of profit given over each particular government assignment.

Nevertheless, it also implicitly grants some cost to SOEs. SOEs that heavily rely on government assignment as the main

source of revenue will be more likely to have unproven competitiveness compared to their private owned peers. In the

long run, it also will harm financially. Moreover, too much business transaction with government and its bureaucrats

may induce political rent seeking activities that undermine SOEs competitiveness.

Furthermore, government has assigned active/retired officers from associated ministries and politicians from ruling

political parties as member of Board of Commissioners (BOC) in most SOEs. It also derives both benefit and cost to

SOEs simultaneously. The presence of an official on the board can be a source of legitimacy and facilitator in passing

government policy to SOEs and in delivering a message from SOEs in an effort of influencing the policymakers that

ultimately benefit SOEs[17]. Even more, this also can provide SOEs access to resources (such as government project)

controlled by department or ministry in which the officials work.

On the other hand, public choice theory states that politicians will maximize their interest in gaining more votes so

that the firm with less political intervention will be more likely in increasing search for better governance [18]. In addi-

tion, as representative of the government, acting officials usually will act on the basis of government interest that is

probably not in line with firm objective. Additionally, as argued in [19] the presences of politician exacerbate the

agency problem. This means that the presence of officials on the BOC may be perceived with significant costs for the

firm. Given those arguments following hypotheses are proposed:

H5: Government assignments through public service obligation will give negative impact toward SOEs

H6: Number of active/retired officers and politicians on BOC will result in negative impact toSOEs

As dependent variable, this study employs Return on Assets (ROA) and Return on Equity (ROE) as performance

measures. Thanks to its simplicity in calculating as well as its explanatory power both measures were used in previous

numerous researches, including for Indonesian SOEscase [20]. For control variable, equity (EQUITY) and firm‟s core

business (CORE) are selected to represent the size of SOEs and the industry where the firm operates consecutively.

4. Data and Methodology

Financial data were collected from the annual report of 114 SOEs (of total 141 SOEs) for the year 2006-2009 (456 ob-

servations). This sample covers almost 97% of population both in term of assets and sales.

The way in giving a score for independent variable as follows:

a. SOEs is scored 1, 2, and 3 if their form is a public agency, company limited, and listed company limited consecu-

tively;

b. Ownership (OWNERS) is represented in percentage of state ownership, range from 0% to 100%;

c. Capital injection (CAPINJ) and government loan (GOVLOAN) are dummy variables. If the SOE did NOT get any

form ofadditional capital injectionwithin last five years score 0 is given and 1 otherwise for CAPINJ. Meanwhile if

there is NO government long term loan balance in the SOEs‟ balance sheet score 0 is provided and 1 otherwise for

GOVLOAN;

d. Number of officers or politicians (OFFBOC) who seat on board of commissioners is expressed in number as it is;

e. PSO is also a dummy variable which is SOEs that conduct government assignment is valued 1 and 0 otherwise;

f. Equity value has been transformed into ln value to reduce the possibility of multicollinearity problem;

g. Type of industry which the SOEs operate is also valued using dummy variable, 0 for good production/manufacture

and 1 for service provider.

Once all data have been identified and inputted, those independent variables are tested to examine the relationship to-

ward dependent variable using ordinary least square method. The regression equations are written as follows:


44

ROE = α0 + α1FORM + α2GOVLOAN + α3OFFBOC + α4PSO + α5CAPINJ + α6OWNERS + α7ln.EQUITY +

α8CORE (1)

ROA = β0 + β1FORM + β2GOVLOAN + β3OFFBOC + β4PSO + β5CAPINJ + β6OWNERS + β7ln.EQUITY +

β8CORE (2)

5. Result and Findings

Table 1 shows the descriptive statistic and correlation. Average ROE of ISOEs, 0,085, is relatively low compared to

their private competitors. Meanwhile, average number of government officers and politicians on the board of commis-

sioner is 3.32. Furthermore, mean of state ownership on SOEs that is 92%, partly because this study doesnot include

SOEs with state-minority ownership, less than 50%, but mainly it shows that the majority of SOEs are still

wholly-owned by the state. With respect to form, most SOEs are in the form of limited corporations. In term of core

business whichSOEs operate, there were more SOEs doing business in the service industry compared to manufacture

industry. The rest of variables are dummy variable so that the means just show the relative proportion over the observa-

tion. For instance, mean of PSO is 0.12 meaning the percentage of SOE executing special government program is

around 12% of population.

Table1: Descriptive Statistics And Correlations with ROE as Dependent Variable

Table 2:

Table 2: Cooefficients, t Statistic, & Collinearity with ROE as Dependent Variable

Table 2 demonstrates roughly 57% variability of the dependent variable can be explained by all combined inde-

pendent variablesemployed. Using 307 observations (after omitting some outliers), this score is considered high. Em-

ploying Variance Inflation Factor (VIF) and Tolerance statistic critical scores that may signal problem with multicollin-

earityhave not been approached by both scores [21].

Looking at the significance, except GOVLOAN and CAPINJ, the rest of independent variables have statistically

significant effect toward ROE. Although FORM and PSO are not significant at 5% confidential level, however, both

variables are quite significant at 10% confidential levels. Therefore, in this paper both variables are still considered as

significant.

Mean SD 1 2 3 4 5 6 7 8 9

1 ROE 0.085 0.092

2 FORM 2.070 0.481 0.299

3 GOVLOAN 0.410 0.492 -0.143 0.067

4 OFFBOC 3.320 1.274 0.289 -0.071 -0.039

5 PSO 0.120 0.322 0.096 0.014 0.089 0.244

6 CAPINJ 0.235 0.424 -0.147 -0.283 0.042 0.110 0.300

7 OWNERS 0.926 0.178 -0.247 -0.450 0.053 0.152 0.023 0.182

8 ln.EQUITY 12.044 4.182 0.711 0.250 -0.095 0.364 0.242 -0.072 -0.132

9 CORE 0.630 0.483 0.136 -0.037 -0.412 -0.007 0.068 0.024 -0.004 0.003

Standardized

Coefficients

B Std. Error Beta Zero-order Partial Part Tolerance VIF

(Constant) -0.093 0.034 -2.714 0.007

FORM 0.015 0.009 0.080 1.764 0.079 0.299 0.102 0.067 0.704 1.420

GOVLOAN -0.002 0.008 -0.009 -0.204 0.839 -0.143 -0.012 -0.008 0.786 1.272

OFFBOC 0.007 0.003 0.102 2.397 0.017 0.289 0.138 0.091 0.794 1.259

PSO -0.023 0.012 -0.080 -1.883 0.061 0.096 -0.108 -0.072 0.799 1.251

CAPINJ -0.010 0.009 -0.045 -1.059 0.290 -0.147 -0.061 -0.040 0.817 1.224

OWNERS -0.067 0.022 -0.129 -2.971 0.003 -0.247 -0.170 -0.113 0.770 1.299

ln.EQUITY 0.014 0.001 0.651 14.733 0.000 0.711 0.649 0.561 0.743 1.345

CORE 0.027 0.008 0.140 3.312 0.001 0.136 0.188 0.126 0.813 1.229

N

F

R Square

Collinearity Statistics

307

48.838

0.567

Unstandardized

Coefficients

t Sig.

Correlations



From the second equation, which is the only difference from table 1 is that ROA is used as dependent variable in-

stead of ROE. The result displayed on the table 3 shows almost similar figure. The ROA score is considerably low at

3.2%. What makes slightly difference is the number of valid observation after taking out the outliers. With regard to

correlation, there is no sharp correlation among variables. It support the argument that multicollinearity problem is neg-

ligible.Table 4 reveals that regression results moderately high r square, 0.450. A couple outliers were identified until

reaching valid observation is 270. After considering F score, Tolerance and VIF score the model is judged statistically

fit. Among predetermined independent variables only CORE was not significant.

Table 3: Descriptive Statistics and Correlations with ROA as Dependent Variable

Table 4: Cooefficients, t Statistic, & Collinearity withROA as Dependent Variable

From the result of two different equations of regression discussed above, the impact of each aspect of government

interventions can be summarized as follows. Overall, as shown by table 5 comparison of the result from two different

tests provide strong argumentto support Hypotheses H2, H3, H5, and reject Hypotheses H6. However, hypotheses re-

garding SOEs form and capital injection are left with unclear answer due to the indecisive results.

6. Discussions

From the findings described earlier, form has a positive impact over ROE, meaning that reducing the government con-

trol signified by transformation of SOEs form is more likely to give good impact of the SOEs‟ performance. However,

the opposite result was found for the second equation which has ROA as dependent variable. The possible reason is

SOEs with less control from the government will have more flexibility in raising capital either through equity capitali-

zation (for instance, through initial public offering) or by leveraging debt. SOEs with less government control seem to

finance their project using more debt rather than equity. As a result, it will keep their equity low so that it can push their

ROE higher. Interestingly, when performance measured by using ROA the opposite result prevails. This paper argues

SOEs with less government control become less conservative in selecting project in the sense that funds obtained from

debt/loan have been invested in the project with low return.

Mean SD 1 2 3 4 5 6 7 8 9

1 ROA 0.032 0.025

2 FORM 2.100 0.469 -0.139

3 GOVLOAN 0.440 0.498 -0.292 0.041

4 OFFBOC 3.250 1.205 0.384 -0.065 0.057

5 PSO 0.120 0.328 -0.107 0.062 0.167 0.271

6 CAPINJ 0.181 0.386 0.288 -0.187 -0.054 0.127 0.323

7 OWNERS 0.923 0.176 -0.082 -0.480 0.049 0.141 0.017 0.129

8 ln.EQUITY 12.119 4.080 0.270 0.210 0.003 0.339 0.269 0.052 -0.046

9 CORE 0.670 0.470 0.042 0.002 -0.412 -0.054 0.042 0.020 -0.020 -0.109

Standardized

Coefficients

B Std. Error Beta Zero-order Partial Part Tolerance VIF

(Constant) 0.046 0.012 3.930 0.000

FORM -0.010 0.003 -0.192 -3.505 0.001 -0.139 -0.212 -0.161 0.703 1.423

GOVLOAN -0.011 0.003 -0.225 -4.291 0.000 -0.292 -0.257 -0.197 0.769 1.301

OFFBOC 0.008 0.001 0.385 7.597 0.000 0.384 0.426 0.349 0.820 1.219

PSO -0.025 0.004 -0.324 -6.102 0.000 -0.107 -0.353 -0.280 0.748 1.336

CAPINJ 0.020 0.003 0.315 6.275 0.000 0.288 0.362 0.288 0.839 1.192

OWNERS -0.034 0.007 -0.241 -4.559 0.000 -0.082 -0.272 -0.209 0.752 1.330

ln.EQUITY 0.001 0.000 0.240 4.621 0.000 0.270 0.275 0.212 0.782 1.279

CORE 0.000 0.003 -0.001 -0.025 0.980 0.042 -0.002 -0.001 0.793 1.261

N

F

R Square

270

26.680

0.450

Unstandardized

Coefficients

t Sig.

Correlations Collinearity Statistics


46

Table 5:The Impact of Each Independent Variable To ROE and ROA

Independent Variable

Dependent Variable

Impact ROE ROA

FORM Positive Negative Indecisive

OWNERS Negative Negative Negative

CAPINJ Negative (insignificant) Positive Indecisive

GOVLOAN Negative (insignificant) Negative Negative

OFFBOC Positive Positive Positive

PSO Negative Negative Negative

Ln.EQUITY Positive Positive Positive

CORE Positive Negative (insignificant) Indecisive

Not surprisingly, both equations show consistent results regarding the impact of ownership toward performance.

The result show higher percentage of government ownership will lead to poorer performance. The presence of other

shareholders other than the government is expected to be able to enhance governance of the firm through improvement

in monitoring, transparency, responsibility, and so on. This is especially true for the case of Indonesian privatized SOEs

as finding of previous research [22]. ListedSOEs may have better governance due to the presence of both internal and

external governance as argued in [3].

With respect to capital injection, the result is mixed. This variable is statistically significant in relation to ROA but

insignificant in the case of ROE with different direction of impact. This finding need to be further investigated by em-

ploying other performance variables or by applying qualitative approach. Similarly, the impact of government loan over

performance is also indecisive. It is because only one test, which ROA as dependent variable, demonstrates significant

result. However, both tests show the same negative impact of such kind of government interference. It can be conclude

that the cost of obtaining and optimizing government loan exceeds the financial benefit that may be able to reaped.

Even possible financial benefit from low interest and low transaction cost of loan acquirement may be offset by illegal

transfer paid to rent seeker in bureaucracy. This finding reinforces the previous research conclusion which is soft budget

constraint will create a conducive environment for spoiled managerial behavior [11]. These managers have no incentive

to run the firm efficiently, reluctant to compete fairly which will severely harm the firm competitiveness in the long run.

Interestingly, the findings with respect to number of government officers occupy seats on board of commissioner ap-

pears to be different from common belief that suspects that the presence of officer on board of commissioners is likely

to worsen the situation and performance. The presence of officer on board commissioner seems to contribute in helping

SOEs in accessing resources that can boost the firm performance. Otherwise, the presence of officers on the supervi-

soryboard can be as an effective tool for conducting “check and balance” among related ministry so that SOEs can op-

erate productively might be the case.

The similar explanation can be relevant for the case of PSO. The financial benefits grasped by SOE in the form of

captive revenue with a certain percentage of normal profit has been outweighed by summation of rent transferred to

official for getting the assignment and potential cost of inefficiency that incurs because of managerial moral hazard.

7. Conclusions

This paper reveals empirical evidence that government intervention in the form of state ownership, government loan,

and government assignmentprovide a negative impact to the corporate performance represented by value of ROA and

ROE. Underlying theories such as property right, soft budget constraint as well as political embeddedness have ex-

planatory power in explaining the findings related to those government actions. The results also seem to be consistent

with previous studies. Surprisingly, the resultassociated with number of government officers seat on the board of com-

missioner appears to be different from common belief.The study demonstrates that the number of officers on supervi-

sory board has positive impact to the firm performance. However, the impacts of SOEs form and capital injection have

not been clear yet. Further test need to be done using some other quantitative performance measure such as efficiency to

clarify the effect or using a qualitative approach instead of quantitative tools.

In addition, the net impact of each government intervention is the resultant of all possible benefits that probably

can be reaped and all potential costs may occur from such government actions. These costs include a rent transferred to

authorities as a cost of interference and potential cost of inefficiency that occurs because of managers‟ moral hazard.

Using the empirical findings of this paper, future related research can examine how institutional structure and incentive

system can play role in making each government intervention favorable not only for SOEs but also for society as a

whole.



8. Acknowledgements

I am grateful for insightful comments and encouragement provided by my PhD supervisor, Professor Yasushi Suzuki,

as well as my colleagues in Ritsumeikan Asia Pacific University, especially Suminto,AliAbidin, and JoshuaYadonfor

their fruitful discussion and suggestions to this research.

9. References

[1] R. H. Coase, "The Problem of Social Cost". Journal of Law and Economics3 (1), 1960, pp. 1–44.

[2] E.F. Fama and M.C. Jensen, “Separation of Ownership and Control”, Journal of Law & Economics, Vol. 26, 1983, pp. 301-25.

[3] A. Shleifer and R.W.Vishny, “A survey of Corporate Governance”, Journal of Finance, Vol. LII, 1997, pp. 737-83.

[4] A. Cuervo and B.Villalonga, “Explaining the Variance in the Performance Effects of Privatization”, Academy of Management

Review, Vol. 25, 2000, pp. 581-90.

[5] J.Y.Linand G. Tan, “Policy Burdens, Accountability and Soft Budget Constraint”. American Economic Review 89 (2), 1999,

pp. 426–431.

[6] J. Kornai, “Vision and Reality: Market and State”, Routledge, New York, 1990.

[7] M..Dewatripont, and E. Maskin, “Credit and Efficiency in Centralized and Decentralized Economies”. Review of Economic

Studies Vol. 62 (4), 1995, pp. 541–555.

[8] D. Li, “Public Ownership as the Cause of a Soft Budget Constraint”. Mimeo, Harvard University, 1992.

[9] I.R. Segal, “Monopoly and Soft Budget Constraint”. RAND Journal of Economics 29 (3), 1998, pp. 596–609.

[10] J.Y. Lin, F. Cai, and Z. Li, “Competition, Policy Burdens, and State-Owned Enterprise Reform”. American Economic Review,

Vol. 88 (2), 1998, pp. 422–427.

[11] B. Jalan, ”India‟s Economic Crisis: The way ahead”, Oxford University Press, New Delhi, 1991.

[12] E. Michelson,”Lawyers, Political Embeddedness, and Institutional Continuity in China‟s Transition from socialism”. American

Journal of Sociology, Vol. 113, 2007, pp. 352–414.

[13] M. Boycko, A. Sheleifer, R. Vishny, “A Theory of Privatization”, Economic Journal 106, 1996, pp. 309–319.

[14] Y. Aharoni, ”The performance of State-Owned Enterprises‟. In Toninelli, P. A. (Ed.), The Rise and Fall of State-Owned Enter-

prise in the Western World”, New York: Cambridge University Press, 2000, pp. 49–72.

[15] I.Okhmatovskiy, “Performance Implications of Ties to the Governmentand SOEs: A Political Embeddedness Perspec-

tive”.Journal of Management Studies, Vol. 47, 2010.

[16] L. D. Alessi, “The Economics of Property Rights: A Review of The Evidence”, Research in Law and Economics, Vol. 2, 1980,

pp. 1-47.

[17] C. R. Xin and J. L.Pearce, “Guanxi: Connections as Substitutes for Formal Institutional Support”. Academy of Management

Journal, Vol. 39, 1996, pp. 1641–58.

[18] J. M. Buchanan, “Theory of Public Choice”, University of Michigan Press, 1972.

[19] A. Cuervo, and B. Villalonga, “Explaining the Variance in the Performance Effects of Privatization”, Academy of Management

Review, Vol. 25, 2000, pp. 581-90.

[20] R. Viverita, and M. Ariff, “Corporate Performance of Indonesian Private and Public Sector Firms: Financial and Production

Efficiency”, University of Queensland, Brisbane, 2004.

[21] J. F. J. Hair, R. E. Anderson, R. L. Tatham, and W. C. Black, “Multivariate Data Analysis”. Englewood Cliffs, NJ: Pren-

tice-Hall, 1998.

[22] E. Yonnedi,“Privatization, Organizational Change and Performance:Evidence from Indonesia”, Journal of Organizational

Change Management, Vol. 23 No. 5, 2010, pp. 537-563.

[23] J. E. Stigliz, “Whiter Socialism” MIT Press, 1996, p. 79.


48



Promoting Collaboration among Stakeholders in

Citarum River Basin Problem

Utomo Sarjono Putro*; Dhanan Sarwo Utomo; Pri Hermawan School of Business and Management, Institut Teknologi Bandung, Bandung, Indonesia

*[email protected]

ABSTRACT

This research aims to develop an agent-based simulation model of the dynamic of negotiation based on interaction

among autonomous agents, who have different interests and act based on their emotion. Agents in this model are

equipped with emotion and ability to learn, and negotiate each other based on drama theory framework. To illustrate

the simulation model, an environmental conflict case in Citarum river basin is discussed in this paper. Qualitative study

is used to gather information regarding the agent‘s historical options, positions and preferences. Based on the qualita-

tive study result, the historical dynamics of common reference frame in the real world is obtained. The simulation

model is then tested and validated by comparing with historical dynamics of the real conflict in Citarum river basin.

Using this simulation, it is possible to describe possible outcomes of the conflict evolution and suggest policy in order

to reduce dilemmas and encourage collaboration among agents in the real world.

Keywords: agent based simulation, drama theory, dilemma, collaboration

1. Introduction

A conflict, from mere difference of opinion to deadly confrontations, is unavoidable in daily life. Negotiation as an

effort to resolve conflict is a very common process in everyday life. This is why negotiation process is studied in

many scientific fields such as economy, political science, psychology, organizational behavior, decision sciences,

operations research and mathematics [1].

One of conflicts in real worldis a conflict occurred in the Citarum River Basin. Citarum River is the longest

river in West Java province. Many people depend on the Citarum River, making it one of the most strategic river in

Indonesia. Unfortunately, the condition Citarum River now has changed completely. Since the industrialization in

the 80s the Citarum River turned into industrial landfills. At present, there are around 500 textile factories that di s-

pose their waste into Citarum River, many of which are conducted without proper waste treatment. Citarum river

condition is worsened by the population explosion in the upstream area. Increasing population has also increased

the number of illegal logging and disposal of household waste. As a result, flood always occurs during the rainy

season due to sedimentation in downstream areas of rivers and increasing number of barren land. Citarum River

Basin problem involves many stakeholders. Based on the literature study and focus group discussion; there are a t

least 33 stakeholders in Citarum River Basin Conflict. These stakeholders have conflicting interest and then, efforts

to restore the condition of Citarum River become more and more difficult.

Negotiation in the real world such as the one in Citarum River Basin conflict posses several characteristics i.e.:

1) Decentralized [1], i.e., parties in a negotiation have different frames and strategy in seeking resolution of co n-

flict; 2) Involving communication among parties [1]; 3) Decisions of negotiators are in terlinked through communi-

cation processes that involve many different levels [2]; 4) Involving incomplete information [1], for example, a

party cannot know for certain utilities from the other parties; 5) Involving repeated interaction with no

well-structured sequences [1]; 6) Emotion is an important device in structuring goals, values and preferences [3]

and affects communication [2].

Negotiation process reflects the characteristics of a complex system since: 1) the elements involved in a neg o-

tiation process are heterogeneous and autonomous agents (parties); 2) agents involved a negotiation process are

bounded rational, so that they may have bias in information and have a misperception toward the other agents; 3)

communication process in a negotiation involves transmission of knowledge that will influence the behavior of its

recipient; 4) negotiation is an iterative process. Such process involves feed-back loops that allows an agent to learn

and revise his/her strategy. Accordingly, the system (condition during the negotiation process in this case) evolves

over time [4]; 5) in general, interactions in a negotiation process are non-linear in the sense of, for an action there

This paper is based on research sponsored by the Air Force Research Laboratory, under agree-

ment number FA2386-10-1-4091. The U.S. Government is authorized to reproduce and distri-

bute reprints for Governmental purposes notwithstanding any copyright notation thereon.

Promoting Collaboration among Stakeholders in Citarum River Basin Problem

50

are many possible outcomes that could be produced and, for an outcome there are many possible actions that may

cause it.

The objective of this study is to construct an agent-based simulation of the dynamics of negotiation based on

drama theory framework. Agents in the simulation model are equipped with emotions and ability to lear n. The

agent-based simulation is chosen because it can minimize the number of simplifications used by its ability to fully

represents individuals and model bounded rational behavior while, drama theory is chosen because, it proposes an

episodic model whereby situations unfold. Using the constructed model, this study will propose strategy that can

promote collaboration among stake holders in Citarum River Basin Conflict.

2. Proposed Mechanism

2.1. Model of Agent’s Options, Position and Threat

In drama theory, there are a number of agents who have options, positions, preferences and threats. Interaction among

agents occurred under the common reference frame that is, the joint perception regarding the conflict that occurs. In

this simulation, an agent is represented as a column in a common reference frame. Each agent i, has a number of options

(Oki) that are represented as rows in common reference frame. At each iteration t, agent i has position to accept or to

reject each its own option. Agent i‟s position toward its own options will generate payoff (Vokit) for the agent i. Agent

i‟s payoff has two dimensions namely accept dimension and reject dimension. If agent i‟s position is to accept option

Oik, then agent i‟s payoff in accept dimension is assigned as x (a real number between 51 to 100) and, agent i‟s payoff in

the reject dimension is assigned as (100 – x). The opposite rule applies if agent i position is to reject option Oik.

Each agent j (j ≠ i), has position to accept, reject or indifferent toward option Oik of agent i. Agent j’s position

toward agent i’s options will generate payoff (Vpokijt) for agent j. Agent j’s payoff consists of two dimensions,

namely accept dimension and reject dimension. If agent j‟s position is to accept option Oik, then agent j‟s payoff in

accept dimension is assigned as x (a real number between 51 and 100) and, agent j‟s payoff in reject dimension is as-

signed as (100 – x). The opposite rule applies if agent j position is to reject option Oik of agent i. If agent j‟s position

is indifferent toward option Oik of agent i then agent j‟s payoff in both dimensions are assigned as 50.

The total real payoff obtained by each agent by adopting its own positions in each iteration t is calculated as fol-

lows:

m

tkij

tki

ti

ti VpoVopVp )(

m = number option and (i ≠ j)

pit = positions of agent i in iteration t

(1)

While, the total payoff obtained by agent i by adopting agent j‟s positions in each iteration t is calculated as follows:

m

tkij

tki

tj

ti VpoVopVpp )(


pjt = positions of agent j in iteration t

(2)

Both payoffs are stored in real payoff matrix. The columns of this matrix represents agent i and the rows of this

matrix represents agent j. The elements on the diagonal of the real payoff matrix represent the payoff that will be ob-

tained by each agent by adopting its own position.

For all options, a set of threats is defined. The total payoff obtained by agent i by adopting threatened future in

each iteration t is calculated as follows:

m

t

kij

t

ki

t

i VpoVoTVpt )(


T = threat (3)

Each agent i has an estimation regarding the payoff that will be obtained by other agents for each of their position.

Agent i‟s estimation toward agent j‟s payoff is also consists of two dimensions, i.e. accept dimension and reject dimen-

sion. If agent j‟s accepting option Oik, then agent i estimates that agent j will obtain payoff equal to x ( x is a random

number from 51 to 100) in accept dimension and 100- x in reject dimension. The opposite rule applies if agent j is re-

jecting option Oik. If agent j is indifferent toward option Oik then, agent i estimates that agent j‟s payoff in both dimen-

sions are equal to 50. All agents store their estimation regarding other agent‟s payoffs in estimated accepting payoff

matrix and estimated rejecting payoff matrix. The columns of agent i‟s estimated payoff matrices represent agent j and

the rows represent option Oik.



2.2. Modeling Agent’s Dilemmas

In each iteration, if agent i and agent j have incompatible position (e.g. agent i accept option Oik while agent j reject the

option) among them, then confrontation dilemmas will emerge. Agent i‟s dilemmas toward agent j are determined by

the payoff that will be obtained by agent i. There are two kinds of dilemmas that are considered in this research, i.e.

rejection dilemma and persuasion dilemma [5]. Those dilemmas are defined as follows:

If agent i‟s payoff by adopting agent j‟s position is greater than or equal to agent i‟s payoff to adopt its own

threat then, agent i has rejection dilemma toward agent j.

If agent i‟s payoff by adopting agent j‟s position is less than or equal to agent i‟s payoff to adopt its own threat

then, agent j has persuasion dilemma toward agent i.

If there are no incompatible positions among agents, the collaboration dilemmas still may occur. The collaboration

dilemma considered in this research is trust dilemma. Agent i who has compatible position with agent j, will have

trust dilemma toward agent j, if agent i‟s estimation regarding agent j‟s payoff is not in accordance with agent j‟s posi-

tion. For example, agent i will have trust dilemma toward agent j if both agent i and j accept option Oik but agent i

estimates that agent j will have greater payoff by rejecting option Oik.

2.3. Negotiation Protocols

In this negotiation protocol, each agent is equipped with emotion that is modeled using PAD temperament model [6].

In this model, emotional state is constructed by three independent dimensions i.e. Pleasure , Arousal and Dominance.

The formulation of agent‟s emotional state is as follows [7].

dapdapij rrrrrrSe )1.(),,( (4)

During the simulation, an agent conducts negotiation with a partner for options on which they have incompatible

positions between them (e.g. agent i accept the option while agent j reject the option). The negotiation protocol in this

research is constructed based on rational negotiation framework in which, agent i will offer certain amount of its payoff

(sti) to agent j in order to influence agent j to change his/her position closer to agent i‟s position. The potency of agent

i‟s offer to shift agent j‟s position (Ovij) is affected by agent i‟s emotional state toward agent j (Seij), and agent j‟s per-

ception toward agent i‟s offer (Ovji) is affected by agent j‟s emotion toward agent i (Seji).

iiijij ststSeOv (5)

ijijjiji OvOvSeOv (6)

Suppose agent i‟s position is to accept option k and agent j‟s position is to reject option Oik, then agent i‟s payoff in

accept dimension will then subtracted by Ovij and agent i‟s payoff in reject dimension is added by Ovij while, agent i‟s

estimation toward agent j‟s payoff in reject dimension is also subtracted by Ovij and agent i‟s estimation toward agent

j‟s payoff in accept dimension is added by Ovij. On the other hand, agent j‟s payoff in accept dimension is added by Ovji

and agent j‟s payoff in reject dimension is subtracted by Ovji while, agent j‟s estimation toward agent i‟s payoff in reject

dimension is added by Ovij and agent j‟s estimation toward agent i‟s payoff in accept dimension is substracted by Ovij..

Similar rules apply for the opposite condition.

For each iteration, an offer from agent i is perceived by agent j, and agent i will compare the response of agent j

with agent j‟s response in the previous iteration. Then, agent i‟s emotion state toward agent j will change according to

the concept of Flow Model of Emotion [8] which then mapped into PAD dimensions.

Table 1. Change in Agent i’s emotional states

Agent i offer (compare to previous

iteration)

Agent j perception (compare to

previous iteration)

Change in agent i emotion

state toward agent j

rp ra rd

Higher higher + + +

Higher lower - + +

Lower higher + + -

Lower lower - - -


52

Through the negotiation process agents will learn to identify the emotional state that can produce the biggest shift

in position of other agents (best emotional state). Learning mechanism which is built in this study is based upon the

assumption that each agent will revise his/her emotional state according to his/her experiences in the previous iterations.

