Eth 41460 02

Research Collection
Doctoral Thesis
Issues of sustainability in engineering decision analysis
Author(s):
Nishijima, Kazuyoshi
Publication Date:
2009
Permanent Link:
https://doi.org/10.3929/ethz-a-005762828
Rights / License:
In Copyright - Non-Commercial Use Permitted
This page was generated automatically upon download from the ETH Zurich Research Collection. For more
information please consult the Terms of use.
ETH Library
DISS. ETH NO. 18238
Issues of sustainability in engineering decision analysis
A dissertation submitted to
ETH ZURICH
for the degree of
Doctor of Sciences
presented by
Kazuyoshi Nishijima
Master of Environmental Studies, University of Tokyo
born 13.08.1978
citizen of
Japan
accepted on the recommendation of
Prof. Dr. Michael Havbro Faber, examiner

Prof. Lucas Bretschger, co-examiner
Prof. Niels Lind, co-examiner
2009
-2-
Acknowledgements
"You are one of the freest PhD students in the world," my supervisor Professor Faber
once said to me - and yes, I am. He gave me this topic and motivated me to conduct
this research work. However, during the course of the work he never gave me any
"fixed" assignments. Instead, he gave me the opportunity to explore different research
disciplines. At the same time, whenever necessary, he gave me exceptional support,
both professionally and personally. Here, I express my deep gratitude to him.
My colleagues at the group of Risk and Safety have created a friendly, pleasant
atmosphere, conductive to research work. It is thanks to them that I could always
concentrate on the research work. Among others, I express my wholehearted thanks to
Matthias Schubert, who was my office mate over the last few years. Whenever I
encountered technical or personal problems, he generously gave of his time to discuss
them with me: he understood the problems and provided me with useful advice. I also
express my gratitude to Patricia Meile, who is a secretary in our research group. She
constantly supported me in setting up and improving the working environment,
including administrative issues such as acquiring working permission for a foreign
student. Dr. Daniel Straub, who is a senior ex-colleague in the group, greatly
influenced my motivation towards research and my research style, directly and
indirectly, which I appreciate very much.
I also want to thank all my friends and especially my family for their support. Their
support has encouraged me to continue with the PhD work. This work is dedicated to
them.
Finally, my gratitude goes to Professors Bretschger and Lind for acting as external
examiners and providing valuable comments.
Zurich, 28.02.2009
Kazuyoshi Nishijima
-3-
-4-
Abstract
Sustainable societal development has become a subject of increased and widespread
societal attention especially during the last two decades. The tremendous economic
development of former developing nations such as China and India and the general
impact of globalization have put even larger pressures on our limited natural resources
and fragile environment. Faced with an ever increasing amount of evidence that the
activities of our own generation might actually impair the possibilities for future
generations to meet their needs, it has become a major political concern that societal
development must be sustainable. The issuing of the famous Brundtland report “Our
Common Future” (1987) formed a political milestone. This important event has
enhanced the public awareness that substantial changes of consumption patterns are
called for and has further significantly influenced research agendas worldwide.
The realization of a sustainable development of society necessitates that a holistic

perspective is taken in operational and strategic societal decision-making. In principle,
a joint consideration of the preferences, needs and capabilities of the present and future
generations across all nations, industrial and public sectors is required if we are to fully
succeed in achieving sustainable societal development. It may be realized that
decisions made to enhance sustainability of societal development not only concern
reduced emissions of pollutants but also directly and indirectly involve a redistribution
of globally available resources and not least a reassessment of the societal affordability
of lifestyle and quality of life. So far, the available research literature in this field has
mainly reported on results relating to individual aspects of sustainable development; as
of yet a general framework that facilitates the joint consideration of the many
dimensions of sustainability in supporting decision-making for sustainable societal
development is still missing.
Whereas the development of a general framework for sustainable decision-making is

one of the most relevant tasks in the research agenda, it is unlikely that this task could
be accomplished in the foreseeable future. However, at the same time, there is an
urgent need for methods that enable societal decision-makers to identify "sustainable"
policies in different sectors of society. Here, the "sustainable" policies imply policies
that conform to current preventive measures, regulations, principles, ethics and
whatever else is regarded as best practice for the realization of the sustainable
development of society.
Motivated by this and focusing on the civil engineering sector, the present thesis has
two aims. The first aim is to reformulate the classical life-cycle cost optimization
concept, which has been advocated in civil engineering as the decision principle, in
such a way that relevant aspects of sustainability can be incorporated into engineering
decision-making. The aspects of sustainability considered in depth in this
-5-
reformulation are intergenerational equity and allocation of limited resources.
Furthermore, for the purpose of facilitating the applications in practical decision
situations, a platform is proposed for the modelling and optimization of decision
problems based on Bayesian probabilistic networks. Thereby, it is possible with the
proposed platform to consider the constraints relating to societal sustainability posed
by present society in the decision problems. The second aim is to present a
fundamental approach for incorporating the reliability of civil infrastructure in general
economic models so that the sustainable policies on design and maintenance of civil
infrastructure can be identified from a macroeconomic perspective.
In the present thesis, two types of engineering decision analyses are differentiated in
order to clarify the extent of the consequence of decisions; marginal engineering
decision analysis and non-marginal engineering decision analysis. In marginal
engineering decision analysis, it is assumed that the economic growth path is
exogenously given and the consequence of decisions does not affect the economic
growth; the life-cycle cost optimization concept corresponds to the marginal
engineering decision analysis; the first aim of the present thesis can be regarded as the
formulation of engineering decision problems from a sustainability perspective in the
context of the marginal decision analysis. In contrast, non-marginal decision analysis
considers the change of economic growth as a consequence of decisions; the second
aim of the present thesis can be regarded as a proposal for a decision framework for the
non-marginal engineering decision analysis.
The present thesis consists of eight chapters. Chapter 1 introduces the background,
aim, scope and outline of the thesis. A literature survey is also provided in the fields of
economics and civil engineering, where the formulation and optimization of
sustainable decision making in civil engineering is dealt with. The core of the present
thesis consists of six chapters (Chapters 2 to 7). Each of the chapters, except Chapter 7,
represents a part of my research work published during the PhD study. Chapter 2
considers the general treatment of uncertainties in engineering decision analysis, which
is the philosophical basis for decision-making subject to uncertainties. Chapters 3 to 5,
respectively, investigate the modelling and optimization of sustainable decision
problems, the issue of intergenerational equity and the issue of allocation of limited
resources in the context of marginal engineering decision analysis. In Chapter 6 the
approach for incorporating the reliability of civil infrastructure in general economic
models is proposed based on economic growth theory. This approach corresponds to
non-marginal engineering decision analysis. The proposed approach is then applied to
a simplistic economic model in Chapter 7 in order to show how the optimal reliability
of civil infrastructure can be identified and the sustainable policy on the design and
maintenance of civil infrastructure can be examined. Thereby, an objective function is
derived in the context of non-marginal decision analysis that is different from the
objective function employed in the classical life-cycle cost optimization concept. The
reason for this is provided by looking at the differences in the formulation of the
-6-
decision problems in marginal and non-marginal decision analysis. In this chapter the
assumptions of the derivation of the classical life-cycle cost optimization and its
limitations are also introduced in order to emphasize the difference between
non-marginal decision analysis and marginal decision analysis. Chapter 8 concludes
the present work.
In the reformulation of the classical life-cycle cost optimization, its practical

applicability is emphasized. Hence, the proposed methods in the corresponding
chapters (Chapters 3 to 5) can be readily applied to practical decision situations.
Practical examples are provided in these chapters. On the other hand, the approach
presented in Chapters 6 and 7 serves as a relevant building block for further
development of the general framework for sustainable decision-making, whereby
scientific insights are provided on how sustainable design and maintenance policies on
infrastructure can be investigated in a macroeconomic context.
-7-
-8-
Zusammenfassung
Die Frage nach einer nachhaltigen gesellschaftlichen Entwicklung hat insbesondere in
den letzten zwei Jahrzehnten zunehmend an Bedeutung gewonnen. Im Fokus stehen
dabei die begrenzten natürlichen Ressourcen und die fragile Umwelt, die durch die
enorme wirtschaftliche Entwicklung von Schwellenländern wie China und Indien noch
stärker unter Druck geraten. Da es immer offensichtlicher wird, dass die Aktivitäten
unserer eigenen Generation die Entwicklungsmöglichkeiten der folgenden
Generationen beeinträchtigen könnten, wurde die Forderung nach einer nachhaltigen
gesellschaftlichen Entwicklung ein wesentliches politisches Ziel. Ein politischer
Meilenstein wurde 1987 durch den Brundtland Report "Unsere gemeinsame Zukunft"
gesetzt. Dieses entscheidende Ereignis verstärkte das öffentliche Bewusstsein, dass
substantielle Änderungen im Konsumverhalten zukünftig notwendig sind. Seit der
Veröffentlichung des Brundlandt Reports beeinflusst das Thema der Nachhaltigkeit
weltweit viele Agenden von Forschergruppen.
Die Umsetzung einer nachhaltigen gesellschaftlichen Entwicklung erfordert eine

Einnahme einer holistischen Perspektive sowohl für die operationelle als auch für die
strategische Entscheidungsfindung in der Gesellschaft. Prinzipiell ist eine integrale
Berücksichtigung der Präferenzen, Bedürfnisse und Fähigkeiten der heutigen und der
zukünftigen Generationen über alle Nationen und alle Sektoren hinweg notwendig,
wenn eine Steuerung hin zu einer nachhaltigen gesellschaftlichen Entwicklung
erfolgreich sein will. Es muss erreicht werden, dass Entscheidungen zur Förderung der
nachhaltigen Entwicklung einer Gesellschaft nicht nur unter Berücksichtigung
monokausaler Zusammenhängegetroffen werden, z.B. die Verringerung von
schädlichen Emissionen, sondern auch unter Berücksichtigung der direkten und
indirekten Umverteilung globaler Ressourcen, der Neubewertung von Lebensstilen und
nicht zuletzt der Qualität des Lebens in der globalen Welt. Der Grossteil der
verfügbaren wissenschaftlichen Literatur zum Thema Nachhaltigkeit fokussiert auf
einzelne Aspekte, die für eine nachhaltige Entwicklung notwendig sind. Ein genereller
Rahmen, der die gemeinsame Betrachtung des mehrdimensionalen Problems der
Nachhaltigkeit erlaubt und gesellschaftliche Entscheidungsträger unterstützen kann,
fehlt bisher noch.
Die Entwicklung eines solchen Rahmens ist die relevanteste Aufgabe, die die Forscher
im Bereich der nachhaltigen Entscheidungsfindung zu bewältigen haben. Es ist nicht
abzusehen, dass in naher Zukunft in diesem Bereich eine Lösung gefunden wird.
Dennoch ist derzeit der Druck gross, Methoden zur Verfügung zu haben, die es
Entscheidungsträgern aus allen Bereichen ermöglicht, die "nachhaltigste"
Handlungsalternative zu identifizieren. Der Ausdruck " nachhaltigste" impliziert, dass
die Handlungsalternativen konform sind zu den Massnahmen, Regulierungen,
Prinzipien, Ethiken und allen anderen Gegebenheiten in einer Gesellschaft, die als
"beste Praxis" für die Umsetzung der nachhaltigen Entwicklung in einer Gesellschaft
gelten.
-9-
Diese vielschichtigen Aspekte waren die Motivation für diese Arbeit, die sich auf den
Bereich des Bauingenieurwesens bezieht. Zwei wesentliche Ziele werden in dieser
Arbeit verfolgt. Das Erste ist, den klassischen Ansatz des Konzeptes zur Optimierung
der Lebenszykluskosten, der im Bereich des Bauingenieurwesens als das
Entscheidungsprinzip betrachtet wird, so umzuformulieren, dass Aspekte der
Nachhaltigkeit im Entscheidungsprozess Berücksichtigung finden können. Die Aspekte
der Nachhaltigkeit, die insbesondere Berücksichtigung in der Neuformulierung finden
sind das Prinzip der intergenerationellen Gleichheit und der Allozierung von
beschränkten Ressourcen. Für die Anwendbarkeit in realen Entscheidungssituationen
wird eine Plattform für die Modellierung und Optimierung von
Entscheidungsproblemen vorgeschlagen, die auf Bayes'schen Probabilistischen Netzen
basiert. Dies ermöglicht es, die Einschränkungen, die durch die Aspekte der
Nachhaltigkeit gegeben sind, im Entscheidungsprozess zu berücksichtigen. Das zweite
Ziel ist, einen fundamentalen Ansatz vorzustellen, der es ermöglicht, strukturelle
Zuverlässigkeit von baulichen Infrastrukturen in allgemeinen ökonomischen Modellen
zu berücksichtigen, so dass nachhaltige Entscheidungen in Bezug auf den Entwurf und
den Unterhalt solcher Anlagen von einer makroökonomischen Perspektive aus
identifiziert werden können.
Zwei Typen von Entscheidungsanalysen im Ingenieurwesen wurden in dieser Arbeit

unterschieden, um das Ausmass der Konsequenzen aus Entscheidungen klar
herauszustellen; es werden sowohl marginale Entscheidungsanalysen als auch
nicht-marginale Entscheidungsanalysen beleuchtet. In der marginalen
Entscheidungsanalyse im Ingenieurwesen wird angenommen, dass das wirtschaftliche
Wachstum exogen gegeben ist und die Konsequenzen, die aus Entscheidungen
resultieren, keinen Einfluss auf das wirtschaftliche Wachstum haben. Das Konzept der
Optimierung der Lebenszykluskosten von baulichen Infrastrukturen ist ein Beispiel für
eine marginale Entscheidungsanalyse. Damit kann das zuvor genannte erste Ziel dieser
Arbeit als Formulierung von Entscheidungsproblemen im Hinblick auf Nachhaltigkeit
im Kontext der marginalen Entscheidungsanalyse gesehen werden. Im Gegensatz dazu
kann das zweite formulierte Ziel als ein Rahmen für Entscheidungen gesehen werden,
die einen nicht-marginalen Einfluss auf das Wirtschaftswachstum haben.
Die vorliegende Arbeit gliedert sich in acht Kapitel. Kapitel 1 stellt die Ziele der Arbeit
vor, grenzt die Arbeit ab und erläutert die Hintergründe zu dieser Arbeit. Im ersten Teil
wird ein Überblick über die Literatur in den relevanten Gebieten der
Wirtschaftswissenschaften und des Bauingenieurwesens, insbesondere in den
Bereichen Formulierung und Optimierung von nachhaltigen Entscheidungsproblemen,
gegeben. Der Kern dieser Arbeit besteht aus sechs Kapiteln (Kapitel 2 bis 7). Jedes
dieser Kapitel (mit Ausnahme von Kapitel 7) repräsentiert einen Teil meiner
Forschungsarbeiten während des Doktorats, die bereits veröffentlicht sind oder zur
Veröffentlichung akzeptiert sind. Kapitel 2 behandelt den allgemeinen Umgang mit
-10-
Unsicherheiten in der Entscheidungsanalyse im Ingenieurwesen und stellt die
philosophische Basis für die Entscheidungsfindung im Ingenieurwesen unter
Unsicherheit dar. Kapitel 3 bis 5 untersucht die Modellierung und Optimierung von
Entscheidungsproblemen unter Berücksichtigung der zuvor genannten Aspekte der
Nachhaltigkeit. Kapitel 6 stellt einen Ansatz vor, mit dem die strukturelle
Zuverlässigkeit baulicher Infrastrukturen in allgemeinen wirtschaftswissenschaftlichen
Modellen und Modellen zur Beschreibung des Wirtschaftswachstums berücksichtigt
werden kann. Dieser Ansatz korrespondiert zu nicht-marginalen
Entscheidungsanalysen. In Kapitel 7 wird dieser Ansatz an einem einfachen
wirtschaftswissenschaftlichen Modell angewendet, um zu zeigen, wie die optimale
Zuverlässigkeit baulicher Infrastrukturen identifiziert werden kann, und eine
nachhaltige Strategie in Bezug auf den Entwurf und den Unterhalt verfolgt werden
kann. Dazu wird eine Zielfunktion in einem nicht-marginalen Kontext hergeleitet, die
grosse Unterschiede zur Zielfunktion aufweist, die im klassischen Ansatz zur
Optimierung der Lebenszykluskosten verwendet wird. Der Grund für diese
Unterschiede liegt in der Formulierung des Problems im marginalen und im
nicht-marginalen Entscheidungsraum. In diesem Kapitel wird auch auf die klassischen
Annahmen und Einschränkungen eingegangen, um die Unterschiede in diesen beiden
Ansätzen beleuchten zu können. Kapitel 8 schliesst die Arbeit.
In der Neuformulierung des klassischen Lebenszyklusansatzes wird die praktische

Anwendbarkeit unterstrichen. Daher können die Methoden, die in den Kapiteln 3 bis 5
vorgestellt werden, direkt in praktischen Problemen angewendet werden. Hierzu
werden in diesen Kapiteln praktische Beispiele gegeben. Auf der anderen Seite ist der
Ansatz, der in Kapitel 6 und 7 vorgestellt wird, ein relevanter Baustein für die weitere
Entwicklung eines allgemeinen Rahmenwerks für die nachhaltige
Entscheidungsfindung, wobei wissenschaftliche Einblicke gegeben werden, wie
nachhaltige Entwurfs- und Unterhaltsstrategien an baulichen Anlagen in einem
makroökonomischen Kontext untersucht werden können.
-11-
TABLE OF CONTENTS
ACKNOWLEDGEMENTS ................................................................................................................... 3
ABSTRACT ............................................................................................................................................ 5
ZUSAMMENFASSUNG ........................................................................................................................ 9
1. INTRODUCTION ....................................................................................................................... 16
1.1. RELEVANCE...................................................................................................................... 16
1.2. AIM OF THE THESIS........................................................................................................... 17
1.3. SCOPE OF THE THESIS ....................................................................................................... 19
1.4. STATE OF THE ART IN RELEVANT RESEARCH TOPICS ........................................................... 20
1.4.1. Structural performance of civil infrastructure.................................................................. 21
1.4.2. Socio-economic role of civil infrastructure ..................................................................... 23
1.4.3. Implication and formulation of sustainability .................................................................. 25
1.5. OUTLINE OF THE THESIS ................................................................................................... 27
2. PROBABILISTIC ASSESSMENT OF EXTREME EVENTS SUBJECT TO EPISTEMIC