Each time agent i gives an offer to agent j, agent i will record emotional state that he/she use and the shift resulted in

agent j‟s position. If in the current iteration the shift in agent j‟s position is higher or equal to the shift in agent j‟s po-

sition in the previous iteration then, agent i will revise his/her best emotional state according to his/her emotional state

in the current iteration [9].

3. Case Study: Citarum River Basin Conflict

The simulation model in this study is constructed by using NetLogo version 4.1.2. In this study, the common reference

frame of the Citarum River Basin conflict is used as the simulation input. This Common reference frame was identi-

fied through observation and focus group discussion with the stake holders in Citarum River Basin conflict. Through

this qualitative study five agents was identified i.e. Government (G), Public Enterprise (PE), Green (GR), Community

Alliance (CA), Enterprise (E). Agent‟s options, positions, and threat are described in Table 2.

Table 2. Common Reference Frame in Citarum River Basin Conflict

During the simulation process, three scenarios are tested. In the first scenario, agents are negotiating by using a

negative emotion toward other agents. In this scenario, the value of pleasure, arousal and dominance of each agent

towards the other agents are set randomly from 0 to -1. In the second scenario, agents are negotiating by using a neutral

emotion toward other agents. In this scenario, the value of pleasure, arousal and dominance of each agent are set as

zero. In the third scenario, agents are negotiating by using a positive emotion toward other agents. In this scenario,

the value of pleasure, arousal and dominance of each agent towards the other agents are set randomly from 0 to 1. The

random assignment of the emotional dimensions is conducted because it is not possible to conduct empirical measure-

ment because of many stake holders in the real world.



Figure 1. Simulation Interface

Each scenario is run thirty times. In every run, the number of iterations needed to eliminate confrontation dilemmas and

the number collaboration dilemmas that occur when the position of all agents have been compatible are observed.

Assuming each run as a sample then, the simulation results can be tabulated and tested using ANOVA to observe the

differences among scenarios. The comparison among scenarios is shown in Table 3.

Table 3. Comparison among scenarios

The comparison results shows that if agents use negative emotions to other agents during the negotiation process then,

in average the time required to eliminate the confrontation dilemmas will be longer than if they use neutral or positive

emotions. In addition, the numbers of collaboration dilemmas that arise when agents use negative emotions are sig-

nificantly higher than if they use neutral or positive emotions.

4. Conclusions

Through this study, an agent-based simulation of the dynamics of negotiation using drama theory frame-work have been

constructed. The simulation model has involved agent‟s emotions and learning ability in the negotiation protocol.

This model is able to show the evolution of common references, and show the effect of agent‟s emotional states toward

the number of dilemmas resulted in the given common reference frame, the time required to eliminate confrontation

dilemmas and the collaboration dilemmas that potentially arise after all agents reach compatible positions.


54

The proposed model is applied to analyze the conflict in Citarum River Basin. Based on the simulation results it

can be concluded that if agents use negative emotions to other agents during the negotiation process then, the time re-

quired to eliminate the confrontation dilemmas will be longer than if they use neutral or positive emotions. In addition

the simulation results also show that the numbers of collaboration dilemmas arise when agents use negative emotions

are significantly higher than if they use neutral or positive emotions. In the real world, positive emotions can be imple-

mented in several forms, for examples, willingness to compromise, giving empathy to others and convincing others etc.

Agents who have positive emotion will not threat partners, and impose the will or the anarchist protest .

In the future, the model in this study needs to be improved by integrating other dilemmas such as threat dilemma

and positioning dilemma. The feasibility and accuracy of this simulation to represents the evolution real world con-

flict is important to be investigated.

5. References

[1] K. Sycara and T. Dai, “Agent Reasoning in Negotiation” In D. M. Kilgour and C. Eden, Eds. Advances in Group Decision and

Negotiation 4 : Handbook of Group Decision and Negotiation , New York: Springer, 2010, pp. 437-451.

[2] S. T. Koeszegi and R. Vetschera, “Analysis of Negotiataion Process” In D. M. Kilgour and C. Eden, Eds. Advances in Group

Decision and Negotiation 4 : Handbook of Group Decision and Negotiation , New York: Springer, 2010, pp. 121-138.

[3] B. Martinovski, “Emotion in Negotiation” In D. M. Kilgour and C. Eden, Eds. Advances in Group Decision and Negotiation 4 :

Handbook of Group Decision and Negotiation , New York: Springer, 2010, pp. 65-86.

[4] E. R. Smith and F. R. Conrey, “Agent-Based Modeling: A New Approach for Theory Building in Social Psychology”,

Personality and Social Psychology Review , Vol. 11, 2007, pp.87-104.

[5] U.S. Putro, M. Siallagan, and S. Novani, “Agen based simulation of negotiation process using drama theory”. Proceedings of

the 51st Annual Meeting of the International Society for the Systems Sciences. Tokyo, 2007.

[6] A. Mehrabian, “Pleasure-Arousal-Dominace: a general framework for describing and measuring individual differences in

temperament”. Current Psychology: Developmental, Learning, Personality, Social , vol 14, no 4, pp. 261-292, 1996

[7] H. Jiang, “From rational to emotional agents” PhD Thesis , University of South Carolina, Department of Computer Science and

Engineering, 2007

[8] L. Morgado and G. Gaspar, “Emotion in intelligent virtual agents:the flow model of emotion”. Proceeding Intelligent virtual

agents: 4th International Workshop , 2003.

[9] U.S Putro, P. Hermawan, M. Siallagan, S. Novani, D.S. Utomo “Agent-Based Simulation of Negotiation Process in Citarum

River Basin Conflict”. Proceedings PAN-PACIFIC Conference XXVII. Bali , 2010



Supplier Selection Model Based On Tolerance

Allocation To Minimize Purchasing Cost And

Quality Loss

Noviasari Sabatini1, Wakhid Ahmad Jauhari

2, Cucuk Nur Rosyidi

3

Production System Laboratory, Industrial Engineering Sebelas Maret University, Surakarta, Indonesia

E-mail : 1) [email protected] 2) [email protected] 3) [email protected]

ABSTRACT

Manufacturing companies do not produce some or even all of the components that make up their final product. The

components are obtained by outsourcing to suppliers. There are some benefits that comes from outsourcing activities,

such as reducing manufacturing cost, doubling before tax income, increasing company‘s performance, and helping the

company for being more focus to their core-business. However, selecting suppliers is a critical, difficult, and

time-consuming activity because if it was not done carefully, it can make the company suffered tangible and intangible

losses. The main problem related to outsourcing is the quality and variability of outsourced-material and components to

be used in the assembly process. In selecting suppliers, manufacturing company has to consider quality and purchasing

cost. Variation of components will affect the final assembly of product. On the other hand, company has its own speci-

fication of final product. The tolerance of the final product has to be allocated to components tolerances such that the

accumulated component tolerances do not exceed the product tolerance. Quality loss and purchasing cost are impor-

tant criteria because they trade-off each other. The tighter tolerance resulting in higher purchasing cost lower quality

loss. In this paper we developed a supplier selection model considering quality, demand, and supplier‘s capacity based

on tolerance allocation. The objective function of the model is to minimize purchasing cost and quality loss. A numeri-

cal example is provided using linear tolerance chain. The product consists of three components where each component

can be supplied from different suppliers and each supplier can supply more than one component. The result of numeri-

cal example shows that particular coefficient of quality loss cost and capacity of supplier has an impact for the

selection of suppliers and tolerances.

Keywords: supplier selection, tolerance allocation, purchasing cost, quality loss.

1. Introduction

In today‟s manufacturing environtment, manufacturing companies do not produce some or even all of components

that make up their final product. Components are obtained by outsourcing to suppliers. According to [1], 30%

company‟s saving comes from 50% lower procurements by outsourcing activities. There are some benefits that

come from outsourcing activities such as reducing manufacturing cost, doubling before tax income , increasing

company‟s performance, and helping the company for being more focus to their core -business [2]. However,

selecting suppliers is a critical, difficult, and time-consuming activity because if it was not done carefully, it can

make the company suffered tangible and intangible losses.

The main problem related to outsourcing is the quality and variability of outsourced-material of components to

be used in the assembly process [3]. The fact is, more than 50% manufacturing cost for non-conformance product

comes from outsourced material including cost of rework and scrap which are tangible cost of quality loss [4].

Further more, there are intangible quality loss cost which is more difficult to be measured. It happens when the

product has been received by costumer and it‟s known as loss to society [5]. This kind of quality loss has various

impacts from losing costumers to loss of the company reputation [6].

One of the critical quality indicators of a product is the tolerance. There are two approach in tolerance design

which are tolerace analysis and tolerance synthesis or tolerance allocation. In tolerance analysis, designer determine

the component tolerances first and check whether the components exceed the assembly tolerance. If the tolerance

components exceed the product tolerance, then the designer must redefine the components tolerance. In tolerance

allocation, the designer determine the assembly tolerance first and then allocate the assembly tolerance to the

tolerance of its components. When the tolerance of the assembly product is not conformed to the specification, there




Supplier Selection Model Based On Tolerance Allocation To Minimize Purchasing Cost And Quality Loss

56

will be quality loss. In selecting suppliers, manufacturing company has to consider purchasing cost and quality

loss since both are trade off each other. The tighter tolerance will result in higher purchasing cost and lower quality

loss.

Many researches have been conducted in supplier selection problem with many criteria and constraints.

Reference [3] developed a model for supplier selection to minimize purchasing cost and quality loss. The research

considered two constraints which are the tolerance of assembly product and the number of selected supplier for each

component which is only one (binary integer). Another research has been conducted by [7]. In, the research, the criteria

of selecting supplier is based on maximization of weight preference using Analytic Hierarchy Process (AHP). Three

constraints are considere in the model: (1) minimum requirement of the number of supplier for each product, (2)

maximum permissible number of products allocated to each supplier, and (3) total number of supplier assignments.

Reference [5] has been developed a supplier selection model based on taguchi loss function. There are five elements of

quality loss in the objective function which are loss of quality, loss of speed, los of flexibiliy, loss of dependability, and

cost of manufacture (if parts are made by the company) or cost of purchase (if parts are outsourced). There are three

constraints in the model, which are company‟s demand of parts, production capacity of the company, and capacity of

suppliers.

Reference [3] did not consider the technological capacity of suppliers in producing various components, while

Reference [7] considered the capacity but did not include quality loss in the criteria. Reference [5] has considered both

quality loss and technological capacity in the model with no assembly tolerance constraint. In this research, we develop

a mathematical model for selecting supplier to minimize purchasing cost and quality loss, considering tolerance of

assembly product, technological capacity of suppliers in producing various components, and allowing more than one

supplier selected.

2. Model Development

2.1. Objective Function

There are two elements of objective function, which are purchasing cost and quality loss. The objective function can be

expressed as in (1) which taken from [3]. In the equation, cij denotes purchasing cost of component i from supplier j, Q

denotes the quality loss, and 𝑥𝑖𝑗 is the decision variable. The quality loss can be expressed as in (2), where A denotes

failure cost, 𝑇𝑘 denotes the k-th asssembly tolerance, and σSij is the standard deviation of component i from supplier j.

f xij = (cijxij+Q xij )

J

j=1

I

i=1

Where

Q xij = A

Tk2

σsij2

K

k=1

2.2. Constraints

We consider three constraints as follows:

1. The constraint of tolerance specification of assembly product is taken from [3]. The constraint states that the

accumulation of component‟s tolerance must not exceed the assembly tolerance. This constraint is considered

as quality requirement of the product and can be expressed as in (3). The variance can be expressed in terms of

tolerance using (4) and (5) for components and assembly respectively.

∂f

∂𝑥𝑖

2

σSij

2xij

Ji

j=1

Ik

i=1

≤σk2 (∀i,j)

(3)

𝜎𝑆𝑖𝑗2=

tij

3Cpk

2

Ji

j=1

Ik

i=1

(4 )

(1)

(2)



σk2=

Tk

3Cp

2

(5 )

2. The minimum number of supplier for component i can be stated as (6). The constraint is used to ensure the

minimum number of supplier for each component.

xij ≥Ni

Ji

j=1

(6)

3. The maximum number of permissible component supplied from one supplier. Equation (7) shows the

constraint which is used to represents the technological capacity of the supplier in producing various

components to allow one supplier supplied more than one component.

xij ≤Oj

I

i=1

(7)

4. Binary variables for supplier selection, 1 if supplier j is selected to supply component i and 0 otherwise.

xij=0, 1 (∀ i,j)

(8)

2.3. Complete Model

The complete model developed in this research can be expressed as follows:

Minimize

f xij = cijxij+ A

Tk2

∂f

∂𝑥𝑖

2

tij

3Cpk

2

Ji

j=1

Ik

i=1

K

k=1

Jt

j=1

I

i=1

Subject to

∂f

∂𝑥𝑖

2

𝑡𝑖𝑗

3𝐶𝑝𝑘

2

xij

Ji

j=1

Ik

i=1

≤ Tk

3Cp

2

(∀i,j)

xij ≥Ni

Ji

j=1

xij ≤Oj

I

i=1

xij=0, 1 (∀i,j)

3. Numerical Example and Analysis

A numerical example is given to illustrate the implementation of the model. As in[3], we consider an assembly consists

of 3 components and has an assembly function as shown in (9).

𝑦 = 𝑓 𝑥𝑖 = 𝑥1 + 𝑥2 + 𝑥3 (9)

(12)


58

Each component x1, x2, and x3 are assumed to be normally distributed with mean μ1 = 10.0000 mm, μ2 = 30.0000 mm,

and μ3 = 20.0000 mm. Each component can be supplied from more than one supplier, but different supplier has different

tolerance and price as shown in Table 1.

Table 1. Prices and tolerances for each component on each supplier.

Component no. Supplier no.

1 2 3 4

Component 1 Tolerance (mm) 0.0020 0.0025 0.0030 0.0035

Price (IDR) 61,000 30,500 17,400 13,000


Price (IDR) 82,700 56,600 34,800 26,100


Price (IDR) 69,700 39,200 21,700 17,400

The optimization result is shown in Table 2. From the Table we can make the following observation. The quality loss

coefficient A and the capacity of suppliers impact the selected suppliers and allocation of tolerances. When A = 0, the

model will only considering purchasing cost. The model will find the cheapest prices for selecting suppliers. Supplier 4

will be selected in such circumstances since supplier 4 has the cheapest purchasing cost for all of components among all

of the available suppliers. When the capacity of supplier 4 is reduced, the model will find the second-cheapest prices,

which is supplier 3.

Table 2. Optimization result for numerical example

Capacity Of Supplier Suppier selected

A = 0 A = Rp 2,134,000

S1 S2 S3 S4 Comp.1 Comp.2 Comp.3 Comp.1 Comp.2 Comp.3

3 3 3 3 S4 S4 S4 S2 S3 S3

3 3 3 2 S4 S4 S3 S2 S3 S3

3 3 3 1 S3 S4 S3 S2 S3 S3

3 3 1 3 S4 S4 S4 S2 S3 S2

3 1 1 3 S4 S4 S4 S2 S4 S3

The price gaps between suppliers for each component will deterrmine the selected suppliers due to the reduction of

suppliers capacity. For example, when the capacity of supplier 4 is reduced from 3 to 2, the selected supplier for

component 3 will switch from supplier 4 to supplier 3 since the price gap between supplier 4 and supplier 3 for this

component is IDR 4,300 which is the least among the gap prices for the two other components. When the capacity of

supplier 4 reduced again from 2 to 1, the next selected supplier is for component 1 since the price gap between supplier

4 and supplier 3 for this component is IDR 4,400 which is the second-least of the gap prices among the other

components (see Table 3).

Table 3. Gap of prices between supplier 4 and supplier 3

Components Prices (Rp)

Supplier 4 Supplier 3 Gap

1 13,000 17,400 4,400

2 26,100 34,800 8,700

3 17,400 21,700 4,300

When A = 2,134,000, which is 10 times of the actual purchasing cost from supplier 1 for each of three components, the

selection of suppliers and tolerances are changed. If all of suppliers can supplied all of components, then the suppliers

selected are supplier 3 for the two of components and supplier 2 for the other one. Any changing of the suppliers‟

capacity excepts the capacity of those two suppliers selected will not affect the chosen suppliers for each components.

The capacity reduction of supplier will still follow the rule of the price gaps between suppliers as when A=0.



4. Conclusions

This paper presented a model for selecting suppliers to minimize purchasing cost and quality loss. There are two main

constraint considered in this paper, which are tolerance allocation and technological capacity of suppliers in producing

various components. From this research we can conclude that the quality loss coefficient A and the capacity of suppliers

impact the selection of tolerances and suppliers which follows the rule of the prices gaps between suppliers. Future

research is directed to involve the components allocation for each selected suppliers and extending the model to make

or buy analysis problem which now currently under investigation.

5. Acknowledgements

Part of this work is supported by BPI from Faculty of Engineering Sebelas Maret University.

6. References

[1] Accenture Consulting, “Achieving High Performance through Outsourcing and Procurement Mastery Podcast”, Accenture,

2008.

[2] J. Barthelemy, “The Seven Deadly Sins of Outsourcing”, Academy Of Management Executive, Vol. 17, No. 2 , 2003, pp. 87-98.

[3] C. X. Feng, J. Wang, and J. S. Wang, “An Optimization Model For Concurrent Selection Of Tolerances And Supplier”,

Computers & Industrial Engineering 40 , 2001, pp. 15-33.

[4] R. Plante, “Allocation of Variance Reduction Targets Under the Influence of Supplier Interaction”, International Journal of

Production Research, Vol. 38, No. 12 , 2000, pp. 2815-2827.

[5] J. Teeravaraprug, “Outsourcing and Vendor Selection Model Based On Taguchi Loss Function”, Songklanakarin Journal of

Science and Technology 30(4), July-Augustus 2008, pp. 523-530.

[6] R. S. Kumar, N. Alagumurthi, and R. Ramesh, “Calculation of Total Cost, Tolerance Based on Taguchi's Assymetric Quality

Loss Function Approach”, ISSN 1941-70 20 American Journal of Engineering and Applied Sciences 2 (4), 2009, pp. 628-634.

[7] A. J. Rajan, K. Ganesh, K. V. Narayanan, “Application of Integer Linear Programming Model for Vendor Selection in a Two

Stage Supply Chain”, International Conference on Industrial Engineering and Operation Management, Dhaka, 9-10 January

2010.


60



Relationship of Entrepreneurial Traits, Eagerness

to Start a Business, And Firm Performances: An

Exploratory Study in Small and Medium

Enterprises In Indonesia

Henry Pribadi and Kazuyori Kanai GraduateSchool of Economics, Osaka University, Osaka, Japan

Email: [email protected]

ABSTRACT

This paper briefly explains about the progress of our research in light of relationship between entrepreneurial traits

with firm creation and firm performances, especially in small family business in Indonesia. Two major entrepreneurial

traits models are used; entrepreneurial intention model for examining the future potential business owner and entre-

preneurial orientation for examining relationship between entrepreneurial trait with environment and firm perform-

ances. Two phases of research are taken: first phase about relationship between entrepreneurial intention theorem with

eagerness of starting or continuing a business through empirical study and second phase about relationship between

entrepreneurial orientations with firm performance through empirical and qualitative studies. Some important findings

acquired, such as important factors that determined eagerness to conduct a business, strong relationship between en-

trepreneurial intentions with higher firm performances, and how aspects of entrepreneurial orientations give strong

positive impact in leveraging firm performances.

Keywords:Entrepreneurial intention, entrepreneurial orientation, small family business, Indonesia Small and Medium Enterprises

1.Introduction

Subject of Small and Medium Enterprise (SME) and entrepreneurship will always be an important factor in building

economy of developing country, such as Indonesia. Indonesian Ministry of Small and Medium Enterprise announced

that up until 2009, there were almost 53 million unit of SME in Indonesia and those units provide jobs to almost 100

million citizen of Indonesia [1]. Reference [2] noted that Indonesian SME comprised for almost 90% of all business unit

that was founded in Indonesia. Those figures reflect how Indonesia really depends on SME growth and entrepreneur-

ship will become a key factor to develop Indonesian economy. On the other hand, researchers show that even though for

developing countries SME is a vital key to promote economic growth; evidences show that to ensure the sustainability

of a SME business is not an easy feat. Reference [3] clearly pointed out that entrepreneurs in Laos faced numerous hur-

dles in their struggle to keep the business intact. Technological barrier, lack of good human resources, lack of focus, and

harsh treatment from unfair policy of government clearly slow the development business in Laos. Reference [4] also

pointed out similar situation in Uganda, where SME unit survivability is really low in the first year of their founding

and they focused more about problems in supply chain and performances. Those finding clearly shows that good SME

business performance is a vital necessity in order to survive and more attention needed in understanding how to increase

SME firm performance through entrepreneurial action and conduct.

We believe that in order to understand more about SME firm performance, one should consider examining rela-

tionship between entrepreneurship factors of firms and successful firms‟ performance. Sustainability and survivability

of a firm will depend majorly on how good the owner of the firm can harness entrepreneurial factors and integrate them

to firm strategy and action [5]. By examining more about firm owner entrepreneurial conduct, we hope to gain more

insight about how a firm operates and which factors that really important in building and operating a business, espe-

cially a small business. Through this paper, we would like to present a brief report about our finding and action

throughout our researches in examining entrepreneurial factors in small business in Indonesia. We will present our

finding through two research phases. First phase about our research on the very base of entrepreneurship: the entrepre-

neur intention. Here we conduct a research on „future entrepreneur‟ that still in their education college ages and try to

Relationship of Entrepreneurial Traits, Eagerness to Start a Business, and Firm Performances:

An Exploratory Study in Small and Medium Enterprises in Indonesia

62

find any significant factors that contribute to the success of founding a business or succession of a business. Second

phase on this paper is about our research on entrepreneurial orientation (EO) of small family business. We conduct a

qualitative research on several small firms to explore and examine how owners‟ entrepreneurial orientation, entrepre-

neurial intention, and relation with previous generation affect their firm performance. We believe that there are linkages

of entrepreneurial intention when one still in preparation stage of conducting a business with entrepreneurial orientation

in operating stage of conducting a business. Our findings reveal some interesting facts about this relationship. Future

step and our next plan also briefly discussed after the report.

2. Entrepreneurial intention among university student

2.1.Entrepreneurial intention

There is plenty of literature about entrepreneurship that has attempted to define characteristics of entrepreneurs. One of

the earlier mainstreams of entrepreneurial research that focused on the characteristics of entrepreneurs is called the trait

approach. This approach was introduced by McClelland‟s[6] who tried to relate entrepreneurship to psychology. In the

trait approach or sometimes called personal characteristics-oriented, there is an implicit assumption that an entrepreneur

is a key actor. He is an individual who identifies opportunities, develops strategies, assembles resources and takes an

action. McClelland‟s study [6] found that most of the laid-off workers stayed at home for a while before finding similar

jobs. Yet, a small number of workers behaved differently. They tried to find a better job or started their own businesses.

McClelland [6] came out with the theory of the need of achievement. He discovered that the need of achievement was a

crucial factor for personal career decision. He further mentioned the role of family education in shaping the entrepre-

neurs‟ character traits. McClelland [6] also postulated that the propensity of individual motivation to go into business is

a force of entrepreneurship. Accordingly, competitiveness was found as the most important variable in Lynn‟s [7] study

of the relationship between national culture and economic growth. A high valuation of money was the second most im-

portant variable in Lynn‟s [7] study although the prospect of making money typically ranks low in entrepreneurs‟ stated

motivation. On the contrary, the need to be one‟s own boss or to have independence is the most significant factor [8].

Self-efficacy has been linked theoretically and empirically with many managerial and entrepreneurial phenomena.

Self-efficacy is linked to initiation and persistence at behavior under uncertainty, setting of higher goals, and reducing

of threat-rigidity and learned helplessness. This is important because opportunity recognition depends on situational

perceptions of controllability and self-efficacy [9].Over decades, the trait approach has been challenged by the envi-

ronmental approach. The environmental approach studies the most influential factors outside entrepreneurs that contrib-

ute to entrepreneurs‟ success. A number of hypotheses have also been proposed about the influence of entrepreneurs‟

families on their willingness to start their own businesses. The previous results concerning the relationship between

education and entrepreneurship are very mixed. In the US, Reynolds [10] indicates that groups with lower education

showed less interest in entrepreneurial career. In the case of university, some evidence shows that a high intelligence

student prefers to pursue his career in education or research. It means hindering the entrepreneurial intention among

high intelligence student.

2.2.Research finding

This research is based on the survey carried out in 2007 on the students at Faculty of Industrial Technology at Petra

Christian University, Surabaya, Indonesia. A random sample of students completed the questionnaire. With the ap-

proval and cooperation of the lecturers, the questionnaire distributed during class sessions. Most students completed and

returned them during the sessions. The participation was voluntary and 140 students completed and submitted the ques-

tionnaire, resulting in a response rate of over 60%. The survey consisted of a two-page structured questionnaire. The

students answered items that addressed their entrepreneurial intentions, perceived feasibility of starting a business, per-

sonal characteristics and effect of entrepreneurship education. Response options included five-point Likert scales, ap-

propriate categorical and dichotomous scales. The information obtained was analyzed using the statistical software

package STATA. In this study, OLS regression was used as an analytical tool. The post regression evaluation concerns

with the existence of multicollinearity among independent variables. To check this problem, the so-called variance in-

flation factor (VIF) was used, which is the reciprocal of tolerance. VIF increases and so does the variance of the regres-

sion coefficients, making it unstable to estimate. Large VIFs are an indication that reflects the presence of multicollin-

earity. The VIFs found in the estimates ranged from 1.24 to 1.58, meaning that no multicollinerity problems occurred.

Our regression result in Table 1 showed that self efficacy; inspiration of role model, and government bureaucracy

gave a positive and strong effect on one‟s decision and intention in starting a business. Lack of self confidence, un-

certainty on external environment, and job offer from prestigious companies gave significant negative effect in starting

a business.




Our finding shows that there is a strong evidence of how subject with strong family background in business will

give significant motivation in building or succession of a business in the future, significant negative effect on lack of

self confidence also signal the real difference between subject with family business background and not. In short, our

finding suggests examining more deeply into entrepreneurial factors of family business subject and give some thought

about how to create a good entrepreneurship curriculum in university at general [11].

Table 1. Regression result

Factors Result

Group 0.092

Sex -0.013

GPA -0.022

Self efficacy .212**a

Family ranking 0.022

Have an business experience 0.069

Inspiration from role model .172**

Motivation to be independence -0.066

Personal achievement and talents -0.034

Money-related motivation -0.062

Market-related motivation 0.046

Uncertainty on politic and economic growth -.234**

Difficulty on government bureaucracy .144**

Lack of guidelines on starting a new venture -0.002

Personal reason (e.g. married, pursue a higher degree) .132*

Receive job offering from big companies -.175**

Lack of initial investment 0.09

Lack of family support 0.001

Lack of university support 0.061

Lack of self confidence -.261**

Uncertainty on market and tight competition 0.043

F 11.01

Significance of F (Prob<F) 0

R2 0.6819

Adjusted R2 0.62

a ***significant at 1%, **significant at 5%, *significant at 10%

3. Small family business: Relationship of Entrepreneurial orientation and firm performances

The result on our first phase research clearly showed on how entrepreneurial intention in subject of future business

owner resides in strong ties of family background in business and self efficacy. These findings suggest continuing our

research of entrepreneurial activity more in family business field. Kanai [12] pointed out that entrepreneurship should

be consists of entrepreneur intention; ability to concept; and power to mobilize various resources and how important the

network effect on entrepreneurship. Therefore the next step should be to examine how business owners conceptualize

their business and the action of harnessing various resources during their business activity. Entrepreneurial orientation

concept is a good model to assist in examining about these kinds of factors [13]. Thus we should direct our second

phase research on examining the relationship of small family business, entrepreneurial orientation, and firm succession

through generation to understand more about entrepreneurial activity in Indonesia.

Entrepreneurial Orientation (EO) refers to a firm's strategic orientation, capturing specific entrepreneurial aspects

of decision-making styles, methods, and practices. As such, it reflects how a firm operates rather than what it does



64

([14]; [15]). Reference [16] summarizes the characteristics of an entrepreneurial firm as one that engages in product

market innovation, undertakes somewhat risky ventures, and is first to come up with proactive innovations, beating

competitors to the punch. Based on this, several researchers have agreed that EO is a combination of the three dimen-

sions: innovativeness, proactiveness, and risk taking. Thus, EO involves a willingness to innovate to rejuvenate market

offerings, take risks to try out new and uncertain products, services, and markets, and be more proactive than competi-

tors toward new marketplace opportunities.

Numerous studies in EO generally agreed that universally EO will give a positive effect on small business per-

formance, but it should be accepted with a grain of salt. Lumpkin and Dess[15] proposed that EO could help a small

firm to leverage its performance if certain condition of internal factors and external factors of the firms can be met.

Wiklund and Sheperd[14] argued that certain configuration of EO, network, and external environment play important

aspect in explaining small business variance in performances.