UNCERTAINTIES (PAPER I) ............................................................................................................ 30
ABSTRACT ...................................................................................................................................... 31
2.1. INTRODUCTION ................................................................................................................ 31
2.1.1. Aleatory and epistemic uncertainties ............................................................................... 32
2.1.2. Probabilistic modeling approach in practice .................................................................... 33
2.2. GENERAL PRINCIPLES FOR THE PROBABILISTIC MODELING OF EVENTS SUBJECT TO
ALEATORY AND EPISTEMIC UNCERTAINTY ....................................................................................... 34
2.3. EXAMPLES ....................................................................................................................... 36
2.3.1. N-year maxima ................................................................................................................ 36
2.3.2. Return period ................................................................................................................... 39
2.3.3. Hazard curve ................................................................................................................... 41
2.4. DISCUSSION ..................................................................................................................... 42
2.5. CONCLUSION.................................................................................................................... 44
2.6. APPENDIX ........................................................................................................................ 44
3. CONSTRAINED OPTIMIZATION OF COMPONENT RELIABILITY IN COMPLEX

SYSTEMS (PAPER II) ......................................................................................................................... 46
ABSTRACT ...................................................................................................................................... 47
KEYWORDS .................................................................................................................................... 47
3.1. INTRODUCTION ................................................................................................................ 47
3.2. PROBLEM SETTING ........................................................................................................... 49
3.2.1. Modelling of complex systems ........................................................................................ 49
3.2.2. Bayesian hierarchical modelling...................................................................................... 50
3.2.3. Optimization of engineering decisions under constraints ................................................ 51
3.2.4. Objective of proposed approach ...................................................................................... 52
3.3. PROPOSED APPROACH ...................................................................................................... 53
-12-
3.3.1. Hierarchical system modelling with Bayesian probabilistic networks ............................. 53
3.3.2. Objective function and constraints .................................................................................. 55
3.3.3. Optimization of actions for components of complex system ........................................... 56
3.4. EXAMPLE 1 ...................................................................................................................... 56
3.4.1. Model description ............................................................................................................ 57
3.4.2. Results ............................................................................................................................. 61
3.4.3. Discussion ....................................................................................................................... 63
3.5. EXAMPLE 2 ...................................................................................................................... 64
3.5.1. Optimization of target reliability for welded joints in components ................................. 66
3.5.2. Results and discussion ..................................................................................................... 68
3.6. CONCLUSIONS .................................................................................................................. 68
4. INTER-GENERATIONAL DISTRIBUTION OF THE LIFE-CYCLE COST OF AN

ENGINEERING FACILITY (PAPER III).......................................................................................... 70
ABSTRACT ...................................................................................................................................... 71
KEYWORDS .................................................................................................................................... 71
4.1. INTRODUCTION ................................................................................................................ 71
4.2. MULTI-DECISION-MAKERS AND CRITERIA FOR SUSTAINABILITY ........................................ 72
4.3. EQUIVALENT SUSTAINABLE DISCOUNT RATE ..................................................................... 75
4.4. EXAMPLE ......................................................................................................................... 76
4.4.1. Cost distribution over time .............................................................................................. 77
4.4.2. Optimization of the concrete cover thickness .................................................................. 80
4.5. DISCUSSION ..................................................................................................................... 82
4.6. CONCLUSIONS .................................................................................................................. 83
4.7. ANNEX A ......................................................................................................................... 84
5. A BUDGET MANAGEMENT APPROACH FOR SOCIETAL INFRASTRUCTURE

PROJECTS (PAPER IV) ..................................................................................................................... 86
ABSTRACT ...................................................................................................................................... 87
KEYWORDS .................................................................................................................................... 87
5.1. INTRODUCTION ................................................................................................................ 87
5.2. BUDGET MANAGEMENT APPROACH .................................................................................. 88
5.2.1. Resource allocation ......................................................................................................... 88
5.2.2. Net benefit maximization ................................................................................................ 89
5.3. EXAMPLE ......................................................................................................................... 90
5.3.1. Maintenance planning for a portfolio of RC structures.................................................... 90
5.3.2. Inspection, repair and failure ........................................................................................... 91
5.3.3. Probabilistic corrosion model .......................................................................................... 92
5.3.4. Cost model ...................................................................................................................... 93
5.3.5. Numerical results ............................................................................................................. 94
5.4. DISCUSSIONS ................................................................................................................... 96
5.5. CONCLUSIONS .................................................................................................................. 97
-13-
6. SOCIETAL PERFORMANCE OF INFRASTRUCTURE SUBJECT TO NATURAL
HAZARDS (PAPER V) ........................................................................................................................ 98
ABSTRACT ...................................................................................................................................... 99
KEYWORDS .................................................................................................................................... 99
6.1. INTRODUCTION ................................................................................................................ 99
6.2. PROBLEM SETTING ......................................................................................................... 101
6.3. ROLE OF INFRASTRUCTURE IN ECONOMIC CONTEXT ....................................................... 102
6.4. PROPOSED METHODOLOGY ............................................................................................. 104
6.4.1. Definition of infrastructure failure................................................................................. 104
6.4.2. Equation of capital accumulation .................................................................................. 105
6.5. ILLUSTRATIVE EXAMPLE ................................................................................................ 106
6.6. DISCUSSION ................................................................................................................... 108
6.7. CONCLUSION.................................................................................................................. 109
7. OPTIMAL DESIGN AND MAINTENANCE POLICY ON INFRASTRUCTURE FROM A

MACROECONOMIC PERSPECTIVE............................................................................................ 110
7.1. INTRODUCTION .............................................................................................................. 110

7.2. PRINCIPLE OF LIFE-CYCLE COST OPTIMIZATION CONCEPT ............................................... 111
7.2.1. Derivation, ..................................................................................................................... 111
7.2.2. Assumption and limitation ............................................................................................. 113
7.3. AVAILABLE ECONOMIC MODELS FOR INFRASTRUCTURE .................................................. 114
7.4. ANALYSIS WITH SIMPLISTIC ECONOMIC MODEL .............................................................. 115
7.4.1. Economic model ............................................................................................................ 115
7.4.2. Steady state analysis ...................................................................................................... 118
7.4.3. Transition state analysis ................................................................................................. 121
7.4.4. Discussion and conclusion............................................................................................. 123
8. CONCLUSIONS AND OUTLOOK.......................................................................................... 125
8.1. CONCLUSIONS ................................................................................................................ 125

8.2. SCIENTIFIC ACHIEVEMENTS AND LIMITATIONS ................................................................ 128
8.3. OUTLOOK....................................................................................................................... 129
8.3.1. Assessment of the boundary conditions in marginal decision analysis .......................... 129
8.3.2. Further development of non-marginal decision framework ........................................... 131
REFERENCES ................................................................................................................................... 132
CURRICULUM VITAE ..................................................................................................................... 138
-14-
-15-
Introduction
1. Introduction
1.1. Relevance
Sustainable design and maintenance policies on civil infrastructure have become a
relevant subject in both developed and developing countries. Many developed
countries are presently experiencing severe deterioration of older infrastructure.
Developing countries are repeatedly faced with the losses of infrastructure due to
natural hazards. In addition, these countries continuously suffer from losses of
infrastructure due to deterioration that arises from the lack of appropriate maintenance
work.
In some developed countries, a considerable amount of economic resources is allocated

for maintenance work for civil infrastructure. For instance, in 2006 Switzerland
allocated 2.3% of its GDP to the investment into civil infrastructure and 54% of this
investment was used for maintenance work1. This ratio is high in comparison to the
average ratio for the European countries, which was found to be 31.4%2. However, in
other developed countries, the resources allocated for maintenance work for civil
infrastructure are not sufficient, and additional resources are urgently called for in
order to restore deteriorated infrastructure to a good condition. The Report Card for
America’s Infrastructure (ASCE (2005)) estimates that US$1.6 trillion is needed over
the next five-year period in the United States, which amounts to approximately 10% of
the country’s annual GDP. JSCE (2008) reports that Japan is expected to experience
severe deterioration of infrastructure by 2025 like what the United States is presently
experiencing, since infrastructure in Japan was mainly constructed between 1970s and
1980s, and the infrastructure constructed in this period will exhibit severe deterioration
in the near future. Developing countries have the same problem, i.e. lack of resources
for maintenance work. However, they are faced with an even more difficult situation,
since they also suffer from the lack of resources for the construction of new
infrastructure. In these countries the optimal balance of resource allocation between
construction and maintenance work is not yet obvious, whereas the World Bank (1994)
assesses that an additional US$12 billion spent for maintenance work for road
networks in African countries could save US$45 billion which otherwise have to be
spent on the reconstruction of the severely deteriorated road networks.
1 These numbers are calculated based on the statistics provided by EUROCONSTRUCT (2007).
2 The average over Austria, Belgium, Czech republic, Denmark, Finland, France, Germany, Hungry,
Ireland, Italy, the Netherlands, Norway, Poland, Portugal, Slovakia, Spain, Sweden, Switzerland and
the United Kingdom, which are included in EUROCONSTRUCT (2007).
-16-
Introduction
The statistics on economic losses due to natural hazards are summarized by

Guha-Sapir et al. (2004). These statistics show that during the period 1974 to 2003 the
highest economic losses due to natural hazards were brought about by: an earthquake
in Japan in 1995, US$159 billion3; flooding in China in 1998, US$22.6 billion; a
hurricane in the United States in 1992, US$39.4 billion. However, the same statistics
show the opposite story if the economic impact is measured in terms of a proportion of
GDP 4 . For instance, the greatest economic impact was caused by: earthquake in
Guatemala in 1976, 27% of the GDP; flood in Yemen in 1996, 28% of the GDP; wind
storm in St Lucia in 1988, 413% of the GDP. These countries are small in economic
terms and/or geographical size. Other developing countries have suffered from major
natural hazards, e.g. the flood event of 1987 in Bangladesh, the earthquake event of
1990 in Iran, the earthquake and associated tsunami event of 2004 in Southeast Asian
countries5.
Deterioration of infrastructure and losses of infrastructure due to natural hazards are

inevitable. However, these are manageable to a large extent by means of design and
maintenance policies on civil infrastructure. Thus, the statistics shown above raise the
question: were the past policies on design and maintenance of infrastructure optimal?
And if this is not the case, which are the optimal policies for the long-term
development of societies, i.e. what policies are sustainable?
Today, due consideration of sustainability is required in almost all civil engineering

decision situations. These decision situations include appraisal of new civil
infrastructure projects, ranking of rehabilitation measures for deteriorating
infrastructure, preparing design codes, donations and investments by international
organizations to civil infrastructure projects in poor countries. Since these activities are
supported by the public and undertaken on behalf of society, it is of the utmost
importance that the process of decision-making in such activities is clear, transparent
and consistent.
1.2. Aim of the thesis

The issue of sustainability is a complex issue that concerns many different aspects of
society and the environment, involving different stakeholders. Thus, it is unlikely that
a commonly agreed, general framework for sustainable decision-making can be
established in the near future. On the other hand, there is an urgent need for methods
that enable societal decision-makers to identify "sustainable" policies for civil
infrastructure projects. Herein, the "sustainable" policies imply the policies that
conform to current preventive measures, regulations, principles, ethics and whatever
3 Adjusted to US dollar in 2003. The same applies in the following unless otherwise stated.
4 GDP in the previous year of the hazard event occurrence.
5 Note that economic loss induced by Hurricane Katharina in 2005 is estimated at US$125 billion,
Munich Re (2005). However, this amounts to only slightly more than 1% of the GDP of the United
States in 2004, i.e. US$11 trillion (World Development Indicators Database, World Bank).
-17-
Introduction
are regarded as best practices for the realization of a sustainable development of

society; due to the absence of a general framework for sustainable decision making,
these best practices may be less efficient, but these are often undertaken in preventive
manners to avoid irreversible consequences.
Motivated by this and focusing on the civil engineering sector, the present thesis has
two aims. The first aim is to reformulate the classical life cycle cost optimization
concept advocated in civil engineering as the decision principle, in such a way that
relevant aspects of sustainability can be incorporated in engineering decision-making.
The relevant aspects of sustainability considered in this reformulation are
intergenerational equity and allocation of limited resources. Furthermore, for use in
practical decision situations, a platform is proposed for the modelling and optimization
of decision problems based on Bayesian probabilistic networks. The proposed platform
enables one to consider the constraints dictated by society in terms of, e.g., regulations
for the realization of the sustainable development of society. The second aim is to
provide a fundamental approach for incorporating the reliability of civil infrastructure
in general economic models so that the appropriate policies for design and
maintenance on civil infrastructure can be identified in the context of macroeconomics.
To achieve these aims systematically and also to facilitate a clear focus on individual
problems, the following four issues are identified. In the present thesis, each of these
issues is investigated individually.
Issue 1: Uncertainties
Decisions involving design and maintenance policies on civil infrastructure must be
made subject to significant uncertainties. These uncertainties are associated with the
randomness of natural phenomena such as the physical process of material
deterioration, a change of the environment surrounding the infrastructure and the
occurrence of natural hazards, in two ways. Firstly, the randomness of nature itself is
one of the uncertainties (aleatory uncertainty). By definition, this type of uncertainty
cannot be reduced. Secondly, modelling the characteristics of the randomness of nature
constitutes the other type of uncertainty (epistemic uncertainty). In principle, this type
of uncertainty can be reduced by a better understanding of the phenomena; however,
although some of the epistemic uncertainties may be reduced by merely collecting
more information, for others a reduction may not be possible in the foreseeable future.
Both types of uncertainty are relevant to decision problems when looking at the choice
of optimal policies, and they must be consistently taken into account in the decision
problems.
Issue 2: Adaptation of optimization problems to sustainable decision-making

Economic growth is a societal goal. At the same time, besides economic growth there
are a number of societal preferences. These preferences concern, for instance, the
preservation of natural resources including landscape, biodiversity and non-renewable
-18-
Introduction
resources, degree of homogeneity of welfare between members in society, and human

safety. These preferences must be fully taken into account in societal decision-making.
Thus, as part of such societal decisions, decisions regarding design and maintenance
policies on civil infrastructure often take the form of multi-objective optimization
problems, or otherwise constrained optimization problems where societal preferences
and other boundary conditions such as constraints on the amount of resources available
act as the constraints in optimization problems.
Issue 3: Inter-generational equity

Civil infrastructure provides benefits to society in terms of direct increase of economic
output and direct as well as indirect increase of social welfare over the long period of
its operation, possibly over a number of generations. At the same time, construction
and maintenance work of the civil infrastructure incur costs over the entire operation
period. Since the temporal distribution of such costs depends on the chosen design and
maintenance policies, the optimal choice of the policies is considered as a decision
problem in regard to fair distribution of the benefits and costs over different
generations.
Issue 4: Balance between quality and quantity

Civil infrastructure is important for economic growth. An increase in the quantity of
civil infrastructure capital increases economic output. Thus, given an amount of
investment in civil infrastructure, it is possible to achieve a higher economic output at
least in the short term by reducing the quality of infrastructure. This is because a unit
of infrastructure capital can be constructed and maintained less expensively, and as a
result the amount of constructed infrastructure can be increased. One of the
consequences of this strategy is a higher deterioration rate of the infrastructure in the
long term; this strategy may partly correspond to the strategies taken in the past by
some developed countries that are presently suffering from severe deterioration of civil
infrastructure. In contrast, high-quality infrastructure can be much more durable,
though it can be realized only at higher costs – not only higher costs of construction
and maintenance work but also a lower economic output in the short term due to a
smaller accumulation rate of the capital.
1.3. Scope of the thesis

In the course of investigating these issues, the present work makes several
assumptions. The most critical assumptions are: definition and formulation of
sustainability are assumed to be given; acceptable levels concerning several aspects,
e.g. human safety, environment and use of resources etc., are assumed to be given.
These assumptions effectively mean that the forms of the objective function (utility
function or social welfare function) and constraints, i.e. the general rule set for
sustainability, are assumed to be given. In fact, the general rule set could be established
given general agreement on the implications of sustainability within/between groups in
-19-
Introduction
society, e.g. individuals, communities, scientists and politicians. Therefore, the present
work, which focuses on engineering decision analysis, does not directly discuss these
topics, but instead relies on relevant related research works presently available. The
state of the art in these topics is briefly summarized in the next section, in addition to
research work on the structural performance of civil infrastructure and the
socio-economic role of civil infrastructure.
The present thesis defines two types of engineering decision analysis; marginal
engineering decision analysis and non-marginal engineering decision analysis. An
engineering decision is marginal if the consequence of the decision does not influence
the economic growth of society. As will be discussed in Section 7.2 this condition is
the assumption required for the application of the life-cycle cost optimization concept.
The marginal decision analysis is thus most suitable e.g. for decision situations in
which: private firms optimize individual engineering projects under constraints such as
budget constraints and regulations imposed by authorities; societal decision-makers
optimize the allocation of given resources in a portfolio of public engineering projects
in which the benefits from the projects are not reinvested into capitals but are
consumed. In contrast, an engineering decision is non-marginal if the consequence of
the decision affects the economic growth. An important example of a non-marginal
engineering decision is code-making for civil infrastructure; a higher acceptance
criterion for human safety imposes higher construction and maintenance costs on civil
infrastructure, which results in a smaller rate of capital accumulation.
In principle, any engineering decision-making may affect economic growth. Hence,

marginal decision analysis should be regarded as an approximation of non-marginal
decision analysis, although often formal non-marginal decision analysis may not be
feasible in practical decision situations due to the complexity of the analysis.
The scope of the present thesis is thus to investigate the issues mentioned in the
previous section in these two contexts; Issues 1 to 3 in the context of marginal
engineering decision analysis and Issue 4 in the context of non-marginal engineering
decision analysis.
1.4. State of the art in relevant research topics

Sustainable policy-making on civil infrastructure is interdisciplinary. It necessitates not
only an understanding of the structural performance of civil infrastructure, but also of
the socio-economic role of civil infrastructure. Furthermore, philosophical discussions
and practical agreements on what sustainability implies are required. In the following
sub-sections the state of the art in these areas is examined.
-20-
Introduction
1.4.1. Structural performance of civil infrastructure