Concur with previous paragraph; we had examined the general condition of Indonesia small business in our previous

research [17]. We used an integrated model approach of configuration of internal assets of a firm (EO), external factors,

and firm strategy to explain the variance of performance in Indonesia small business. We define firm strategy as char-

acteristics of how a firm conducts their business in a distinct strategy by describing it into Porter‟s positioning strategy

[18].By referring to [19] we define firm strategy as cost leadership strategy and differentiation strategy. Firm perform-

ances were measured through how a firm performs in term of profit and market growth in last three years. We refer our

definition of firm performances and how to measure them into Spanos and Lioukas [20] works.We conducted an em-

pirical study with 256 samples of small firms in East Java and analyzed our data through structural model equation. We

found out that positive and significant relationship occurred on EO, external factors, and firm strategy together in ex-

plaining the variance of firm performances. Comparing our result in table 2 with Wiklund and Sheperd[14] result, we

confident our scope of research also confirm similar conclusion with previous studies.

Therefore, based on our finding and previous studies, we conducted a qualitative research on several small firms in

Indonesia to examine the relationship between EO and firm performance. We also include questions about entrepreneu-

rial intention, relationship with previous generation and business model to enrich our result and findings. For now, we

succeed in acquiring three samples with different variation of firm performance (high, normal, and low) to be compared

each others.

Our result in table 3 shows interesting finding regarding with relationship of firm performance, entrepreneurial in-

tention and entrepreneurial orientation. Firm with high performance exhibits good trait of the owner; such as high self

efficacy, high confidence, and high entrepreneurial orientation. This finding supports our previous studies about entre-

preneurial intention and relationship of EO with firm performance. Higher firm performance‟s firm shows stronger rela-

tionship with education level and knowledge acquisition, something that support previous study about relationship of

firm performance and knowledge assets [21]. In term of family relationship and family succession, our finding shows

that franchise-like succession model will reduce agency problem inside a family business and good relationship in the

family is one of high performance small family business characteristics.

Table 2. Structural model results

Examined path Standardized path coefficients P value Result

1. External factors → Firm strategy 0.639 0.09 Supported at 10% level

2. External factors → Firm performance -0.068 0.459 Not supported

3. Firm strategy → Firm performance 0.192 0.03 Supported at 5% level

4. EO → Firm performance 0.105 0.093 Supported at 10% level

5. EO→ Firm strategy 0.219 0.05 Supported at 5% level

Table 3. Result of samples interview

Factors A Firm B Firm C Firm

Business Watches and Bags Glassware and Kitchen utensils Gold and silver jewelry

Build 80's 80's 80's

Generation Second Second Second

Firm performance Good Normal Worst

Succession model Franchise-like Normal succession Exit/Spin off

Education level University University Drop out

Business-edu relation High Normal Low




Factors A Firm B Firm C Firm

Previous gen. relation Good Good Bad

External condition Rough Rough Rough

Cultural understanding Good Good Bad

Self efficacy High Normal Low

Self confidence High Normal Low

Agency problem None Middle High

Innovativeness High Low Low

Proactiveness High High Low

Risk taking High Low High

4. Closing remarks and future plan

ur research examines the basic building of small and medium enterprise, entrepreneurial factors that consist of entre-

preneurial intention, orientation, external factors, and firm strategy. We succeed in capturing important aspect and find-

ing of entrepreneurial intention and factors that affect variation in firm performances. We find out that in regards of

entrepreneurial intention, self efficacy and role model will leverage eagerness to start a business while offer from big

company to work and uncertainty external condition will degrade the intention. Those positive factors still hold true

when we examine the business owners, self efficacy, and good relationship with previous owner/family matters, to-

gether with factors of entrepreneurial orientation will help the firm to achieve higher performances. Regardless, our

research still in early stages; there are still many works to be done. Our next step will be to reaffirm our research and

our finding in term with previous studies. While we success in getting some interesting finding, but we believe that it is

still too premature to wrap our research in a conclusion. More data are favorable, especially qualitative data to confirm

our finding and concluded the symptom to be valid across bigger population. When more data and more literature stud-

ies have been conducted, we confident that our finding will be a good contribution to scientific world in business field

in order to give better understanding about the condition of Indonesia small and medium enterprises.

5. Acknowledgements

We wish to express our most gratitude to Monbukagakusho Scholarship of Japan Government and Osaka University

that funded our long term research and academic opportunity. We also wish to express our thanks to Petra Christian

University in Surabaya, Indonesia which provide valuable data for our research, also our fellow colleagues and profes-

sors whose generous input help us in conducting our research.

6. References

[1] KementrianKoperasidan Usaha Kecil danMenengahRepublik Indonesia, Indonesian. SME development 2005-2009.

http://www.depkop.go.id

[2] A. G. Brata, DistribusiSpasial UKM dimasakrisisekonomi. JurnalEkonomi Rakyat, Vol. 2, No.8, 2003.

[3] N. Southiseng, and J. Walsh, Competition and Management Issues of SME Entrepreneurs in Laos: Evidence from Empirical

Studies in Vientiane Municipality, Savannaketh, and LuangPrabang. Asian Journal of Business Management, Vol. 2, No. 3,

2010, pp. 57-72.

[4] S. Eyaa and J. M. Ntayi, Procurement Practices and Supply Chain Performances of SME‟s in Kampala. Asian Journal of Busi-

ness Management, Vol. 2, No. 4, 2010, pp. 82-88.

[5] G. J. Avlonitis and H. E. Salavou, Entrepreneurial orientation of SMEs, product innovativeness, and performance. Journal of

Business Research, Vol. 60, 2007, pp. 566-575.

[6] D. C. McClelland, Achievement and Entrepreneurship : A longitudinal Study, Journal of Personality and Social Psychology,

Vol. 1, No. 4, 1965, pp 389-393

[7] R. Lynn, The Secret of the Miracle Economy: Different National Attitudes to Competitiveness and Money. 1991, London: The

Social Affairs Unit.

[8] D. J. Storey, Understanding the Small Business Sector. 1994, London: Routledge.

[9] N. Krueger and D. V. Brazael, Entrepreneurial Potential and Potential Entrepreneurs, Entrepreneurship Theory & Practice,

Spring 1994, pp. 91-104.

[10] P.D. Reynolds, Who Starts New Firms? Linear Additive versus Interaction Based Models”, paper presented at the 15th Babson

College Entrepreneurship Research Conference, London, April 1995, pp. 13-15.



66

[11] D. Soetanto, H. Pribadi, I. G. A. Widyadana, Determinant factors of entrepreneurial intention among university student. The

IUP Journal of Entrepreneurship Development, Vol. 7, Nos. 1 & 2, March & June 2010, pp. 23-37.

[12] K. Kanai, The mechanism for promoting entrepreneurship. In K. Gonda, F. Sakauchi, T. Higgins. Regionalization of Science

and Technology Resources in the Context of Globalization, Tokyo 1994.

[13] J. C. Casillas, A. M. Moreno, and J. L. Barbero, Entrepreneurial orientation of family firms: Family and environmental dimen-

sions. Journal of Family Business Strategy, Vol. 2, 2010, pp. 90-100.

[14] J. Wilkund, and D. Shepherd, Entrepreneurial orientation and small business performance: a configurational approach. Journal

of Business Venturing, Vol. 20, 2005, pp. 71-91.

[15] G. Lumpkin, and G. G. Dess, Clarifying the entrepreneurial orientation construct and linking it to performance. Academy

Management Review, Vol. 21, No. 1, 1996, pp. 135-172.

[16] D. Miller, The correlates of entrepreneurship in three types of firm. Management Science, Vol. 29, 1983, pp. 770-791.

[17] H. Pribadi and K. Kanai. Examining and Exploring Indonesia Small and Medium Enterprise Performance: An Empirical

Study. Asian Journal of Business Management, Vol. 3, No. 2, 2011, pp. 98-107.

[18] M. E. Porter, Competitive Advantage. 1985. The Free Press.

[19] M. Acquaah and M. Yasai-Ardekani, Does the implementation of a combination competitive strategy yield incremental per-

formance benefit? A new perspective from a transition economy in Sub-Saharan Africa. Journal of Business Research, Vol. 61,

No. 4, 2008, pp. 346-354.

[20] Y.E. Spanos and S. Lioukas. An examination into the causal logic of rent generation: Constrasting Porter‟s competitive strategy

framework and the resource-based perspective. Strategy Management Journal, Vol. 22, 2001, pp. 907-934.

[21] Y. H. Li., J. W. Huang, and M. T. Tsai, Entrepreneurial orientation and firm performance: The role of knowledge creation

process. Industrial Marketing Management, Vol. 38, 2009, pp. 440-449.


Copyright © 2011 IESS

A Framework For Assessment and Validation of

Construction Project Management Performance

Din, Sabariyah* and Abd-Hamid, Zahidy**

UTM Razak School of Engineering and Advanced Technology, Universiti Teknologi Malaysia, International Campus,

Kuala Lumpur, Malaysia. Email: *([email protected]) ** ( [email protected]).

ABSTRACT

There has been a limited framework from which construction companies could confidently apply to derive conclusions

on the relationship between the ISO 9000 Quality Management System (QMS) certification and construction project

performance. In view of this limitation, this paper offers a project management framework on which a model called a

Project Management Performance Assessment for Contractors (PMPAC) was built. Based on the framework, a set of

questionnaire, after being piloted was mailed to project managers to gather sample data. Sample was drawn from

Grade G7 ISO certified and non-ISO certified Malaysian construction companies. Data were analyzed to test three

hypotheses. The study revealed that there is a significant difference in Project Management Practices, no significant

difference in Project Success and a significant difference in Financial Returns between the two data sets, encompassing

ISO 9000 certified and non certified construction companies. The framework was validated by practitioners who

testified that the PMPAC Model is an effective tool for assessing construction project management performance.

Keywords: Construction, Project , Management Performance Framework, ISO 9000

1. Introduction

The construction industry is a huge industry, accounting for around 10% of the world‟s gross domestic product (GDP),

7 % of employment and up to 40% of energy usage [19]. Nevertheless, it is often criticized for being inefficient [3] and

that the industry failed to meet clients‟ requirements [9]. The Malaysian construction industry is no exception where a

number of large projects were abandoned, mainly due to financial problems [1]. In the effort to eliminate those negative

reputations, the Construction Industry Development Board Malaysia (CIDB) had introduced a compulsory measure for

Grade G7 contractors to be certified with the ISO 9000 QMS (by January 1st 2009), before they could undertake any

business operations in Malaysia.

This paper seeks to explore the relationship between ISO 9000 certification and construction project performance.

It firstly details the formulation of the conceptual framework, then the application of the framework and brief research

methodologies followed by the research findings, discussion and conclusion.

2. Previous Studies of ISO 9000 Certification and Project Management Performance

A number of past studies focused on the motives of gaining the ISO 9000 QMS certification. The most frequently men-

tioned were to expose the image of the organization, to improve the business performance [5], and to capture project re-

lated benefits [12; 21] from the ISO 9000 QMS certification, through internal changes in the operational functions. Later

Benner and Veloso [4] highlighted improvements in revenue through wider access to new customers, after adopting the

ISO 9000 QMS. Lo and Humphreys [14] suggested that project management techniques could be used in developing a

project network and in resources loading profile to ensure an effective and efficient implementation of the QMS.

Orwig and Brennan [17] noticed that many of the elements of quality management systems were applied on key

business processes involving repetitive, steady-state and standardized manufacturing operations. Serpell [20], while

recognising that the QMS has its origin in manufacturing, the concept could be effectively applied to construction pro-

ject environments. Construction is however unique; in that no two projects are exactly the same. Construction projects

are characterized by their complexity and by an evolving non-standardized nature of the management processes. Due to

the fundamental differences of the two sectors, Kazaz and Birgonul [13] viewed that the manufacturing-oriented quality

concept cannot be directly applied to the construction industry.

With regards to Project Success (PS), Heerkens [10] suggested that PS could be measured in four levels: Meeting pro-

ject targets; Project management efficiency; User utility and Organizational improvement. PS can also be measured in

mailto:%[email protected]

A Framework For Assessment and Validation of Construction Project Management Performance

68

the form of lessons learnt from prior failures or successes [8]. Lock [15] noticed that the construction industry has a

long record of adopting project management methods effectively and the success of projects is said to be more depend-

ent on the people who managed, rather than the specialized equipment applied to affect the standardization.

In a study on project financial returns (FR), Beatham [3] noted that traditionally, company‟s performance was

measured solely in financial terms, profit and turnover. Manoochehri [16], however found that traditional financial

measures based on accounting concepts and practices are often inappropriate and insufficient. Since FR are categorized

as lagging indicators, they are considered to be poor predictors of tomorrow‟s performance [18]. A survey of 114 pro-

ject managers, Cook [7] concluded that financial returns (FR) had a positive impact on Project Success.

3. Conceptual Framework, Model Development, Hypotheses and Methodology

3.1 Conceptual Framework

It is proposed that the effects of ISO 9000 certification efforts on the project‟s management (PM) performance can be

evaluated by obtaining measurements from three broad components: Project Management Practices (PMP), Project

Success (PS), and Project Financial Returns (FR). These three components are treated as dependent variables and the

independent variable being the „certification‟. The hypothesis (see 3.3) against each component is marked in Figure 1 as

H1, H2, and H3 respectively.

Figure 1: The Framework of Project Management (PM) Performance.

3.2 Model Development

The Project Management Performance Assessment (PMPA) Model of Bryde [6] was referred. The Model conforms

with the framework of the European Foundation for Quality Management (EFQM) Business Excellence. Hillman [11]

noted that the EFQM Model provides a tried and tested framework. Figure 2 represents the PMPA‟s Model, showing

the enablers (inputs) such as PM Leadership on the left, and output such as PM KPI on the right.

Figure 2: The PMPA model from Bryde [6].

Some variables related to QMS certification are not measured in the above PMPA‟s model. It is therefore extended and

now called the „Project Management Performance Assessment for Contractor or „The PMPAC Model‟. Figure 3 intro-

duces some additional business performance indicators. On the left are the enablers (inputs) of PM Leadership and on

the right are results: Project Success (PS) and Financial Returns (FR) to be assessed using this PMPAC Model.

Project Management

(PM)

Leadership

PM Staff

PM Policy

& Strategy

PM Partnerships

& Resources

Project Life Cycle

Management

Process

PM Key Performance

Indicators

(KPIs)

RESULTS ENABLERS

H3

H2

H1

ISO-certified

Construction

Companies

Non-certified

Construction

Companies

PM Practices

(PMP)

Project Success

(PS)

Project Financial

Returns

(FR)

Performance Management

Outcomes from the ISO vs.

Non ISO certified companies



Figure 3: The PMPAC Model

3.3 Hypotheses on Project Management (PM) Performance

i. PM Practices: Given that ISO 9000 QMS is used to assure the quality of management processes derived from

the PM Practices of construction projects, it is expected that ISO 9000 certification will result in enhanced

PM Practices. The first hypothesis is:

H10: There is no difference in PM Practices (PMP) between ISO certified and non-ISO certified construction companies

ii. Project Success: In relation to suggestions put forward by Heerkens [10] and Forsberg [8] led to the second

hypothesis:

H20: There is no difference in Project Success (PS) between ISO-certified and non-certified construction companies.

iii. Financial Returns: Given some evidences reported in Manoochehri [16], Parker [18], Beatham [3] and Cook [7]

this hypothesis is postulated:

H30: There is no difference in Financial Returns (FR) between ISO-certified and non-certified construction companies.

3.4 Methodology

The PMPAC Model was applied to measure PM Performance (PMP). A set of questionnaire was designed, containing

the following enablers: Under PMP, questions were arranged in these categories: PM Leadership (5 questions);

PM Staff (2 questions); PM Policy and Strategy (3 questions); PM Partnerships and Resources (2 questions); Project

Life Cycle Management Process (4 questions); PM Key Performance Indicators (4 questions). Under Financial Returns

(6 questions) and Project Success (10 questions). Data were collected on PM practices, PS, FR by using structured

questionnaire divided into four parts. Part 1: Descriptive data on the respondents‟ organization. Part 2 : Enablers (PM

Practices), arranged in these categories: PM Leadership, PM Staff, PM Policy and Strategy, PM Partnerships and

Resources. Part 3: Perceptions of PS and facts on FR; Part 4: Demographic data of respondents. A five-point Likert

scale (1 = strongly disagree, 2 = disagree, 3 = neutral, 4 = agree and 5 = strongly agree) was used to measure PMP. The

questionnaire was piloted with 20 project managers, 10 from ISO-9000-certified and 10 from non-certified construction

companies. Tests for questionnaire‟s reliability confirmed the appropriateness of the data, without any items deleted.

The internal and external validity showed that an additional refinement of the survey instrument was not needed.

The sample was drawn from approximately 130,000 companies listed in the CIDB Directory 2006-2007. Table of

random numbers was used. The population for the ISO certified companies were limited to those which had been certi-

fied from year 2004 and had completed at least one project since first obtaining certification. A total of 151

ISO-certified construction companies and 3437 non-ISO certified companies matching the criterion were identified. An

approximately 20% of the non- ISO certified companies were selected as the sample, resulted in 806 companies being

systematically selected.Table1below shows the distribution of completed questionnaires from where the response data

were tabulated. Assumptions for the multivariate MANOVA test as suggested by Tabachnick and Fidell [22], were

evaluated. These include: unequal sample size, multivariate normality, linearity, outliers, homogeneity of vari-

ance-covariance matrices, reliability of covariates, and multi-collinearity and singularity. The normality test confirmed

that the original data set was approximately normal. Bias test for non response showed no significant difference.

RESULTS

PM Staff

Project Life

Cycle Man-

agement Proc-

ess

PM Project,

Success

PM Finan-

cial Returns

PM Key Per-

formance Indi-

cators (KPIs)

Project Manage-

ment (PM) Lead-

ership

PM Policy

& Strategy

PM Partner-

ships & Re-

sources

ENABLERS


70

Table 1: Number of Completed Questionnaire

Type of Samples Response Response (Rate) Incomplete Complete (Rate) Total sample

ISO Companies 151 73 (48.3%) 2 71 (47%)

Non- ISO Companies 806 263(32.6%) 18 245(30.4%) 336

4. Data Analysis and Model Validation

4.1 Descriptive

There were 57 (80.3%) male and 14 (19.7%) female respondents of ISO-certified companies, and 184 (75.1%) male

and 61 (24.9%) female respondents of non-ISO companies. Most had more than 15 years experience in PM (n=27,

38.0% for ISO-certified companies and n=86, 35.1% for non-certified companies). A t-test and MANOVA was run to

check if there were any respondent bias in respect of the sex or the level of experience. The result of the t-test, based on

mean scores for PM Practices showed no significant difference between practices by genders. There was no significant

difference between different levels of experience and PM Practices, Financial Returns and Project Success mean scores

[Pillai‟s Trace (0.377) >Alpha (0.01)]. The proportion having ISO 9000 certification amongst larger organizations was

higher than that for smaller organizations.

4.2 Hypotheses Testing

Results from the F test tabulated in Table 2(a), suggest that there is a significant difference in PM Practices between the

ISO-certified and non-certified construction companies. Table 2(b), MANOVA results show a significant difference

between the two groups for each of PM Policy and Strategy, Project Life Cycle Management Process, and KPIs.

The other three factors - Leadership, Staff, and Partnerships and Resources show no significant difference at p < 0.01,

between the two groups of companies.

Table 2(a) Summary of MANOVA Test for Project Management Practices

ISO-certified

(N = 71)

Non-certified

(N = 245) Tests of Between-Subjects Effects

Variables Mean SD Mean SD η2 F Sig.

PM Practices

3.8127

0.39113

3.5431

0.48501

4.002

18.449

0.000

Table 2(b) Summary of MANOVA Test on Factors of Project Management Practices

ISO-certified

(N = 71)

Non-certified

(N = 245)

Tests of Between-Subjects

Effects

Mean SD Mean SD η2 F Sig.

Leadership

Staff

Policy and strategy

Partnerships and resources

Project life cycle

management process

Key performance indicators

3.6845

3.9577

4.0516

3.7394

3.8697

3.7007

0.49067

0.56535

0.46683

0.60273

0.55719

0.56001

3.5469

3.7449

3.7918

3.5796

3.4633

3.3122

0.43074

0.74742

0.57950

0.82158

0.69959

0.75235

1.042

2.494

3.716

1.406

9.094

8.307

5.266

4.935

12.004

2.323

20.230

16.295

0.022

0.027

0.001

0.128

0.000

0.000

Table 3(a) shows the F test yielded a p-value of 0.038. With an alpha of 0.01, H20 is not rejected. Therefore, one

might conclude that there was no significant difference in Project Success between the ISO-certified and non-certified

construction companies however, as shown in Table 3(b), the MANOVA results reveal that four factors of PM Success

namely: Within Budget, Efficient Management, Benefit to Intended User, and Impact on Company's Business show a

significant difference between the two groups.



Table 3(a) Summary of MANOVA Test for Project Success

ISO-certified

(N = 71)

Non-certified



Project Success

4.1958

0.56327

4.0482

0.51435

1.199

4.341

0.038

Table 3(b) Summary of MANOVA Test on Factors of Project Success

ISO-certified

(N = 71)

Non-certified

(N = 245)


Effects


Within schedule

Within budget

Efficient management

Within quality

Works accordingly

Use by Intended User

Benefit to Intended User

Impact on Client‟s performance

Impact on company‟s business

results

Lessons learned

3.96

4.00

4.01

4.23

4.21

4.27

4.35

4.23

3.31

4.39

0.977

0.956

0.853

0.721

0.607

0.585

0.588

0.741

0.600

0.686

3.96

3.83

3.90

4.03

4.00

4.02

4.04

4.01

4.29

4.40

0.879

0.884

0.783

0.636

0.668

0.689

0.664

0.698

0.660

0.582

0.002

1.618

0.691

2.132

2.363

3.476

5.334

2.500

0.022

0.000

0.002

1.994

1.082

4.951

5.503

7.804

12.710

4.989

0.053

0.000

0.000

0.006

0.003

0.016

0.017

0.024

0.004

0.016

0.000

0.000

The F test of Table 4(a) led to rejecting H30 and concluding that there was a difference in Financial Returns between the

certified and non-certified construction companies. As shown in Table 4(b), the MANOVA results revealed that three

factors of Financial Returns namely: Financial Calculation Procedure, Financial Contingency Plan, and Effect of Price

Escalation show significant differences, between the two groups.

Table 4(a) Summary of MANOVA Test for Financial Returns

ISO-certified

(N = 71)

Non-certified



Financial Returns

3.7441

0.49828

3.4476

0.60043

4.840

14.426

0.000

Table 4(b) Summary of MANOVA Test on Factors of Financial Returns

ISO-certified

(N = 71)

Non-certified

(N = 245)


Effects


Financial calculation procedure

Financial contingency plan

Amount loan used

Inflation allowance and price es-

calation

Effect of price escalation

Availability of positive financial

returns

4.10

3.96

3.59

3.61

3.70

3.51

0.539

0.685

0.785

0.746

0.782

0.826

3.69

3.59

3.34

3.39

3.41

3.26

0.764

0.823

0.917

0.893

0.853

0.968

9.199

7.370

3.405

2.516

4.693

3.438

17.750

11.685

4.304

3.386

6.693

3.903

0.000

0.001

0.039

0.067

0.010

0.049

The PMPAC Model has been validated to determine whether or not the Model which was built based on the framework

shown in Figure 1, could become an effective tool for assessing PM Performance in the construction industry. The

Model validation was carried out in 2 phases and interview method was adopted: In Phase 1, twenty two (22) randomly

selected construction companies of various grades and three (3) developers were involved The total score of each re-

spondent was calculated. The results indicate that 6 or (24.0% ) of the respondents had less than 50% of the PMPAC

scores, thus were below the average project management performance. In Phase 2, the Public Works Department

(PWD) of the State of Pahang (client to construction contractors) was consulted. The District Engineer, representing

PWD was asked to rank the corresponding contractor‟s Project Management Performance. Likert Scale was applied: 1=


72

Very Good, 2= Good, 3= Fair, 4=Poor and 5= Very Poor. The rank given by the District Engineer was compared

against the level of project management performance reported by the corresponding contractor from the same district.

Five (5) most recently completed projects from each district were chosen. The validation test result showed in Table 5

had indicated that there was no significant difference (t = 2.679, p > 0.01) in PM Performance mean scores (at a 1 per-

cent level of significance) reported between contractors and the clients (the District Engineer). It thus suggests, that the

validated PMPAC Model could be used for assessing PM performance in the Malaysian construction industry.

Table 5 Summary of t-Test Analysis (Second Phase of PMPAC Model Validation Test)

N Mean Standard Deviation df t Sig. (2-tailed)

Contractor

PWD

17

17

77.0235

66.1765

6.28499

15.46462

32

2.679

0.012*

*Significant at the 0.01 level

5. Discussions and Conclusion

Results from the analysis indicated that the ISO 9000 QMS certification had a positive effect on PM Practices and

Financial Returns (FR), but not on Project Success (PS). Findings on the PM Practices seemed consistence with the

views forwarded by Brown [5], Serpell [20], Lo and Humphreys [14], Karim [12] and Benner and Veloso[4]. Findings

also showed that ISO 9000 QMS did not enhance PM Partnership and Resources. The absence of a significant differ-

ence between ISO-certified and non-certified construction companies in the PM Project Success involved some en-

ablers; such as „Use by its Intended User‟ and „Lessons Learned from Completed Project‟. These results might be re-

lated to the current practices of quality assurance procedures adopted by the companies. Added Au and Yu [2] that the

contract was good enough to control the quality procedures adopted by construction companies. In Malaysia however,

the quality of the construction work was totally relied on the supervision of the project manager and site supervisor,

while the activities of quality assurance in most construction companies were oriented to meet technical specifications

of the final product, often expressed through owner‟s inspection.

Findings reported in this paper suggest that the PMPAC Model could be used as an effective tool for assessing

construction project management performance. The framework had shed lights for a symbiotic relationship between the

ISO 9000 QMS certification effort and a project management practices in the construction industry. Some limitations

may be raised since data analyzed in this study were based on replies from what project managers could recall from his

experience in managing the most recent completed project. Nevertheless, the rich experience from construction project

had been utilized in designing a project management performance framework, if applied could counter the negative

reputations of the construction industry. It is suggested that a wider project stakeholders, such as contractors from

various grades and their clients should be included, in order to enhance the present findings.

In synthesis, improvement in the QMS may have to be industry tailored to warrant successful application of the

system. Many related companies should also focus on systematic project management activities, while applying the

quality management systems as catalyst to achieve better project performance, financial returns and project success.

6. References

[1] Alaghbari, W., Kadir, M. R. A., Salim, A., and Ernawati (2007). The significant factors causing delay of building construc-

tion projects in Malaysia. Engineering, Construction and Architectural Management, 14(2), 192-206.

[2] Au, J. C. W., and Yu, W. W. M. (1999). Quality management for an infrastructure construction project in Hong Kong. Lo-

gistics Information Management, 12(4), 309-314.

[3] Beatham, S., Anumba, C., Thorpe, T., and Hedges, I. (2004). KPIs: a critical appraisal of their use in construction. Bench-

marking: An International Journal, 11(1), 93-117.

[4] Benner, M. J., and Veloso, F. M. (2008). ISO 9000 practices and financial performance: a technology coherence perspective.

Journal of Operations Management, 26, 611-629.

[5] Brown, A., Van der Wiele, and Loughton, K. (1998). Smaller enterprises‘ experiences with ISO 9000. International Journal of

Quality & Reliability Management, 15(3), 273-285.

[6] Bryde, D.J., 2003. Modelling project management performance. Int. J. of Qual. & Rel. Mgt. 20, 2, 225-229.

[7] Cook, B. W. (2004). Measuring the value of success in project management organizations. Argosy University-Orange

County, USA: DBA Dissertation.

[8] Forsberg, K., Mooz, H., and Cotterman, H. (2000). A model for business and technical success, second edition. John Wiley &

Sons, Inc. N.Y., USA.

[9] Giles, R. (1997). ISO 9000 perspective for the construction industry in the UK. Training For Quality 5(4), 178-181.



[10] Heerkens, G. R. (2002). Project management. McGraw-Hill, N.Y., USA.

[11] Hillman, G. P. (1994). Making self-assessment successful. The TQM Magazine, 6(3), 29-31.

[12] Karim, K., Marosszeky, M., and Davis, S. (2006). Managing subcontractor supply chain for quality in construction. Engi-

neering, Construction and Architectural Management, 13(1), 27-42.

[13] Kazaz, A., and Birgonul, M. T. (2005). The evidence of poor quality in high rise and medium rise housing units: a case study

of mass housing projects in Turkey. Building and Environment, 40, 1548-1556.

[14] Lo, V., and Humphreys, P. (2000). Project management benchmarks for SMEs implementing ISO 9000. Benchmarking: An

International Journal, 7(4), 247-259.

[15] Lock, D. (2004). Project management in construction. Gower Publishing Limited, Hants, UK.

[16] Manoochehri, G. (1999). Overcoming obstacles to developing effective performance measures. Work Study, 48(6),223-229.

[17] Orwig, R. A., and Brennan, L. L. (2000). An integrated view of project and quality management for project-based organiza-

tion. International Journal of Quality & Reliability Management, 17 (4/5), 351-363.

[18] Parker, C. (2000). Performance measurement. Work Study, 49(2), 63-66.

[19] PricewaterhouseCoopers (2008). Engineering & construction industry sector. PricewaterhouseCoopers International Limited

[20] Serpell, A. (1999). Integrating quality systems in construction projects: the Chilean Case. International Journal of Project

Management, 17(5), 317-322.

[21] Singels, J., Ruel, G., and Van de Water, H. (2001). ISO 9000 series-Certification and performance.International Journal of

Quality & Reliability Management, 18(1), 62-75.

[22] Tabachnick, B. G., Fidell, L. S.(2007), Using Multivariate Statistics, 5th Edition. Boston: Pearson Education, Inc.


74



Hybrid Neural Network-Genetic Algorithms

Approach for Fault Diagnosis of Bearing System

1L.A. Wulandhari ,

2A. Wibowo,

3M.I. Desa

Faculty of Computer Science and Information System, Universiti Teknologi Malaysia, 81310 UTM Skudai, Johor Bahru, Malaysia [email protected], [email protected],3 [email protected]

ABSTRACT

Fault diagnosis of critical systems such as bearing systems require some concern. Unexpected breakdown of one

component in such systems can induce failure of the whole system. With effective diagnosis, faults can be detected

much earlier and unacceptable consequences from total system failure can be avoided. In this paper, we present the

development of fault diagnosis techniques for bearing systems based on time series vibration data using hybrid of back

propagation neural networks (BPNN) and genetic algorithms (GAs) and is called BPNN-GAs. The GAs are used in

BPNN to increase the performance of condition classification in the bearing systems. Here, we consider a bearing sys-

tem that consists of two bearings called Drive End Bearing (DE) and Fan End Bearing (FE). Three accelerometers are

attached to the DE, FE, and Baseline (BA) through which the vibration data are captured. These vibration data need

to be analyzed to obtain information on conditions of the bearing systems. We extract ten features from the vibration

data as input and use sixteen classes as target output. The results between standard BPNN and BPNN-GAs are com-

pared and it clearly shows that BPNN-GAs give better classification accuracy in less CPU time and number of itera-

tions.

Keywords: Back-Propagation Neural Network, Genetic Algorithms, Fault Diagnosis, Bearing System

1. Introduction

Bearings are parts in machine that are used to support rotating shaft. Appropriate bearing design can minimize the

friction and its failure may cause expensive loss of production [1]. Unfortunately, bearing is one of machine parts

which has a high percentage of defect compared to the other component such as stator winding and rotor [2].

Therefore, an early and effective fault diagnosis of bearing is an essential task.

Actual fault diagnosis could be executed by examining and analyzing the vibration signal of t he bearing. The

vibration signal data contains frequency, time or time-frequency domain [3]. In this paper, we use time-frequency

domain as the data to diagnose the bearing fault. However, it is not easy to identify the condition of the bearing

system directly from the vibration signal especially if it involves more than one bearing in a system. Artificial I n-

telligence is one of the techniques that can provide an automated procedure for fault diagnosis [4]. Some previous

researchers used fuzzy neural network [5], radial basis function (RBF) network [4], and genetic-based neural net-

works (GNNs) [6] for this purpose.

This paper presents the hybrid technique that combines back-propagation neural networks (BPNN) and genetic

algorithms (GAs) in identifying the condition of the bearing system. GAs is applied in BPNN to obtain acceptable

weights for the BPNN training. In this paper, we improve bearing fault data representation from previous work by

combining and modifying available vibration signal data to obtain more specific condition diagnosis. Ten features

are extracted from these vibration signals that are used as the input of BPNN training. Those features are standard

deviation, skewness, kurtosis, the maximum peak value, absolute mean value, root mean square value, crest fac-

tor, shape factor, impulse factor and clearance factor [7]. These non-dimensional features are effective and practical

in fault diagnosis due to their relative sensitivity to early faults, and robustness to various loads and sp eeds [4].

These features are used as the input of BPNN training, whereas the target outputs are sixteen conditions of the

bearing system. In the result section we will show the comparison of performance between BPNN and hybrid

BPNN-GAs. Detail steps for fault diagnosis of the bearing system are presented in the next section.






Hybrid Neural-Network-Genetic Algorithms Approach for Fault Diagnosis in Bearing System

76

2. Bearing Data Structure

In this paper, vibrations signal data are captured from a bearing system which consists of Drive End bearing (DE) and

Fan End bearing (FE). Three accelerometers are attached on the bearings and baseline respectively as the tools to record

the vibration signals of the bearing. The structure of the bearings and accelerometer are shown in Figure 1.

Figure 1. Bearing and Accelerometer Structure

Bearing vibration data were collected under seven different conditions; (1) FE and DE normal, (2) FE normal and

DE Inner race fault (IRF), (3) FE normal and DE Ball fault (BF), (4) FE normal and DE Outer race fault (ORF), (5) FE

IRF and DE normal, (6) FE BF and DE normal and (7) FE ORF and DE normal. Normally, based on the available data

we have seven condition classes of bearing as the output of the diagnosis. However, we combine and modify the data to

improve the class of bearing conditions into sixteen classes. The sixteen classes of the bearing condition are presented

in Table 1.

Table 1. Sixteen conditions of bearing

ON noitidnoC ON noitidnoC ON noitidnoC ON noitidnoC

1C ED dna EF

lamroN 5C

ED dna FRI EF

lamroN 9 C

ED dna FRI EF

FRO 13 C FB ED dna FB EF

2C lamroN EF

FRI ED dna 6C

ED dna FRO

EF lamroN 10 C

ED dna FRI EF

FB 14 C

ED dna FRO EF

FRI

3C lamroN EF

FRO ED dna 7C

ED dna FB EF

lamroN 11 C

ED dna FB EF

FRI 15 C

ED dna FRO EF

FRO

4C lamroN EF

FB ED dna 8 C

ED dna FRI EF

FRI 12 C

ED dna FB EF

FRO 16 C

ED dna FRO EF

FB

The 320 samples of time series data are used in BPNN and BPNN-GAs. These samples are split into two sets: 240

samples for training and 80 samples for testing. BPNN training uses 30 neurons for input which is composed based on

ten features extraction from three accelerometers. The topology of BPNN and BPNN-GAs is explained in the next sec-

tion.

3. Hybrid BPNN-GAs

Standard BPNN is one of the supervised training algorithms that are widely used in defect diagnosis. However, BPNN

has conflict between overfitting and generalization which leads to a low learning training speed and the easiness of

converging to local optimum point of the network [8-9]. This problem can be tackled by applying GAs in standard

BPNN. GAs are global search methods which are based on principles like selection, crossover and mutation [10]. In this

paper, GAs are applied to find the acceptable weights and they are used in BPNN training. By using the acceptable

weight, minimum mean square error (MSE) can be obtained in less iteration. In the next subsection we briefly introduce

standard BPNN, GAs and BPNN-GAs.

FE Bearing DE Bearing

FE Accelerometer DE Accelerometer

BA Accelerometer



3.1. Back Propagation Neural Network (BPNN)

Here, we assume BPNN has a training input vector 𝒙 = 𝑥1 , 𝑥2,… , 𝑥𝑛 𝑇 and target 𝒚 = 𝑦1 , 𝑦2 ,… , 𝑦𝑚

𝑇 which

implies that input and output layer of the BPNN consist of 𝑛 and 𝑚 neurons, respectively. The input layer and the

output layer are related by several hidden layers and they are adjusted in advance. The 𝑖th input and the 𝑗th output

neuron are connected by a weight 𝑤𝑖𝑗 which satisfy the following function:

𝑂𝑗 = 𝑥𝑖𝑛𝑖=1 𝑤𝑖𝑗 , (1)

𝑓 𝑂𝑗 =1

1+exp 𝑂𝑗 . (2)

The error value between the output and target is calculated using Mean Square Error (MSE) formulation:

𝐸𝑖 =1

2 𝑦𝑗𝑖 − 𝑜𝑗𝑖

2𝑛𝑗=1 (3)

and BPNN updates the weights for obtaining the desired MSE value.

3.2. Genetic Algorithms (GAs)

GAs are adaptive search and optimization algorithms which ease in operation because involve genetic nature principles,

minimal requirements and global perspective. GAs are good in finding good acceptable solution, however, they do not

guarantee global optimum solution [11]. The GAs are performed by the following steps [12]:

1. Generate an initial population of 𝑝 chromosomes randomly.

2. Calculate the fitness value of the population using equation

.

1

ii

EF

(4)

3. Form a mating pool which contains the best genes that are selected using roulette selection method.

4. Select parents pair from mating pool

5. Combine respective pair of parents using crossover operator to obtain offsprings.

6. Create a new population of 𝑝 chromosomes by combining the selected parents and their offsprings

7. Evaluate the fitness value of new population. If the fitness values converge, stop, and return the best solution in

current population. Otherwise, go to step 3 for the new population

3.3. BPNN-GAs

The hybrid of BPNN-GAs is conducted in the following steps:

1. Assume BPNN numbers of weights are 𝑛 + 𝑚 𝑙 where n is the number of neurons input, m is the number of

neurons output, l is the number of neurons in hidden layer and each weight (gene) is a real number.

2. Generate an initial population of chromosome which consists of 1380 (m=30, n=16, l=30) genes from BPNN

random weight.

3. Calculate fitness value of p chromosomes using equation (4).

4. Generate the mating pool by selecting the best genes using roulette selection methods.

5. Select parent pairs from mating pool for crossover mechanism.

6. Create a new population by combining the selected parents and their offsprings.

7. Evaluate the fitness value of new population. If the fitness values converge, stop, and return the best solution in

current population. Otherwise, go to step 3 for the new population

8. Apply the best solution in current population as the initial weights for BPNN training.

The scheme of hybrid BPNN-GAs is shown in Figure 2.


78

Figure 2. The Scheme of Hybrid BPNN-GAs Algorithm

4. Result and Analysis

We conduct the experiment of standard BPNN and BPNN-GAs to obtain the fault diagnosis of the bearing system. In

BPNN-GAs approach, we used BPNN with the topology: 30 neurons of input, 30 neurons of one hidden layer and 16

neurons of output layer. In this experiment, we set GAs parameters as follows: 100 chromosomes of each population,

each chromosome contains1380 genes, crossover rate is 0.6 and maximum generation is 200.

For standard BPNN we use three topologies as follows: (1) 30 neurons of input, 30 neurons of the first hidden

layer and 16 neurons of output layers, (2) 30 neurons of input, 30 neurons of the first hidden layer, 30 neurons of the

second hidden layer and 16 neurons of output layer and (3) 30 neurons of input, 30 neurons of the first hidden layer, 30

neurons of the second hidden layer, 30 neurons of the third hidden layers and 16 neurons of output layer. We refer

m-l1-l2-l3-n as BPNN with m neurons input, l1 neurons of the first hidden layers, l2 neurons of the second hidden layers,

l3 neurons of the third hidden layers and n neurons output.

BPNN and BPNN-GAs are implemented in MATLAB to obtain the desired classification of bearing system condi-

tion and performed on a computer with Intel Core 2 Quad processor Q8200, 2.33 GHz and 1.96 GHz and RAM 3.46

GB. The accuracy of classification is calculated using the following equation:

%100output Total

class output true totalaccuracy tionclassifica (5)

The accuracy of classification is shown in a confusion matrix which represents true output and target class in diagonal

of the matrix. As the results, Figure 3 presents confusion matrix of the best results of standard BPNN and hybrid of

BPNN-GAs. It shows that both of approaches achieve 93.3% classification accuracy, however, standard BPNN needs

100000 iterations whereas BPNN-GA just needs 15000 iterations. The comparison between standard BPNN and

BPNN-GAs is given in Table 2:

Table 2. Comparison of performance between standard BPNN and hybrid BPNN- GAs Approach in fault diagnosis

dohteM noitaretI

gniniarT gnitseT

noitacifissalC

ycaruccA CPU Time

(sec) noitacifissalC

ycaruccA CPU Time

(sec)

BPNN 30-30-16 100000 85.8% 6289.6 60.3% 0.053 BPNN

30-30-30-16 100000 93.3% 7261.8 71.6% 0.047

BPNN

30-30-30-30-16 61542 88.8% 3991.8 72.2% 0.066

BPNN-GA

2000 85.8% 657.9 80.0% 0.034

5000 89.6% 645.5 87.5% 0.034

15000 93.3% 765.2 86.3% 0.034

NO YES

Generate initial population of

chromosome (Random BPNN

weights)

Calculate fitness value

Selection

Mating pool Crossover

Do fitness values of current popula-

tion converge?

New population

(new weights)

BPNN training



A B

Figure 3. Confusion Matrix of BPNN-GAs (A) and Standard BPNN (B)

5. Conclusion

This paper presented hybrid BPNN-GAs approach to diagnose condition of bearing system. The result shows that

BPNN-GAs gives better result than standard BPNN in the bearing system diagnosis. The hybrid of BPNN-GAs

achieves higher classification accuracy in less iteration and shorter CPU time compared to the standard BPNN.

6. Acknowledgements

The authors thank to Universiti Teknologi Malaysia (UTM) and Ministry of High Education (MOHE) for Research

University Grant (RUG) Vote No. Q.J. 130000.7128.00J96 and the Research Management Center (RMC) - UTM for

supporting this research project. The first author sincerely thank to UTM for awarding International Doctoral Fellow-

ship (IDF).

7. References

[1] A. Harnoy, Bearing Design in Machinery : Engineering Tribology and Lubrication. 2003: Marcel Dekker, Inc.

[2] P.V. J. Rodriguez and A. Arkkio Detection of Stator Winding Fault in Induction Motor Using Fuzzy Logic. Applied Soft

Computing 8. 2008: p. 1112-1120.

[3] N. Saravanan, V. N. S. K. Siddabattuni, and K. I. Ramachandran, Fault Diagnosis of Spur Bevel Gear Box Using Artificial

Neural Network (ANN), and Proximal Support Vector Machine (PSVM). Applied Soft Computing 10. 2010: p. 344-360.

[4] Y. Lei, Z. He, and Y. Zi, Application of An Intelligent Classification Method to Mechanical Fault Diagnosis. Expert Systems

with Applications 36. 2009: p. 9941-9948.

[5] H. Wang and P. Chen, Fault Diagnosis for A Rolling Bearing Used in A Reciprocating Machine by Adaptive Filtering

Technique and Fuzzy Neural Network. WSEAS TRANSACTON on SYSTEMS Issue 1, Vol. 7. 2008: p. 1-6.

[6] Y. C. Huang, C. M. Huang, H. C. Sun, and L. S. Liao, Fault Diagnosis Using Hybrid Artificial Intelligent Methods. 5th IEEE

Conference on Industrial Electronics and Applicationsis. 2010: p. 41-44.

[7] W. Li, T. Shi, G. Liao, and S. Yang, Feature Extraction and Classification of Gear Faults Usig Principal Component Analysis.

Journal of Quality in Maintenance Engineering Vol. 9 No.2. 2003: p. 132-143.

[8] J. Tetteh, E. Metcalfe, and S. L. Howells, Optimisation of radial basis and backpropagation neural networks for modelling

auto-ignition temperature by quantitative-structutre property relationship. Chemometrics and intelligent laboratory systems 32.

1996: p. 177-191.


80

[9] J. Rafiee, F. Arvani, A. Harifi, and M. H. Sadeghi, Intelligent condition monitoring of a gearbox using artificial neural network.

Mechanical systems and signal processing 21. 2007: p. 1746-1754.

[10] J. L. Tang, Q. R. Cai, and Y. J. Liu, Gear Fault Diagnosis with Neural Network based on Niche Genetic Algorithm

International Conference on Machine Vision and Human-machine Interface. 2010: p. 596-599.

[11] S. Rajasekaran and G. A.V. Pai, Neural networks, fuzzy logic and genetic algorithms: synthesis and applications. 2007: New

Delhi, II : Prentice-Hall of India.

[12] Y.J. Cao and Q. H. Wu, Teaching Genetic Algorithm Using MATLAB. Int. J. Elect. Enging. Educ., Vol.36. 1999: p. 139-153.

Proceeding of Industrial Engineering and Service Science, 2011, September, 20-21


The Effects of Trade Location: The Case of Dual

Listing Telkom in NYSE and IDX

SitiArfiah Arifin*, Deddy P. Koesrindartoto**

School of Business and Management, Institut Teknologi Bandung, Indonesia

*[email protected], **[email protected]

ABSTRACT

Although some stocks, in some sectors, affected in the short term by irrational behavior, the stock market as a whole fol-

lows fundamental laws, grounded in economic growth and returns on investment. This is following the classical finance

paradigm predicts that an asset‘s price is unaffected by its location of trade and other factors. In the second half of the

1990s, the S&P 500 Index more than tripled in value to an all-time high of almost 1,500. The stocks, such as Amazon and

AOL, became stock market superstars. Then the market crashed, and many stars flickered out. Moreover, people began to

question whether classical finance theories could really explain such dramatic swings in share prices. This is the basic

point of view of this essay that relative price of the stock is correlated with the relative stock market indexes of the coun-

tries where the stock are traded most actively. The essay hypothesizes the single stock of Telkom that is most intensively

traded on a given market will co-move excessively with that market‘s return and currency. The measurement for the

co-movement of stock price through the regression of stock log return differential on Indonesia and U.S. market index log

returns plus the relevant log currency changes. Finally, this essay will provide an example in which location of trade and

ownership appears to influence prices. Then, a similar sort of phenomenon occurs with closed-end country funds.

Key words: Telkom, Trade Location, Market Index, Currency Change

1. Introduction

Following the classical finance paradigm predicts that an asset‟s price is unaffected by its location of trade. This condition

happened when international financial markets are perfectly integrated, then a given set of risky cash flows will have the

same value and risk characteristics when its trade is being redistributed across markets and investors [2].

Then, sometimes the market crashed, and many stars of stock flickered out, people began to question the situation

that classical finance theories could really explain such dramatic moves in stock prices [1]. Some would even assert that

stock markets lead lives of their own, detached from the basics of economic growth and business profitability. Although

some stocks, in some sectors, affected in the short term by irrational behavior, the stock market as a whole follows fun-

damental laws, grounded in economic growth and returns on investment [3].

This essay provides an example in which location of trade and ownership appears to influence prices. It shows that

the stock price of the biggest state telecommunication company in Indonesia, Telkom, is influenced by location factor.

This condition seem happened in general because it is also shows to the stock prices of three world‟s largest and most

liquid multinational companies [2]. Furthermore, the main contribution of the essay is to show that therelative price of the

stocks is correlated with the relative stock market indexes of the countries when the stocks are traded most actively. Spe-

cifically, it tests whether location matters by examining single company stock whose charter fixes the division of past and

current equity cash flow of price stock.

The stock of Telkom provides a clearly example of co-movement for several reasons. First, the single stock that

examined in the paper is one of the pioneer that being tradedabroad beside in Indonesia. Second, the Telkom stock are

majority owned by the government of Indonesia thatcan be influences the national market index directly. Third, the stocks

are traded on world stock exchanges, and many investors can purchase the stock locally.

The objective of this essay is to show that the relative price of the stock is correlated with the relative stock market

indexes of the countries where the stock are traded most actively. For example, when the Indonesian market moves rela-

tive to the U.S. market, the price of Telkom (which trades relatively more in Jakarta) tends to move relative to the price in

New York. Similarly, when the rupiah appreciates against the dollar, the price of Telkom tends to move relative too. A

similar sort of phenomenon occurs with closed-end country funds. In particular, it appears that closed-end fund share

prices co-move most strongly with the stock market on which they trade. Then, while net asset values co-move most

strongly with their local stock markets [2].

mailto:*[email protected]

mailto:**[email protected]

The Effects of Trade Location: The Case of Dual Listing Telkom in NYSE and JKSE

82

Finally, the rest of the chapter is organized as follows. Section 2 describes the company profile structure of Telkom

and its stock. Section 3 presents the empirical hypotheses and tests of the single stock price. Section 4 discusses the origin

of data. Section 5 shows the findings and results on co-movement because of the trade location. Then, section 6 offers

conclusions of the paper.

2. Company Profile

PT Telekomunikasi Indonesia, tbk or Telkom as operator of TIME (Telecommunication, Information, Multimedia, and

Edu-taintment) business has performed telecommunication operation in the form of telephone (fixed wire line, fixed

wireless, and cellular), data and internet, network service and interconnection, and content/application. As of December 31,

2009, number of subscribers has grown at 21.2% or 105.1 million compared to the previous year. For telephone only,

TELKOM serves 8.4 million of fixed wire line subscribers, 15.1 million of fixed wireless subscribers, and 81.6 million of

cellular phone subscribers [8].

As of December 31, 2009, the Indonesian Government (52.4%) owns TELKOM common share and public share-

holders (47.53%) own the rest. TELKOM share is traded at Indonesia Stock Exchange (“IDX”), New York Stock Exchange

(“NYSE”), then at London Stock Exchange (“LSE”) and Tokyo Stock Exchange (“TSE”), both in form of Publicly Offering

without Listing (POWL) [8].

Figure 1. Sequence Plot of Log Return Telkom are traded on the NYSE and IDX

3. Empirical Hypotheses and Tests

This essay hypothesizes that stocks are most intensively traded on a given market will co-move excessively with that

market‟s return and currency. The null hypothesis is that the relative stock price should be uncorrelated with everything.

Then, alternative hypothesis is that markets are segmented, so that relative markets shocks explain movements in the

price differential.

Regression of stock log return differential on Indonesia and U.S. market index log returns plus the relevant log cur-

rency changes is the measurement for the co-movement of stock price:

1 1 1 1

ΔrTelkom (NYSE-IDX), t = α + ∑ βi NYSE t+i+ ∑ δj S&P 500 t+j + ∑ λkIDXt+k + ∑ γl g$/Rpt+l + εt, i=-1 j=-1 k=-1 l=-1 (1)

Because of the cross-border aspects of the markets, it is include currency changes as well as local-currency stock

returns as market factor in Eq. Therefore, at the null hypothesis isthe entire slope coefficients is zero. Otherwise, on the

alternative hypothesis, the more a stock trades on a given market, the higher its estimated slope.

The Effects of Trade Location: The Case of Dual Listing Telkom in NYSE and IDX


4. Data

Indonesian stock prices for Telkom (Tlkm) are taken from the Jakarta Stock Exchange (IDX) [5], while the U.S. stock

prices for Telkom (Tlk) are taken from the New York Stock Exchange (NYSE) [6]. Moreover, index from S&P 500 is

used for US market return [7], and the currency change of US Dollar and Rupiah [4], respectively. The sample monthly

period is January 1, 2005 to December 1, 2010. All returns are expressed in log form.

Other important consideration is where returns are measured; the essay also estimates the relative return on the stock

price by taking the difference of the log returns in the markets where they trade most actively. For example, it is uses the

Δ returns of Telkom in New York and Jakarta. Finally, issue concerns the currency denomination of returns. The eq.

mentions all return variables in local currencies and then add exchange rate changes as separate independent variables on

the right-hand side of the regressions.

5. Findings and Results

The result from the first regression estimates of Eq. (1) for Telkom stock, respectively. First, the regression uses return

log return differential on Indonesia and U.S. market index log returns plus the relevant log currency changes at time t.

The result is that the R2: 0.440 shows the correlation between Δ log return TlkTlkm can be explained about 44% with the

others independent variables. Then Standard Error Estimate (SEE) is 0.013, determine that less amount of the number

make the regression model is worth to predict Δ log return TlkTlkm. Moreover, from the F test, the amount is 12.980 with

significant 0.000, it predicts that the regression model can be used to determine Δ log return TlkTlkm or all of independ-

ent variables together can predict the Δ log return TlkTlkm. In significance column (Sig.), variables log return NYSE, log

return S&P 500 and log return IDX have amount above 0.05, so it is not affect the Δ log return TlkTlkm, respectively.

Moreover, only variable log return USD/IDR affect the Δ log return TlkTlkm because the Sig. amount is 0.000.

Table 1. Result of the regression at period time t-1, t, and t+1

Specification Return

Period R2 SEE F Sig. Sig.

NYSE Sig.

S&P 500 Sig. IDX Sig.

g $/Rp 2005-2010 t-1 0.438 0.014 13.055 .000a .204 .703 .193 .000 2005-2010 t 0.440 0.013 12.980 .000a .177 .694 .215 .000

2005-2010 t+1 0.580 0.013 6.456 .000a .489 .403 .588 .467

Because the result in Table 1 is not too significant respectively, the other variable is added to the Eq. (1), in order to

get better result, and then the regression process replied again. The second multiple regression test shows the result that

the R2 amount is 0.580, then the correlation between Δ log return TlkTlkm and other variables is strong. The F test

(6.456) and significance (.000) shows that the independent variables exposure together is significant. Moreover, variables

log return USD/IDR at time t (sig .000) and log return USD/IDR at time t-1 (.000) affect Δ log return TlkTlkm signifi-

cantly.

The result shows that log return USD/IDR at time t and log return USD/IDR at time t-1 are the most affected vari-

ables to Δ log return TlkTlkm than others. The new estimation of Δ log return differential TlkTlkm and other variables:

r Telkom(NYSE-IDX),t = α + βrNYSE,t + δr S&P 500,t + λrIDX,t + γg$/Rp,t-1 + γg$/Rp,t

+ ε Telkom(NYSE-IDX),t (2)

Table 2. Result of the next regression at period time t-1 and t

Specification Return

Period R2 DW F Sig. B (β)

NYSE B (δ)

S&P 500 B (λ) IDX

B (γ) g $/Rp

2005-2010 t-1 0.438 2.80 13.055 .000a .140 .001 -.096 .623 2005-2010 t 0.440 2.78 12.980 .000a .152 .001 -.092 .614

The result in table 1 and table 2, reject the perfect-integration hypotheses that explained in classical finance para-

digm. The signs of virtually all coefficients line up with the alternative hypothesis, and all are significantly different from

zero at the 1 percent level. In table 2, for example, at period t of Δ log return TlkTlkm differentials yields coefficients of

about of about 0.152 on the NYSE, 0.001 on the S&P500, -0.92 on the Indonesian index (IDX). The coefficients on the

exchange rate changes are also large, at 0.623 at return period t-1 and 0.614 for the dollar/rupiah exchange rates. At 1

percent appreciation of the dollar against the rupiah, influences the relative price of Telkom stock by about 60 basis

points. Then, this coefficient values also describe that at 1 percent appreciation of the dollar relative to the rupiah influ-

The Effects of Trade Location: The Case of Dual Listing Telkom in NYSE and JKSE

84

ence the relative price of Telkom stock by about 60 basis points. The R2 in the table 2 are quite high, about 40 percent for

return of period t-1 and period t.

6. Conclusions

The essay presents evidence that stock price of Telkom is affected by the location of trade, especially with the change of

currency where the place of stock is traded, NYSE in United State of America and IDX in Republic of Indonesia. There-

fore, the location of trade therefore appears to matter of pricing. The co-movement between the return price differentials

and market indexes are present too. The similar result is happened to the twin stocks that have nearly identical cash flows,

move more similar to the markets where they trade most intensively than they should. The co-movements between the

price differentials and market indexes are present at short and long horizons [2].

Moreover, this essay presents the possible sources about the cause of the result. The first source of the change ex-

planation is noise, irrational traders made market-wide noise shocks, which affect locally traded stocks more than foreign

traded stocks, but the main problem with this is that thesource of noise or persistent irrationality is difficult to identify.

The second possible source might institutional inefficiencies, because the stock might beclassified as a “domestic” stock,

so the classification is needed in practice also could help resolve informational asymmetries and agency problems in the

investment process. Third, the source of the change is tax-induced investor heterogeneity, but still incomplete for explain

more about investor behavior. Finally, the future research for other local stocks from Republic of Indonesia, which trade

in the foreign area is important to be determined. This is important to know whether there are anomalies from these stocks

trade, especially when the stocks are correlated with the relative stock market indexes of the countries where the stock are

traded most actively, and also with the currency change between the countries.

7. References

[1] Chopra, N. et al, 1993. “Yes, discounts on closed-end funds are a sentiment index”, Journal of Finance, Vol.48, pp 801-8.

[2] Froot, K.A., E. Dabora, 199., “How are stock prices affected by the location of trade”. Working Paper no. 6572. National Bureau

of Economic Research, Cambridge, MA.

[3] Hardouvelis, G., R. La Porta, T. Wizman, 1995. “What moves the discount on country equity funds”. Working Paper no. 4571.

National Bureau of Economic Research, Cambridge, MA.

[4] Historical Exchange Rates, 2011.http://www.oanda.com/currency/historical-rates/

[5] Index Jakarta, 2011. http://finance.yahoo.com/q?s=^jkse&ql=1

[6] Index New York, 2011. http://finance.yahoo.com/q?s=^NYA&ql=0

[7] Index S&P 500, 2011. http://finance.yahoo.com/q?s=^GSPC&ql=0

[8] Info Perusahaan, 2011.http://www.telkom.co.id/info-perusahaan/

http://www.oanda.com/currency/historical-rates/

http://finance.yahoo.com/q?s=%5ejkse&ql=1

http://finance.yahoo.com/q?s=%5eNYA&ql=0

http://finance.yahoo.com/q?s=%5eGSPC&ql=0

http://www.telkom.co.id/info-perusahaan/



A Scheduling Model for Production System

Considering Material Handling Operations

Dwi Kurniawan*, Rispianda, Isa Setiasyah Toha

Industrial Engineering Department, Institut Teknologi Nasional, Bandung, Indonesia

*E-mail: [email protected]

ABSTRACT

In most techniques, production scheduling only consider production machines as resources. These techniques generally

do not consider the material handling operations by assuming that the material handling equipments are always avail-

able and the handling time can be ignored. In systems with significant material handling time, ignoring the material

handling operations may cause the material handling equipment allocation that does not fit the need for part transpor-

tation. In this study, a scheduling model considering material handling equipments as resource will be developed. The

developed model will start from priority dispatching technique, by adding necessary steps to consider the material han-

dling equipments.