Modelling the performance of structures has a long history. Until now, significant
effort has been directed towards the development of theories that describe the
performance of structures. Here, one of the most important paradigm shifts is the
introduction of the concept of probability: the concept that the performance of
structures can/should be evaluated in a probabilistic manner. This concept is especially
suited to the evaluation of the structural performance of civil infrastructure, since civil
infrastructure is typically exposed to random natural phenomena, e.g. earthquakes,
storms and floods, and the structural capacity of infrastructure and its modelling
involves large uncertainties.
Whereas some attempts were made to base structural performance on probability (see
Mayer (1926), Wierzbicky (1936) and Freudenthal (1947)), this important concept was
clearly formulated by Freudenthal (1954), wherein the failure and unserviceability of
structures are defined with due consideration given to uncertainties associated with
both loading on and the resistance of structures. Subsequently, the theory was extended
in many directions, which presently constitute the structural reliability theory. The
so-called second-moment concept gained its reputation at an earlier stage in the
development of structural reliability theory. This concept does not assume the form of
a probability distribution function to measure the reliability of structures (reliability
index), but only requires the first two orders of moments of the random variables that
characterize the reliability of structures. Due to this relatively simple way of measuring
the reliability, and also enhanced by the work by Cornell (1969), the concept was
widely accepted.
However, for the same reason, the concept has several disadvantages. One of the most
significant disadvantages is that the reliability index measured in accordance with this
concept is not invariant; the measured reliability index can differ in the algebraic
reformulations of the equations that mathematically represent the failure of structures,
i.e. limit state functions. This "invariance" problem was solved by Hasofer and Lind
(1974) with the introduction of the geometrical definition of reliability index.
Thereafter, a number of its extended variants have been proposed to incorporate more
information on the distributions of the random variables that characterize the reliability
of structures, e.g. the first order reliability methods (FORM) and the second order
reliability methods (SORM), see Ditlevsen and Madsen (2005) for an overview.
Other extensions are directed at application to the analyses for cases where the
reliability of structures may change over time, see e.g. Lin (1967), Ferry-Borges and
Castanheta (1971) and Vanmarcke (1983). The techniques developed for time-variant
reliability analysis have been widely applied to examine e.g. the reliability of
deteriorating structures and the dynamic response of structures in a probabilistic
manner. However, the techniques practically applicable for these analyses are highly
-21-
Introduction
dependent on the nature of the stochastic processes that characterize the resistances of
structures and the loads on the structures.
The structural reliability theory has also been extended to investigate the reliability of
structural systems. Earlier contributions to this extension primarily focus on the
development of algorithms for evaluating the probability of system failure defined by a
set of limit state functions, see e.g. Hohenbichler and Rackwitz (1982), Der Kiureghian
and Moghtaderi-Zadeh (1982), Ditlevsen and Bjerager (1986). Later, based on these
earlier contributions, more systematic and realistic approaches have been developed
for evaluating the reliability of structural systems. These approaches include the
consideration of the statistical dependence of the performance of structural system
components, e.g. Straub and Der Kiureghian (2008), Song and Kang (2008) and Der
Kiureghian and Ditlevsen (2008).
Today, some generic software tools for the reliability analysis of structures and
structural systems are available, e.g. STRUREL/COMREL (RCP GmbH) and
CalREL/FERUM (Der Kiureghian et al. (2006)).
The probability-based concept for the evaluation of structural performance has been
applied to the design optimization of structures within the framework of life-cycle cost
analysis. Therein, the optimal design is obtained by minimizing the sum of the initial
cost and the expected future costs due to possible failures. This life-cycle cost
optimization concept was first introduced by Rosenblueth and Mendoza (1971) in civil
engineering. At the same time, Bayesian decision theory was developed, see e.g. Raiffa
and Schlaifer (1961), Lindley (1965) in general and Benjamin and Cornell (1970) for
the application to civil engineering in particular. Later, the life-cycle cost optimization
concept was formally integrated into the framework of Bayesian decision theory.
Presently, the life-cycle cost optimization concept and Bayesian decision theory are
widely accepted and employed as the guiding philosophical principles in a variety of
engineering decision problems. The most important and successful applications of the
concept and the theory in civil engineering include: risk-based inspection planning e.g.
Tang (1973), Thoft-Christensen and Sørensen (1987), Faber et al. (2000) and Straub
(2004); reassessment of existing structures, e.g. JCSS (2001a); code making, e.g. JCSS
(2001b) and Rackwitz (2000).
Recently, the life-cycle cost optimization concept has been applied in the context of
sustainable societal development. However, most of these applications do not
explicitly consider intergenerational aspects; the utility function assumed in these
applications corresponds to the utility of one representative individual who is assumed
to live for an infinite time. The exception is Rackwitz et al. (2005), who consistently
consider the intergenerational aspect and apply discounting accordingly for the
marginal cost-benefit analysis of individual civil infrastructure projects. However, no
-22-
Introduction
general framework for sustainable decision-making on civil infrastructure in a

macroeconomic context, i.e. a non-marginal manner, has been developed.
1.4.2. Socio-economic role of civil infrastructure

In the last two decades, the role which civil infrastructure plays in the economy has
been intensively discussed. One of the relevant research questions in the discussion is
the social return rate of investment in civil infrastructure. The social return rates have
been estimated using a variety of historical datasets from different time periods and
different countries/regions. These estimates have then been utilized to discuss the
effectiveness of investment in civil infrastructure. Meanwhile, significant research
efforts have been made to develop economic models, within the framework of the
growth theory, that incorporate the role of civil infrastructure capital in the economy.
The primary goal of these efforts is to describe the effect of investment in civil
infrastructure on the long-term development of the economy, and to facilitate societal
policy-making on civil infrastructure.
A pioneering work on the effectiveness of investment in civil infrastructure is that of

Aschauer (1989). Based on statistics from the USA, he reveals that investment in civil
infrastructure has strong explanatory power for economic productivity. Subsequently, a
number of studies confirmed and generalized this observation, see review papers by
Munnell (1992) and Gramlich (1994). However, this observation is critically analyzed
by, among others, Holtz-Eakin and Schwartz (1995), arguing that there is little support
for claims of drastic productivity boost from increased infrastructure capital. Further
investigation was made by Canning and Bennathan (2000), focusing on the
complementarities of civil infrastructure capital to other types of capital, e.g. physical
and human capital. The results suggest that the investments in civil infrastructure are
not sufficient by themselves, and the investments should be undertaken in coordination
with investments in other types of capitals. Presently, whether or not current policies
on investment in civil infrastructure are effective is still a controversial question, and a
considerable amount of literature is available, see the review paper by Nijkamp and
Poot (2004).
The assessment of the social return rate is mostly made by relying on statistical
analysis techniques, especially regression analysis, see e.g. Chapters 11 and 12 in
Barro and Sala-i-Martin (2004). One of the problems of standard regression analysis is
that it is difficult to identify the causality between economic growth and infrastructure
investment; whether economic growth demands more infrastructure capital, or whether
increased infrastructure capital leads to an increase of economic output, see e.g.
Duffy-Deno and Eberts (1991) and Canning and Bennathan (2000). In order to avoid
the causality problem, several techniques have been developed, e.g. Engle and Granger
(1987) and Canning (1999), and applied to the estimation of the social return rate.
-23-
Introduction
Using these techniques, Canning and Bennathan (2000) show that investment in civil
infrastructure can result in an increase of economic output.
The results of these assessments on the productivity of civil infrastructure are useful
not only in discussing the effectiveness of investment in civil infrastructure, but also
serve as building blocks of economic models that represent the productivity of the civil
infrastructure.
The development of economic models for the economic role of civil infrastructure
capital is often based on the growth theory. The growth theory aims, in general, at
describing the long-term development of the economy in which different stakeholders,
e.g. households, firms and governments, maximize their own objective functions. The
original work on the growth theory is by Ramsey (1928). It investigates the optimal
saving rate of households to achieve their maximum utility in an infinite time horizon.
Today the theory presented therein forms the fundamental basis for a variety of
economic theories, ranging from consumption theory, asset pricing and business-cycle
theory (Barro and Sala-i-Martin (2004)). This work was later refined by Cass (1965)
and Koopmans (1965). Meanwhile, Solow (1956) and Swan (1956) propose a model
known today as the Solow-Swan model, which employs the neoclassical form of
production function and the assumption that saving rate is constant and exogenously
given. These conditions result in a very simple representation of the general
equilibrium of the economy. For this reason, the Solow-Swan model is widely used, in
spite of claims that the assumptions are not realistic and consistent with actual
observations.
Whereas these classical models involve labor and (aggregated) capital as factors of
production, modern models have been proposed that explicitly incorporate specific
factors of production, e.g. technology (e.g. Arrow (1962)) and natural resources (e.g.
Stiglitz (1974), Dasgupta and Heal (1974) and Solow (1974)). More recently, so-called
endogenous models have been developed, which enable the long-term growth of the
economy to be described without relying on exogenous growth factors (e.g. Romer
(1986) and Lucas (1988)). Today, both these modern and classical models are widely
applied as tools to investigate the sustainability of the economy, see e.g. Pezzey and
Withagen (1998), Krautkraemer (1999) and Valente (2005).
Within the framework of the growth theory, several directions have been proposed to
incorporate civil infrastructure capital in economic models as one of the production
factors. For instance, Glomm and Ravikumar (1994) implement civil infrastructure
capital into the production function of private firms as an external input. Duggal et al.
(1999) incorporate civil infrastructure capital in the production function as part of the
technological constraints. These production functions can then be employed to discuss
sustainable policies for investment in civil infrastructure and sustainability of the
economy.
-24-
Introduction
However, most of the economic models that incorporate civil infrastructure capitals
assume that the deterioration rate of the infrastructure capital is exogenously given and
constant; the deterioration rate is not considered as a variable. This means that the
average reliability of the infrastructure remains constant over the entire time period,
being independent of the growing economic states – the reliability remains the same
when the economy is in a poor state and in a richer state. However, this is not realistic
since the deterioration rate of infrastructure can be dynamically controlled by means of
the design and maintenance policies on civil infrastructure. There are only a few
research studies available that consider the deterioration rate as a variable. Rioja
(2003) proposes a dynamic general equilibrium model that explicitly considers
investment into maintenance work of civil infrastructure, thereby incorporating the
effect of the maintenance works on the deterioration rate of infrastructure. This model
is extended by Kalaitzidakis and Kalyvitis (2004), which endogenizes the decision of
budget allocation into both investment in the construction of new infrastructure and
investment in maintenance work on existing infrastructure.
The use of these models is a promising way to investigate the optimal reliability level
of infrastructure as a function of economic growth, thereby to identify the optimal
policies for the design and maintenance work on civil infrastructure in a
macroeconomic context. However, the assumptions made in these models are too
simplistic in regard to the relations between the amount of investment in maintenance
work and the deterioration rate; for instance, the investment in maintenance work at
one particular time influence the deterioration rate at the same time but not for the
deterioration rate in the future. Realistic models and a methodology that can
incorporate engineering knowledge into the models are still missing, and thus need to
be developed.
1.4.3. Implication and formulation of sustainability

Sustainable development is development that meets the needs of the present without
compromising the ability of future generations to meet their own needs (Brundtland
(1987)). The intuitive implication of this statement seems clear: increasing energy
consumption efficiency, less dependence on non-renewable resources, preserving
biodiversity, etc. However, when it comes to the formulation of sustainability, there are
a huge variety of opinions, approaches, methodologies and philosophies between
researchers in different disciplines, and even among researchers within the same
disciplines. In this section, instead of identifying the best formulation among them,
relevant discussions of three aspects of sustainability in the field of economics are
briefly summarized.
The first aspect is the substitutability of different types of capital in production

functions. Especially, the substitutability between man-made capital (e.g. physical
-25-
Introduction
capital, human capital) and natural capital (e.g. non-renewable resources) is the focus
of discussion, see Chapter 4 in Perman et al. (2003). Therein a distinction of the
concept of sustainability is made; weak sustainability and strong sustainability. The
perspective of weak sustainability is that man-made capital can substitute natural
capital, thus a certain production level can be kept by maintaining the level of the sum
of both types of capital. On the other hand, the perspective of strong sustainability is
that the level of production can be sustained only if natural capital is provided at a
certain level. If the strong sustainability perspective is taken, the level of production
can be maintained in an infinite time horizon only by exploiting natural capitals
indefinitely, which seems unfeasible at least for non-renewable resources. In contrast,
based on the weak sustainability perspective, the (feasible) conditions under which a
certain level of production and thus consumption can be maintained are derived by
Hartwick (1977) and Hartwick (1978).
The second aspect concerns the economic concepts of sustainability; opportunity-based

concept or consumption-based concept. The opportunity-based concept considers that
the sustainability should be based on the opportunities, i.e. opportunities to use capitals
should be maintained. On the other hand, the consumption-based concept assumes that
the sustainability is realized as long as the same level of (aggregated) consumption is
maintained. Seen in this light, the famous sentence in the Brundtland report:
Sustainable development is development that meets the needs of the present without
compromising the ability of future generations to meet their own needs, stands for the
opportunity-based concept. The opportunity-based concept is also supported by
ecologists, since resources which ecologists focus on are primarily renewable, thus the
preservation of the opportunities is feasible. Furthermore, the concept fits well with the
preservation of biodiversity. One of the proponents of the consumption-based concept
is Solow (1986), who argues that we have no obligation to our successors to bequeath
a share of this or that resources. Our obligation refers to generalized productive
capacity or, even wider, to certain standards of consumption/living possibilities over
time. Although this distinction poses an important philosophical question, practically it
makes little difference in economic models. This is because the economic models
presently employed in the discussions on sustainability are so simple that each type of
capital is nothing other than an input to the production functions. Consequently, within
these economic models the opportunities to use capitals are limited to production; then,
to maintain the opportunities is much the same as to maintain the production level, and
thus, the consumption level.
Today, no general agreement on the definition and criteria for sustainability is made; a
steady increase of consumption or utility over time is often considered as the criterion
for sustainability, see e.g. Withagen (1996), in which relevant works that employ this
definition are listed. However, this is not a unique criterion, see e.g. Pezzey (1992) and
-26-
Introduction
Pezzey (1997). Some concepts which are widely used and discussed in economics are
stated as6:
• A sustainable state is one in which utility (or consumption) is non-declining

through time
• A sustainable state is one in which resources are managed so as to maintain
production opportunities for the future
• A sustainable state is one in which the natural capital stock is non-declining
through time.
Other concepts, which originate in ecology, are stated as:
• A sustainable state is one in which resources are managed so as to maintain a

sustainable yield of resource services
• A sustainable state is one which satisfies minimum conditions for ecosystem
resilience through time.
1.5. Outline of the thesis

The core of the present thesis consists of six chapters. The next five chapters (Chapters
2 – 6) represent research work published or accepted for publication in four
peer-reviewed journal papers and a conference paper during the PhD study7. Chapter 7
is devoted to illustrating the approach proposed in Chapter 6 with a simplistic example.
Each of the chapters focuses on one of the four issues mentioned in the previous
section.
Chapter 2 (Paper I) focuses on the treatment of aleatory and epistemic uncertainties in

probabilistic assessments of extreme events. This chapter first reviews the general
principle for the treatment of these uncertainties. Then, focusing on the probabilistic
assessment of extreme events, it is pointed out that the general principle is often
violated in practice, and it is shown that such violations can lead to biased assessments
of probabilistic characteristics of extreme events. Since a consistent treatment of
aleatory and epistemic uncertainties is essential for risk-based decision analysis in
general, and the probabilistic assessment of extreme events is especially relevant to the
risk assessment of long-term structural performance of infrastructure, the principle
presented in this chapter constitutes a basis for the treatment of uncertainties in
sustainable policy making for civil infrastructure.
Chapter 3 (Paper II) proposes a method for optimizing decisions for complex
engineering systems under constraints. Constrained optimization problems are often
encountered in engineering decision analysis, especially where societal preferences
6 From Table 4.2 in Perman et al. (2003).

7 Therein, minor modifications such as grammatical corrections are made. Also, errata in the original
papers are, if any, corrected.
-27-
Introduction
must be taken into account. For instance, transport networks have to be designed and
maintained by satisfying requirements on human safety over their entire operation
periods. An engineering facility may have to satisfy the regulations imposed for
environmental protection, e.g. in terms of maximum leakage of harmful biochemical
agents. The proposed method employs Bayesian probabilistic networks for the
probabilistic representation of the structural performance of complex systems, and
generic algorithms for solving constrained optimization problems. Since these
techniques are commonly available in terms of software tools, the proposed method is
directly facilitated in practical decision situations.
Chapter 4 (Paper III) considers the issue of discounting in the context of

intergenerational equity. A large amount of research literature is available on the issue
of discounting, focusing on different types of discount factors. Among others, the most
relevant discount factors in civil infrastructure projects are the factors of pure-time
preference and long-term economic growth. The former concerns the preference of
individuals regarding the timing of consumption. The latter is related to the relative
wealth of the members of the society at different point in time. Incorporating these two
types of discount factors with due consideration of the finite lifespan of individuals, a
logically consistent concept for discounting (generation-adjusted discounting) is
proposed by Bayer and Cansier (1999). However, the application of the concept
requires tedious calculation. Thus, based on the consideration similar but independent
from Bayer and Cansier (1999), this chapter proposes a formula for deriving an
equivalent discount rate which, if applied to a decision problem with the classical
perspective where one decision-maker who is assumed to have an infinite lifespan,
yields the same total expected utility as when the decision problem is analyzed in
accordance with the consistent consideration of discounting over generations.
Chapter 5 (Paper IV) reformulates optimization problems of civil infrastructure

projects from a different perspective. The classical perspective is that the projects
should be optimized by minimizing the (discounted) life cycle costs. In this chapter
instead the optimization of projects is seen from the perspective of optimal budget
allocation. The shift of the perspective naturally introduces costs incurred by the delay
of actions which in turn is caused by the lack of a budget. In the reformulated
optimization problems, ultimate decision variables to be optimized are the amount of
budget that needs to be allocated to individual projects. This perspective is especially
useful for societal decision-makers who have to decide on the allocation of limited
resources.
In Chapters 3 to 5, the primary objective is to optimize individual civil infrastructure

projects. One of the underlying assumptions therein is that decisions made regarding
individual projects do not influence long-term economic growth in society, i.e., the
economic consequence of the projects is marginal – this assumption is required in
order to justify the assumption that the discount factor for economic growth is
-28-
Introduction
exogenous, independent of the decisions regarding individual projects. However,

whenever this assumption is violated, the marginal perspective mentioned above may
be invalid and the non-marginal (macroeconomic) perspective should be chosen. In
Chapters 6 and 7, a new conceptual approach for this is proposed and illustrated.
Chapter 6 (Paper V) proposes an approach for how the reliability of infrastructure can
be treated in the context of macroeconomics. The proposed approach consists of two
steps: (1) defining infrastructure failure by limit state representations; (2)
implementing the reliability concept into economic models. The first step takes basis in
the structural reliability theory and the second step employs the economic growth
theory. Thus, the proposed approach can incorporate knowledge of civil engineering
concerning structural performance into economic models. In order to show how the
proposed approach can be applied an illustrative example is provided. Therein, a
simplistic economy is assumed, which solely depends on civil infrastructure as the
production factor and is subject to natural hazards, and the economic growth path is
examined as a function of the policy on the design and maintenance of civil
infrastructure.
In Chapter 7, the proposed approach is applied to another simplistic economy, and the
steady and transition states of the economy are examined as a function of the policy on
the design and maintenance of civil infrastructure. By analyzing the steady state a
decision principle is derived, which differs from the decision principle adopted in the
life cycle cost optimization concept. Furthermore, it is shown that by analyzing the
transition state the optimal policy at each point in time depends on the current
economic output level.
-29-
Probabilistic assessment of extreme events subject to epistemic uncertainties (Paper I)
2. Probabilistic assessment of extreme events subject to

epistemic uncertainties (Paper I)
Kazuyoshi Nishijima
Institute of Structural Engineering, ETH Zurich, ETH Hönggerberg, HIL E 22.3,

Zurich 8093, Switzerland.
Michael Havbro Faber

Marc A. Maes
Schulich School of Engineering, University of Calgary, 2500 University Ave. N.W.,

Calgary, AB T2N1N4 Canada.
Proceedings of the ASME 27th International Conference on Offshore

Mechanics and Arctic Engineering, OMAE2008, Estoril, Portugal, June
15-20, 2008.
-30-
Copyright © 2008 by ASME

Abstract
Over the years the modeling and treatment of aleatory and epistemic uncertainties in
probabilistic assessments has repeatedly been an issue of discussion and also some
controversy. The philosophical and mathematical aspects may be said to be well
appreciated; however, there are cases in practice where principles seem to be violated
and frequently the effects of the epistemic uncertainty are treated inconsistently in the
probabilistic modeling. The present paper first reviews the general principles for the
modeling and treatment of uncertain characteristics subject to both aleatory and
epistemic uncertainties. Thereafter, the general principles are applied considering three
examples concerning the probabilistic modeling of extreme events; 1) the n-year
maximum distribution, 2) the corresponding return period and 3) the exceedance
probability in hazard analysis. Through these examples typical inconsistencies made in
practical probabilistic assessments are pointed out. The results from the examples are
interpreted and discussed from a structural design perspective and from a rational
risk-based decision perspective. Finally, a practical solution to avoid the
inconsistencies is suggested emphasizing the analogy of the analysis of extreme events
with the analysis of portfolios.
2.1. Introduction
The probabilistic modeling of events and not least extreme events forms a crucial
corner stone in risk based decision making concerning the design, assessment,
inspection and maintenance planning for engineering structures and facilities. The
assessment of probabilities can be performed based on probabilistic models that
describe the events of interest; extreme wave heights, current and wind velocities, etc.
In general, such probabilistic models are established through the joint consideration of
knowledge, experience and observations; combining statistical assessments with
subjective judgments. Consequently, very often the resulting probabilistic models are
associated with not only aleatory uncertainties, i.e. the inherent natural variability
associated with the phenomenon of interest but moreover with significant epistemic
uncertainties. It is of utmost importance that both of these two contributions to
uncertainty are treated correctly in the probabilistic assessments.
In the literature a number of discussions have been made on how uncertainties arising
from different sources may be categorized and how these different categories should
and/or can be considered in probabilistic risk assessment and risk-based decision
making, e.g. Raiffa and Schlaifer (1961), Pate-Cornell (1996), Faber (2003), Wen et al.
(2003), Faber and Maes (2005) and Der Kiureghian and Ditlevsen (2007). It can be
said that the relevance of epistemic uncertainties in risk assessments is well recognized
and also the general principles for modeling and assessing the relevant probabilistic
characteristics seem well understood. However, there are still several situations where
the general principles are violated in practice. The present paper considers the
treatment of aleatory and epistemic uncertainties especially in the probabilistic
-31-

modeling and assessment of extreme events. The probabilistic modeling of extreme

events often requires that several probabilistic models are applied jointly and that some
logical framework is assumed for extrapolation of knowledge concerning e.g. the
probabilistic characteristics of annual events to the corresponding characteristics of
events with much longer return periods, e.g. 100 years. If in this process the aleatory
and epistemic uncertainties are inconsistently mixed up the probabilistic characteristics
of the extreme events of interest are assessed incorrectly.
The present paper first reviews the general principles for the probabilistic modeling of
uncertain characteristics subject to both aleatory and epistemic uncertainties.
Thereafter, three examples are considered pointing out in parallel the typical
inconsistent assessments often made in practice and the results of a correct assessment
following the general principles. Finally, a practical procedure to avoid inconsistent
probabilistic assessments of extreme events is presented based on an analogy to the
probabilistic modeling and treatment of portfolio loss assessments.
2.1.1. Aleatory and epistemic uncertainties

Without going into detailed and philosophical discussions, it is taken for granted in the
present paper that the probability measure is sufficient to represent any type of
uncertainty e.g. O'Hagan and Oakley (2004) and that the Bayesian statistics provides a
consistent basis for representing both aleatory and epistemic uncertainties, see e.g. De
Groot (1970) and Lindley (1980).
Generally, it is understood that aleatory uncertainty reflects the variability of events

subject to inherent natural variability and epistemic uncertainty represents imprecise
models, lack of data and insufficient knowledge, e.g. Pate-Cornell (1996), Wen et al.
(2003) and Der Kiureghian and Ditlevsen (2007). Pate-Cornell (1996) provides a
general overview on the treatment of the uncertainties in risk assessment over different
engineering applications identifying different levels of analytical sophistication.
Therein, the explicit consideration of epistemic uncertainty in risk assessment is
qualified as the highest level of risk assessment.
In engineering decision making the treatment and categorization of the two

components of uncertainties has received attention for mainly two reasons. The first
reason is that the categorization of uncertainties allows for the optimization of resource
allocations aiming to reduce uncertainty and thereby to enhance ranking of options for
the purpose of risk management; epistemic uncertainty can be reduced by
accumulating data and knowledge. In this context the pre-posterior decision analysis
provides the theoretical basis, see Raiffa and Schlaifer (1961). The pre-posterior
analysis has been extensively applied in the field of engineering in general, e.g. Faber
(2003) and Faber and Maes (2005) and in risk-based inspection planning in particular,
e.g. Straub and Faber (2005). The second reason is that the epistemic uncertainty often
-32-

may have a profound effect on the probabilistic characteristics of systems. In Nishijima

and Faber (2007a) systems with quasi-identical components subjected to epistemic
uncertainties are considered. There it is shown that the epistemic uncertainty can be
utilized for the reduction of the uncertainty of a whole system performance by
inspecting the states of some of the components in the system. Faber et al. (2007a)
considers the effect of epistemic uncertainties on the portfolio loss analyses subject to
seismic hazards; epistemic uncertainties concerning the resistance of types of buildings
commonly affect all buildings that belong to the same type. Thus, the quantile values
of the distribution of failure costs are highly dependent on the extent of the epistemic
uncertainties. The present paper is strongly related to the latter considerations as
discussed in more detail in the subsequent sections.
2.1.2. Probabilistic modeling approach in practice

Within the framework of probabilistic hazard analysis, the probabilistic modeling of
hazards, such as the seismic ground motion, wind speed and wave height, can be
established by either pure statistical modeling relying only on available relevant data
or by means of engineering probabilistic models which also facilitate for the utilization
of subjective information such as experience and physical understanding.
The pure statistical approach has been preferred by classical statisticians since the
results of such models are coherent with the frequentistic interpretation of
probabilities; there is a one to one correspondence between observations and model
predictions. Typically the statistical models are formulated as annual extreme value
distributions, and the extreme value theory thus provides the justification for assuming
either one of the three extreme value distributions or the generalized extreme value
distribution, e.g. Leadbetter et al. (1983) and Coles (2001). This approach may be a
reasonable solution for cases where the detailed physical mechanisms that govern the
hazard events are not well understood or too complex to represent in a practically
manageable effort. However, this approach also has drawbacks; 1) direct observations
of extreme events are by definition rare why the parameter estimation of the
distributions generally involves large statistical uncertainties (epistemic uncertainty),
and 2) the potentially available scientific knowledge and/or engineering experiences
cannot be included in the modeling. To overcome these drawbacks, engineering
probabilistic approaches have been developed for different types of hazards, which
enables one to integrate into the hazard analysis the available knowledge and
engineering experience. For instance, in Nishijima and Faber (2007b) hurricane
simulation techniques have been developed for wind hazard analysis integrating
several probabilistic model components each of which represents individual parts of
the involved physical mechanisms, e.g. the transition of hurricanes and development of
the pressure fields.
-33-

In the pure statistical modeling approach the distinction between epistemic uncertainty
and aleatory uncertainty is relatively clear, since the epistemic uncertainty is primarily
statistical uncertainty that is involved in the parameter estimation of the distributions
(including uncertainty on the choice of distribution family). The epistemic uncertainty
can be integrated into the probabilistic assessments within the Bayesian statistical
framework, e.g. Coles et al. (2003), although in practice it is often neglected. On the
other hand, in the engineering approach taking basis in the Bayesian framework the
epistemic uncertainties are associated with each individual probabilistic model
components that jointly comprise the probabilistic assessment model in terms of model
uncertainty and statistical uncertainty.
As is discussed in more detail later, the integration of aleatory and epistemic

uncertainties at the level of the individual probabilistic models may lead to inconsistent
assessments of the probabilistic characteristics of extreme events, see Maes and
Jordaan (1985) and Maes (1990). This can be seen through a simple example: consider
throwing two different dice. One die is a fair die which has six numbers (one to six)
and the probability of the outcome of each number is assumed equal to 1/6 (pure
aleatory uncertainty). Therefore, the probability that a six comes out in one trial is 1/6.
The other die is an unfair die which has an identical number, between one and six, on
all six faces, yet the number is unknown. Thus, it is assumed that the probability that
the number is i ( i = 1, 2,..., 6 ) is equal to 1/6 (pure epistemic uncertainty). Therefore,
the probability that a six comes out in a trial is 1/6, which is the same as with the fair
die. Now consider throwing each of two dice 100 times. The probability that the six
comes out at least once with the fair die is equal to 1 − (1 − 1/ 6)100 ≈ 1 , while the
probability that the six comes out at least once with the unfair die remains 1/6. When
the different origins and/or types of uncertainty is not identified and differentiated in
the probabilistic assessments, it may not be possible to assess the probability of
extreme events correctly. Thus, it is of utmost importance to distinguish between
aleatory and epistemic uncertainties in the probabilistic modeling of extreme events for
both statistical and engineering-based approach.
2.2. General principles for the probabilistic modeling of events

subject to aleatory and epistemic uncertainty
This section reviews the general principles for assessing the probabilistic
characteristics of events in general and provides remarks which are relevant for the
probabilistic assessment of extreme events in particular. The probabilistic models for
assessing probabilistic characteristics of extreme events are assumed to have been
developed aiming at describing the random nature of phenomena of interest in e.g.
offshore engineering. Hence, the probabilistic models specifically focus on the aleatory
uncertainties associated with e.g. extreme wave heights. However, due to the lack of
data and/or knowledge the developed probabilistic models do not precisely represent
-34-

the random phenomena of the real world, why epistemic uncertainty is introduced to
account for such model uncertainties.
In the context of engineering decision making or reliability assessments the

probabilistic modeling problem can in general be represented as a problem involving
the expectation operation (in some cases a conditional expectation) over a function
g ( X) of aleatory random variables X = ( X 1 , X 2 ,..., X n ) as:
E[ g ( X)] = EΘ ⎡⎣ EX [ g ( X) | Θ ]⎤⎦ (2.1)
The random variables X are characterized by the joint probability distribution

function FX (x | θ) conditional on the epistemic random variables Θ =
(Θ1 , Θ2 ,..., Θm ) , which in turn are characterized by the probability distribution function
FΘ (θ) . Thus, FX (x | θ) corresponds to the developed probabilistic model and together
with FΘ (θ) constitutes the probabilistic assessment model, see Figure 2.1. From a
probability theoretical viewpoint the expectation operation of g ( X ) may be
performed in any manners as long as it is integrated over the domain of the joint
probability density function of ( X, Θ) . However, the hierarchical expression on the
expectation given by Equation (2.1) is useful especially for probabilistic modeling of
extreme events since some of the aleatory random variables often can be assumed to be
conditionally independent given the epistemic uncertainties Θ ; this can reduce the
computational effort required to evaluate the expectation significantly.
Figure 2.1. Probabilistic assessment subject to aleatory and epistemic uncertainties.
Figure 2.1 illustrates the roles of aleatory and epistemic uncertainties in probabilistic
modeling. Probabilistic characteristics of extreme events are first assessed conditional
on the epistemic uncertainty θ then integrated over possible realizations of epistemic
random variables Θ . The epistemic random variables Θ should be interpreted
heuristically; the epistemic random variables represents not only the uncertainties of
the parameters of distributions but also the likelihood or degree of belief associated
-35-

with different distribution families and even pre-assumptions for probabilistic

calculations etc. The pre-assumptions reflect the modeler’s perception of the
phenomena of interest, for instance what concern causal relations and boundaries for
the considered phenomena. Although these pre-assumptions are often precluded in the
probabilistic modeling and simply assumed certain, it should be kept in mind that these
may be significant for the probabilistic modeling. It should be also mentioned that the
categorization of epistemic uncertainty and aleatory uncertainty is dependent on these
pre-assumptions, a process which in itself is subject to the modeler’s choice and taste
why in a certain sense any assignment of aleatory uncertainties is conditional on
factors or variables which are associated with epistemic uncertainty.
2.3. Examples
Three examples are now considered in order to illustrate how the general principle
introduced in the previous section might be utilized in practice. Through the examples,
pointing out the typical inconsistent probabilistic assessments of characteristics of
extreme events which are commonly utilized in engineering design and assessment, the
probabilistic models which follow from the application of the general principle are also
provided. The discussion on the implications of the results is provided subsequently in
Section 2.4.
2.3.1. N-year maxima

The first example considers the derivation of the cumulative distribution of the n-year
maxima from the annual maximum distribution. It is pre-assumed that the annual
maxima are statistically independent and identically distributed. The cumulative
distribution function of n-year maxima can be calculated in accordance with Equation
(2.1) by defining:
g ( X) = I ⎡ max { X i } ≤ x ⎤ (2.2)
⎢⎣ i =1,2,..,n ⎥⎦
where I [⋅] is an indicator function that returns the value one if the condition in the
bracket is satisfied and zero otherwise and X i is the i th year maximum. By
substituting Equation (2.2) into Equation (2.1) the cumulative distribution function is
obtained as:
FX ,n ( x) = ∫ { F ( x | θ )} p (θ ) dθ
n
(2.3)
where F ( x | θ ) is the conditional cumulative distribution function of the annual

maxima and p (θ ) is the probability density function of the epistemic random
variable Θ . The epistemic random variables may be represented by a scalar or a
vector. The possible sources of the epistemic uncertainty are the statistical
-36-

uncertainties when the cumulative distribution function is established by a pure

statistical approach and the model and statistical uncertainties when the cumulative
distribution function is established based on engineering probabilistic models.
In practice deviations from the general principle are observed. One example for this
concerns the utilization of probabilistic hazard maps or load recommendations for risk
management purposes. Hazard maps usually provide characteristic values, e.g. quantile
values including the effect of the epistemic uncertainty, e.g. in the form of
conservatively assessed fractile values or median values of the fractile values relative
to the epistemic uncertainties. Based on these characteristic values a distribution
function of annual maxima F ( x) is established and based on this finally the
distribution of the n-year maximum distribution is calculated as:
{ }
n
FX ,n* ( x) = F ( x) (2.4)
Since the annual maximum distribution F ( x) that is established utilizing the

probabilistic hazard map or load recommendations already contains the effect of
epistemic uncertainty, F ( x) can be written as:
F ( x ) = ∫ F ( x | θ ) p (θ ) dθ (2.5)
Obviously, FX ,n ( x) and FX ,n* ( x) are in general not identical. Furthermore, for n > 1
it can be shown by applying Jensen’s inequality that
FX ,n ( x) = EΘ ⎡{ F ( x | Θ)} ⎤
n
⎣ ⎦
(2.6)
{
≥ { EΘ [ F ( x | Θ) ]} = F ( x) }
n n
*
= FX ,n ( x)
The equality holds if there is no epistemic uncertainty. Thus, for any given quantile the
corresponding value is larger when FX ,n* ( x) is employed instead of FX ,n ( x) ; n-year
maximum events are overestimated when FX ,n* ( x) is employed.
A numerical example is shown to illustrate the degree of the difference between

FX ,n ( x) and FX ,n* ( x) , considering the case of wind hazard analysis. For this purpose
it is assumed that the conditional annual maximum wind speed X follows the
Gumbel distribution as:
F ( x | θ ) = exp ( − exp ( −α ( x − θ ) ) ) (2.7)
-37-
Copyright ©2008 by ASME

Figure 2.2. Probability density functions of maximum wind speed.
Figure 2.3. Exceedance probabilities of 50-year maximum wind speed.
where θ represents the epistemic uncertainty and α = 0.257 (this corresponds to the
standard deviation of 5 [m/s] given θ ). The epistemic uncertainty represented by the
random variable Θ is assumed to follow the Normal distribution with mean and
standard deviation being equal to 20 [m/s] and 5 [m/s] respectively. Figure 2.2 shows
the assessed probability density functions of the 50-year maximum in accordance with
Equations (2.3) (denoted as “consistent”) and (2.4) (denote as “inconsistent”)
respectively. It is seen that the probability density function looks significantly different
-38-

and that the mean value of the 50-year maximum wind speed is overestimated when it
is evaluated using Equation (2.4).
Figure 2.3 shows the corresponding exceedance probabilities of the 50-year maximum
wind speed. Whereas the (inconsistent) Equation (2.4) overestimates the exceedance
probability at the range between 10−1 and 1, the tendency diminishes for the range of
lower probabilities. These results should be appreciated depending on the context as
will be discussed further in the subsequent section.
2.3.2. Return period