Keywords: Scheduling, Material Handling, Priority Dispatching Technique

1. Background

Scheduling is the allocation of resources to perform a set of tasks based on time. To determine this allocation, various

techniques have been developed using optimization and heuristic approaches. In the techniques currently available,

scheduling only consider production machinery as a resource [1]. These techniques do not consider the material han-

dling process by assuming that the process can be done using material handling equipments available and the processing

time can be ignored.

In certain production systems, such as production system in Balai Yasa Jembatan Kereta Api (railway bridge

workshop), Bandung, the time required to perform the material handling process has a significant proportion to the total

processing time. Thus, the material handling time in this type production systems can not be ignored. The common rea-

son of this condition is the size or mass of material, so that most or all of the material handling process must use special

equipment, and the material handling time can not be ignored.

In the production systems with significant proportion of material handling time, the material handling equipment

should be regarded as a source as production machinery, and should be considered in scheduling. Without considering

the scheduling of material handling equipment, it is possible that at a time, work-in-process parts are waiting to be

transported from one station to another station because of all material handling equipments are being used to transport

components, while at the other time, all material handling equipments are idle. Therefore, in this type production sys-

tem, scheduling needs to consider the material handling time and equipments.

This research aims to:

1. Create a flow shop production scheduling model for systems with significant proportion of material handling time.

2. Apply the model to solve the problems occurred in Balai Yasa Jembatan Kereta Api, Bandung.

2. Literature Review

There are some recent developments in scheduling considering material handling operations. Lei and Wang [2] consid-

ered the problem of cyclic scheduling of two hoists. Bilge and Ulusoy [3] exploited the interactions between the ma-

chine scheduling and the scheduling of the material handling system in an FMS by addressing them simultaneously.

Das and Spasovic [4] presented a straddle scheduling procedure that can be used by a terminal scheduler to control the

movement of straddle carriers. Khayat et al. [5] proposed an integrated formulation of the combined production and

material handling scheduling problems. Babiceanu et al. [6] presented a solution for scheduling material handling de-

vices in the cellular manufacturing environment using the holonic control approach. Finally, Anwar and Nagi [7] con-

sidered the simultaneous scheduling of material handling transporters (such as automatic guided vehicles or AGVs) and

manufacturing equipment (such as machines and work centers) in the production of complex assembled product.


A Scheduling Model for Production System Considering Material Handling Operations

86

This paper will develop a scheduling model considering material handling operations. It will be developed from

Baker‟s job shop scheduling [1]. The material handling consideration will refer to the system designed by Apple [8].

3. Model Development

3.1 Problem Modelling

Problem of scheduling production machines and material handling equipments will be developed gradually. The

problem is developed from simple to complex in several stages:

1. Scheduling one production machine and one material handling equipment.

2. Scheduling m production machines and one material handling equipment.

3. Scheduling m production machines and h independent material handling equipments.

4. Scheduling m production machines and h dependent material handling equipments.

A. Scheduling one production machine and one material handling equipment

Examples of scheduling one production machine and one material handling equipment can be seen in Figure 1. All

products are processed by one machine and supported by one material handling equipment, but each has different num-

ber of repetitions and operation time. Problems like this have a general form as shown in Figure 2. The routing of this

problem is shown in Table 1. Handling Boring Handling Boring Handling Boring Handling Boring Handling

Figure 1: An example of scheduling one material handling equipment and one machine

Figure 2: An example of scheduling one machine and one material handling

Table 1. Routing of problem in Figure 2

Job

(i)

Operation (j)

1 2 3 2p 2p+1

1 H M H M H

⁞ H M H ⁞ M H

n H M H M H

M: machine; H: material handling

B. Scheduling m production machines and one material handling equipment

Examples of scheduling m production machines and one material handling equipment can be viewed in Figure 3. All

jobs are processed by some machines and transported by one material handling equipment, with different sequence and

operation time. The problem has a common model as shown in Figure 4. The routing of this problem is shown in Table

2. Handling Boring Handling Lathe Handling Sewing Handling Painting Handling

Figure 3: Examples of scheduling m machines and one material handling

Product with

4 holes

material

handling

n job

p repetition

machine



Figure 4: General form of scheduling m machines and one material handling equipment


Job

(i)

Operation (j)

1 2 3 2k 2k+1

1 H Mij H Mij H

⁞ H Mij H ⁞ Mij H

n H Mij H Mij H

Mij: machine used in job i operation j; H: material handling

C. Scheduling m production machines and h independent material handling equipments

Examples of scheduling m production machines and h independent material handling equipments can be viewed in

Figure 5. All jobs are processed by some machines and transported by some material handling equipments, with

different sequence and operation time. The “independent” term means that the material handling equipments transport

the jobs without any dependence or collaborative action with other. The problem has a common model as shown in

Figure 6. The routing of this problem is shown in Table 3. Forklift Boring Crane Lathe Crane Sewing Crane Painting Forklift

Figure 5: Examples of scheduling m machines and h independent material handling

Figure 6: General form of scheduling m machines and one material handling equipment

material handling

n job

k operations

Machine 1

⁞

Machine m

Material handling 1

n job

⁞

k operations

Material handling h

Machine 1

⁞

Machine m


88


Job

(i)

Operation (j)

1 2 3 2k 2k+1

1 Hij Mij Hij Mij Hij

⁞ Hij Mij Hij ⁞ Mij Hij

n Hij Mij Hij Mij Hij

Mij: machine used in job i operation j; H: material handling used in job i operation j

D. Scheduling m production machines and h dependent material handling equipments

In the previous section, material handling equipments are assumed to work independently, do not affect each other. In

the real system, it is possible that several material handling equipments have a mutual dependency (dependent).

Interdependence between some material handling equipments can occur in some conditions:

1. Selection of material handling equipment.

Generally, machining and material handling operation have designated machines or material handling equipment

required. However, in certain material handling operations, materials handling equipment

which will be used is not determined before (some equipments are possible). If a material handling operation can

be handled by more than one material handling equipment, the equipment used to handle this operation is the one

that can perform operations earlier. The earliest completion of the operation depends on:

• Ready time of each material handling equipment to be used.

• Arrival time of each materials handling equipment in the displacement starting location.

2. Simultaneous use of resources.

Generally, each operation, either machining or material handling operation, uses only one resource. However, it is

possible that certain operations require more than one resource simultaneously. The examples are:

• Use of some material handling equipments simultaneously. To lift a locomotive body weighing over 70 tons

from its wheels, two overhead cranes of 36 tons haul powered are used simultaneously.

• Use of materials handling equipment and production machines simultaneously. To drill a bar of bridge

component, a drill machine is used assisted by an overhead crane to hold the bar.

3. Shared path.

Movement of material handling equipments may use a path simultaneously (or alternately). In double girder over-

head crane, one crane path (axis movement) is used by the two cranes. Therefore, when a crane will move from one

location to another, another crane should not be in a location that will be passed through. In transport vehicles such

as forklifts and tractors, track dependencies can also occur on paths that are used together by several equipments. If

an equipment will be used, but the required path is being used by other equipment, then it should wait until the path

they need is available.

The general form and the routing of scheduling m production machines and h dependent material handling

equipments are the same with the independent one, as can be seen in Figure 6 and Table 3.

3.2 Scheduling Algorithm

An algorithm is developed to solve the scheduling problem described in Section 3.1. The algorithm refers to the prob-

lem of scheduling m machine scheduling production and h dependent material handling equipments, due to this problem

is the most complex problem of the four problem types. This means that the algorithm is applicable for the three other

simpler problems.

The scheduling process is based on notations and steps in the Priority Dispatching Technique [1]. The notations

used in the model develompent are:

PSt = partial schedule consists of t scheduled operations;

St = operations ready to be scheduled at stage t;

rj = earliest time at which operation j St can be started;

cj = earliest time at which operation j St can be completed;

Dij = arrival time of job i in the j-th material handling operation;

t‘ij = total time of job i in the j-th material handling operation (including the arrival time).

The algorithm is described as follows.

Step 1. Suppose t = 0, PSi = 0 and S0 = set of operations without predecessors.

Step 2. If there are material handling operations in St:

a. If there are more than one material handling equipment that can be used, select a material handling equipment that

capable to complete the operation earlier.



b. Determine rj, the time when all material handling equipments can be used, regarding the use of paths by the

equipments.

Step 3. For material handling operations at St, determine the arrival time Dij based on the last location of the required

material handling equipment. If some resources are used, the arrival time is the latest arrival time of all resources. Then,

specify t‟ij by summing Dij with material handling time specified in the routing.

Step 4. Considering rj, determine c* = minjSt {cj} and the resource r* where c* would be done.

Step 5. For each operation j St requires the resource r* and has rj < c*, select a priority using the following stages:

a. Prioritize operations using less resource.

b. Select an operation using certain priority rule.

Add selected operation to PSt so the partial schedule PSt+1 is obtained.

Step 6. For the partial schedule PSt+1 obtained from Step 5, update the following data.

a. Remove the scheduled operation j from St.

b. Create St+1 by adding the operation succeeding the scheduled operation j.

c. Add t by one.

Step 7. Return to Step 2 to review PSt+1 and continue all the steps until all jobs are scheduled (St = {}).

4. Model Application

A problem is shown in the following. Table 4 and 5 shows the original routing and processing time of four jobs without

the material handling operations.

Table 4. Original routing

Table 5. Original processing time

Job Operation Job Operation

1 2 3 1 2 3

1 1 2 3 1 4 3 2

2 2 1 3 2 1 4 4

3 3 2 1 3 3 2 3

4 2 1 3 4 3 3 1

Materials handling operations in this system is performed based on the following description.

The material handling equipments consist of 2 forklifts (number 41 and 42) and 2 overhead cranes (number 51 and

52).

Transportation between two machines will use crane, and transportation between storage and machines will use

forklift.

Operations on Machine 3 need assistance of one crane, and material handling operations of Job 4 requires two ma-

terial handling equipments (forklift or crane) simultaneously.

Aisles in the plant are wide enough to pass by 2 forklifts as well. Meanwhile, two cranes are located on one line so

that the position of each crane in the scheduling must be considered.

After including material handling operations, the job routing is updated as shown in Table 6. Further, Table 7

shows the transportation time between machines and storages in the plant.

Table 6. New routing

Table 7. Transportation time

Job Operation From To

1 2 3 4 5 6 7 R. mat.

storage

Machine

1

Machine

2

Machine

3

End pr.

storage

1 4 1 5 2 5 3+5 4 Raw material storage

- 3 3 4 2

2 4 2 5 1 5 3+5 4

3 4 3+5 5 2 5 1 4 Machine 1 3 - 1 3 4

4 412 2 512 1 512 3+512 412 Machine 2 3 1 - 2 3

Notes: Shaded cells are material handling operations

4 or 5: one material handling is required

412 or 512: two forklifts or two cranes are required 3+5: machine 3 is required assisted by a crane

3+512: machine 3 is required assisted by two cranes

Machine 3 4 3 2 - 2

End product storage

2 4 3 2 -

Forklift park 2 3 4 5 2

Crane 1 park - 1 1 2 -

Crane 2 park - 2 1 1 -

Considering Table 6 and Table 7, and the machining time in Table 5, an updated processing time for both machin-

ing and material handling operations is then summarized in Table 8.


90

Table 8. Updated processing time

Job Operation

1 2 3 4 5 6 7

1 9 4 5 3 9 2 5

2 7 1 6 4 13 4 6

3 8 3 6 2 9 3 8

4 7 3 5 3 13 1 7

Note: Shaded cells are material handling operations

Applying the algorithm developed in Section 3, a schedule of machinery and material handling equipments is

shown in Figure 7.

Figure 7: Final schedule for the given problem

The Gantt-chart in Figure 7 is important to analyze to understand how the model works. For example, see Job 2

with the operation sequence 214, 222, 235, 241, 255, 26(35) and 274. Remember that the odd sequence operations are

material handling ones. So, the operation sequence of Job 2 switches alternately between machining operations (lower

region) and material handling operations (upper region).

Further more, two-resource operation can be seen in the notation 26(35) in the Gantt-chart. The operation 26(35)

appears in Machine 3 and crane 2, due to the operation requires both resources. The same appearance occurs in opera-

tion 32(35) and 16(35). Other two-resource operations with different notation are 41412, 43512 and 45512. Finally, a

three-resource operation occurs in 46(3512) as can be seen in the Gantt-chart.

5. Concluding Remarks

The model developed in this paper works properly. The model is applicable for production systems with significant

proportion of material handling operations. The model works by combining both machining operations and material

handling operations in one routing table and one processing time table. The model can adopt special conditions of mate-

rial handling operations, such as multi-resource operation, consideration of material handling equipment locations, the

usage of machinery and material handling equipment simultaneously, and consideration of the route of material han-

dling equipment movements. The model can also be extended to production systems using material handling equipment

such as Automatic Guided Vehicle (AGV) and Robotic Guided Vehicle (RGV).

6. References

[1] K. R. Baker, “Introduction to Sequencing and Scheduling,” John Wiley & Sons Ltd., 1974.

[2] L. Lei and T. Wang, “The Minimum Common-Cycle Algorithm for Cyclic Scheduling of Two Material Handling Hoists with

Time Window Constraints,” Management Science, Vol. 37, Issue 12, 1991, pp. 1629-1639.

[3] U. Bilge and G. Ulusoy, “A Time Window Approach to Simultaneous Scheduling of Machines and Material Handling System

in an FMS, Operations Research, Vol. 43, Issue 6, 1995, pp. 1058-1070

[4] S. K. Das and L. Spasovic, “Scheduling Material Handling Vehicles in A Container Terminal,” Production Planning & Con-

trol: The Management of Operations, Vol. 14, Issue 7, 2003, pp. 623 – 633.

[5] G. E. Khayat, A. Langevin and D. Riope, “Integrated Production and Material Handling Scheduling Using Mathematical Pro-

gramming and Constraint Programming,” European Journal of Operational Research, Vol. 175, Issue 3, 2006, pp. 1818-1832.

[6] R. F. Babiceanu, F. F. Chen and R. H. Sturges, “Real-Time Holonic Scheduling of Material Handling Operations in A Dynamic

Manufacturing Environment,” Robotics and Computer-Integrated Manufacturing, Vol. 21, Issues 4-5, 2005, pp. 328-337.

[7] M. F. Anwar and R. Nagi, “Integrated Scheduling of Material Handling and Manufacturing Activities For Just-In-Time Pro-

duction of Complex Assemblies,” International Journal of Production Research, Vol. 36, Issue 3, 1998, pp. 653- 681.

[8] J. M. Apple, “Material Handling Systems Design,” The Ronald Press Company, New York, 1972.

http://www.informaworld.com/smpp/title~db=all~content=t713737146~tab=issueslist~branches=14#v14



http://www.sciencedirect.com/science/journal/03772217

http://www.sciencedirect.com/science?_ob=PublicationURL&_tockey=%23TOC%235963%232006%23998249996%23634422%23FLA%23&_cdi=5963&_pubType=J&view=c&_auth=y&_acct=C000050221&_version=1&_urlVersion=0&_userid=10&md5=feebd97e5af876cfff4b2f94ca633cd0

http://www.sciencedirect.com/science/journal/07365845

http://www.sciencedirect.com/science?_ob=PublicationURL&_tockey=%23TOC%235764%232005%23999789995%23599344%23FLA%23&_cdi=5764&_pubType=J&view=c&_auth=y&_acct=C000050221&_version=1&_urlVersion=0&_userid=10&md5=2a654d4381153aab5e7595e329799243






An Improved Fuzzy Number Ranking Method

Based on the Centroid-index

Shuo-Yan Chou, Vincent F. Yu*, Luu Quoc Dat

Department of Industrial Management, National Taiwan University of Science and Technology,43, Section 4, Keelung Road, Taipei

10607, Taiwan. E-mail: [email protected] (V.F. Yu); Tel: +886-2-2737-6333; Fax: +886-2-2737-6344

ABSTRACT

Ranking fuzzy numbersplays a very important role in decision process, data analysis and applications. Ranking

indieces based on the centroids of fuzzy numbers are commonly used approaches for raking fuzzy numbers. However,

there are some weaknesses associated with these indieces. This paper reviews several fuzzy number ranking methods

based on centroid-indiecesand proposes a new centroid-index ranking method which is capable of ranking various

types of fuzzy numbers effectively. Acomparative example is presented to demonstrate the usage and advantages of the

proposed centroid-index ranking method for fuzzy numbers.

Keywords: Fuzzy Numbers, centroid index, centroid of fuzzy numbers, ranking.

1. Introduction

Ranking fuzzy numbers plays a very important role in decision making, optimization, and other usages. Ever since

Yager [1], who presented the centroid concept in the ranking method, numerous ranking techniques using the centroid

concept have been proposed and investigated [2-19]. Some of them have been compared and contrasted in Wang and

Lee [16] and more recently in Ramli and Mohamad [11].

Cheng [6] in 1998 used a centroid-based distance method to rank fuzzy numbers. For a trapezoidal fuzzy number

( , , , ; ),A a b c d the distance index can be defined as 2 2

( ) AA

R A x y , with ,

b c dL R

A Aa b c

A b c dL R

A Aa b c

xf dx xdx xf dxx

f dx dx f dx

1 1

0 0

1 1

0 0

.

L R

A A

AL R

A A

yg dy yf dyy

g dy g dy

, and

R

Af and L

Af are the respective right and left membership functions of A , and R

Ag and

L

Ag are the inverse of R

Af and L

Af , respectively. The larger the value is of ( )R A , the better the ranking will be of A .

Cheng [6] further proposed a coefficient of variation (CV) index that improves the concept of ranking fuzzy numbers,

using fuzzy mean and fuzzy spread as presented by Lee and Li [8].

Chu and Tsao [7] in 2002 found that the distance method and CV index proposed by Cheng [6] still have some

shortcomings. Hence, to overcome these problems, Chu and Tsao [7] proposed a new ranking index function

,AA

S x y where Ax is similar to Ax in Cheng [6] and 0 0

0 0

.

w wL R

A A

w wAL R

A A

yg dy yg dyy

g dy g dy

The larger the value is of

( )S A , the better the ranking will be of A .

In some special cases, the method proposed by Chu and Tsao also hasthe same shortcomings as that in Cheng‟s

method [6]. The shortcomings of Cheng‟s and Chu and Tsao‟s centroid-index are presented as follows. For example, for

fuzzy numbers , ,A B C and , , ,A B C according to Cheng‟s centroid-index 2 2

R x y , whereby the same

results are obtained - that is, if A B C , then .A B C This is clearly inconsistent with the mathematical

logic. For Chu and Tsao‟s centroid-index S xy , if 0x , then the value of S xy is a constant zero. In other

words, the fuzzy numbers with centroids 1(0, )y and 1 1 2(0, ), ( )y y y are considered the same. This is also obviously

unreasonable.

An Improved Fuzzy Number Ranking Method Based on the Centroid-Index

92

Wang, Yang, Xu, and Chin [18] found that the centroid formulae proposed by Cheng [6] and Chu and Tsao [7] are

incorrect. Therefore, to avoid any more misapplication, Wang, Yang, Xu, and Chin [18] presented the correct centroid

formulae as:

/b c d b c d

L R L RA A A A A

a b c a b c

x xf dx xwdx xf dx f dx wdx f dx ,

and,

0 0

[ ( ) ( )] / [( ( ) ( )]w w

R L R L

A A A AAy y g y g y dy g y g y dy .

The correct formula proposed by Wang, Yang, Xu, and Chin [18] is only limited to trapezoidal fuzzy numbers

with invertible membership functions [11]. Shieh [13] presented the correct centroid formula, which can cater to both

invertible and non-invertible fuzzy numbers. The formula of the horizontal point is similar to Wang, Yang, Xu, and

Chin [18], while the vertical point is defined as 0 0

| | / | |w w

Ay A d A d , where | |A is the length of the

cut .A In particular, for a trapezoidal fuzzy number ( , , , ; ),A a b c d the value of

( )( ) ( ) 1

3 ( ) ( )

c by A

d c a b

, which coincides with Wang, Yang, Xu, and Chin‟s [18] formula.

To overcome the shortcomings of these existing fuzzy numberranking methods, this paper proposes a new cen-

troid-index ranking method based upon centroid formulae of Wang, Yang, Xu, and Chin [18] and Shieh [13]. The paper

further presents acomparative example demonstrating the efficiencies and advantages of the proposed centroid-index.

2. Fuzzy numbers

There are various ways of defining fuzzy numbers. This paper defines the concept of fuzzy numbers as follows.

Definition 1. A real fuzzy number A is described as any fuzzy subset of the real line R with membership function

( )A x that can be generally be defined as [20]:

( ),

,( )

( ),

0, otherwise,

L

R

A x a x b

b x cA x

A x c x d

(1)

where , ,a b cand d are real numbers. Unless elsewhere specified, it is assumed that A is convex and bounded (i.e.

, ).a d :[ , ) [0, ]LA a b is monotonic increasing continuous from the right function, and

: ( , ] [0, ]RA c d is monotonic decreasing continuous from the left function. If 1, then A is a normal fuzzy

number; otherwise, it is said to be a non-normal fuzzy number. If the membership function ( )A x is piecewise linear

and continuous, then A is referred to as a trapezoidal fuzzy number and is usually denoted by ( , , , ; )A a b c d or

simply ( , , , )A a b c d if 1.

Figure 1 is an illustration of the trapezoidal fuzzy number ( , , , ; ).A a b c d In this case, ( )

( ) ,L

x aA x

b a

a x b and ( )

( ) ,R

w x dA x

c d

c x d . In particular, when ,b c the trapezoidal fuzzy number is reduced to

a triangular fuzzy number and can be denoted by ( , , ; )A a b d or ( , , )A a b d if 1 . Thus, triangular fuzzy

numbers are special cases of trapezoidal fuzzy numbers.



A

y

a b c d

( )LA x ( )RA x

Figure 1. Trapezoidal fuzzy number.

Definition 2. The cut of fuzzy number Acan be defined as [21]

| ( ) ,AA x f x where , [0,1].x R

The symbol Arepresents a non-empty bounded closed interval contained in R. It can be denoted by

lA and

uA as the lower and upper bounds of the closed interval, respectively.

3. Improved Ranking Method Based on the Centroid-index of Fuzzy Numbers

In this section the centroid point of a fuzzy number corresponds to x value on the horizontal axis and y value on

the vertical axis. The centroid point ( , )x y for a fuzzy number A in definition 1 is defined as [13]:

( ) / ( )Ax xA x dx A x dx

(2)

0 0

| | / | |w w

Ay A d A d , (3)

where A is a fuzzy number with sup ( )x R

A x

, and | |A is the length of the cut ,0 1A , and

| | u lA A A . If A is a crisp set with 0( )A x and ( ) 0A x if 0x x , then its centroid is defined by

0( , )x .

For a trapezoidal fuzzy number ( , , , ; )A a b c d , the centroid point ( , )A Ax y is defined as follows [13, 18].

0 ( ) / ( ) ( ) / 3x A a b c d dc ab d c a b (4)

0( ) / 3 1 / ( ) ( )y A c b d c a b (5)

Remark. It is clear that 0

( / 3) ( ) ( / 2)y A .

Proof.

0( ) [1 ]

3 ( ) ( ) 3

c by A

d c a b

1 1

( ) ( )

c b

d c a b

0

( ) ( )

c b

d c a b

c b (6)

In the case of a triangular fuzzy number, b c so 0( ) ( / 3)y A .

0( ) [1 ]

3 ( ) ( ) 2

c by A

d c a b

2( )1

( ) ( )

c b

d c a b

( ) ( )0

( ) ( )

c d a b

d c a b

( ) ( ) 0c d a b c a b d . (7)

Because ,c a c b b d hence (6) is satisfied.

The new centroid-index is now proposed as follows. Suppose 1 2, ,..., nA A A are fuzzy numbers. First, we calculate

the centroid point of all fuzzy numbers ( , ), 1,2,..., .ii

i A AA x y i n We then define min min( , ),G x y such that


94

min inf ,x S 1 ,n

i iS U S { / ( ) 0},ii AS x f x

min inf ,y Y 1 ,n

i iY U Y { / 0 ( ) }.ii AY y Y x The distance between

the centroid point ( , ), 1,2,...,ii

i A AA x y i n and the minimum point min min( , )G x y , is proposed as follows.

2 2min min( , ) ( ) ( )

3i

ii A A

D A G x x y y

(8)

If ,i jA A are two fuzzy numbers, then the ranking will be done as follows.

(1) ( , ) ( , )i j i jA A D A G D A G

(2) ( , ) ( , )i j i jA A D A G D A G

(3) ( , ) ( , )i j i jA A D A G D A G

4. Numerical Example

This section uses numerical example to illustrate the validity and advantages of the proposed centroid-index ranking-

method.Numerical example demonstrates that the proposed centroid-index ranking method can rank a mix of fuzzy

numbers.

Example. Consider a mix of normal and non-normal fuzzy numbers using the proposed centroid-index ranking method.

The normal triangular fuzzy number is 1 ( 2, 1,3;1),A the non-normal triangular fuzzy number is

2 ( 2, 1,3;0.8),A and the non-normal trapezoidal fuzzy number is 3 ( 3, 2, 1,0;1).A Figure2 shows the pictures

of the three fuzzy numbers. Table 1 presents the results obtained by applying Cheng‟s [6] centroid-index, Chu and

Tsao‟s [7] centroid-index, and the proposed centroid-index (8). The final ranking result obtained by using (8) is

3 2 1.A A A It is worth mentioning that Chu and Tsao‟s centroid-index[7] cannot differentiate between1A and

2A -

that is, their rankings are always the same. On the other hand, the ranking order by using Cheng‟s [6] centroid-index

leads to an incorrect ranking order 2 1 3.A A A This example demonstrates one of the advantages of the proposed

centroid-index ranking method - it effectively ranks a mix of various types of fuzzy numbers.

A3

1

-3 -2 -1 0 1 2 3

y

x

A1

A2

0.8

Figure 2. Fuzzy numbers 1 2,A A and

3A .

Table 1. Comparative between fuzzy numbers 1,A 2A , and

3A .

Fuzzy number

Centroid points Cheng‟s ranking index Chu and Tsao‟s ranking index

ii

S x y

Minimum points G

Centroid by formulae (8)

iAx iA

y 2 2

R x y minx miny

1A 0 1/3 0.3333 0 -3 0.8 3.0091

2A 0 4/15 0.2667 0 -3 0.8 3.0005

3A

-3/2 7/18 1.9 1.5496 -3 0.8 1.5049

5. Conclusion

This paper proposes a new centroid-index method for ranking fuzzy numbers. The proposed formulae are simple and

have consistent expressions on the horizontal axis and vertical axis. Because the proposed centroid-index ranking



method is based on Wang, Yang, Xu, and Chin‟s [18]and Shieh‟s [13] centroid formulae, it can be used to rank both

invertible and non-invertible fuzzy numbers. The paper presenteda comparative example to illustrate the validity and

advantages of the proposed centroid-index ranking method. It shows that the ranking order obtained by the proposed

centroid-index ranking method is more consistent with human intuitions than those obtained byexistingmethods. Fur-

thermore, the proposed ranking method can effectively rank a mix of various types of fuzzy numbers (invertible and

non-invertible, normal, non-normal, triangular, and trapezoidal), which is another advantage of the proposed method

over other existing ranking approaches.

References

[1] R. R. Yager, “On a general class of fuzzy connectives,” Fuzzy Sets and Systems, Vol. 4, No. 6, 1980, pp. 235-242.

[2] L. Abdullah and N. J. Jamal, “Centroid-point of ranking fuzzy numbers and its application to health related quality of life indi-

cators,” International on Computer Science and Engineering, Vol. 02, No. 08, 2010, pp. 2773-2777.

[3] S. M. Chen and J. H. Chen, “Fuzzy risk analysis based on ranking generalized fuzzy numbers with different heights and differ-

ent spreads,” Expert Systems with Application, Vol. 36, 2009, pp. 6833-6842.

[4] S. J. Chen and S.M. Chen, “A new method for handling multi-criteria fuzzy decision making problems using FN-IOWA opera-

tors,” Cybernatics and Systems, Vol. 34, 2003, pp. 109-137.

[5] S. J. Chen and S. M. Chen, “Fuzzy risk analysis based on the ranking of generalized trapezoidal fuzzy numbers,” Applied Intel-

ligence, Vol. 26, 2007, pp. 1-11.

[6] C. H. Cheng, “A new approach for ranking fuzzy numbers by distance method,” Fuzzy Sets and Systems, Vol. 95, 1998, pp.

307-317.

[7] T. C. Chu and C. T. Tsao, “Ranking fuzzy numbers with an area between the centroid point and original point,” Computers &

Mathematics with Application, Vol. 43, 2002, pp. 111-117.

[8] E. S. Lee and R. L. Li, “A method for ranking fuzzy numbers and its application to decision making,” IEEE Transactions on

Fuzzy Systems, Vol. 7, No. 6, 1988, pp. 677-685.

[9] E. Mehdizadeh, “Ranking of customer requirements using the fuzzy centroid-based method,” International Journal of Quality

& Reliability Management, Vol. 27, No. 2, 2010, pp. 201-216.

[10] S. Murakami, H. Maeda, and S. Imamura, “Fuzzy decision analysis on the development of centralized regional energy control

system,” Proceedings of the IFAC SymposiumMarseille, 1983, pp. 363-368.

[11] N. Ramli and D. Mohamad, “A comparative analysis of centroid methods in ranking fuzzy numbers,” European Journal of

Science Research, Vol. 28, No. 3, 2009a, pp. 492-501.