In this example first the definition of the return period of events is briefly revisited and
thereafter the effect of epistemic uncertainties on the return period is assessed.
The return period may be defined as the expected value of the arrival time of the event
of interest, see e.g. Benjamin and Cornell (1970). Assuming that the probability of
occurrence of an event in a Bernoulli sequence of trials is p , then the arrival time
follows the geometric distribution. The expected value of the arrival time E[T ] is
then calculated as 1/ p . When the event is characterized by its intensity X , e.g. a
given wind speed or a given precipitation, the probability p is represented by the
cumulative distribution function F ( x ) of the maximum within a given period (e.g.
one year). Thus, the return period is a function of the intensity x and may be written
as:
1
E[T ( x )]* = (2.8)
1 − F ( x)
However, when the epistemic uncertainty represented through Θ is involved, the

assumption of independence between the intensities at different times does not hold,
even if this might be a reasonable assumption considering observations from the real
world; the intensities are independent only conditional on the realization of epistemic
uncertainty θ . Thus, the return period defined by Equation (2.8) should be
reformulated as:
⎡ 1 ⎤ p(θ )
E[T ( x)] = EΘ ⎢ ⎥ =∫ dθ (2.9)
⎣ 1 − F ( x | Θ) ⎦ 1− F (x | θ )
where F ( x | θ ) is the conditional cumulative distribution function on the epistemic
uncertainty θ and p(θ ) is the probability density function of θ . This formulation
is coherent with the general principle given in Equation (2.1).
Probabilistic engineering models are often employed where the cumulative distribution
function of the maximum intensity within a given reference period is established by
combination of probabilistic models that represent the natural random nature (aleatory
-39-

uncertainty) yet subject to model/statistical uncertainties (epistemic uncertainty), as

e.g. in hurricane simulation for wind hazards analyses. The cumulative distribution
function obtained in this manner already considers the epistemic uncertainty and can
thus be written as:
F ( x) = ∫ F ( x | θ ) p (θ ) dθ (2.10)
The return period is often assessed by combining Equations (2.8) and (2.10) as:
1
E[T ( x )]** = (2.11)
1 − F ( x)
This is obviously not the same as Equation (2.9) and it can be shown by applying
Jensen’s inequality that:
⎡ 1 ⎤ 1
E[T ( x)] = EΘ ⎢ ⎥ ≥
⎣1 − F ( x | Θ) ⎦ 1 − EΘ [ F ( x | Θ) ] (2.12)
1
= = E[T ( x)]**
1 − F ( x)
Figure 2.4. Comparison of return periods.
The equality in Equation (2.12) holds if there is no epistemic uncertainty; in that case
E[T ( x)] and E[T ( x)]** coincide. From this inequality, it can be said that the return
period assessed by Equation (2.11) underestimates the expected arrival time.
-40-

In Figure 2.4 the results of a probabilistic assessment of the relation between extreme
wind speeds and corresponding return periods are shown. Based on the same
assumption as in the first example it is seen that the application of Equation (2.9) and
(2.11) respectively result in different return periods. For instance, based on the
application of Equation (2.11) a wind speed of 40 m/s corresponds to a return period of
80 years, whereas the correct return period using Equation (2.9) is in fact 400 years.
2.3.3. Hazard curve

In the following example it is investigated how hazard curves, i.e. the relationships
between the exceedance probabilities for a given uncertain phenomenon represented by
the random variable X should be calculated according to the general principle given
by Equation (2.1). For illustrative purposes an example considering an earthquake
hazard analysis is selected and for simplicity, only one seismic zone is considered in
this example.
Seismic hazard analysis aims at assessing the probability of exceedance of any given
seismic hazard intensity x for a specified reference period, e.g. one year, (seismic
hazard curve). In the assessment of this probability several assumptions and
probabilistic models are required; e.g. the occurrence of earthquake in the seismic
zone, the magnitude of the earthquake, the distance between the epicenter and the site
for which the hazard analysis is performed and the so-called attenuation law that
relates the relevant parameters and the seismic hazard intensity. Essentially such
assumptions and probabilistic models involve epistemic uncertainty due to the
imperfection of the postulated models and scarce data available for estimating
parameters in the models. Whereas the presence of epistemic uncertainty in general is
appreciated and some epistemic uncertainties are considered correctly, other epistemic
uncertainties are often inconsistently considered. Examples of cases where epistemic
uncertainties are consistently accounted for include the epistemic uncertainty
associated with the choice of attenuation law and the choice of the range of the
possible magnitudes. For instance, a typical attenuation law is represented in the form
of X = ε ⋅ g ( a, b, c,...) , where X denotes the hazard index, e.g. peak ground motion,
and a, b, c,... represent the relevant parameters in the attenuation law, e.g. magnitude
and distance from the epicenter, and ε represents the residual term. Different
attenuation laws are proposed by different experts. These differences are often ascribed
to expert judgments, for each of which a probability is assigned in order to incorporate
the different expert judgments into one unified seismic hazard curve. Such
incorporations are consistent with Equation (2.1), since the inner expectation in
Equation (2.1) corresponds to each hazard curve conditional on each expert judgment
and the outer expectations correspond to the uncertainties associated with the expert
judgments. An example of the inconsistent consideration of the epistemic uncertainties
corresponds to the residual term of the attenuation law. The random variable ε can be
-41-

considered to involve epistemic uncertainty, since obviously this uncertainty can be

reduced by updating using data on the seismic hazard intensity from the site for which
the seismic hazard analysis is performed.
Denote by q( x | θ ) the probability that the seismic hazard intensity X exceeds x

given the occurrence of an earthquake. The probability q( x | θ ) is conditioned by the
epistemic uncertainty θ , e.g. the uncertainty associated with the attenuation law.
Hence, the probability that the seismic hazard intensity X exceeds x may be
written in accordance with Equation (2.1) as:
P[ X > x] = ∫ (1 − exp [ −ν q ( x | θ ) ]) p (θ ) dθ (2.13)
Here it is assumed that the occurrence of an earthquake follows a Poisson process with
intensity ν . However, in some practices the probability is calculated as:
P[ X > x]∗ = 1 − exp ⎡⎣ −ν ∫ q( x | θ ) p(θ )dθ ⎤⎦ (2.14)
where the conditional probability of the seismic hazard intensity given the occurrence
of an earthquake is first marginalized by integrating over the epistemic uncertainty θ ,
thereafter the assumption of the Poisson process is applied to calculate the probability
of exceedance x ; Equation (2.14) is inconsistent with the general principle given by
Equation (2.1). Generally, Equation (2.14) does not provide the same value as Equation
(2.13), although if ν is small enough both equations can be approximated as
ν ∫ q ( x | θ ) p (θ )dθ . In this sense, the evaluation of the probability with Equation (2.14)
can be seen as a numerical approximation and this may justify the use of Equation
(2.14) in practice. Furthermore, by applying Jensen’s inequality, it can be shown:
P[ X > x] = ∫ (1 − exp [ −ν q( x | θ ) ]) p(θ )dθ

= EΘ ⎡⎣1 − exp [ −ν q( x | Θ) ]⎤⎦
(2.15)
≤ 1 − exp ⎡⎣ −ν EΘ [ q( x | Θ) ]⎤⎦
= P[ X > x]*
A similar discussion may apply to cases where non-Poisson processes are assumed for
the occurrence of earthquake and for cases where two or more seismic zones are
considered.
2.4. Discussion
Three examples considering the n-year maximum distribution, the return period and
the exceedance probability respectively have been considered. For each of these
examples typical inconsistent treatments of epistemic uncertainties found to occur in
-42-

practical applications have been considered and analyzed. The results from these
examples should be interpreted corresponding to the contexts: structural design in
practice and optimal decision making. In the context of structural design in practice the
results of the examples may be understood such that the inconsistent probabilistic
assessments often made in practice are conservative and hence can be justified.
Furthermore, the inconsistent probabilistic assessments are in general less complicated
compared with the consistent assessments, since they allow for incorporation of the
epistemic uncertainties at earlier stages of the assessments. However, in the context of
optimal risk-based decision making the inconsistent probabilistic assessment should be
circumvented as it leads to sub-optimal decisions.
The first example reveals that the information provided in typical hazard maps and
load recommendations are not sufficient to use directly in the context of optimal
decision making, since they do not differentiate the sources of uncertainties; hence the
distributions of maximum values for a given reference period cannot be correctly
established. The second example shows that the return period that provides the basis
for structural design as well as for validation of the established probabilistic models
based on observations does not correspond to the expected value of the arrival time.
Therefore, the return period assessed by Equation (2.11) should not be used for these
purposes. The third example justifies the seismic hazard analyses presently made in
practice in a numerical sense, although it is important to realize that the analyses are
not conceptually consistent with the general principle for the probabilistic assessments.
Figure 2.5. Graphical representation for interrelation between random variables.
In order to circumvent inconsistent probabilistic assessments, a causal representation,

e.g. through Bayesian probabilistic networks Jensen (2001) may be useful just for the
purpose to explicitly understand the interrelations between all random variables in the
probabilistic assessment models. Figure 2.5 shows the causal representation
corresponding to the first example, where Θ represents the epistemic uncertainty,
X i represents the annual maximum wind speed at the i th year and Y represents the
50-year maximum wind speed ( Y = max in=1 X i ). When the interrelation between all the
variables are explicit, it is clear that X i ( i = 1, 2,..., n ) are not independent but are
instead exchangeable, see Maes and Jordaan (1985). Thereby it is also clear how to
calculate the marginal distribution of Y according to the general graphical
-43-

representation theory e.g. Jensen (2001). It is worthwhile mentioning that the random
variables X i can be seen as the components of a temporarily distributed portfolio
with the analogy of a spatially distributed portfolio – the graphical representation in
Figure 2.5 can be also understood to represent a spatially distributed portfolio, the
component of which are subject to epistemic uncertainty, see Faber et al. (2007a).
Then, it is obvious that the probabilistic characteristics of identical components X i
are subject to epistemic uncertainty Θ that simultaneous affects all the components.
In this regard the distinction between aleatory and epistemic uncertainty might be
useful simply to make clear which variables affects other variables. For completeness
the incorporation of epistemic uncertainty in seismic hazard analysis as discussed in
the third example is shown in detail in the Appendix.
2.5. Conclusion
The present paper first provides general principles on how aleatory and epistemic
uncertainties should be considered in the probabilistic modeling and assessments for
risk based decision making. Focusing on the probabilistic modeling of extreme events,
several inconsistencies often made in practical probabilistic assessments for extreme
events are pointed out; i.e. the n-year maximum distribution, the return period and the
exceedance probability in hazard analysis. For the considered examples it is shown that
such inconsistent probabilistic assessments overestimate the probabilistic
characteristics of the extreme events. From the perspective of structural design it can
be seen as a conservative assessment and thus may be justified. However, from the
perspective of optimal decision making the inconsistent assessments lead to
sub-optimal decisions and should thus be avoided.
2.6. Appendix
The exceedance probability is calculated assuming that the occurrence of earthquakes
over time follows a Poisson process as:
∞
P[ X > x] = ∑ P ⎡ N = k ∩ max X i > x ⎤
k =1
⎢⎣ i =1,2,.., k ⎥⎦
∞
(2.16)
= ∑ P ⎡ max X i > x | N = k ⎤ P [ N = k ]
k =1
⎢⎣ i =1,2,..,k ⎥⎦
where N is the number of occurrence of earthquake and X i is the peak ground

intensity due to the i th earthquake. When the intensities can be assumed independent,
the calculation proceeds as:
-44-

∞
νk
P[ X > x] = ∑ ⎡⎣1 − (1 − q ( x))k ⎤⎦ e−ν
k =1 k! (2.17)
= 1 − exp [ −ν q ( x)]
where q ( x) is the probability that the intensity exceeds x given the occurrence of
an earthquake and ν is the occurrence rate. This is the same form as Equation (2.14)
using that q ( x) = ∫ q ( x | θ ) p (θ ) dθ . However, when epistemic uncertainties which
affect all X i are present, the calculation should proceed as:
∞
P[ X > x] = ∫ ∑ P ⎡ max X i > x | N = k , θ ⎤ P [ N = k ] p (θ )dθ
k =1 ⎣⎢ i =1,2,.., k ⎦⎥
ν k e −ν
( )
∞
= ∫ ∑ 1 − (1 − q ( x | θ ) )
k
p (θ )dθ (2.18)
k =1 k!
= ∫ (1 − exp [ −ν q ( x | θ ) ]) p (θ )dθ
which is equivalent to Equation (2.13). In this way, the fact that the epistemic
uncertainty affects the ground motion intensities for all earthquakes over time plays a
crucial role.
-45-
Copyright ©2008 by ASME

Constrained optimization of component reliabilities in complex systems (Paper II)
3. Constrained optimization of component reliability in

complex systems (Paper II)
Kazuyoshi Nishijima

Marc A. Maes
Civil Engineering Department, Schulich School of Engineering, University of Calgary,

2500 University Avenue N.W., Calgary, Canada T2N1N4.
Jean Goyet
Bureau Veritas, Marine Division, Research Department, 17 bis Place des Reflets, La
Defense 2, 92400 Courbevoie, France.

Structural Safety, Vol. 31, pp. 168-178, 2009,

doi: 10.1016/j.strusafe.2008.06.016.
-46-
Abstract
The present paper proposes an approach for identifying target reliabilities for
components of complex engineered systems with given acceptance criteria for system
performance. The target reliabilities for components must be consistent in the sense
that the system performance resulting from the choice of the components’reliabilities
satisfy the given acceptance criteria, and should be optimal in the sense that the
expected utility associated with the system is maximized. To this end, the present paper
first describes how complex engineered systems may be modelled hierarchically by
use of Bayesian probabilistic networks and influence diagrams. They serve as
functions relating the reliabilities of the individual components of the system to the
overall system performance. Thereafter, a constrained optimization problem is
formulated for the optimization of the component reliabilities. In this optimization
problem the acceptance criteria for the system performance define the constraints, and
the expected utility from the system is considered as the objective function. Two
examples are shown: (1) optimization of design of bridges in a transportation network
subjected to an earthquake, and (2) optimization of target reliabilities of welded joints
in a ship hull structure subjected to fatigue deterioration in the context of maintenance
planning.
Keywords
Constrained optimization, complex system, acceptance criteria, Bayesian probabilistic
network, influence diagram.
3.1. Introduction
Typically engineered systems are complex systems comprised of geographically
distributed and/or functionally interrelated components, which through their
connections with other components provide the desired functionality of the system
expressed in terms of one or more attributes. This perspective may indeed be useful for
interpreting and modelling a broad range of engineered systems ranging from
construction processes over water and electricity distribution systems to structural
systems. One of the characteristics of engineered systems is that, while the individual
components may be standardized in regard to quality and reliability, the systems
themselves often cannot be standardized due to their uniqueness. The performance of
the systems will depend on the way their components are interconnected to provide the
functionalities of the systems as well as on the choice of reliabilities of their
components. Thus, the design and maintenance of such systems effectively concern the
requirements to the reliability of their components, which can be translated from given
requirements to the attributes of the performance of systems in accordance with the
way the components are connected.
-47-
Due to the complex nature of the problem, modelling and optimization of such systems
generally require that different levels of analyses provided by different experts and
supported by data are integrated interdisciplinary. Taking basis in engineered
structures, at component level physical failure mechanisms may be analyzed, such as
yielding, fracture and corrosion. The component failure modes now constitute the
building stones for the development of systems failure modes including the formation
of failure modes for sequences of sub-systems, for which the corresponding
consequences may be assessed. An optimization of the target reliability for components
of a given system, i.e., a system with a given interrelation between its components,
must take basis in such analyses. Seen in this light, it is useful to hierarchically
establish models for complex engineered systems which accommodate for the
integration of the different levels of analyses. Such a hierarchical approach may also
prove to be beneficial as a mean of communication between professionals representing
the expertise required for the modelling of the performance of the different types of
components, sub-systems and systems.
The present paper addresses the problem outlined in the foregoing in the context of a
hierarchical system modelling developed for risk assessment of engineered systems by
the Joint Committee on Structural Safety (Faber et al. (2007b)), where taking basis in
structural systems a framework is formalized in regard to how the hierarchical system
model can be established and then applied to optimize the reliability for components of
structures based on specified requirements to the acceptable risks for the considered
structural system.
The present paper first provides a short summary of available techniques on the
modelling of complex systems. Following this, a general approach for the optimization
of the reliability of system components with given criteria to the acceptable system risk
is proposed. The proposed approach is composed of three steps; (1) adaptation of
Bayesian probabilistic network and influence diagram representation for hierarchical
system modelling, (2) linking of acceptance criteria for system level to component
level through the Bayesian probabilistic networks and the influence diagrams, and (3)
optimizing the target reliabilities of individual components. The original contribution
of the presented approach is the effective use of the commonly available techniques,
i.e. Bayesian probabilistic networks, influence diagrams and generic algorithms for
constrained optimization problems. The approach suggested allows for the assessment
of optimal target reliabilities for the individual components of systems for which the
risk acceptance criteria are specified in regard to the system performance. The
proposed approach is most useful in cases where (1) the components that constitute the
system or the sub-system can be categorized into groups with identical probabilistic
characteristics and/or (2) the components are hierarchically related. Finally, two
illustrative examples are provided. The first example addresses the design of bridges in
a transportation network subject to earthquake hazards. Through this example the
individual steps of the proposed approach are explained. The second example
-48-
considers a floating production storage and offloading unit (FPSO), which constitutes a
typical complex engineered system. In this example, the target reliabilities of welded
joints subject to fatigue deterioration in the framework of inspection and maintenance
planning are optimized with given acceptance criteria for the performance of the ship
hull structure as a whole.
3.2. Problem setting
3.2.1. Modelling of complex systems

The requirements to the probabilistic modelling of complex engineered systems in the
context of risk based decision making concern the consistent and tractable
representation of the physical characteristics of the considered system and the
appropriate detailing to facilitate the assessment of the benefit associated with different
decision alternatives. In addition, of course the modelling should facilitate an efficient
analysis of the probabilities and consequences required for the ranking of decision
alternatives. Fault tree analyses comprise classical techniques for the representation
and analysis of systems failure modes, see e.g. Vesely et al. (1981). Assuming that
components in a system have only two states (failure and success) and that the
component failures are statistically independent, the probability that a predefined state
of the system (top-event) occurs may be quantitatively assessed (Bobbio et al. (2003)).
Fault tree analyses have been applied to a variety of fields, e.g., among others, risk
assessments of nuclear power plants (USNRC (1975) and USNRC (1990)) and the
reliability analysis of control systems for gas turbine plants (Bobbio et al. (2003)).
Fault tree analysis is from a technical perspective relatively simple, and hence in many
ways attractive, however, for the same reason subject to important limitations. Among
these limitations, the difficulty in representing dependencies between basic events as
well as the problems associated with updating based on new information should be
mentioned. Bayesian probabilistic networks (BPNs) and influence diagrams (IDs)
seem to provide an interesting and promising alternative to the classical techniques for
system analysis. Any fault trees can be mapped into BPNs as is shown in Bobbio et al.
(2001). The BPN approach for systems modelling has been utilized for the analysis of
structural systems, see e.g. Baker et al. (2007). The applications of BPNs in the context
of hierarchical modelling are briefly reviewed in the subsequent section.
When modelling the performance of systems it is important to consider temporal

aspects. Petri Nets provide a powerful platform based on which temporal dependencies
associated with e.g. repair or replacement actions which may provoke cyclic references
to states of the components in the model can be accounted for, see Volovoi (2004).
However, the evaluation of the reliability of a given system through a Petri Net often
takes basis in Monte Carlo simulation, which in general requires a considerable
amount of computational effort, and the generic algorithms applicable to a broader
range of problems are not yet available. BPNs are not immediately appropriate for the
-49-
representation of cyclic effects; however, by introducing time slices in a BPN

(so-called dynamic BPN), BPNs may also be applied for such analysis. Several
efficient time slice BPN algorithms have been developed for calculating probabilistic
characteristics of state variables of BPNs, e.g. expected values and conditional
probabilities, see e.g. Kjaerulff (1995). It should be noted that a dynamic BPN
representation is equivalent to a Markov chain representation (Smyth (1997)).
Another approach for the probabilistic modelling and analysis of complex systems is
proposed by Der Kiureghian and Song (2008). In this approach, the probability of an
event of interest (related to the system performance) is formulated as a sum of the
probabilities of the mutually exclusive combinations of the component states that
govern this event. Upper and lower probability bounds on the system performance are
calculated based on an out-crossing formulation and using linear programming
techniques. Moreover, it is shown in Der Kiureghian and Song (2008) that by
aggregating several components as "super-components" and applying the linear
programming method in a hierarchical way, the approach provides reasonable
probability bounds on the system performance with a manageable computational effort.
However, the applied scheme for component aggregation affects the efficiency of the
computation and the width of the obtained probability bounds. An optimization of the
aggregation scheme in principle requires trial and error, although general guidelines
are provided in Der Kiureghian and Song (2008).
3.2.2. Bayesian hierarchical modelling

The applications of the Bayesian hierarchical models range from, for instance,
sociology, biology, environmental studies to engineering. In experiments in sociology,
e.g., experiments for studying school effect in educational research, it is difficult to
control all the experimental conditions. Ignoring dependences between the
uncontrolled experimental conditions at different levels - for the example of school
effect, student level, classroom level and school level - and applying simple statistical
analysis are proven to produce misleading results as is summarized in Raudenbush and
Bryk (1986). Raudenbush and Bryk (1986) propose a hierarchical approach for
studying school effect taking basis in the Bayesian multi-level linear model proposed
by Lindley and Smith (1972). It provides a flexible statistical tool for estimating how
variations in school policies and practices influence educational processes, whereby
the different levels of interrelations are taken into account. Environmental sciences
face similar situations where due to the complex nature of processes and interactions
between systems, observing all the relevant variables that may influence the process of
interest is not realistic. Furthermore, it is difficult to realize the identical conditions in
different experiments. Thus, the comprehensive use of data obtained for different
conditions is necessary for efficiently estimating the parameters of the models, see
Clark and Gelfand (2006). In these contexts the Bayesian hierarchical models are
employed in such ways that the causal relation or interrelation of variables at different
-50-
levels in whole systems are first established based on scientific knowledge without
specifying the probabilistic characteristics of the variables or assuming weak prior
distributions. The parameters of the variables are then estimated or updated using
observed data. Other applications of Bayesian hierarchical models can be found in the
area of pattern categorization/recognition, see e.g. Li and Pietro (2005) and George and
Hawkins (2005). Due to the characteristics of the applications of the models for the
pattern categorizations or recognitions, it is important that these models allow for
promptly updating the parameters in the models for a broader range of objects. To this
end, flexible representations and systematic learning algorithms which the BPN
approach provides are extensively utilized. The Bayesian hierarchical approach has
been applied also for engineered complex systems. Among others, Johnson et al.
(2002) apply the hierarchical model for estimating the reliability of missile systems,
where the fault tree analysis is extended using the Bayesian approach to accommodate
the integration of available expert knowledge and data.
Emphasizing the difference of the use of the Bayesian hierarchical models, the present
paper appreciates the fact that input-output relations of phenomena in engineering at
different levels are often quantitatively available in probabilistic terms. For instance,
given the geometry and material properties of an engineered component, it is possible
to calculate the probability of failure of the component using data and by physical
modeling and analysis techniques, e.g. finite element methods. Fatigue deterioration
can be probabilistically modelled for given environments, using physical models and
data, see Straub (2004). As the events of interest such as component failure and fatigue
degradation are subject to given circumstances, which themselves might be associated
with uncertainty, the probabilities of the events are appropriately represented in terms
of conditional probabilities. Therefore, in the context of modelling of complex
engineered systems, the main focus is how the system can be hierarchically modelled
using these conditional probabilities of components at different levels.
As observed in the above the applications of Bayesian hierarchical models are rather
diverse. However, all Bayesian hierarchical models utilize generic algorithms
developed for estimating parameters and/or obtaining conditional or posterior
distributions. The algorithms themselves are indifferent to the contexts where the
Bayesian hierarchical models are employed.
3.2.3. Optimization of engineering decisions under constraints

It is often the case that the optimization of decisions for engineering systems must be
performed under constraints. These constraints are typically given a priori to the
decision problems in terms of acceptance criteria regarding risks and/or practical
operational limitations. Acceptance criteria are generally defined for the attributes of
the performance of systems considering the consequences due to possible failures.
Recent design codes e.g. ASCE7-98 (2000) provide acceptance criteria in terms of
-51-
minimum requirements to structural performance. The Joint Committee on Structural

Safety (JCSS (2001b)) recommends different target reliabilities for engineered
structures in accordance with the different magnitude of the consequence of failure as
well as the relative cost of safety measures. Also, safety to personnel must be
considered. Recently, a general principle for evaluating the acceptability of a life
saving measure has been proposed using the concept of life quality index (LQI), e.g.
Nathwani et al. (1997) and Rackwitz (2002). Based on the LQI principle it is possible
to optimize and specify requirements for the reliability of engineered systems based on
the costs of improving their reliability. Additionally, several practical constraints, e.g.,
available budget, cost-benefit ratios and allowable environmental impacts, may be
given for projects involving design and maintenance of engineered systems. Together
with acceptance criteria given from normative perspectives, these exogenously given
constraints constitute important boundary conditions for the optimization of the
performance of engineered systems.
A number of approaches have been proposed for optimizing decisions under

constraints in engineering (e.g. Royset et al. (2003), Guikema and Pate-Cornell (2002)
and Salazar et al. (2006)). Thereby, one of the central issues is how the optimization
process can be transformed in such ways that it allows for the utilization of commonly
available techniques for the probability calculations as well as for numerical
optimization. Royset et al. (2003) propose algorithms for reliability-based optimal
design problems with which the required calculations of reliability and optimizations
are completely decoupled, hence, allowing for a flexible choice of the optimization
algorithm and the reliability calculation method. Guikema and Pate-Cornell (2002)
propose a method for the optimization whereby the performances of engineered
systems are related discontinuously to decision variables. These approaches are in fact
highly sophisticated and also efficient in the treatment of some optimization problems.
However, for the same reason they may be cumbersome to apply in practical situations
where complex engineered systems are of interest, since different levels of models
established by different experts must be reformulated to fit the format which these
approaches require. To overcome this difficulty Bayesian probabilistic network and
influence diagram representations are employed in the present paper as is described in
the following sections.
3.2.4. Objective of proposed approach

The acceptance criteria mentioned in the foregoing may be seen to constitute the
boundary conditions, which any engineered system must satisfy during its service life.
The present paper takes the standpoint that the acceptance criteria for systems are a
priori given. This situation is often the situation encountered in practice. The goal of
the present paper is to establish an approach for the optimization of the target
reliability for components of systems for given system performance requirements in
terms of acceptance criteria, by minimizing life cycle costs for the design and
-52-
operation of the system, or more generally by maximizing the service life expected
utility.
3.3. Proposed approach
3.3.1. Hierarchical system modelling with Bayesian probabilistic networks

A hierarchical system modelling for complex systems facilitates the representation of
complex systems at an early stage of risk analysis, e.g. at the concept evaluation, but
may also serve to optimize the final design as well as the management of the risk
during operation. Hierarchical BPN models appear suitable as a platform for modelling
complex systems, since they provide a causal and mind mapping representation of the
system characteristics and functionalities. In Figure 3.1 it is illustrated how the system
functions are represented in terms of a hierarchical aggregation of components and
their interrelations. At the same time the requirements to the system performance may
be disaggregated into reliability performance requirements for the components. In what
follows, the proposed approach is explained in accordance with Figure 3.1.
Figure 3.1. Hierarchical modelling and translation of acceptance criteria.
Let A and E = ( E1 , E2 ,..., En ) denote the sets of possible actions and possible states
of a system respectively. The combination of a ∈ A and e ∈ E specifies the joint
probability conditional on the action P[e | a ] and the consequences
C(a, e) = (C1 (a, e), C2 (a, e),..., Cm (a, e)) . In general these quantities are the functions
describing how the components and the sub-systems in the system are interconnected.
However, in the following it is assumed that the interconnectivity is fixed. Note that
the consequences C( a, e) may be a vector when two or more attributes of the system
performance are considered, e.g. financial cost, fatalities and damages to the qualities
of the environment. It is assumed that the consequences C( a, e) can be represented as
an attribute-wise sum of the consequences C A (a) associated with action a and the
consequences CE (e) associated with event e , namely
C(a, e) = C A (a) + CE (e) (3.1)
A Bayesian probabilistic network is a probabilistic model representation in terms of a

directed acyclic graph that consists of nodes representing uncertain state variables,
-53-
so-called chance nodes and edges that logically link the nodes, and conditional
probability assignments, see Figure 3.2 for example, and see e.g. Jensen (2001) for
general introduction. An influence diagram (ID) is an extension of a Bayesian
probabilistic network that includes so-called decision nodes and utility nodes in a
graph in addition to chance nodes. Using the chain rule for Bayesian probabilistic
networks (Jensen (2001)), the joint probability P ( E | a ) can be decomposed as
P ( E | a ) = ∏ P ( Ei | pa ( Ei ), a ) (3.2)
i
where pa( Ei ) is the parent set of Ei . From Equation (3.2) it can be seen that the
joint probability P ( E | a ) can be built up by conditional probabilities. Any marginal
probabilities of the states of the subset of E can be efficiently calculated from the
joint probability P ( E | a ) with generic algorithms and software tools commonly
available, see the appendix of Korb and Nicholson (2004). For the BPN shown in
Figure 3.2, the parents of E3 are the nodes E1 and E2 , and the node E2 is a
function of action A . The joint probability is then written as
P( E | a) = P( E3 | E1 , E2 ) P( E1 ) P( E2 | a) (3.3)
Each term in Equation (3.3) thus the joint probability is fully characterized by the
conditional probability tables shown in Figure 3.2.
Figure 3.2. Example of a BPN and conditional probability tables.
Let F(C, P) = ( F1 (C, P), F2 (C, P),..., Fl (C, P)) denote a vector function of C(a, e)
and P ( E | a ) . For instance, the expected total cost, may be one of the attribute of a
system performance to be considered, and is written as one element of F (C, P ) as
Fi (C, P ) = ∑ Ci (e, a ) P (e | a ) (3.4)

e∈E
-54-
where Ci (⋅, ⋅) represent the cost. The probability that the damage to environmental
quality exceeds a given threshold cacc may be another element of F (C, P ) and is
written as
F j (C, P ) = ∑ I ⎡⎣C j (e, a ) > cacc ⎤⎦ P (e | a ) (3.5)

e∈E
where C j (⋅, ⋅) represents the environmental damage and I [⋅] is the indicator
function, which returns unity if the condition in the bracket is satisfied and zero
otherwise. Such environmental damages may be represented e.g. in terms of release
volumes, the geographical release extent and/or temporal release extent of agents. The
conditional expected value of the number of fatalities given the state Em = em may be
other element of F (C, P ) and is written as
∑
e '∈E \ Em
Ck (a, (em , e ')) P((em , e ') | a )
Fk (C, P) = (3.6)
∑
e '∈E \ Em
P((em , e ') | a)
where Ck (⋅, ⋅) represents the number of fatalities and e ' ∈ E \ Em

= { E1 , E2 ,..., Em−1 , Em +1 ,..., En } . Note that any functions represented in terms the
elements of F (C, P ) can be systematically calculated by the algorithms developed for
the analyses of BPNs and IDs when the state variables E = ( E1 , E2 ,..., En ) and their
interrelations and the (conditional) probabilities corresponding to the interrelations of
the variables are defined in an ID, see e.g. Jensen (2001). Thus, the remaining task for
developing models for engineered complex systems is to represent the physical
understanding, the relevant experience and the data available at different hierarchical
levels in terms of (conditional) probabilities of states of variables or in terms of
decision nodes or utility nodes, and then link them together. Thereby, the general
characteristic that engineered systems are comprised and built up by components,
which are standardized by codes and industrial standards in regard to quality and
reliability may add value to the use of object-oriented BPN representations. This
special type of BPN models allows for creating classes of BPNs, which are
representative for sub-systems that have identical characteristics, see e.g. Bangso et al.
(2003) and Bangso and Olesen (2003).
3.3.2. Objective function and constraints

Having established the hierarchical system model in terms of IDs, the objective
function such as service life utility or expected total cost may be assessed from the ID
as a function of the chosen action utilizing the functional representation of F (C, P ) as
shown in the previous section, i.e.:
-55-
u (a) = F1 (C(a, ⋅), P(⋅ | a)) (3.7)
Acceptance criteria are typically defined in regard to the functionality or performance

of the considered system measured in terms of risks and/or probability of failure. Since
the design and maintenance of a system usually specifically addresses the components
of the system, it is of interest how the acceptance criteria for the components may be
derived from the acceptance criteria specified for the system performance. Thus, the
optimization of reliabilities for components in a system constitutes an inverse problem,
see Figure 3.1. The acceptance criteria for the system performance can be related to the
target reliabilities for the components using the function type of F (C, P ) as is shown in
the previous section as
Fi (C(a, ⋅), P(⋅ | a)) ≤ ci , ( i = 2,3,..., m ) (3.8)
where Fi ( i = 2,3,..., m ) represent the functions on the ID calculating the quantities

for which the acceptance criteria for the system are defined, and ci are acceptance
levels for the corresponding quantities.
3.3.3. Optimization of actions for components of complex system

Since several combinations of target reliabilities for different components in a system
may satisfy the prescribed acceptance criteria for the system, the optimal combination
of target reliabilities for components may be identified as the combination of the target
component reliabilities associated with action a which maximizes the expected
utility u using Equations (3.7) and (3.8) formulated in accordance with the previous
sections as
Maximize u (a ) = F1 (C(a, ⋅), P(⋅ | a)) s.t.