[12] N. Ramli and D. Mohamad, “A centroid-based performance evaluation using aggregated fuzzy numbers,” Applied Mathemati-

cal Science, Vol. 3, No. 48, 2009b, pp. 2369-2381.

[13] B. S. Shieh, “An approach to centroids of fuzzy numbers,” International Journal of Fuzzy Systems, Vol. 9, No. 1, 2007, pp.

51-54.

[14] A. H. Vencheh and M. Allame, “On the relation between a fuzzy number and its centroid,” Computers and Mathematics with

Applications, Vol. 59, 2010, pp. 3578-3582.

[15] A. H. Vencheh and M. N. Mokhtarian, “A new fuzzy MCDM approach based on centroid of fuzzy numbers,” Expert Systems

with Application, Vol. 38, 2011, pp. 5226-5230.

[16] Y. J. Wang and H. S. Lee, “The revised method of ranking fuzzy numbers with an area between the centroid and original

points,” Computers and Mathematics with Applications, Vol. 55, No. 9, 2008, pp. 2033-2042.

[17] Y. M. Wang, “Centroid defuzzification and the maximizing set and minimizing set ranking based on alpha level sets,” Com-

puters & Industrial Engineering, Vol. 57, 2009, pp. 228-236.

[18] Y. M. Wang, J. B. Yang, D. L, Xu, K. S. Chin, “On centroids of fuzzy numbers,” Fuzzy Sets and Systems, Vol. 157, 2006, pp.

919-926.

[19] Z. X. Wang, J. Li, S. L. Gao, “The method for ranking fuzzy numbers based on the centroid index and the fuzziness degree,”

Fuzzy Information and Engineering, Vol. 2, 2009, pp. 1335-1342.

[20] G. J. Klir and B. Yuan, Fuzzy Sets and Fuzzy Logic Theory and Applications, Prentice Hall PTR, 1995.

[21] A. Kaufmann and M. M. Gupta, “Introduction to Fuzzy Arithmetic: Theory and Application,” VanNostrand Reinhold, New

York, 1991.


96



Development of an Automatic Cruise Control

Simulator

Hendro Nurhadi Mechanical Eng. Dept., Faculty of Industrial Technology, Institut Teknologi Sepuluh Nopember (ITS), Surabaya, INDONESIA

E-mail: [email protected]

ABSTRACT

An intelligent transportation system (ITS) is a subject to meet the need of an automated vehicle. The automatic cruise

control (ACC) is a driver-assisting device, used to control the headway with respect to the vehicle in front, according to

a given control law. There are various kinds of ACC systems of different complexity. This paper presents a systematical

approach to develop an automatic cruise control simulator in order to assist design engineers in obtaining the suitable

control parameters for desired performance. In the proposed approach, the standard unified modelling language

(UML) is adopted to design the system. Simulation results show that the developed system successfully helps us to de-

sign the automatic cruise controller.

Keywords: UML, ACC, simulator design, ITS, driver-assisting device

1. Introduction

Intelligent transportation systems (ITS), formerly called intelligent vehicle-highway systems (IVHS), aim to improve

the efficiency of current transportation systems by applying modern technology. One part of the ITS program, the

automated highway system (AHS), promises to reduce traffic congestion and increase the safety, efficiency and capac-

ity of highway systems without building additional highways [1], [2]. It does this by adding intelligence to both the ve-

hicle and the roadside. In the automated vehicle, the automatic cruise control (ACC) is a driver-assisting device, used to

control the headway with respect to the vehicle in front, according to a given control law. There are various kinds of

ACC systems of different complexity [3]-[7]. In general, it should provide the basic functionality for keeping a constant

speed, typically during a long journey of highway driving. It allows the driver to use the default speed (e.g. 100 km/hr

in this paper) or accelerate to a desired speed, and then activate the system so as to make the car maintain this cruising

speed without driver interventions.

This paper develops an automatic cruise control simulator so that the design engineer can further use it to evaluate

and compare various designed ACC systems and further modify the control parameters to achieve the desired perform-

ance, such as the transient overshoot, response time, and steady state error. The unified modeling language (UML) is a

language for specifying, constructing, visualizing, and documenting the artifacts of a software-intensive system [8]. It

defines the notation and semantics for modeling systems using object-oriented concepts. In this paper, we design the

ACC simulator based on the object-oriented technology with UML.

Generally, the UML consists of nine main diagrams corresponding to standard static and dynamic aspects. The de-

signer can freely choose a subset of the diagrams and their order is not constrained in UML. Although UML does not

define the development process and how to do object-oriented analysis and design, a use-case driven, architec-

ture-centric, iterative, and incremental development process [8] is recommended by using UML, as shown in Fig. 1.

First, the use-case diagram in UML is modeled to capture the requirements in the functional analysis stage. Then, in the

static structural design stage, the class diagram is used to describe the static relationship of the system. Subsequently,

the state chart is constructed according to above models to describe the dynamic behaviors. Finally, implementation of

the above models is performed by using the Java language. Each constructed model in Fig. 1 may be modified in an

iterative fashion, through a repeated cycle of analysis, design, and implementation, and then back to the beginning of

the cycle again (i.e. so-called round-trip engineering). In this paper, the ACC simulator is developed through this de-

velopment procedure.

The main goal of the system modeling, analysis, and design in previous stages is to provide standard models for

system implementation. Although the UML modeling is not restricted to any particular language for implementation,

we prefer Java as the target language due to its object-orientation, portability, safety, and built-in support for network-

ing and concurrency [9]. Java also possesses several features for real-time development [10]-[12]. During the imple-

Development of an Automatic Cruise Control Simulator

98

mentation, it concerns translating information from multiple UML models and the Petri net into the code and database

structure. This translation is not straightforward, however, there is a close correspondence between Java and UML, and

a standard mapping between UML and Java is described in [13].

Round-trip engineering

Figure 1. Systematic development procedure

2. Development of the Automatic Cruise Control (ACC) simulator

In this section, the UML models will be used to design the ACC simulator. Then, Java language is adopted to imple-

ment the system. Note that the models described later in the remaining paper have been simplified for illustration pur-

poses.

2.1. System specification

The vehicle specification used in this paper is obtained from the Formosa Magnus, which is a domestic car produced by

Formosa Plastic Croup in Taiwan. The cruise controller is based on the PID controller scheme, as shown in Fig. 2.

Some major parameters of the vehicle dynamic model and the designed controller are shown as Table 1. The maximum

adjustable output is designed in order to prevent rapid acceleration, to avoid engine damage, and to keep passenger

safety and comfort (the passenger would encounter bounded force 0.44 G for the maximum adjustable output). Note

that our purpose of this simulation is to obtain the suitable control parameters (P, I and D) so as to make the system

have the acceptable performance.

Figure 2. Block diagram of PID control scheme

Table 1. System specification for simulated ACC system

Parameter Value

For Vehicle Vehicle mass (including passengers) 1800 kg

Maximum engine output 2500 kg.m/s2

Aerodynamic drag coef. 0.5 kg/m

Mechanical drag 4 kg.m/s2



Parameter Value

For Control Maximum cruising speed 180 km/hr

Maximum adjustable output 800 kg.m/s2 (0.44G) Control param. 1: P N/A kg/sec

Control param. 2: I N/A kg.m/s2

Control param. 2: D N/Akg

2.2. Functional analysis with the Use-Case diagram

A use-case diagram is used to capture the basic functional requirements of the system. It consists of actors and use

cases. Actors, drawn as stick figures, represent users and other external systems that interact with the described system.

Use cases, drawn as ellipses, represent the scenarios of the system. A scenario is a sequence of steps describing an in-

teraction between a user and a system. Fig. 3 shows use cases for the automatic cruise control system, in which there are

2 actors and 8 use cases.

The actor, Driver, can perform use cases: Power ON/OFF, Pause/Resume, Stop, Set Speed, Default, Dial Up, and

Dial Down to manipulate the ACC system. The ACC status of the performed use cases will be displayed by using the

Display Status. After the driver powers on the system, he/she can use the Set Speed to set the cruising speed and further

use the extended use cases: Default, Dial Up, and Dial Down to adjust the cruising speed. In the cruising mode, the

driver may re-engage the manual control by using the Pause (in Pause/Resume) or Stop use cases. Then, the driver can

resume a previously set speed by using Resume (in Pause/Resume) after performing the Pause, or can set a new cruising

speed after performing the Stop.

Figure 3. Functional analysis with the use-case diagram

2.3. Static structural design with the class diagram

The class diagram is the main static structural analysis and design model for a system. It is developed through informa-

tion collected in the use-case diagram. A class diagram describes the types of objects in the system and the various

kinds of static relationships that exist among them. It also shows the attributes and operations of a class and the con-

straints that apply to the way objects are connects.

Fig. 4 represents the static structure and object relations of the ACC system. The Vehicle has the composition rela-

tion (represented as a black diamond) with the ThrottleActuator, VehicleDynamic, and SpeedSensor classes. The com-

position relation indicates that the composite is explicitly responsible for the creation and destruction of the contained

objects. Other relations in the diagram are associations, indicating loosely coupled classes that send messages to each

other in order to collaborate. The Driver can manually control the Vehicle directly through the ThrottleActuator or may

automatically control it by using the AutoCruiseCtrl with ThrottleCtrl class through the ThrottleActuator. The Speed-

Sensor detects the speed of the VehicleDynamic and then feedbacks it to the ThrottleCtrl and exports it to the

AutoCruiseCtrl for displaying on UserInterface.


100

2.4. Dynamic behavioral analysis with statechart

The statechart in the UML is the main dynamic behavior analysis model for a system. Fig. 5 shows the simplified

statechart of the ACC system. The ACC system has two major states: ON (a super-state) and OFF. When the driver

turns on the system power, the system will lies in the ON state, in which the system will display status continuously.

Two main threads are processed concurrently in the ON state.

On one hand, the system initially stays in Waiting state. When the driver set the cruising speed, it will transfer to the

Keeping Speed state to maintain the vehicle speed as closely as possible to the cruising speed. Then, if the driver stops

the ACC, it will transfer back to the Waiting state. If he pause the ACC, it will change to the Pausing state and await the

resume command. On the other hand, the ACC lies in the Setting Cruise Speed and processes the default set, dial up and

dial down commands to change the cruising speed.

Figure 4. Static structural design with the class diagram

Figure 5. Dynamic behavioral analysis with the statechart

2.5. Implementation with Java language

The system modeling and design developed in previous stages provide ACC models for implementation. The developed

graphical human/machine interface (HMI), shown in Fig. 6, is designed with a Java Applet. The human user can push

the buttons to issue commands and interact with the ACC system. Also, the status feedback is displayed on the HMI.



3. Simulation result

After simulating the ACC system with various control parameters (P, I and D), the suitable control parameters are ob-

tained, as shown in Table 2. The corresponding performance indices of such PID controller are also shown in the table.

It takes less than 26 seconds to accelerate from 0 km/hr to 100 km/hr without negatively impacting passenger safety and

comfort by using the ACC system. Furthermore, the resulted PID controller has good transient and steady responses,

less than 1.7 % overshoot and 0.2 % steady state error, respectively.

4. Discussion

The present work leads to the following discussions:

1. The design procedure in Fig. 1 is a round-trip engineering, in which all models may be developed in an itera-

tive and incremental way through a repeated cycle of analysis, design, implementation and test. This approach

admits the possibility of making some alterations, such as changing the requirements or discovering a flaw in

the original design.

Table 2. Simulation result

Item Value

Controller param. 1: P 5200 kg/sec

Controller param. 2: I 4200 kg/sec2

Controller param. 3: D 1 kg

Response Time < 26 sec (0-100 km/hr) Transient overshoot < 1.7 %

Steady state error < 0.2 %

2. The plug-in feature of the object-orientation helps achieve better modularity and further evaluate various ACC

systems. For example, after we set up the scenario with vehicles, ACC, and other components, we may want to

run the same simulation with different ACC models (or with another vehicle models) in order to make com-

parisons. From the class diagram in Fig. 4, we can easily change and plug-in the new AutoCruiseCtrl object (or

Vehicle object) to run other simulations without significantly changing other components.

Figure 6. The implemented ACC system

3. Since the UML is based on object-oriented concept, reusable models in the resulted models can be grouped

into a design library so as to saving time for the similar case design.

5. Conclusion

This paper presents an object-oriented approach to systematically design and implement the ACC by using the UML

and Java. First, the use-case diagram is adopted to describe the functionalities of the system. Then, the class diagram is

used to model the static structures, and the state chart is further applied to describe the dynamic behaviors of the system.

Finally, the implementation has been accomplished by using Java language with a Java Applet. The developed system

has been successfully useful for obtaining the control parameters of the ACC system through the simulation.

6. References

[1] P. Varaiya, “Smart cars on smart roads: Problems of control,” IEEE Trans. Automat. Contr., vol. 38, no. 2, pp. 195-207, 1993.

[2] J. S. Lee and P. L. Hsu, “Statechart-based representation of hybrid controllers for vehicle automation,” IEE Proc. Intelligent


102

Transport Systems, vol. 153, no. 4, pp. 253-258, Dec. 2006.

[3] Y. Zhang, E. B. Kosmatopoulos, P. A. Ioannou, and C. C. Chien, “Autonomous intelligent cruise control using front and back

information for tight vehicle following maneuvers,” IEEE Trans. Veh. Tech., vol. 48, no. 1, pp. 319-328, 1999.

[4] P. Li, L. Alvarez, and R. Horowitz, “AHS safe control laws for platoon leaders,” IEEE Trans. Contr. Syst. Tech., vol. 5, no. 6,

pp. 614-628, 1997.

[5] D. N. Godbole and J. Lygeros, “Longitudinal control of the lead car of a platoon,” IEEE Trans. Veh. Tech., vol. 43, no. 4, pp.

1125-1135, 1994.

[6] P. A. Ioannou, and C. C. Chien, “Autonomous intelligent cruise control,” IEEE Trans. Veh. Tech., vol. 42, no. 4, pp. 657-672,

1993.

[7] S. E. Shladover, “Longitudinal control of automotive vehicles in close-formation platoons,” ASME J. Dyn. Syst., Meas., Contr.,

vol. 113, pp. 231-241, 1991.

[8] G. Booch, J. Rumbaugh, and I. Jacobson, The Unified Modeling Language User Guide. Reading, MA: Addison-Wesley, 1999.

[9] E. Bertolissi and C. Preece, “Java in real-time applications,” IEEE Trans. Nuclear Science, vol. 45, no. 4, pp 1965-1972, 1998.

[10] Sun Microsystems, The Java Tutorials, December 2010. [Online]. Available: http://java.sun.com/docs/books/tutorial/

[11] K. Nilsen, “Real-time programming with Java technologies,” in Proc. IEEE Int. Symp. On Object-Oriented Real-Time Distri.

Comput., 2001, pp. 5-12.

[12] R. F. Mello and C. E. Moron, “A Java real-time kernel,” in Proc. IEEE Int. Conf. on Indu. Elec., vol. 2, 1999, pp. 728-734.

[13] J. Greenfield, “Unified Modeling Language/Enterprise JavaBeans (UML/EJB) Mapping Specification,” Rational Software

Corporation Document, May, 2001.



Risks Analysis on Yield Curve of Indonesian

Sharia Mortgage Financing Versus Conventional

Home Loans: Utilizing Vasicek Approach

Sudarso Kaderi Wiryono1; Barli Suryanta

2; Oktofa Yudha Sudrajad

3; Aulia Nurul Huda

4; Ana Nove-

ria5

Sub Interest Group of Business Risk and Finance, School and Business Management ITB, Bandung, Indonesia

[email protected], [email protected], [email protected], [email protected],

[email protected]

ABSTRACT

Murabaha is one of the popular sharia banking instruments for mortgage financing which usually has 15 year maxi-

mum duration. The ultimate competitor of sharia mortgage financing comes from conventional home loans. The concept

of sharia mortgage financing is different with conventional home loans including how to manage crucial risks and de-

termining its yields. Therefore, this study has purpose to compare yield curve of Indonesian sharia mortgage financing

and conventional home loans yield curve. Comparative analysis will be conducted from perspective of risk analysis. To

obtain displaying of yield curve, this study utilizes Vasicek approach to forecast yield of each. And the data on this

study are obtained from real data of sharia mortgage financing and conventional home loans with 15 year maturity.

The contribution of this study is viewing the substance differentiation between Indonesian sharia mortgage financing

and conventional home loans explicable of managing risks relying to yield curve.

Keywords: Murabaha, Vasicek model, Indonesian sharia mortgage financing yield curve, Conventional home loans

yield curve

1. Introduction

Islam prohibits Muslims involving interest (riba), defined as any predetermined or fixed return from financial

transactions including both deposits and loans, although the purpose for which such loans are made or how low the

rate of interest charged is [1]. Meanwhile, debt financing is a trade based financing engaging related parties with

buying and selling of good under sharia principles [2]. Murabaha is one of scheme of a trade based financing that

involves the banks buying what the merchant wants and then selling to customer later at an agreed price [3].

Then, Ismail [2] expressed in his study that as a trade-based contract, Murabaha total payment contract will be

treated as an opportunity cost concept related to present and future value where to calculate both of them will adopt

rate of return. Rate of return term in sharia banks can be called as an equivalent rate. An equivalent rate is according

to distribution of profit sharing between sharia bank and its deposit. In Indonesian case, an eclectic range of

equivalent rates are utilized in mortgage financing under Murabaha scheme. A portion of customers to pay their

burden at specified maturity linkage to Murabaha contract could be mentioned as customer equivalent rates of

mortgage financing.

After discussing an equivalent rate, there is another terminology which sharia banks always reveal in their

revenue, named as a yield. Yield in sharia mortgage financing is the discrepancy between equivalent rates erected to

customer who took a mortgage and sharia deposit funding rates. Consideration of yield, this study seeks to compare

yield curve of Indonesian sharia mortgage financing and conventional home loans. This is very important to observe

phenomena both of them, especially from risk analysis framework.

2. Methodology

2.1. Vasicek Approach

Vasicek [4] proposed model that avoids the certainty of negative yields and eliminates the need for a potentially infi-

nitely large extension factor. Yields in sharia mortgage financing usually have positive yields to avert loss and also to






Risks Analysis on Yield Curve of Indonesian Sharia Mortgage Financing Versus Conventional Home Loans:

Utilizing Vasicek Approach

104

minimize large extension factor by conducted prudent codes. Thereby, sharia mortgage financing confirms to the as-

sumption of Vasicek approach closely. The original Vasicek approach can be written as [5] :

dzdtrbadr .)( (1)

Where, r is the instantaneous short rate of interest; a is the speed mean reversion; b is the long term expected value for

r , and; is the instantaneous standard deviation r ; and a , b , and are constants. Vasicek elaborates (1) more

detail into the price at time t of a zero-coupon bond that pays $1 at time T :

)(),(),(),( trTtBeTtATtP (2)

In this equation )(tr is the value or r at time t ,

a

eTtB

tTa )(1),(

(3)

Then,

a

TtB

a

batTTtBTtA

4

),()2/)(),((exp),(

22

2

22 (4)

And, the formula on yield:

)(),(1

),(ln1

),( trTtBtT

TtAtT

TtR

(5)

Vasicek also expressed that r ( t ) is a stochastic process, subject to two requirements: first, r ( t ) is continuous func-

tion of time, that is, it does change value by instantaneous jump; second, it is assumed that r ( t ) follows a Markov

process. Under this assumption, the future development of spot rate given its present value is independent of the past de-

velopment that has led to present level. One of very critical from Vasicek model is thus made: the Markov property im-

plies that the spot rate process is characterized by a single state variable, namely its current value. The probability distri-

bution of the segment }({ tr is thus completely determined by the value of r ( t ). In another word, Vasicek model

can be utilized to forecast forward rate by knowing the spot rate or current value.

Therefore the existence of Markov process in Vasicek approach, this study proposes some vital procedures related to

Markov process. The procedures will be:

For sharia mortgage financing: first, using the assumption of Markov process to forecast sharia banks deposit rate for

15 year maturity; second, to appoint of previous first, this study needs to utilize real sharia deposit rate of 1 month, 2

month, 6 month and 12 month; third, in order to get yields calculation, the real data of sharia mortgage financing will

be applied to construct sharia mortgage financing.

For sharia mortgage financing: first, similar with sharia mortgage financing that using the assumption of Markov

process to forecast conventional banks deposit rate for 15 year maturity; second, to conduct appropriate forecast, it

takes real conventional banks deposit rate of 1 month, 2 month, 6 month and 12 month; third, the real data of con-

ventional home loans will be employed to get yield curve of conventional home loans.

Noted that their customers will subtract 15 year maturity and the data are collected on period of March and April 2011

from the largest of three state owned Indonesian sharia banks and conventional banks based on parameter of asset, i.e.

Bank of Sharia BNI, Bank of Sharia Mandiri, Bank of Sharia BRI, Bank of BNI, Bank of Mandiri, and Bank of BRI.




3. Risk Analysis on Yield Curve: Sharia Mortgage Financing Versus Conventional Home

Loans

Figure 1. Yield curve of Sharia BNI mortgage financing by utilizing Vasicek approach

Figure 2. Yield curve of conventional BNI home loans by utilizing Vasicek approach

Figure 3. Yield curve of Sharia Mandiri mortgage financing by utilizing Vasicek approach

0%

2%

4%

6%

8%

10%

12%

14%

16%

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Maturity

YT

M

Equivalent sharia BNI's rate of mortgage financing

Deposit sharia BNI's equivalent rate

4.0%

5.0%

6.0%

7.0%

8.0%

9.0%

10.0%

11.0%

12.0%

13.0%

14.0%

15.0%

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Maturity

YT

M

BNI deposit rate BNI home loans

4.0%

5.0%

6.0%

7.0%

8.0%

9.0%

10.0%

11.0%

12.0%

13.0%

14.0%

15.0%

16.0%

17.0%

18.0%

19.0%

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Maturity

YT

M

Mandiri deposit rates Mandiri home loans



106

Figure 4. Yield curve of conventional mandiri home loans by utilizing Vasicek approach

Figure 5. Yield curve of Sharia BRI mortgage financing by utilizing Vasicek approach

Figure 6. Yield curve of conventional BRI home loans by utilizing Vasicek approach

0%

2%

4%

6%

8%

10%

12%

14%

16%

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Maturity

YT

M

Equivalent Mandiri's rate of mortgage financing

Deposit sharia Mandiri's equivalent rate

0%

2%

4%

6%

8%

10%

12%

14%

16%

18%

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Maturity

YT

M

Equivalent sharia BRI's rate of mortgage financing

Deposit sharia BRI's equivalent rate

4.0%

5.0%

6.0%

7.0%

8.0%

9.0%

10.0%

11.0%

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15Maturity

YT

M

BRI deposit rates BRI home loans




Figure 1, figure 3, and figure 5 shows that Sharia BNI, Sharia Mandiri, and Sharia BRI mortgage financing adopts

positive yield curve. Positive yield curve in which yields with long rates substantially greater than short rates [6]. Con-

trast with conventional home loans of BNI, Mandiri, and BRI where they expose normal yield curve that the curve

slopes gently upwards as maturity increases, all the way to the longest maturity. Therefore interest rate is prohibited by

sharia principles so that yield curve in sharia banks influenced by factor of inflation, maturity, and operational. Inflation

risk reflects that sharia banks in this context expect inflationary pressures in the future. Thereby to anticipate such thing,

sharia banks induces higher yield than conventional home loans in order to obtain optimum return from mortgage fi-

nancing. Maturity risk in sharia banks linkage to mark up risk therefore 15 year of sharia mortgage financing under

Murabaha scheme can not be re-priced and can not be used swaps to transfer the risk [7]. Regarding liquidity risk, even

though in the process way the mortgage is increasing, as long as the contract has not been terminated, the customers can

not sell to get arbitrage gain on their mortgage unless all the obligation is already settled to sharia bank. This infers

sharia mortgage asset is not liquid. Fact of operational risk of sharia banks which they need to be instituted [7] and cur-

rently the number of their operational branch still under conventional banks. The much more operational branch of

banks, the much more customer can acquire to take loans, the more efficient rates on loans. Right now, sharia banks in

Indonesia are relatively new with little branch, indeed they need high yield to hedge their less efficient on operational.

Sharia banks are having low risk such credit and market. From mortgage financing, sharia banks enjoy certain income

and definitely assets of sharia mortgage can be treated as trusted collateral [7].

Figure 2 and figure 6 displays normal yield curve for conventional home loans of BNI and BRI. Compared to

positive yield curve, normal yield curve provides yields at average levels and the curve slope gently upwards as matur-

ity increases, all the way to the longest maturity [6]. Only conventional Mandiri exposes positive yield curve. In their

business process conventional home loans so much depend on dynamics of interest rates. This study detects rates of

BNI and BRI home loans are more efficient than sharia mortgage financing. To determine the rates for home loans,

conventional BNI and BRI calculate urgent factors prudently, i.e. delivering deposit rates on routine schedule, how

much risk free rate announced by central bank, and inflation rate expectation in the future. By concerning to these, they

earn ease credit spread, otherwise they will get loss.

This condition is supported by a lot of operational branches in nearly all major cities in Indonesia, established in-

stitution, and well risk management. It implies that their operational risk is low. And conventional home loans can be

countenanced as liquid assets. Customers that already owned the house, it can be sold anytime to a third party when its

price increases as long as payment running well to the banks. Another advantage thing to take home loans is customers

pick up second home loans by using their first house as trusted collateral and so on. BNI and BRI can bundle their as-

sets backed-mortgages becoming a valuable bond easily, then issuing to primary market in order to receive a significant

funding. To minimize credit risk and market risk of 15 year home loans maturities, both conventional BNI and BRI

utilize credit debt swap instruments. They can transfer credit and market risk to derivative instruments or in scheme of

efficient portfolio or combination between them to tackle interest rate risk. It is very clearly that normal yield curve of

conventional BNI and BRI is taking advantage of their business process with the customer. Caused the rate is competi-

tive and efficient, and by that, both can deliver more home loans to reasonable customers in significantly.

Positive sloping yield curve of conventional Mandiri (see figure 4) has different character compared to BNI and

BRI. It is interesting because of the similarity between Sharia Mandiri and conventional Mandiri. Positive yield curve of

Mandiri indicates that short term interest rates are expected to rise, then the longer yields should be higher than shorter

ones [6]. As the grandstanding bank in Indonesian, Mandiri has wide and steady operational branches throughout Indo-

nesia. Considering the large size of its branches, Mandiri can easily minimize the operational risk of its housing loans.

To attain optimum credit spread, Mandiri respects with country outlook in the future by applying positive yield curve in

order to compensate inflationary risk and maturity risk. Inflationary risk usually is stimulated by constructive growth of

a country. To intercept inflationary risk plus maturity risk, Mandiri entails significant yields. Mandiri also appoints de-

rivative instruments and portfolio strategy as a credit debt swaps mechanism to secure its home loans from probability

of default. Mandiri gets beneficial from liquidity of its mortgages bond to collect a huge funding. Not only Mandiri but

also its customers enjoy liquidity same as customers of BNI and BRI that already discussed before.

4. Conclusions

Sharia BNI, Sharia Mandiri, and Sharia BRI applied positive yield curve to hedge inflation, maturity and operational

risk. Sharia mortgage financing is not liquid neither for sharia banks nor its customer because of the Murabaha

agreement. Mark up risk is correlated with maturity where the longer maturity means the greater mark up risk. Sharia

Banks do not provide a set of alternative to hedge mark up risk because they do not allow credit debt swaps mechanism

especially established derivative instruments. However they have low credit and market risk. Sharia banks earn certain



108

income and asset of sharia mortgage can be treated as trusted collateral. Conventional BNI and BRI attribute normal

yield curve. They consider efficient rate as their weapon to invite potential customers to take home loans on them. They

have capacity to cover operational risk easily based on wide range operational branch throughout Indonesia. Their

housing loans are so liquid and very beneficial whether for the banks or its customers. To minimize credit and market

risk, they play derivative instruments to transfer risk (credit debt swaps) and implement portfolio strategy to manage

interest rate risk. Conventional Mandiri is a little bit different by inducing positive yield curve. But in its business proc-

ess actually has same pattern as conventional BRI and BNI.

5. References

[1] J.C.Y. How, M.A.Karim, and P. Verhoeven, “Islamic Financing and Bank Risks: The Case of Malaysia,” Thunderbird Interna-

tional Business Review, vol. 47(1) 75-94, Wiley periodicals, Inc, 2005.

[2] R. Ismail, “Assessing Moral Hazard Problem in Murabaha Financing,” Journal of Islamic Economics, Banking, Finance, vol-

ume-5 Number-2, p.102-112, Unknown Year of Published.

[3] F.F. Ghannadian, “Developing economy Baking: The Case of Islamics Banks,” International Journal of Social Economics, vol.

31 no. 8, Emerald Group Publishing, 2004, pp. 740-752

[4] O. Vasicek, ”An Equilibrium Characterization of The Term Structure,” Journal of Financial Economics 5, North Holland Publi-

hing Company, 1977, p.177-188.

[5] H.C. John, “Options, Futures, and Other Derivatives, Fourth Edition,” Prentice Hall Upper Saddle River, NJ 07458, 2000,

p.567.

[6] F.J. Fabozzy,”Interest Rate, Term Structure, Valuation Modelling,” John Wiley and Sons, Inc, 2002, p.74.

[7] T.K.B. Ahmad, “Risk Management An Analysis of Issues in Islamic Financial Industry,” IDB Islamic Research and Training

Institute, Occasional Paper No.5, Jeddah, 2001.