(3.9)
Fi (C(a, ⋅), P(⋅ | a)) ≤ ci , (i = 2,3,..., m)
Since the functions Fi ( i = 1, 2,..., m ) are readily calculated, the problem is reduced to
a standard non-linear constrained optimization problem for which efficient algorithms
are available, see e.g. Press et al. (1988).
3.4. Example 1
This example considers the simple optimization of the design of bridges subject to
earthquake hazards. The aim of this example is to explain in detail how the proposed
approach may be applied in practical situations. The bridges b1 , b2 and b3
geographically connect the location a with c , and thus constitute the system
components in a transportation network system, see Figure 3.3. It is assumed that the
state of the system is fully described through the combinations of the states of the three
-56-
bridges, and hence, the failures of e.g. the road sections besides the bridges in the
network are not considered. The system failure is assumed to be defined as the joint
failures of all three bridges. The objective function to be minimized is the expected
discounted total cost, which consists of the initial cost and the expected cost associated
with the failures of bridges. The acceptance criteria are assumed to be given for (1) the
expected number of fatalities in the system given that an earthquake occurs as 10, and
(2) the conditional probability that the system fails given that an earthquake occurs as
1%. The life time considered in the design of the bridges is 100 years, and it is
assumed that an earthquake occurs at most once in the system's life time. The
discounting rate applied for evaluating the future costs is assumed equal to 3% per
annum.
Figure 3.3. Transportation network system.
3.4.1. Model description

The earthquake hazard is modelled in the earthquake class BPN as is shown in Figure
3.4 (left). It consists of five nodes, namely, "Scenario", "Time", "V1", "V2" and "V3".
The node "Scenario" contains different possible earthquake scenarios with
corresponding probabilities. The term scenario may refer to an earthquake occurring at
different seismic zones and different faults, or more specifically, different
combinations of the values of ground motion intensities at different locations. The
latter corresponds to the cases where the joint probability density of ground motion
intensities at different sites is identified by seismic hazard analyses and thereafter the
joint probability density is discretized into a finite number of probabilities
corresponding to the intervals of the ground motion intensities at different sites. When
the different combinations of the values of ground motion intensities are taken as the
identifiers of the scenarios, the spatial correlations between the intensities at different
locations can be suitably considered in the earthquake hazard model. In this example,
however, for illustrative purposes only one scenario "eq1" is considered.
The node "Time" specifies the probability of the yearly discretized time T when the
scenario eq1 occurs. T is assumed to follow a geometric distribution with an
occurrence probability for each year given as νΔt = 0.01 . The nodes "V1", "V2" and
"V3" represent the logarithms of the peak ground accelerations ( cm / s 2 ) at the
locations where the bridges b1 , b2 and b3 are to be built, and are assumed to follow
normal distributions given the scenario eq1 with the parameters shown in Table 3.1.
-57-
Figure 3.4. Classes of BPNs for Earthquake hazard (left) and for Bridge (right).
Table 3.1. Assumed distributions of nodes in BPNs and ID.
Variables Distributions Bounds

Earthquake class BPN
Scenario P[Scenario=eq1]=1
V1|eq1 Normal (ln200, 0.5) [0, 9]
V2|eq1 Normal (ln300, 0.5) [0, 9]
V3|eq1 Normal (ln400, 0.5) [0, 9]
Time|eq1 Geometric (0.01) [1, 100]
Bridge class BPN
A Normal (ln2, 0.1) [0, 2]
Theta1 Normal (ln1, 0.1) [-0.5, 0.5]
ID for transportation network system
Theta2 Normal (ln1, 0.1) [-0.5, 0.5]
X1,X2 and X3 given design alternative a1 Normal (ln600, 0.1) [0, 9]
(Normal ( μ , σ ) abbreviates the normal distribution with the mean μ and the standard deviation σ ,
and Geometric ( p ) abbreviates the geometric distribution with occurrence probability p . The
geometric distribution is discretized by the interval of 1 and the Normal distributions are discretized by
the interval of 0.1 when implemented into the conditional probability tables in the BPNs. The last
column shows the upper and lower bounds in the corresponding conditional probability tables.)
When the probabilistic characteristics are implemented into the conditional probability
table in BPNs they have to be discretized. The intervals and the upper and lower
bounds must be chosen carefully assuring the efficiency and accuracy of the
discretization. They are chosen in this example as shown in Table 3.1. Note that the
BPN in Figure 3.4 (left) assumes that "V1", "V2" and "V3" are conditionally
independent given the scenario. The nodes "Time", "V1", "V2" and "V3" (surrounded
by the bold line) are output nodes, and are connected to other nodes in the BPN for the
transportation network system, Figure 3.5.
-58-
Figure 3.5. ID for transportation network system (cost).
The bridges are modelled in the Bridge class BPN as shown in Figure 3.4 (right). The
bridges b1 , b2 and b3 are assumed to be identically modelled through the Bridge
class BPN. However, the different probabilities in the input nodes "V", "X" and
"Theta2" (highlighted with bold dashed line) facilitate the differentiation between the
resistances of the bridges and the corresponding probabilities of failure. In the Bridge
class BPN, S denotes the load effect, which is represented by
S =V + A (3.10)
where A represents the logarithm of the soil amplification factor. A is assumed to
follow a normal distribution with the parameters given in Table 3.1. The resistance R
of the bridge is modelled as
R = X + Θ = X + (Θ1 + Θ2 ) (3.11)
where X specifies the design of the bridges and Θ represents the uncertainties
associated with the resistance of the bridge. Θ can be decomposed into two types of
uncertainties, Θ1 and Θ2 . Θ1 is the uncertainty associated with individual
realizations of bridges, and can be assumed independent between the different bridges,
whereas Θ2 denotes the common uncertainty that affects all realizations of bridges
thus introduces the statistical dependence. For example, uncertainty on material
geometry or uncertainties associated with construction work may belong to the former
type of uncertainty. Modelling and statistical uncertainties belong to the latter type of
-59-
uncertainty. The assumed probabilistic characteristics of Θ1 and Θ2 are shown in

Table 3.1. The failure of a bridge, which is defined as the event R < S , is denoted by
the Boolean node "F", and the probability of failure is expressed as
P[ F = ' true '] = P[ R < S ] (3.12)

The node "F" is the output node from the Bridge class BPN and is utilized for the
assessment of consequences in the ID, see Figure 3.5. Figure 3.5 shows the ID for the
whole transportation network system. "Earthquake" is an instance of the Earthquake
class BPN, and "Bridge_1", "Bridge_2" and "Bridge_3" corresponding to b1 , b2 and
b3 , respectively, are instances of the Bridge class BPN, for which only input and output
nodes are shown. The node "Fsys" represents system failure, which is connected with
the nodes "F1", "F2" and "F3" representing the individual failures of the bridges b1 , b2
and b3 , respectively. These are required for checking if the acceptance criterion is
satisfied for the conditional probability of system failure given that an earthquake
occurs. The node "Theta2" specifies the probability distribution of the common
uncertainty Θ2 , see Table 3.1. Finally, the decision node "D" represents the set of
design alternatives for the three bridges. Three design alternatives a1 , a2 and a3 are
considered for each bridge, hence, there are 33 = 27 possible actions in the decision
node. The nodes "X1", "X2" and "X3" represent the probability distribution of state of
the bridges b1 , b2 and b3 respectively, corresponding to the choice of the design
alternatives, see in Table 3.1. For each action, the corresponding initial cost is defined
in the utility node "Cx" whose values are shown in Table 3.2. The utility node "Ce"
defines the discounted failure costs for all combinations of the states of the three
bridges for each year up to 100 years. The failure costs assumed in the example are
shown in Table 3.3. From the utility nodes "Ce" and "Cx" the expected discounted total
cost is calculated. Similarly, the expected number of fatalities in the system given that
an earthquake occurs can be calculated with a similar ID as the one shown in Figure
3.6. In the figure input and output nodes of the instances of the class BPNs (earthquake
class, bridge class, design class and consequence class) are abbreviated. The summary
of the magnitudes of the consequences are given in Table 3. Failure costs and fatalities
shown in the tables should be considered as the expected values over possible
consequences given the states of the bridges when an earthquake occurs. In practice the
development of the table requires that the consequences must be analyzed for all
possible combinations of the states of all bridges in the network. While it requires
considerable efforts, it allows for flexibility considering the significance of each bridge
in the network, e.g. consideration of the topology of network.
Table 3.2. Initial costs.
Design alternative Initial cost (Monetary unit)

Design alternative a1 10
-60-
Table 3.3. Failure costs and fatalities.
State of Bridge
Bridge 1 NF F
Bridge 2 NF F NF F
Bridge 3 NF F NF F NF F NF F
Failure cost (Monetary
unit) 0 10 10 50 10 50 50 200
Fatality 0 10 10 20 10 20 20 30
(Failure costs are not discounted. F and NF are abbreviations for failure and no failure, respectively.)
Figure 3.6. ID for transportation network system (fatality).
3.4.2. Results
The expected discounted total costs, the expected number of fatalities and the
probabilities of system failure given that an earthquake occurs for the 27 possible
actions are calculated using the established IDs. The result is shown in Figure 3.7. At
the bottom of the figure the correspondence between the actions and the combinations
of the design alternatives for the three bridges is also shown. The optimal action
consistent with the two acceptance criteria regarding the expected number of fatalities
and conditional probability of system failure given the occurrence of an earthquake is
identified as action 25 (design alternative a3 for the bridges b1 and b2 , and design
alternative a1 for the bridge b3 ); action 17 results in the minimum expected
discounted total cost, but it does not satisfy the acceptance criteria. The strategy behind
action 25 may be interpreted as follows; considering the non-linear relation between
the number of failed bridges and the failure costs, a sound strategy may be to avoid, by
all means, the simultaneous failures of the three bridges in an economically efficient
way, which may be realized with higher reliabilities for one or two of the three bridges
and comparatively low reliability for the other bridge(s). Since the earthquake hazard
-61-
is smallest for bridge b1 , the highest reliability of the system can be realized most
efficiently through bridge b1 and be realized relatively efficiently for the bridge b2 ,
by adopting the design alternative a3 for the bridges b1 and b2 ; corresponding to
the highest design resistance in the three design alternatives. At the same time, by
accepting a relatively higher failure probability for bridge b3 , the expected discounted
total cost can be reduced. This becomes clearer by comparing action 25 with action 9,
which is composed of the same set of design alternatives but applied for different
bridges, i.e., a1 for the bridge b1 and a3 for the bridges b2 and b3 . Action 9
requires the same initial cost as action 25, and results in almost the same amount of the
expected discounted total cost, but significantly high conditional probability of system
failure given an earthquake. This strategy seems tricky, and may not be considered in
practical situations where typically the resistances of structures may be designed in a
proportional way to the magnitudes of hazards. However, from a system optimization
point of view, this is the best strategy that satisfies the acceptance criteria given for the
system. It should be noted that in practical situations decision makers might accept
slightly higher costs to further reduce the risk of fatalities (e.g. Action 27 instead of
Action 25 in this example). However, if the objective function and the constraints are
established to fully represent the decision maker's preference, such a subjective
decision may lead to sub-optimal decisions.
-62-
Figure 3.7. Expected discounted total cost, and expected fatality and probability of
system failure given that an earthquake occurs.
3.4.3. Discussion
The hierarchical Bayesian approach provides a clear perspective of how the whole
system should be built up using the modules representative of different levels of
analyses. In this example, the transportation network system can be built up with four
modules, i.e., earthquake module represented by the earthquake class BPN, a bridge
module represented by the bridge class BPN, a design module and a consequence
-63-
module, see Figure 3.5. These modules can be built up separately, whereas the
interfaces between the modules must be specified. Such a module oriented modelling
in the hierarchical Bayesian approach not only enhances the integration of the
knowledge of different experts, experience and data available at different levels, but
also increases the productivity of risk assessments, since the modules are re-useable.
Updating of the probabilistic characteristics in BPNs is of practical use, although this

aspect is not emphasized in the example. For instance, when the data on the damage
states of the bridges and the load effects are obtained after the occurrence of an
earthquake, the uncertainties associated with the resistance of the bridges can be
updated by conditioning the corresponding nodes. Hence, the updated probability can
be used for future risk assessment.
While only a small number of discrete action alternatives are considered in this
example, there are other cases where a large number of discrete action alternatives or
continuous action alternatives are to be considered. In such cases it is not feasible to
perform the ID analysis for every action, thus adaptation of efficient algorithms for
solving optimization problems under constraints are needed. In this context, IDs serve
as the function in the process of calculating the value of the expected utility and the
values of the quantities for which acceptance criteria are defined which then in turn can
be implemented into optimization algorithms. In the next example, it is shown how this
may be realized using commonly available software tools.
3.5. Example 2
Optimal reliability for components in Floating Production Storage and Offloading
Units (FPSOs) subject to fatigue deteriorations is considered in this example. The main
function of FPSOs is to produce and store oil at offshore oil fields with given
requirements to reliability in production and safety to persons and environment.
Typically considered events of system failure for FPSOs are:
• Loss or damage of ship due to loss of buoyancy or explosions/fires.

• Loss of production due to reduced functionality.
• Loss of lives due to foundering or explosion/fires.
• Leaks and other damages to the quality of the environment.
Considering the hull as an assembly of components, the hull may be considered to

comprise an assembly of tanks tied together with deck plates, tank partitions, and
bottom and side plates. The individual components are furthermore stiffened by girders
and web frames to ensure a sufficient structural integrity of the hull, see Figure 3.8.
The corresponding hierarchical model representation is shown in Figure 3.9.
-64-
Figure 3.8. Hierarchy of ship hull structure considered.
Figure 3.9. Hierarchical modelling of hull structure.
The hull components as described above have basically two functions, namely, to
ensure that the overall ship has a sufficient structural integrity and provide the means
for containing cargo and ballast. Failure of the components of the hull at this level can
be assumed as the events of:
• Loss of or reduced structural integrity.

• Loss of containment due to explosion.
• Leaks of the individual tanks.
Considering now the individual components as outlined in the above these may be
viewed upon as assembly of plates connected by welded joints. Failure of these
components may lead to:
• Crack or pit through plate thickness.

• Reduced overall plate thickness.
• Joint stiffness reduction or failure.
-65-
Thus, the losses or damages at component level may lead to the hull failure or
undesired economic and environmental losses as well as loss of lives given the way the
components are interconnected. The problem in this example is to optimize the target
reliabilities for the welded joints in plate and tank partition components given the
requirements to the functionality/consequence of the ship hull, e.g. the probability of
hull failure. It is emphasized in this example how commonly available software tools
can be used in accordance with the proposed approach. For this purpose a software tool
is developed using Hugin ® for BPN/ID representation and Microsoft Excel ®
(hereafter Excel) for the optimization algorithm as well as the user interface. In the
subsequent section, the overview of the software tool development is illustrated.
3.5.1. Optimization of target reliability for welded joints in components

The developed software tool provides an easy interface to obtain the optimal target
reliabilities for welded joints subjected to fatigue deterioration. Excel is used as a
platform for integrating the various computational modules and storing information
required for calculations. The Excel platform is linked dynamically to the Hugin
ActiveX server (hereafter Hugin). In order to use the software tool the user has to
define, through Hugin files, the BPNs corresponding to the hierarchical model of the
hull structure as described above. The outputs, i.e. optimized target reliabilities for all
welded joints, are written into the Excel file.
In Figure 3.10, the illustration of the hierarchical Bayesian representation of the ship
hull structure is given. Two BPNs in the top of the figure represent the performances of
tanks. The tank performances are characterized by the states of the plates that
constitute the tanks. As is described above, at this level the possible consequences due
to component failures are capacity reduction, explosion and environmental damage due
to leaks. The ID in the bottom of the figure concerns how the component failures may
propagate and lead to further consequences at system level. Here, three attributes of the
consequences are identified, i.e. economic loss, loss of lives and environmental
damage measured in terms of leak intensities. These BPNs and ID are interconnected
as shown in the figure. In the entire ID the conditional probability tables are assumed
established with the help of experts, see e.g. Figure 3.11 (which is the conditional
probability table for node "Explosion_1" as implemented into a Hugin file), whereas
the nodes that represent the components serve as root nodes whose probabilities are
represented in terms of unconditional probabilities, which are derived from the target
reliabilities for welded joints in each components. Therefore, by changing the target
reliabilities for the welded joints which are set in the Excel file, the unconditional
probabilities for the components are changed accordingly. In turn, the corresponding
probabilistic characteristics, e.g. expected total cost or probability of ship hull failure
are changed and stored in the Excel file, see Figure 3.12. This process is made
automatically through ActiveX. The design and service life maintenance cost for the
different welded joints is in general a function of the target reliability in regard to
-66-
fatigue failure, and this is implemented as a VBA code in the Excel file. For the
assessment of the relationship between the reliability of the welded joints subjected to
fatigue failure and the service life cost, the iPlan software described in Straub and
Faber (2006) may be utilized. Finally, the optimal target reliabilities for welded joints
are obtained using the Solver add-in provided in Excel – target reliabilities
correspond to "changing cells", and acceptance criteria for the ship hull correspond to
"constraints" in the Solver add-in.
Figure 3.10. ID for the tanks and the hull structure.
Figure 3.11. Illustration of conditional probability table.
-67-
Figure 3.12. User interface of developed software tool.
3.5.2. Results and discussion

In this illustrative example, the acceptable probability of system failure is set as 10−3
per annum which constitutes the boundary condition in the optimization problem. The
objective function is the expected total cost including the inspection cost, the repair
cost and the failure cost due to ship hull failure. As is shown in Figure 3.12, different
optimal target reliabilities are obtained for the components in different tanks, reflecting
the different contribution to the system failure. The set of these optimal target
reliabilities correspond to the set of the target reliabilities that satisfy the acceptance
for the probability of system failure and that minimizes the expected total cost.
Although in this example, the exposure to the ship hull structure, e.g. wave load, is not
directly considered and thus the failures of the individual tanks are assumed to be
independent, it is possible to take into account the exposures which may introduce the
correlation between the failures of the components and/or subsystems by adding the
node for the exposure scenario in the ID as is found in the previous example.
3.6. Conclusions
The present paper proposes a framework for the modeling and the optimization of
reliabilities for components in complex engineered systems subject to requirements
specified in terms of system performance. It is shown how the identification of the
target component reliabilities that are optimal and consistent with given acceptance
criteria for system performance can be treated as an optimization problem with
constraints. Appreciating the perspective that engineered systems are built up by
-68-
standardized components which through their connections with other components

provide the desired functionality and that the system performance will depend on the
way the components are interconnected, the proposed framework takes basis in a
hierarchical system modelling facilitated by use of (object-oriented) BPNs and IDs.
Using the established BPNs and IDs it is possible to calculate the objective function
such as service life utility, and the quantities for which the acceptance criteria are
given, both of which are required for solving the optimization problems with
constraints. Two examples are shown: (1) optimization of the design of bridges in a
transportation network subject to earthquake hazards, and (2) optimization of target
reliabilities of welded joints in a ship hull structure subject to fatigue deterioration in
the context of maintenance planning. The first example serves as the introduction how
the proposed approach is implemented step by step. The second example illustrates
how complex engineered system may be modelled and how the target component
reliabilities may be optimized using commonly available software tools.
-69-
Inter-generational distribution of the life-cycle cost of an engineering facility (Paper III)
4. Inter-generational distribution of the life-cycle cost of an

engineering facility (Paper III)
Kazuyoshi Nishijima

Daniel Straub
Institute of Structural Engineering, ETH Zurich, Switzerland8.

Journal of Reliability of Structures and Materials, Vol. 3, Issue 1, pp.