Understanding the Potential Effects of Queue

Information on Visitors’ Behavior and the Factors

that Influence Their Decisions: Case Study at

Dufan Theme Park

I Putu Wisnu Saputra1; Yos Sunitiyoso

2

School of Business and Management, Institut Teknologi Bandung, Indonesia

[email protected], [email protected]

ABSTRACT

In a theme park, it is expected that information about queue and the availability of rides would be beneficial for the

visitors and thus increasing the level of service of the theme park. The information provides the visitors with options to

choose what rides in that theme park that are available with shorter queue. To understand the potential effects of such

information on visitors‘ choice behavior, a research is conducted in Dufan, the biggest theme park in South East Asia.

The study utilizes a questionnaire survey involving over 200 respondents who are visitors at Dufan during its peak days.

The questionnaire data are being analyzed using descriptive and statistical analysis. The study found that nearly half of

respondents will move to other line when they get the information about the queue and the information in the form of

digital board is the most preferred. Furthermore, the most influential factor that affects a visitor‘s decision of whether

to stay or to move to another ride when having to queue in his/her current location is the distance between rides. Level

of favorite is also an important factor. The more favorite the ride to the visitor, the more he or she wants to go to that

ride. The study implies the need for such queue information and its potential positive effect shows that this would add to

customer satisfaction. People are given a choice to move or stay than just wondering when their time to play comes.

People can calculate the time of waiting to gain the best option: to stay in the current line, move to other ride or just

move out from the queue and wait outside while eat in a restaurant. It means that giving people various options would

make them enjoying the amusement park more than just waiting in line.

Keywords: theme park, queue information, visitors‘ behavior.

1. Introduction

People react differently on information they get. Their behaviors are studied in this research in order to understand

the factors that influence people ride choice or movement from one ride to another. In every ride in Dufan there are

no information that would provide answer to customers on (a) how many people those stand in line with them? (b)

How many minutes that they have to wait to get into the rides? and (c) are there any other rides with fewer queues than

they stand on now? The answer of these questions would be very helpful to all visitors of Dufan in choosing the

shortest queue to stand on. In this study, this information is named as Queue Information.

Dufan in total has 25 rides which are separated in eight different theme areas (see Figure 1). Those rides have

different characteristics. These characteristics, whether they are exciting, adventurous or fun may drive people to go

to one ride or not. If we try to compile all of the characteristics of one ride and match them to each of the visitor of

Dufan, we can get the Level of Favorite from visitor to that ride. Dufan is a 9.5 hectare area. If someone has an av-

erage 1 m/s (meter/second) of walking time, he or she would require 26.4 hours, more than one day, to cover whole

the area of Dufan just walking not playing the rides. Distance is another important factor that influences people to

go from one ride to another ride. The origin where visitor come from is also hypothesized as a factor for visitor to

choose any of rides. People from Jakarta would only choose rides that he or she intended to, while people outside

Jakarta would like to choose all the rides because they couldn‟t go to Dufan often. People come to Dufan individu-

ally or in group. Those who come to Dufan in group can be separated into two: small group (consist of less than 25

people) and large group (consist of more than 25 people). People who come in group may follow wher e their group

going. But, people who come individually or in relatively small group have their independent choice to choose the

Understanding the Potential Effects of Queue Information on Visitors‟ Behavior and the Factors that Influence Their Decisions: Case

Study at Dufan Theme Park

110

ride they wanted. In this study, both people who come in small group and individually are considered as individual

visitors. These five factors: queue information, level of favorite, distance, origin, and group, are the focus of this

study and are investigated to understand their influences to visitors‟ behavior.

Figure 1. Dufan Rides and Facilities

Differences on the queue length from one ride to another and the absence of information regarding the queue

makes the people only concentrated in lining for a ride; especially their favorites. It can make other rides have only

a few visitors. People who wanted to go to Dufan for fun can only get their time stuck in queue. In Dufan, there are

many queue spots such as, Simulator Theatre, Kora Kora, Bianglala, Arum Jeram, Halilintar, Istana Boneka, Nia g-

ara gara, Baku Toki, Tornado and Hysteria (see Figure 1 for the location of these rides). The longest queue is in

Simulation Theatre. Furthermore, the other rides that have small number of people who stand in line are: Burung

Tempur, Rajawali, Pontang Pontang, Ombang Ombang, Poci Poci and Ubanga Banga.

An important reason that may cause the queue problems in Dufan is that there is no information about queue in

other rides. It makes those who stand in line become captive or they just do not know the situation in other rider and

have no other choice than staying in the line until they are served. When they receive information, they may change

their choice whether to move to another ride with shorter queue or they may decide to continue waiting for their turn.

The medium of information and queue time are also expected to be important in influencing their choices. This study is

looking to understand the potential effects of such information and the factors that influence visitors‟ decisions.

2. Underlying Theories

Maister [1] studies about the feeling of people while he or she is in queue. He stated that “if managers are to con-

cern themselves with how long their customers or clients wait in line for service, and then they must pay attention

not only to the actual wait times but also to how these are perceived. Managers should also try to put their position

as the customers to understand their feeling while in line. Knowing the characteristic of people in queue is important

to this study because the author would like to understand the feeling of those people and therefore should bring success

in the questionnaire process.

Hudson [2] stated that there are factors that influence people on choosing what he or she wants to buy a pro d-

uct or service in tourism and hospitality. Therefore understanding the way these factors influence customers‟ dec i-

sions is very important in this industry. There are several factors that may influence consumer behavior: motivation,

culture, age & gender, lifestyle, life cycle, and reference groups [2]. In this study, some of these factors are h y-

pothesized to be influential to behavior of Dufan customers. They are including: level of favorite of rides, distance

between rides, effect of origin of the respondents on choosing rides, and effect of coming to Dufan in a group or

individual. Level of favorite can be influenced by the Dufan visitors‟ culture, age and gender. Women and children

are usually like fun rides such as Istana Boneka and Balada Kera. But young men are usually like challenging rides

such as Kora Kora, Halilintar, Tornado, Arung Jeram and Hysteria. Motivation has in fluence people to go to rides

even though the distance between rides is far away. The relative home location to Dufan may influence people to




chose enjoying only the rides they wanted (their favourite ones) or to try enjoying all ride while they were in Du fan

due to their limited chance to visit Dufan frequently. Reference groups may also influence people who come in

groups, for example by following their group‟s choice in choosing the rides to play.

3. Research Methodology

The way to obtain business solution on the queue problem of Dufan is by understanding visitors‟ preferences and

behavior thorough a survey in Dufan. Starting in November 2010, a questionnaire survey using face -to-face inter-

view is conducted by asking 228 respondents who visit Dufan during its peak periods which are during weekends;

Saturdays and Sundays. The data is then checked and it is found that 200 questionnaires are complete and can be

used in analysis. The questionnaire consists of 25 questions in total. They are divided into three sections (a) first

impression to the queue section which asks respondents about their experience when visiting Dufan and experiencing a

queue; (b) giving information about the queue section which contains question regarding the effect of information to

visitors choices and (c) respondent profile which consists of ages, gender, origin, educational background, and occupa-

tion.

The research is focused only on „ride‟ and not including „show‟. The example of a show is Balada Kera. The di f-

ferent of show and ride is that a show has a schedule for the visitors who want to watch it, while a ride cannot offer

it because the time of playing ride can vary on some rides depending on the length of queue. The number of people

who are standing in queue is the main factor that makes ride cannot give the exact schedule on every play. If the

queue line is short, then ride‟s operators would extent the ride‟s time of play. If the queue line is long, then ride‟s

operators would cut the ride‟s time of play. The questionnaire is only taken on weekends not weekdays when the

theme park is at its peak periods. Weekends were chosen to reflect the real feeling of visitors who are standing in

queue which is usually takes place in weekends. The research is only focusing on the queue problem which happens

in Dufan and does not take any benchmarking with other theme parks.

4. Results and Analysis

From the first section of questionnaire, it is obtained that 69% of respondents choose to stand in current queue, 18%

of respondents choose to move to another ride and the last 14% choose to follow friends or family or to use Fas t-

Trax. When completing this section, the respondents were not informed yet about the questions in the second part.

In the second part of the questionnaire, respondents were asked about their responses to information about the

queue. Respondent were given information about other ride that has shorter queue than he or she has at current time.

There are types of movement that can be chosen by respondents:

1. To move from Favorite ride to Favorite ride that is located near to current location.

2. To move from Favorite ride to Favorite ride that is located far from current location.

3. To move from Favorite ride to Non-Favorite ride that is located near to current location.

4. To move from Non-Favorite ride to Favorite ride that is located near to current location.

5. To move from Non-Favorite ride to Favorite ride that is located far from current location.

6. To move from Non-Favorite ride to Non-Favorite ride that is located near to current location.

These questions are used to identify the effects of preferences on rides and the distance of one ride to another to the

behavior of people to move from one ride to another. Figure 2 shows preferred actions of respondents after they

were provided with queue information.

Figure 2. Preferred Actions After Given Information Figure 3. Value on the Queue Information



112

In Figure 2, The highest percentage is „moving from non-favorite to favorite which located near to the current

queue‟ with 79% of respondents. The lowest percentages are „Favorite – Favorite – Far‟ and „Favorite – Non Fa-

vorite – Near‟ with only 27% of respondents will move to other ride. It can be concluded that people who has been

standing in favorite ride line would be reluctant to move to other ride if that ride is far from the one they are stand-

ing now or not as favorite as the one they are waiting for now. From that figure, we can get the average percentage

of „moving‟ decision is 45%. This means that almost half of the respondents are willing to move when they got the

information about other ride. This is supported by their answers on the value of the queue information as in Figure

3. Around 94% of respondents consider that queue information would be valuable or useful feature of Dufan .

The next proposition relates to the digital information board. More than 50% of respondents are choosing digital

board as the medium for queue information. Digital board is being chosen because it can be seen by all visitors at

the same time. Digital board also will not be interrupted by various noises produced by people, announcement via

loudspeakers or sound of music in Dufan.

The questionnaire also tries to answer how many minutes are acceptable to stand in the line for the respondents.

There are two biggest percentage of queue time, less than 10 minutes (< 10 min) with 46% and between 10 and 20 min-

utes (10-20 min) with 37%. The combination of these percentages is 83% of respondents. The result would mean that

there are most of people accept the queue time less than 20 minutes. This concludes that most people do not want to

wait for a long time. In reality, they may have to wait much longer than that.

To study the factors that influence visitors‟ decisions to move to another ride or to stay in the queue, a statistical

analysis is conducted. The data is reorganized following these rules: level of favorite is assigned into four numbers

based on hypothesized preference of visitors on their movement from a ride to another ride, „0‟ as moving from favorite

ride to non-favorite ride, „1‟ as moving from non-favorite ride to non-favorite ride, „2‟ as moving from favorite ride to

favorite ride, and „3‟ as moving from non-favorite ride to favorite ride; distance is assigned into two numbers, „1‟ as

near and „0‟ as far; origin is assigned into two numbers, „1‟ as from Jakarta and „0‟ as from outside Jakarta; group is

assigned into two numbers, „1‟ as coming to Dufan as group and „0‟ as individual. The categorized data is then exported

to the SPSS and analyzed using Logistic Regression. The method is a specialized form of regression that is formulated

to predict and explain a binary categorical variable [3]. Those factors above become independent variables and the

decision variable, which is assigned into two numbers: „0‟ as stay in the queue and „1‟ as move to another ride becomes

a dependent variable. The objective of logistic regression is to predict the value of dependent variable which is an

binary variable („0‟ or „1‟) using independent variable which has already known the value before [4]. The result is

shown in Table 1.

Table 1. Variables in the Equation

B S.E. Wald df Sig. Exp(B)

Step 1a Favorite .787 .070 127.594 1 .000 2.196

Distance 1.328 .150 78.119 1 .000 3.773

Origin .336 .131 6.638 1 .010 1.400

Group -.245 .129 3.606 1 .058 .783

Constant -2.596 .235 121.976 1 .000 .075

a. Variable(s) entered on step 1: Favorite, Distance, Origin, Group.

It can be seen that there are three factors that are significant with = 0.05, which are favorite, distance and

origin. Group is not significant in influencing people to move to another ride. The negative value of constant

(-2.596) shows that if there are no queue information, customers would like to stay in the queue. From the three

significant factors, it is interesting to find that Distance has the biggest coefficient (1.328) that means that Distance

is the most influential factor for making people move to another ride. Level of Favorite comes second with 0.787

and the last is Origin with 0.336. With this result, Dufan should consider to offer ride that is located not far away

from the queue information board. Showing information of favorite rides on the queue information board would

also influence visitors to move from their current location. It is also concluded from the result above that people

from Jakarta are eager to move to another ride when the queue is too long for them to wait than people from outside

Jakarta. It can be summarized that providing queue information would make people move and the re are three factors

which are significant in influencing visitors‟ decisions of whether to stay or to move from a queue, which are level

of favorite, distance and origin with distance factor to be the most influential factor.

These factors have been tested for multicollinearity which can be shown from the result of their VIF (Variance

Inflation Factor) and Durbin Watson value. VIF value for each factors are: Level of Favorite = 1.242; Distance =

1.242; Group 1.066; Origin = 1.066 and the Durbin Watson value is 1.785. If the VIF is less than 10 there is strong




indication that multicollinearity is not affecting the regression coefficients and therefore they are well estimated.

Furthermore, if Durbin Watson value is close to 2, it means that there is no correlation between factors used in the

analysis.

From the classification result, it was obtained the overall percentage is 66.8%. This means that the regression

morel can give correct answer in 66.8% of possibility, which is higher than the percentage of correct by chance that

is 50%. This percentage is enough to shows that the regression model can be used to answer the research question:

the importance of queue information and how it can influence people.

5. Conclusion and Future Study

From the analysis, it is concluded that there is a significant change of respondents‟ decisions when the information of

queue is provided to them. The percentage of respondents who will move to other ride when the queue is too long for

them to wait before given any information about another ride queue condition is 18%, while the average percentage of

respondents who will move to other ride after given the queue information is 45%. This significant increase of percent-

age would mean that it is important to provided queue information for Dufan visitors.

It is also concluded that the most influential factor is the distance between rides, followed by level of favorite be-

tween rides and then the origin of the visitors. Group is not significant factor in influencing people to move to another

ride. It is also obtained in the statistical analysis that the regression model‟s constant is negative. It means that if the

customers do not have any information about another ride‟s queue, they would not like to move to another ride. Most of

respondents would agree on below 20 minutes as an acceptable queue time. Digital Board would be the best option to

choose because it cannot be distracted by the noise and can be seen to everyone.

The study also found some potential implications to Dufan. Firstly, information about number of people who

standing in line on one ride is required in order to give an informed decision of whether the visitors would like to

move or to stay when they are in queue. He or she will choose the one of ride which has shorter queue. However

some people may only be affected if they just get in a line for less than their acceptable waiting time. The reason is

because people who stand in for long time and get into the queue too deep, would not want to step onto other ride

because they have spent too much time to wait.

Secondly, regarding the medium of the information, the majority of respondents prefer the use of a digital

board to inform them about the queue information. Digital board would be the best option to choose beca use it can

be easily seen to everyone and cannot be distracted by the noise that formed by the huge number of people gathered

in the theme park. This is more preferred than a booking system like FastTrax in Dufan and FastPass in Disney,

Lo-Q, Multi Motions, and Alton Towers as this adds another problem of fairness. People which have more money

to book the ticket so that they do not have to stand in the line would make other people who do not have this ability

to buy the tickets would feel disappointed. Lutz stated that the customers of theme park that do not use the virtual

queue system and wait in the general line see the intrusion as negative [5].

The respondents choose that it is best to place the information medium on every ride in Dufan. This will give the

same chance for everyone to access that important information. Placing the digital board on every ride would require

investment for the theme park management. But for the customer side, it would add benefit because they do not have to

walk to a kiosk to get the information and particularly they do not have to face a secondary queue. Furthermore, the

digital board should be installed in front of every entrance of ride or in the range of 20 minutes of waiting time from the

entrance (considerable time of waiting in line according to the questionnaire result). This would prevent people who

want to go to another ride because of the information given to them get too deep in the queue. Based on the author‟s

experience, it would be very hard for people to move out from a crowded queue. Dufan should put the distance between

rides on its every information board on each ride. To get more effective result, Dufan may give the queue information

of rides limited on the themes areas. For example in Hysteria queue, people would be given the queue information only

about rides which are located in the same theme area, Greeks. This action also could prevent people who are standing

in queues at another theme areas moving to other areas. If all people in other areas know that Hysteria in Greeks has a

very short queue line, it would make them go directly to it. Because the distance between theme area is quite far, when

they come to Hysteria, it would have possibly been full. This incident may add to their dissatisfaction. So, by selec-

tively choosing the queue information only for the rides in one theme area would minimize this problem.

Finally, the study implies the need for information and its potential positive effect shows that this would add to

customer satisfaction. People are given a choice to move or stay than just wondering when their time to play comes.

People can calculate the time of waiting to gain the best option: to stay in the current line, move to other ride or just

move out from the queue and wait outside while eat in a restaurant. It means that giving people various options would

make them enjoying the amusement park more than just waiting in line. Lith mentioned that having information about



114

visitor‟s interests would benefit them [6]. He also said that this valuable information could stimulate cooperation with

sponsors to attract the public to their products.

There are several future studies that can be conducted to follow up on the outcome of this study. These i n-

cludes: simulation and field experiment. The first follow up of this study is a simulation study to model the visitors‟

movement in the theme park. It is found in this study that the information has given benefit to the respondents, however

to estimate the benefit of information in reducing queue a simulation is required. The visitors‟ behaviors are modeled

based on the outcomes of this study. Field experiment is the next step after the simulation, which aims to prove the

benefit of providing queue information to visitor. This can be conducted for example by placing digital board to one of

the ride of Dufan and then studying visitors‟ response to such information. Field experiment will make sure whether the

information show in the digital boards could be useful in the way required by visitors. Another purpose is this field ex-

periment would attract feedback from the visitors of Dufan.

6. References

[1] Maister, D. H. (1985). The Psychology of Waiting Lines. The Service Encounter, 113-123.

[2] Hudson. (2007). Tourism and Hospitality Marketing. University of Calgary, Canada : Sage Publication Ltd.

[3] Hair, J., Black, W., Babin, B., Anderson, R., & Tatham, R. (2006). Multivariate Data Analysis, Sixth Edition. New Jersey:

Pearson Prentice Hall.

[4] Santoso, S. (2010). Statistik Multivariat. Jakarta: Elex Media Komputindo.

[5] Lutz, H. (2008). The Impact of Virtual Queues for Amusement Parks. Proceedings of Decision Sciences Institutes (DSI)

39th Annual Meeting, Baltimore.

[6] Lith, P. v. (2002). Queue Management. Retrieved from http://multimotions.websystems.nl/eng/index.html.

http://multimotions.websystems.nl/eng/index.html



Developing Model to Identify Significant Human

Factors in Aviation Maintenance

Mohd Noor Said1 ,Nooh Abu Bakar

2, Ahmad Zahir Mokhtar

3

1Universiti Kuala Lumpur Lot 2891, JalanJenderamHulu, JenderamHulu, Dengkil, Selangor, Malaysia 2School ofgraduate studies, UTM InternationalCampus, JlnSemarak, Kuala Lumpur, Malaysia 3Universiti Kuala Lumpur Lot 2891, JalanJenderamHulu, JenderamHulu ,Dengkil, Selangor, Malaysia


ABSTRACT

Elimination of aviation accidents is one of the primary goals of the aviation maintenance industry. A leading cause of

aviation accidents is lack of oversight of various human factors issues and organization‘s maintenance operation per-

formance. The technologies used in the industry generate multiple risks, mostly from three domains: systems, hardware

and people. Analysis of existing aviation maintenance data is a crucial step in meeting the aviation industry‘s need to

improve aviation safety. This paper approaches to assess significant human factors impacting human error in aviation

maintenance. We conducted study of Malaysia aviation maintenance industries to determine these significant human

factors and to illustrate how empirical analysis approaches integrate aircraft maintenance personnel opinions about

the relative importance of human factors. We develop model through modified SHELL model to categories the human

factors that we derived from the literature review and the opinions of aviation personnel who involved in maintenance.

The result showed that there are significant human factors impacting human error, and the result also provided ap-

proaches of Structural Equation Modeling (SEM) to verify the hypotheses in the path analysis model. The model helped

to determine the significant human factors underlying aviation maintenance errors, ultimately helping aviation person-

nel to manage human error and safety issues in aviation maintenance.

Keywords:Developing Model, Significant Human Factors, Impacting Human Error, Aviation Maintenance

1.Introduction

Aviation maintenance personnel work on extremely sophisticated aircraft with complex integrated systems which are

continuously upgraded and improved. The technological changes with respect to digital computer system and introduc-

tion of new materials requires that the maintenance personnel be trained to analyze, repair, inspect and certify these

system in accordance with the quality standards defined by the aircraft manufacturers and Aviation Authorities. Aircraft

maintenance is an essential component of the global aviation industry. It involves a complex organization in which each

aircraft maintenance personnel performs varied task with limited time, minimal feedback, and sometimes difficult am-

bient conditions [1]. Maintenance in this context is essentially about keeping aircraft operational within a strict time

schedule. The main role of aviation maintenance personnel is to categorize and judge the important of problem that

could threaten the airworthiness of aircraft [2]. Aircraft contain many rapidly developing advanced technologies, such

as composite material structures, glass cockpits, highly automated systems, and build-in diagnostic and test equipment;

therefore, the need to simultaneously maintain new and old fleets requires aviation maintenance personnel to be knowl-

edgeable and adept in their work than in previous years [3]. However, the complexity of such operations naturally pre-

sents new possibilities for human error and subsequent break-downs in the system‟s safety net [4].

In recent years, the aviation industry has gradually begun to make use of risk management and risk incident analysis

[5], [6], [7], [8]. Many accident reports now include risk factors in their conclusions. For example, on May 25, 2002, a

B747-200 China Airlines passenger aircraft departing Taiwan for Hong Kong broke up in-flight; all 225 people on board

were killed. The accident report by the Aviation Safety Council (ASC) in Taiwan found that the incident involve many

items related to maintenance risks that had the potential to degrade aviation safety [9].The most important step in aviation

management is risk identification. If the risk cannot be accurately identified, it cannot be analyzed or evaluated. Once

actual and potential hazards are identified, an assessment should be made of the cause and contributing factors and a

decision should be made as to whether action is required [10]. We aimed at evaluating the significant human factors in

aviation maintenance industries. The objective is to help aviation industries better understanding their major operational

and managerial weaknesses in order to improve management and aviation maintenance operation. The study of ques-




Developing Model to Identify Significant Human Factors in Aviation Maintenance

116

tionnaire in Malaysia aviation industries was conducted to determine these significant human factors and to illustrate how

empirical evaluation approach integrate expert opinion about the relative importance of these factors.

1.1 The human factors model

Human factor practitioners typically concentrate on the interface among people and the other system elements. The im-

portant point of the system view is that humans cannot be isolated from other system components. The view is similar

to that of ecologist, i.e. that all elements in nature interacts. We can't change one aspect of the system without being

concerned about its effects on other system [11].All aviation accidents are composed of four factors [12], this is known

as the SHEL model: software (e.g. maintenance procedure, maintenance manual, checklist ), hardware (e.g. tools, test

equipment, physical structure of aircraft, and instruments ), environment (physical environment such as condition in the

hangar, work environment such as work patterns, and management structures), and liveware ( the person or people at

the center of the model, including maintenance engineers, supervisors, managers, etc,) [13]. The model which identifies

three kinds of interactive resources, it‟s indicated that the sources of all aviation accidents can be categorized as one

(Liveware) or combination of three major relationships (Liveware-Software, Liveware- Hardware, and Live-

ware-Environment.)

Hawkins [14] modified Edwards‟ model to include the interactive nature of the person to person relationship

(Liveware-Liveware) and called it SHELL. Hawkins used the relationship between liveware and software, liveware and

hardware, liveware and environment and liveware and liveware to describe situations that the people encountered or

what happened to them in the working environment. The model does not cover the interfaces that the outside human

factors (Hardware-Hardware, Hardware-Environment and Software-Hardware) and is intended only a basic aid to un-

derstanding human factors [15].

1.2 The modified model for categorizing the human factors in aviation.

We are in the era of organizational accidents [16]. In recent years, there has been a shift in emphasis within the safety

literature away from the individual-level that might be responsible for accident and incidents, and towards organizational

and organization-related factors [17], [18], [10], [8], [19]. When people are at the center of aviation safety, the quality,

capacity, attitude, perception, and training of personnel are important and therefore highlighted. The organizational cul-

ture, organizational climate, managerial model, decision making pattern and aviation safety culture will also affect an

individual [20], [10], [21].Accidents are usually organizational or managerial issues composed of series of errors that are

sometimes difficult for aviation personal to recognize and control. In practice, the International Civil Aviation Organi-

zation‟s (ICAO) Human Factor Training Manual [22] emphasizes the organizational issues of airline maintenance oper-

ations. Furthermore, the International Air Transport Association [23] has five categories for the accident classification

system: human, technical, environmental, organizational, and insufficient data.

1.3 The extended SHELL model and research hypotheses

To examine the importance of the organizational aspect of the aviation maintenance system, we extended the SHELL

model to explicitly include organization as a mediator factor. This extension enables the role played by the organizational

aspect of the aviation maintenance system to be examined, through its interaction with the aviation maintenance personal.

With the extended SHELL model, an aviation maintenance system is described as human factors interfaces in which the

aviation maintenance personal (liveware) as a human factor interact with other human factors including others (live-

ware), physical resources (hardware), non physical resources (software), physical settings (environment), and non

physical settings (organization). In aviation accident analysis, organizational errors in relation to resource management,

organizational climate, and operational processes have been highlighted in order to better understand and manage human

error. These latent organizational failures can directly impact affect supervisory practices, as well as the conditions and

actions of operators [24]. In aviation maintenance, the efficiency and reliability of human performance are influenced by

working conditions, which stem from the overall organizational process [25]. Organization and management decisions

made in the technical support, policies, workforce, finance and safety have significant impacts on the type of human error

that can appear.

As such, an effective (Liveware) aviation maintenance personal interface with less organizational deficiencies

would better help reduce human errors created by other human performance interfaces of the system. In addition, an

effective (Liveware) aviation maintenance personal interface derived from positive and innovative organizational cli-

mate will help the organization operating in a high-risk environment such as an aviation maintenance system to better

manage and more easily adapt to ongoing changes [21]. Mismatches at the above human performance interfaces have

been regarded as sources of human error in which the aviation maintenance personal (liveware) play a vital role. To

examine how this ideal situation has been achieved, it is thus hypothesized that:



H1 There is a positive & direct relationship human factors and human error in aviation maintenance.

H2 There is a positive & direct relationship human factors and organization in aviation maintenance.

H3 There is a positive & direct relationship organization and human error in aviation maintenance.

H4 The impact of human factors on human error in aviation maintenance increases with mediating role of organiza-

tion effort in aviation maintenance

2. Research Method

2.1 Survey instrument

The survey items on the questionnaire for measuring the three constructs of the research model in Fig. 1 were obtain

from existing literature including SHELL model [26],[13],[10],[8],[23]. The expressions of the items were adjusted,

where appropriate, to the context of aviation maintenance. The total of 84 survey items was considered for measuring

the three constructs (Human Factors, Organization, and Human Error). A pre-test was performed with three aviation

maintenance industries on the 58 survey items for the improvement in the content and appearance. The respondent was

asking to complete the questionnaire. The respondents suggested that all statements were appropriate.

2.2 Data collection

A survey questionnaire containing the measurement items was distributed to aviation maintenance personal of all levels

in 30 aviation maintenance industries, including supervisor, instructor, license aircraft engineer and technician. A total

of 315 effective responses were received.

3. Result and Discussion

Structural Equation Modeling (SEM) was used to test and analyze the hypothesized relationship of the research model

in Fig. 1. SEM aims to examine the inter-related relationships simultaneous between a set of posited constructs, each of

which is measured by one more observed item (measure). SEM involves the analysis of two models: a measurement or

factor analysis model and structural model [27]. The measurement model specifies the relationships between the ob-

served measures and their underlying construct, with the constructs allowed to inter-correlate. The structural model

specifies the posited causal relationships among the constructs.

3.1 The measurement model with reliability analysis

A reliability analysis was first carried out on survey data to ensure the internal consistency of the constructs. For ex-

ploratory research, Cronbach‟s alpha should be at least 0.70 or highest for a set of item to be considered and adequate

scale [28]. An exploratory and confirmatory factor analysis was than conducted on a single and multiple constructs to

extract the factors from the items retained after reliability analysis. The items retained are good indicators of their un-

derlying factors extracted, which are used as the observed variables or indicators in the measurement model for meas-

uring their corresponding constructs


118

3.2 The structural model

Figure 1: The research model showing input and output variables with regression weight.

The structural model with a path diagram shown in Fig.1 with the measurement model in Tables 1 and 2 was con-

structed. Ovals represent the constructs (Latent variables), and rectangles represent the factors (observed variables or

indicators). Single headed arrows represent causal relationships between variables. Goodness of fit test was conducted

with the survey data to examine the efficiency of the structural model. The chi-square of the structural model was sig-

nificant (χ2 = 126.530, df = 50, p = 0.000) with the value of ((χ

2/df = 2.531) smaller than 3, indicating ideal fit [29], the

large chi-square value was not surprising, since the chi-square statistic has proven to be directly related to sample size.

To assess the overall model fit without affected by sample size, alternative standalone fit indices less sensitive to sample

size were used. These indices included the goodness of fit index (GFI) the adjusted goodness of fit index (AGFI), the

comparative fit index (CFI), and the root mean square error (RMSEA) [5]. To have a good model fit, GFI should be

close to 0.90, AGFI more than 0.80, CFI more than 0.90, and RMSEA less than 0.10 [30]. An assessment of the struc-

tural model suggested and acceptable model fit (GFI = 0.939 ; AGFI = 0.905; CFI = 0.936; RMSEA = 0.070).

3.3 Significant Human Factors Impacting Human Error in Aviation Maintenance

The purpose of this research is to determine if there is significant human factor impacting human error in aviation

maintenance. The findings of this research reveal that there are significant human factors impacting human error in Ma-

laysia aviation maintenance industry. The research finding supports findings that the human factors have a positive im-

pact to the aircraft maintenance technician [3]. The results of the study also agree with indicated that human errors were

caused by one or several components failures among Software, Hardware, Environment and Liveware in a system [14].

From the path analysis, we can observe that human factors and organization were significant towards dependent vari-

able human error. The significant level was referring 95% confidence level with p-value < 0.001. With reference to the

significant importance, independent organization factors (0.405) were more significant compared to independent human

factors (0.324) which referring to the estimate value stated in the Table 1.

Based on weight and ranking, the order of significance of the five dimensions when we studied the result as pre-

sented in Table 1, with human factors as dependent variable and variables software, hardware, environment, liveware

(I) and liveware (O) as independent variables, it indicated that variable software, hardware, environment, liveware (I)

and liveware (O) were significant at 95% confidence level (p-value ***) with variable software took as reference group.

In this model, variable hardware (0.871) is the most significant factors, followed by variable liveware (I) (0.818), live-

ware (O) (0.768), software (0.754) and environment (0.714). Furthermore, when we took independent variable quality

support as reference factor, we found that independent variables quality support, company policy, workforce, finance

strategy and safety culture were significant in 95% confidence level (p-value ***). Finance strategy (0.883) is the most

significant factors influence organization. It continues the significant with other independent variables company policy

(0.836), workforce (0.818), quality of support (0.803) and safety culture (0.7.41).



Table 1: Regression weight of the factors in path model.

Unstandardized Estimate p-value Standardized Estimate

SW <------- HF 1.000 *** .754

HW<------- HF 1.214 *** .871

ENV <------- HF 1.069 *** .714

LW(I)<------- HF 1.110 *** .818

LW(O) <------- HF 1.120 *** .768

QS <------- ORG 1.000 *** .803

CP <-------ORG .920 *** .836

WF <------- ORG .908 *** .818

FS <------- ORG .878 *** .883

SC <------- ORG .867 *** .741

INST<------- HE 1.000 *** .781

SR <------- HE .910 *** .848

3.4 Hypotheses Testing

In SEM analysis, the relationships among independent and dependent variables (constructs) are assessed simultaneously

via covariance analysis. Maximum Likelihood (ML) estimation was used to estimate model parameters with the co-

variance matrix as data input. The ML estimation method has been described as being well suited to theory testing and

development [27],[30]. Two sets of independent variables and dependent variables are used for testing research hy-

potheses H1-H4 respectively. The first set has the human factors construct and the organization as independent variable,

and human error as dependent variables. Fig. 1 shows the result of the structural model. The values associated with each

path (hypothesized relationship) are standardized path coefficients. These values represent the amount of change in the

dependent variables for every single unit of change in the independent variable. For example, an increase of one unit in

the human factors construct will cause an increase of one unit in the human error construct. Solid lines indicate sup-

ported relationship respectively.

Table 2: Regression weight between the constructs

Unstandardized Estimate p-value Standardized Estimate

HE <----- HF .569 *** .324

ORG <----- HF .501 .008 .252

HE<----- ORG .632 *** .405

The standardized regression weight and p-values for structural relationships as shown in Table 2.The result shows

that, the standardized regression weight for H1was found to be 0.324 (p-value < 0.001). This result was support to H1

that the HF has direct and strong impact on HE . Table 1 also presents the relationship between HF and ORG efforts.

The standardized regression weights for the hypothesized relationship between HF and ORG was found positive (0.252)

and insignificant (p-value > 0.001), the result does not provide support to H2 that the HF have a direct and strong im-

pact on ORG effort. The standardized regression weight for the direct relationship between ORG effort as found to be

positive (0.405) and significant (p-value < 0.001) confirming H3 that ORG had strong positive direct impact on HE.

The empirical support for mediating role of organization efforts in the context of relationship between human fac-

tors and human error is hardly found. In the case of aviation industry, organization may intermediate between HF and

HE. This discussion leads to the following hypotheses: H4: The impact of significant human factors on human error in

aviation maintenance increases with mediating role of organization effort in Malaysia aviation maintenance industries.

These theoretical discussion and proposed hypothesized relationships are deliberating in Figure 1. Human error is also

indirectly affected by HF through ORG efforts. In order to test whether ORG efforts are an important mediator of HF

with HE relationship the following rule of thumb will be followed [30][31].

i. IE < 0.085 and IE => Non mediator

ii. IE > 0.085 and IE ~ DE => Partial Mediator (HF -> HE relationship, p < 0.05)

iii. IE > 0.085 and IE > DE => Total Mediator (HF -> HE relationship, p > 0.05)

The standard Indirect Effect (IE) of HF to HE is 0.103 which is more than 0.085 (Table 3). Thus, ORG efforts me-

diate the relationship between HF and HE. Since, the p-value for Direct Effect (DE) between HF to HE is less than 0.05

thus, ORG efforts are a partial mediator. In conclusion, this finding provides support to H4 hypotheses which is the im-

pact of HF on HE increases with a mediating role of ORG efforts in Malaysia aviation maintenance industry.


120

Table 3: Direct effect (DE) and Indirect Effect (IE) analysis for Malaysia aviation maintenance industry

Note: Std. Total Effect = Std. Direct Effect (DE) + Std. Indirect Effect (IE)

4. Conclusion

The empirical findings of a questionnaire survey of 315 aviation personnel in Malaysia show that the model and ap-

proach are both strategically effective and practically acceptable for categorizing the significant human factors.The re-

sult reveal that the aviation maintenance companies may want to propose management strategies related to the signifi-

cant factors to minimize the human error. Our findings also suggest that the Civil Aviation Authority may consider

asking management level groups in aviation companies such as human recourses and maintenance departments, to focus

on significant factors to improve aircraft maintenance performance and reducing error. Specifically, these significant

factors are related to the hardware, liveware (I), environment, and liveware (O). Aviation maintenance companies also

have to focus other significant factors under organizational such as financial strategy, policies, and manpower and

safety culture. When employee professionalism is protected and the individual staff members have the company atten-

tion, safety and human error costs should be reduced.

5. References

[1] Latorella, K.A.,Prabhu, P.V., 2000. A review of human error in aviation maintenance and inspection. International Journal of

Industrial Ergonomics 26 (2), 133–161.

[2] Pettersen, K.A.,Aase, K., 2008. Explaining safe work practices in aviation line maintenance. Safety Science 46, 510–519.

[3] Y.-H. Chang, Y.-C. Wang 2010.Significant human factors in aircraft maintenance technician, Safety Science 48, 54-62

[4] CAA, 2002a. CAP 715: An Introduction to Aircraft Maintenance Engineering Human Factors for JAR 66. UK Civil Avia-

tion Authority.<http://www.caa.co.uk/docs/33/CAP715.PDF (accessed on 15 May 2009).

[5] Janic, M., 2000. An assessment of risk and safety in civil aviation. Journal of Air Transport Management 6, 43–50

[6] Lee, W.K., 2006. Risk assessment modeling in aviation safety management. Journal of Air Transport Management 12 (5),

267–273.

[7] Wong, D.K.Y.,Pitfield, D.E., Caves, R.E., Appleyard, A.J., 2006. Quantifying and characterizing aviation accident risk factors.

Journal of Air Transport Management 12, 352–357.

[8] CAA, 2007. Aircraft Maintenance Incident Analysis. UK Civil Aviation Author-

ity.<http://www.caa.co.uk/docs/33/Paper2007_04.pdf. (accessed on 15 May 2009).

[9] ASC, 2005. Aviation Occurrence Report I, In-flight breakup over the Taiwan Strait, Northeast of Makung, Penghu Island,

China Airlines Flight CI611, Boeing 747- 200, B-18255, May 25, 2002, (ASC-AOR-05-02-001). Aviation Safety Council,

Taipei, Taiwan

[10] CAA, 2003. CAP716: Aviation Maintenance Human Factors. UK Civil AviationAuthor-

ity.<http://www.caa.co.uk/docs/33/CAP71PDF (accessed on 15 May 2009).

[11] Endsley, M.R., Robertson, M.M., 1996. Team situation awareness in aviation maintenance. In: Proceedings of the Human Fac-

tors and Ergonomics Society 40th Annual Meeting. Human Factors and Ergonomics Society, Santa Monica, CA, pp.

1077-108

[12] Edwards, E., 1995. Aviation ergonomic: whence and whither? Ergonomics 38 (3),565–569

[13] CAA, 2002b. CAP 718: Human Factors in Aircraft Maintenance and Inspection. Civil Aviation Author-

ity.<http://www.caa.co.uk/docs/33/CAP718.PDF (accessed on 15 May 2009).

[14] Hawkins, F.H., 1993. Human Factors in Flight.Ashgate, Aldershot, England.I ATA, 2006. Safety Report. International Air

Transport Association. Geneva, Switzerland/Montreal, Canada.

[15] ICAO, 2003. Human Factors Guidelines for Aircraft Maintenance Manual, first ed. international Civil Aviation Organization.

Doc. 9824-AN/450<http://www2.hf.faa.gov/opsManual/assets/pdfs/ICAOHF.pdf(accessed on 15 May 2009).

[16] Reason, J., 1990. Human Error. Cambridge University Press, Cambridge, England

[17] Westrum, R., 1996. Human factors experts beginning to focus on organizationalfactors in safety. ICAO Journal (October).

[18] Neal, A., Griffin, M.A., Hart, P.M., 2000. The impact of organizational climate onsafety climate and individual behavior.

Safety Science 34, 99–109.

Std. Total Effect Std. Direct Effect Std. Indirect Effect.

HF ORG HE HF ORG HE HF ORG HE

ORG 0.324 0.000 0.000 0.324 0.000 0.000 0.000 0.000 0.000

HE 0.423 0.405 0.000 0.324 0.405 0.000 0.103 0.000 0.000



[19] Parker, D., Lawrie, M., Hudson, P., 2006. A framework for understanding thedevelopment of organizational safety culture.

Safety Science 44, 551–562

[20] McDonald, N., Corrigan, S., Daly, C., Cromie, S., 2000. Safety management systems and safety culture in aircraft maintenance

organizations. Safety Science 34 (1–3), 151–176.

[21] Arvidsson, M., Johansson, C.R.,Ek, A., Akselsson, R., 2006. Organizational climate in air traffic control: innovative prepared-

ness for implementation of new technology and organizational development in a rule governed organization. Applied Ergo-

nomics 37, 119–129.

[22] ICAO, 1998. Human Factors Training Manual, first ed. International Civil AviationOrganization, Doc. 9683-AN/950.

[23] IATA, 2006. Safety Report. International Air Transport Association. Geneva, Switzerland/Montreal, Canada.

[24] Wiegmann, D.A.,Shappell, S.A., 2003. A Human Error Approach to Aviation Accident Analysis: the Human Factors Analysis

and Classification System. Ashgate, Burlington, VT.

[25] Isaac, R., Ruitenberg, B., 1999. Air Traffic Control: Human Performance Factors. Ashgate, Aldershot, England.

[26] ICAO, 2003. Human Factors Guidelines for Aircraft Maintenance Manual, first ed. international Civil Aviation Organization.

Doc. 9824-AN/450. <http://www2.hf.faa.gov/opsManual/assets/pdfs/ICAOHF.pdf (accessed on 15 May 2009).

[27] Anderson, J.C.,Gerbing, D.W., 1988. Structural equation modeling in practice: a review and recommended two-step

approach. Psychological Bulletin 103 (3), 411–423.

[28] Nunnelly, J.C., 1978. Psychometric Theory, second ed. McGraw Hill, New York

[29] Bentler, P.M., 1988. Theory and Implementation of EQS: a Structural Equations Program. Sage, Newbury Park, CA.

[30] Hair Jr., J.F., Anderson, R.E., Tatham, R.L., Black, W.C., 1998. Multivariate Data Analysis with Readings, fourth ed. Prentice

Hall, New Jersey.

[31] N.M. Zakuan, 2009. Structural Analysis of Total Quality Management, ISO/TS16949 and Organization Performance in Malay-

sia and Thailand Automotive Industry. Unpublished ,UniversitiTeknologi Malaysia.


122



Improving Process Performance Through Quality

Engineering Using TAGUCHI Robust Design

Arum Sari*;Halida Intan Industrial Engineering, Pasundan University, Bandung, Indonesia

*[email protected]

ABSTRACT

Experiment design is a powerful technique for understanding a process and studying the impact of potential variable

affecting a process. Robust design is a methodology for making process performance insensitive to variation in manu-

facturing condition and environment. Taguchi method of robust design in optimizing process parameter setting has

widely used. Although it has been widely accepted for improving variability in manufacturing process, research has

shown very little has been done on the application of such powerful methodology. This paper investigates how the ro-

bust design of parameter setting can improve the quality performance of a product. The research has shown that the

proposed robust design of setting parameter, has significantly improved the probability of nonconforming unit. The

paper also discusses the quality performance of the robust design compares to non robust design. The result has shown

that the robust design has produced better probability of nonconforming units compared to non-robust design. Benefit

of the proposed parameter setting to cost reduction is also investigated.

Keywords:Taguchi, DOE, Robust Design, Parameter setting, TQM, Quality Improvement, S/N Ratio

1. Introduction

The quality of a process is technically measured from variance. High variance is potential to produce high nonconforming

unit.Settingthe parameters is one of the ways on reducing the variation. The tool for achieving parameter design is the

design of experiment. Effort should be directed toward determining the best design at the least cost. Focus on reducing

variability without adding cost can be achieved by defining a strategy to minimize the effect of these causes. Taguchi

experiment is one of the ways to carry out an effective experiment and robust design is a methodology in Taguchi ex-

periment for making product‟s performance insensitive to variation. Although it has been widely accepted for tackling

variability problems in manufacturing processes, research has shownthat very little has been done in the manufacturing

sector in Indonesia. This paper will explain how to improve quality performance through robust design of parameter

setting. PT. Mitrametal Perkasa will be used as a case of this study.

2. Problem Statement

PT. Mitrametal Perkasa produces the components of motor vehicle. Brake hose is the main product which has very poor

quality performance. Most of all is out of specification and it should be reworked to meet the specification. The quality

improvement program was still focused on quality control, while the parameter settings were still conducted by trial with

no significant result. A redesigned parameter setting is an alternative to provide a significant improvement. Parameter

settings will be design robustly based on Taguchi experiment. The research questions arehow to design a robust parameter

setting, how muchtheimprovement result, and whether the performance of arobust design is outperformed. The research

methodology is described in Section 3, planning and designing the experiment in Section 4, conducting the experiment in

Section 5. Analysis of the experiment is offered in Section 6, while analysis of the robust design is inSection 7. The

conclusion is presented at the end of the paper.

3. Research Methodology

The experiment will be performed through four distinct phases. The first phase is planning the experiment. It includes

forming the team, determining the objective, identifying quality characteristics, determining measurement methods,

selecting factors and level factors, specifying variable setting, and identifying potential interaction. The second phase is

designing the experiment. It involves calculating degree of freedom, selecting the orthogonal array(s), construction of

the experiment layout, assigning of the factors and interactions of interest to the array(s). The third phase is conducting

Improving Process Perfomance Through Quality Engineering Using TAGUCHI Robust Design

124

the experiment. It includes execution of the experiment as developed in the planning and design phases. This phase in-

cludes development of the test plan and performing the experimental runs. The last phase is analyzing the experiment. It

relates to calculation for converting raw data to the S/N ratio to get a robust design of the parameter. Analysis includes

determining the most important factors, selecting the optimal levels for those factor settings using tabular and graphical

techniques. The last step at this phase is conducting the confirmation run at the optimal settings for checking the repro-

ducibility and to identify the improvement result of the program.

4. Planning And Designing The Experiment

After the appropriate team has been selected, they are working together to improve the quality. The objective of the

study is to minimize percent defective of the product in order to reduce reworked units. Pareto diagram of historical

data of percent defective is used to select the department, the product, the process and the quality characteristic to be

considered. There are 6 departments available in the plant namely(1) Casting, (2)Machining, (3) Painting, (4) Brake Hose,

(5) Lining Bonding, and (6) Stamping. Pareto diagram showsthat Brake Hose department that has the largest percent

defective (38%) is selected. There are four type of products produce in Brake Hose department namely (1) “2W Front

Brake Hose All Type”, (2) “2W Rear Brake Hose All Type”(3)” 4W Front Brake Hose All type” and (4) “ 4W Rear

Brake Hose All Type”. The Pareto diagram define that product “2W Front Brake Hose all types” whichhas the largest

percent defective (29%) is selected. Based on the flow process diagram of the “2W Front Brake Hose all types”, the

pareto diagram define that “Crimping 2”is selected. There are five quality characteristics of the “Crimping 2” process.The

Pareto diagram for each quality characteristics of the “Crimping 2 ” processdefine that there are three quality character-

istics which has 86% contribution. They are (1)Crimping diameter, (2) Skirt Width, and (3) Side Line Out. Another

important in the objective of the experiment determination is:”Not to try to solve the whole problem in one experiment”

(Glenn,S.P.,1993). For this reason, the “Crimping Diameter” which is categorized as a Nominal Is The Bestcharacter-

istic, is put into first priority. It is clearly define that the objective of the experiment is to reduce the defect of

“Crimping Diameter” in the “Crimping 2” process for product type “2W Front Brake Hose all types “ in “Brake hose”

department as a program for quality improvement at PT. Mitrametal Perkasa. Cause and effect diagram is used to select

the factors and the level factors and check sheet is used to define the root causes of the problem. Analysis of the root

causes results that there are 4 factors to be considered. They are (1) heat input, (2) heating rate, (3) pressure, and (4)

humidity. Heat input‟s parameters are electrical current and voltage, and heating rate‟s parameter is heating time.

Analysis of the root causes has identified 5 factors to be studied. The factors are, (A) electrical current, (B) voltage, (C)

heating time, (D) pressure and (E) humidity. Factor A,B,C,and D are classified as control factors while factor E is a noise

factor. Each factor has two levels.It is also interested in the interaction of AxB, AxC, and BxC. Summary of the factors,

level factors and the associate setting is presented in Table 1, Selection of the factor and level factor resulting 7 degree

of freedom for the control factor. The appropriate orthogonal array for the selected factors is L8. The control factor is

placed in an inner array and the noise factor is placed in an outer array. The layout of such design is shown in Table 2,

Noise factor ofE that has two levels is treated as a replication and placed in the outer array. L8 means that the experi-

ments consists of 8 experiment runs that are repeated 2 times each according the presence of one noise factor at two levels.

There will be a total of 16 experimentsto be conducted. For each setting of the design factor, a mean of the experiment

responses is calculated. Assignment of factor and the interaction into Columnin the Orthogonal Array is done using

standard Linear Graph of L8. Using the Linier Graph, the assignment of factor A, factor B, interaction AxB, Factor C,

Interaction AxC, Interaction BxC and factor D is placed consecutively in column 1to7 as shown in Table2.

Table 1. Factors and Level factors

Types Factors Symbol 1st Level 2nd level

Control Factor Voltage A 12 Volt 9 Volt

Current electric B 60A 90A

Heating Time C 5 sec 3 sec

Pressure D 50 MPa 40 MPa

Noise Factor Humidity E Low High

5. Conducting The Experiment

The experiment is conducted in 8 experimental runs. For each experimental runs or combination of factors, two repeti-

tions are performed. The first repetition is conducted at the first levelof noise factorand the second repetition is conducted

at thesecond level of noise factor. Since there are two repetitionsper experimental runs, there will be two responses for

each run as presented in Table 2 column 9 and 10, then the mean responses for each run experiment is calculated and

presented in Table 2 column 11. Due to the objective of the experiment is to design a setting parameter robustly, thenthe



mean responses(column 11) must be converted into the S/N ratio.Because the characteristic of Crimping Diameter is

nominal the best, then the suitable S/N ratio is calculated as presented in column 12 of Table2.

Tabel 2. Responses of The Experiment

Eksperimen Outer Array (Noise Factor) Resp. 1

Resp. 2

Mean response

S/N Ratio Inner Array (Control Factors)

A B AXB C AXC BXC D

1 1 1 1 1 1 1 1 10.40 10.30 10.350 43.31

2 1 1 1 2 2 2 2 10.45 10.50 10.475 49.43

3 1 2 2 1 1 2 2 10.55 10.45 10.500 43.43

4 1 2 2 2 2 1 1 10.65 10.40 10.525 35.50

5 2 1 2 1 2 1 2 1055 10.60 10.575 49.53

6 2 1 2 2 1 2 1 10.50 10.45 10.475 49.43

7 2 2 1 1 2 2 1 10.45 10.35 10.480 43.35

8 2 2 1 2 1 1 2 10.55 10.65 10.600 43.52

To define the effect of each factor and the associate levels, a response table is developed. This is performed by

grouping the mean responses by factor level for each column in the array, taking the sum and dividing by the number of

responses. For example, the effect of factor A level 1 (A1 ) is the average of the S/N ratio resulted from experiment 1,2,3,

and 4, while the effect of factor B level 2 (B2) is the average of S/N ratio resulted from experiment 1,2,5 and 6. The

overall result of the effect factor is presented in Table3. Based on the average response computed for each factor and

interaction, a means response graph is constructed (Figure 1). The absolute difference between the two average re-

sult(from 2 levels) is the effect of the factors and interaction.Based on Table 3, the proposed robust design of setting

parameters of Crimping 2” process are set as follows:electric current at 1st

level (60 A), pressure at 2nd

level (40 MPa),

voltage at2nd

level (9 Volt), heating time at 1stlevel (5 sec) or 2

ndlevel (3 sec).

Tabel 3. Response Table of The Effect Factor

Factor A B AxB C AxC BxC D

LEVEL 1 42.918 47.923 44.903 44.903 44.923 42.959 42.897

LEVEL 2 46.454 41.449 44.470 44.470 44.449 46.413 46.475

RANK 3 1 7 6 5 4 2

(a) (b) (c) (d) (e) (f) (g)

Figure 1. Graphic Factor Effects based on the S / N Ratio

6. Analysis Of Variance

Analysis of variance is conducted to examine the effect of the factors under study. Analysis of variance begin with the

calculation ofthe sum of square andthe mean square of the S/N ratio. A pooling up approach is used to find the significant

effect of the factors. Pooling up is done starting from the factor or interaction factor which has the smallest sum of

square. Column 3 of Table 4 shows that the smallest sum of squareis 0.37 and it belongs to the interaction factorAxB, so

that thesum of square (AxB) is combined with the sum of square error.The overall result of recalculation of the sum of

square, themean square and the F ratio for all factors except for factors that have been pooling up (AxB interaction)is

shown in Table 4. For this case, pooling up is done three times and the result is summarized in Table 4.

30

40

50

1 2

S/N

LEVEL

Faktor A

30

40

50

1 2

S/N

LEVEL

Faktor B

30

40

50

1 2

S/N

LEVEL

Faktor AXB

30

40

50

1 2

S/N

LEVEL

Faktor C

30

40

50

1 2

S/N

LEVEL

Faktor AXC

30

40

50

1 2

S/N

LEVEL

Faktor BXC

30

40

50

1 2

S/N

LEVEL

Faktor D


126

Table 4. Analysis of Variance

` POOLING I POOLING II POOLING III

Factor V SS MS SS MS F SS MS F SS MS F

A 1 25.01 25.01 25.01 25.01 66.82 25.01 25.01 66.82 25.01 25.01 62.61

B 1 83.83 83.83 83.83 83.83 223.9 83.83 83.83 223.9 83.83 83.83 209.86

AXB 1 0.37 0.37 Pooling I

C 1 0.37 0.37 0.37 0.37 1.00 Pooling II

AXC 1 0.45 0.45 0.45 0.45 1.20 0.45 0.45 1.20 Pooling III

BXC 1 23.86 23.86 23.86 23.86 63.75 23.86 23.86 63.75 23.86 23.86 59.73

D 1 25.60 25.60 25.60 25.60 68.40 25.60 25.60 68.40 25.60 25.60 64.90

Error 0 0.37 0.37 0.37 0.37 0.75 0.37 1.20 0.40

total 7 159.5 159.5 159.5 159.5

Using 90% confidence intervals, thevalue of the F table (F (0.10, 1, 1))is 39.9. That means after pooling up the in-

teraction factor AxB, there is influence of factor A to the CrimpingDiameter. The process is recalculated to find the

influence of all factors. The end result for each pooling is summarized in Table5 while the contribution percentage for

each factor and interaction is presented in Table6. It isconcluded that factors A,B,Dand the interaction factor of BxC has

a significant effect. The proposed setting parameter are as follow: electric current is set at first level ( 60 A), pressure

is set at second Level (40 MPa), voltage is set at the second level ( 9 V) andheating time is set at first level( 5sec).A

number of units built according to the recommended setting should be tested for confirmation of the result.For that

purpose, 25 observations of four subgroup sizeis conducted. The same calculation is done for the new experiment re-

sponses resulting μ = 10.548, variance 0.016 and probability non nonconforming unit almost zero.The confirmation

results compare against the initial condition which has μ=10.477 and σ=0.051 and the probability ofnonconforming unit

is 96.4%. This means that the proposeddesign of setting parameter has significantly improve the quality performance.

Table 5. Test of Hypothesis of Factors

Effect Factor POOLING 1 POOLING 2 POOLING 3

FT FC Ho FT FC Dec FT FC Dec

A 39.9 66.8 Re 8.5 66.8 Re 8.5 62.6 Re

B 224 Re 224 Re 209.9 Re

AX B Polling I

C 1.0 Acc Polling II

A X C 1.2 Acc 1.2 Acc Polling III

B X C 63.7 Re 63.7 Re 59.7 Re

D 68.4 Re 68.4 Re 64.90 Re

Table 6. Percent Contribution (r) of factors

Factors v SS MS SS‟ R(%)

A 1 25.01 25.01 24,61 15

B 1 83.83 83.83 83.43 52

BXC 1 23.86 23.86 23.46 15

D 1 25.60 25.60 25.20 16

Error 3 1.20 0.40

Total 7 159.50

7. Analysis of Robust Design

To conclude that the robust design performs better than the non robust design, the same calculation is repeated instead of

based on the value of S / N ratio but it is based on the mean response as listed in columns 11 ofTable 2. The propose

setting parameters of the non robust design resulting a slightly different as shown in row 2 Table 7

Table 7. Confirmation Results of Robust Design and Non Robust Design

CONTROL FACTORS CONFIRMA-

TION RESULT

PROB OF

REJECTIN

(%) A B C D µ σ

tsuboR

ngiseD

Effect √ √ x √ 10.55 0.033 0.0002

Level 2 1 1 2



CONTROL FACTORS CONFIRMA-

TION RESULT

PROB OF

REJECTIN

(%) A B C D µ σ

9 V 60 A 5 Sec 40 Mpa

Non-Robust

Design

Effect x x x √ 10.59 0.052 27.000

Level 2 2 2 2

9 V 90 A 3 sec 40 Mpa

It is clear that a robust design outperforms the quality performance of non - robust design. Probability of noncon-

forming unit of the robust design is 27% greater than the non-robust design. Assuming that the daily throughput is 700,

then non-robust parameter settingswill produce 189 units nonconforming. While a robust design only produced one unit

non-conforming. Assumed that the costs of scrap unit is 210,000 and reworks unit is 3000, then the daily savings from the

proposed robust design of the setting parameter is 24.300.000or 680.400.000 monthly

After the improvement at the“Crimping 2” processis completed, the next process to beimproved based on Pareto

Diagram is the process of “Cutting”.Afterallprocessof product 2WFrontBrakeHoseAllType has been improved,the next

product to be improved based on Paretodiagram is “2WRearBrakeHoseAllType”and then moving forward to depart-

ment of machining, casting, and stamping.

8. Conclusion

It is concluded that the proposed robust design of parameter settings has improved the quality significantly. The prob-

ability of nonconforming units was reduced from 96.44% to 0002%. Robust design has produced a better quality per-

formance, compared to nonrobust design.

9. Reference

[1] Fawlkes, W.Y.,Creveling, C. M., 1995, “Engineering Method For Robust Design Using Taguchi Method In Technology And

Product Development”. AdisonWesly Publishing Company.

[2] Mitra , A. ,1998, “Fundamental Of Quality Control And Improvement”, 2nd

edition Prentice-Hall Inc

[3] Peace, G.S., 1993, “Taguchi Method: A Hands On Approach”, AdisonWesly Publishing Company.

[4] Ross, J.E., 1994, “Total Quality Management, Text, Cases And Reading”, 2nd

edition, London, Kogan Page Limited.

[5] Ross, P. J., 1996, “Taguchi Technique For Quality Engineering”, 2nd

edition, Mc Graw Hill

[6] Taguchi, G., Elsayed, E.A. Hang Siuang, T. C., 1989, “Quality Engineering In Production System”, Mc GrawHilSimak

Baca secara fonetik

Simak

Baca secara fonetik

Simak

Baca secara fonetik


128

bagian 1-128_2

Documents