33-46, 2007.
8 As of the date of the paper acceptance.
-70-
Abstract
In decision making for civil engineering facilities, as well as other societal activities,
the criteria for sustainability are inter-generational equity and optimality. Two
challenging questions must be addressed in this context: How to compare the benefits
and costs among different generations and how to compensate and adjust for the
in-homogeneously distributed benefits and costs between the generations. To address
and answer these questions for engineering facilities, first of all the temporal
distribution of the life-cycle benefits must be assessed. To ensure optimality, the total
life-cycle benefits for the facility must be maximized. In the present paper initially the
normative criteria for sustainability are presented. Thereafter it is demonstrated how
the criteria may be implemented for the purpose of optimization of structural design.
The inter-generational distribution of benefits and the implications for sustainable
decision-making are then illustrated by an example considering the optimal design of
the concrete cover thickness of a RC structure subject to chloride-induced corrosion of
the reinforcement.
Keywords
Sustainability, discounting, life-cycle cost, chloride-induced corrosion,
cover thickness.
4.1. Introduction
A significant amount of research has been devoted to life-cycle analysis for civil
engineering facilities. In recognition of the significant uncertainties associated with the
performance of structures over their service life, decision-theoretical approaches have
been applied for the optimization of structural design, e.g., Rosenblueth and Mendoza
(1971) and Rackwitz (2000). The developed methodological framework facilitates the
optimization of the design of structures such that a balance is achieved between the
benefits achieved through the facility and the costs associated with design and
construction, future costs of inspection and maintenance as well as costs associated
with possible repairs, replacements and failures. Recently, life-cycle analysis has been
utilized to enhance a sustainable development of the built environment, e.g. Rackwitz
et al. (2005), Faber and Rackwitz (2004), Nishijima et al. (2004) and Nishijima et al.
(2005). In this context, focus is shifted from the facilities to a sequence of decision
makers and stake holders, each of which represents a subsequent generation that
benefits from the facility while paying the costs of maintenance, repair, replacement
and other adverse consequences. Although life-cycle analysis is well advanced in the
civil engineering field and has been applied within the context of sustainability, less
attention has been paid to the distribution of costs over time. This distribution is
essential, since it allows for assessing the burden of each generation, and thus indicates
the necessity for an inter-generational compensation when the aggregation of benefits
and costs is not uniformly distributed over time.
-71-
The present paper initially formulates the criteria for sustainability and thereafter sets
up a multi-decision-maker framework for inter-generational sustainable decision
making. As it will be discussed this framework may also provide a useful basis in any
intra-generational context for organizations involved in decision making concerning
activities with life times significantly exceeding the budgeting periods or the life time
of the individuals responsible for the decision making within the organization. The
optimization of structural design using the suggested framework is illustrated by an
example considering the optimal design of the cover thickness for a RC structure
subject to chloride-induced corrosion. Finally, the temporal distribution of the
life-cycle costs is explicitly assessed, clearly illustrating how the benefits and costs are
unevenly distributed over the generations.
4.2. Multi-decision-makers and criteria for sustainability

Sustainability is interpreted in accordance with the Rio convention in 1992, following
the report by Brundtland (1987). To facilitate sustainable decision making, two criteria
are provided: 1) inter-generational equity and 2) optimality. Inter-generational equity
dictates equal treatment of the present and all future generations. Optimality can be
interpreted as the maximization of an idealized utility function, considering all
generations and their preferences. These two criteria are strongly interrelated and this
must be taken into account in the decision making. In order to set up the utility
function aggregating the benefits and costs for all generations, the equal treatment of
the individual generations in accordance with the inter-generational equity criterion is
required. Once optimality is obtained by maximizing the idealized utility function, the
temporal in-homogeneity of the utilities among the different generations must be
reconsidered to ensure inter-generational equity.
Basically any kind of activity at present has consequences for the future in terms of
benefits and costs. The benefits and costs may not necessarily be expressed in
monetary terms and there are controversial discussions on whether all societal and
environmental consequences can be measured comprehensively in monetary terms, as
discussed in Turner (1992) and Ayres et al. (1998). However, in the present paper,
benefits and costs are assumed to be represented by monetary values for the
convenience of discussion. The temporal distribution of consequences associated with
different activities differs significantly; however, it is difficult to identify activities
which do not have some effect for the future generations. In case of exploitation of
natural resources the benefit is more or less immediate – but the resources exploited
are no longer available for future generations. In case of disposal of toxic waste the
situation is much the same – the benefit is achieved by the present generation but the
potential adverse consequences are likely to be transferred to future generations.
Sustainability is an issue which always has to be kept in mind.
-72-
Figure 4.1. Schematic distribution of benefit or cost over time.
The schematic benefit or cost path is illustrated in Figure 4.1. A sequence of decision
makers is assumed along with the time, each representing one generation. Since each
generation considers the benefits and costs and makes decisions from its point of view,
an explicit modeling of the different subsequent decision makers is indispensable,
especially when the pure time preference or loss of life are considered in the utility
function.
The benefits and the costs illustrated in Figure 4.1 correspond to the gross values at
each point in time, i.e., they are not discounted. The i th generation enjoys the benefit
or carries the cost of the hatched area. Since this is the gross value, the same values at
different points in time do not necessarily have the same perceived influence to
different generations, mainly because of the economic growth. Therefore, benefits and
costs should be discounted by the economic growth to ensure the equal treatment
between generations in accordance with the inter-generational equity. Taking into
account the economic growth and disregarding the effect of overlapping generations,
the total utility aggregating benefits and costs can be expressed as:
∞
U = ∑ δ (ti )U i (4.1)
i =1
where U is the total utility for all generations, δ (⋅) is the discounting factor
representing economic growth and U i is the utility for i th generation which begins
at t = ti . Extension of Equation (4.1) to cover also the case of overlapping generations
may be performed as shown in Bayer and Cansier (1999), Bayer (2003) and Rackwitz
et al. (2005), however, the effect of this is of minor importance for the overall
life-cycle benefit assessment. When decision making is subject to uncertainty, the
utilities in Equation (4.1) should be interpreted as the expected utilities. The utility for
the i th generation may be written as:
-73-
Figure 4.2. Transfer of benefits, costs and resources.
ti +1
U i = ∫ u (t )γ (t − ti )dt (4.2)
ti
where u (⋅) is the utility per unit time and γ (⋅) is the discounting factor within one
generation. The utility within one generation may be discounted by pure time
preference as well as by economic growth, thus
γ (t ) = δ (t ) ρ (t ) (4.3)
where ρ (⋅) is the discounting factor representing pure time preference. Note that the
discounting factor is related to the discount rate, e.g., for δ (⋅) as:
δ (t ) = exp(−δ t ) (4.4)
where δ is the discount rate per unit time.
Each decision in regard to a civil engineering facility results in one specific temporal
distribution of expected utility and thus enables the calculation of the total utility
according to Equation (4.1). To comply with the second criterion for sustainability, i.e.,
optimality, the total utility must be maximized, which in the case where the benefit
function does not depend directly on the decision corresponds to a minimization of the
total cost. However, even if the maximization is performed under consideration of
inter-generational equity in terms of proper discounting as applied in Equations (4.1) -
(4.3), it does not necessarily imply that each generation obtains the same utility from
the facility, as illustrated in Figure 4.1. It is unlikely that each single activity optimized
in the above sense results in a uniform distribution of the utility among the current and
all future generations. Therefore, the transfer of the benefits in terms of, for instance,
man-made capital or natural resources is essential to achieve inter-generational equity,
see Figure 4.2. The distribution of costs over time provides the basic information
required to achieve inter-generational equity, enabling a comparison and a
compensation between the generations through societal activities which are not
necessarily within the civil engineering field.
-74-
4.3. Equivalent sustainable discount rate

Classical life-cycle cost analysis approaches the discounting problem from the
perspective of the anticipated duration of the considered activity, e.g. the anticipated
service life when a given structure is considered. Furthermore, decision making in
classical life-cycle analysis takes basis in a utility modeling where only the preferences
of the present generation are directly accounted for. This includes also the aspects of
valuation of future benefits and costs through discounting. For a given activity it is
possible to assess a discount rate which if applied in a classical life-cycle analysis
yields the same total expected utility as resulting from the proposed
multi-decision-maker framework (Equations (4.1) - (4.3)). This discount rate is
denoted the equivalent sustainable discount rate γ * by:
∞ ∞
e−γ t u (t )dt = ∑ e−δ tn ∫
tn+1
∫
*
e−γ (t −tn )u (t )dt (4.5)
0 tn
n =1
where u (t ) is the (expected) utility per unit time at time t . The equivalent
sustainable discount rate may be interpreted as the one which, if applied to a decision
problem with the classical one-decision-maker perspective, yields the same total
expected utility as when the decision problem is analyzed from the
multi-decision-maker perspective. In general, it is not possible to obtain an analytical
expression for γ * . However, in the case where consequences are invariant at any time,
the durations of generations τ = tn +1 − tn ( n = 1, 2,3,... ) are constant and the
occurrences of events associated with consequences follow a stationary Poisson
process, the equivalent sustainable discount rate is given as follows:
1 − e −δτ
γ* = γ (4.6)
1 − e −γτ
where δ is the discount rate per unit time by economic growth , ρ is the discount
rate per unit time by pure time preference and γ = δ + ρ , see Faber and Nishijima
(2004). The equivalent sustainable discount rates for several cases are illustrated in
Figure 4.3, where for ρ kept constant at 3% per year or 0% per year for comparison,
the equivalent sustainable discount rates are given as functions of the duration of the
generation τ for several values of δ . The equivalent sustainable discount rate γ * is
smaller than the total discount rate γ , except for the case where ρ = 0 . If the
discount rate consists only of pure time preference ( δ = 0 ), the equivalent sustainable
discount rate is zero, i.e., within the classical framework, the benefits and costs should
not be discounted at all to obtain the same utility function as with the
multi-decision-maker framework. If the discount rate by pure time preference is set
equal to zero and the discount rate by economic growth is set to equal to 5%, the
equivalent sustainable discount rate is equal to 5%, regardless of the duration of the
generation. This means that if the discount rate is only due to economic growth, the
multi-decision-maker framework is identical to the classical framework. In general, the
-75-
Figure 4.3. Equivalent sustainable discount rate γ * for the case of constant utility per
unit time.
discount rates which have been applied so far in the classical framework are too large,
i.e., are leading to non-optimal solutions from the view-point of sustainability.
4.4. Example
Optimal life-cycle cost based design of the concrete cover thickness of a RC structure
subject to chloride-induced corrosion of the reinforcement is considered. The intended
service life time is assumed to be infinite, meaning that the desired function of the
structure is unlimited in time. The applied probabilistic modeling of the degradation
over time is included in Annex A for simple reference and more details are provided in
Faber et al. (2005). The expected life-cycle costs are assumed to consist of the initial
costs CI , the expected repair costs E[CR ] and the expected failure costs E[CF ] ,
which all depend on the optimization variable dnom , i.e. the concrete cover thickness.
It is assumed that visual inspections are made every ΔtI = 5yr and that an indication
of visible corrosion automatically triggers a repair. In accordance with the
renewal-theoretical approach outlined in Faber and Rackwitz (2004), it is assumed that
in case the structure fails, it is reconstructed. Following a repair or a reconstruction, the
structure is assumed brought back to its original state, i.e., described using the same
probabilistic model as a new structure. The realization of the structure after repair or
reconstruction is assumed to be independent from previous structures. Furthermore,
inspections are modeled as being perfect, i.e., visible corrosion is detected with
probability 1 at an inspection. The costs of initial design, repairs and failures are
modeled as:
CI = (1 + aI d nom )C0 (4.7)
-76-
CR = aRCI (4.8)
CF = aF CI (4.9)
with parameter values in Table 4.1, where also the assumed discount rates are
summarized. The initial cost CI is assumed to consist of a fixed cost and the cost
depending on the cover thickness, and the repair cost CR and the failure cost CF are
assumed to be proportional to the initial cost.
Table 4.1. Cost and discount model.
Discount rate for time preference: ρ 3% per year

Discount rate for economic growth: δ 2% per year
Normalizing cost C0 1
Cost ratio for cover thickness aI 0.002
Coefficient of repair cost aR 0.5
Coefficient of failure cost aF 5
4.4.1. Cost distribution over time

In order to calculate the distribution of life-cycle costs over time, an efficient algorithm
is required, since the number of branches in the decision tree develops exponentially
with time. In Nishijima et al. (2004), these costs are calculated by using a recursive
formulation; in the following, a different recursive formulation is provided which
facilitates the explicit calculation of the expected cost of repair and failure at each
point in time. After specifying the decision rule, which defines in which situations a
repair is made, the probability of repair qR (t ) and the probability of failure qF (t ) at
time t ( t = 1yr, 2yr,3yr,... ) for a given realization of the structure are readily
available, see e.g. Faber et al. (2005), Nishijima et al. (2004) and Nishijima et al.
(2005). In accordance with the above, the decision rule adopted in this example is that
the structure is repaired if and only if corrosion is visibly observed at the inspection.
Whether or not this decision rule is optimal is beyond the scope of this paper, which
focuses on the design optimization. However, by consideration of the deterioration
model and the possible actions it is easily seen that, for the present example, there are
only a few reasonable alternative rules. When optimizing the inspection/maintenance
strategy, these alternatives can be compared and the one leading to minimal costs can
be selected, e.g., Straub (2004). According to the probabilistic model in Annex A,
qR (t ) and qF (t ) are estimated by Monte Carlo simulation with 10−6 samples for
each cover thickness, see Figure 4.4 and Figure 4.5. Note that the probability of repair
can be different from zero only at t = iΔtI ( i = 1, 2,3,... ), since the repairs are
associated with inspections which are made at intervals of ΔtI = 5yr ; failure can occur
in any year, but its probability is increasing with time and thus more likely to occur
-77-
Interr-generationnal distributiion of the liife-cycle cost of an enggineering facility (Papeer III)
whenn approachhing the inspections. With thee probabiliities qR (t ) and qF (t ) , the

probability of repair PR (t ) and thhe probabillity of faillure PF (t ) at time t are
calcuulated baseed on the renewal thheory (see,, e.g., Felller (1966) in generall and
Rackkwitz (20000) considerinng applicatiions to civill engineerinng facilitiess) as:
t −1
PR (t ) = qR (t ) + ∑ ( PR ( s) + PF ( s) ) ⋅ qR (t − s ) (4.10
0)
s =1
t −1
PF (t ) = qF (t ) + ∑ ( PR ( s) + PF ( s) ) ⋅ qF (t − s ) (4.11)
s =1
for t = 2yr,3yr,, 4yr,... andd
PR (1yr) = qR (1yr)
( (4.12
2)
PF (1yr) = qF (1yr) (4.13)
for t = 1yr .
Figuure 4.4. Probbability of repair

r qR (t ) at time t for a givenn realizationn of the struucture
(cover thhickness = 50 mm).
-78-
Figure 4.5. Probbability of failure

f qF (t ) at time t for a givenn realizationn of the struucture
(cover thhickness = 50 mm).
The recursive
r foormulationss Equationss (4.10) to (4.13)
( are obtained
o as follows. Th he set
of poossible diffferent evennts leading to a repairr at time t can be spplit into sub bsets:
Thesse subsets are
a differenttiated by thhe time of th he last repaair or reconnstruction, which
w
can occur
o mes t − 1yr , t − 2yr , ettc. until 0yyr ; the latteer corresponnding to thee case
at tim
wherre no repairr or reconstruction haas been perfformed prevviously. Thhe probabiliity of
failure at time t is obtainned analoggously. As the t decisionn rule just specifies qR (t )
and qF (t ) , this recursive formulation
f n can be app plied for anny kind of decision ru ule, as
long as the struucture is reppaired at soome point in n time and reconstructed after faailure,
resullting in idenntical but stochasticallyy independeent structurees.
Oncee the probaability of repair and thhe probabiliity of failurre at each ppoint in tim me are
obtaiined, the calculation of the expected
e coosts is strraightforwaard. Repairr and
reconnstruction after
a failuree can be caarried out ata each insppection timee, the inspeection
interval being 5 years. Figure 4.6 shoows the disttribution off costs over time for seeveral
coveer thicknesses. These costs
c are not discounted. The expeected cost ffor each point in
time consists off the expectted repair costs and the expected failure costts. The exp pected
failure costs aree much smaaller than thet expected d repair costs in the ppresent exam mple,
that is why thhe expectedd total costts are closse to the expectede reepair costs.. The
(non-discountedd) expectedd costs decrrease with time for all cases in F Figure 4.6. This
tendeency is duee to the factt that the faailure rate, which is thhe probabiliity of failurre per
unit time condditional on survival up u to timee t, is decrreasing witth time fo or the
considered deteerioration mechanism.
m . When thee structure performs ppoorly (i.e., the
realizzations of the randoom variables are un nfavorable), it will be repaireed or
reconnstructed allready afterr a few yeaars. After each
e repair or reconstrruction, thee new
-79-
structures are identical but stochastically independent of the old ones. A structure with
an initially bad performance will thus eventually be replaced by one with a good
performance. The expected value of the performance of the structure is therefore
increasing with time and the expected costs of failures and repairs are decreasing. It
should be realized that this tendency depends strongly on the assumed dependency
between subsequent realizations of the structure as well as the characteristics of the
failure rate function.
Figure 4.6. Temporal distribution of expected costs (at 5 year intervals), not
discounted.
4.4.2. Optimization of the concrete cover thickness

Taking basis in the multi-decision-maker framework presented in the previous section,
the total expected costs to be minimized are calculated for each decision alternative
(i.e., for different cover thicknesses). For this example the total expected costs reduce
to:
− E[U (d nom )] = E[C (d nom )]

τ / ΔtI
∞ (4.14)
( )
= CI (d nom ) + ∑ δ (ti ) ∑ E[CR ,ti + jΔtI (d nom )] + E[CF ,ti + jΔtI (d nom )] γ ( jΔt I )
i =1 j =1
which should be minimized. CR ,t and CF ,t are the costs of repair and failure at time
t respectively. Since different discount rates are applied within the generations and
between the generations, the duration of each generation τ must be specified. Figure
7 shows optimal cover thicknesses for different values of the durations of the
generations. With increasing duration of generations, the optimal cover thickness
becomes smaller. This is because the “equivalent sustainable discount rate” becomes
-80-
largeer as the durration of thee generationn becomes longer, see Equation ((4.6) and Figure
F
4.3, and conseqquences in the future are, thereffore, valuedd less. The case wherre the
durattion of a generation
g is infinite correspondds to the classical
c liffe-cycle anaalysis
wherre only onee decision maker
m is asssumed. Ass observed in Figure 4.7, the op ptimal
coveer thickness varies signnificantly wiith the duraation of the generationss, pointing to
t the
impoortance of consideering the problem from the viewppoint of the
multi-decision-m makers.
In thhe followinng, the durration of a generation n is assum

med to be 225 years. When W
applyying the muulti-decisionn-maker fraamework, th he optimal cover
c thicknness is 52m
mm, in
accorrdance withh Figure 4.77. By applyying the classsical frameework (the infinite durration
of generations
g in Figure 4.7), the optimum is at 44m mm. Figure 4.8 shows the
correesponding expected
e coosts with time.
t Thesee costs are discountedd to time t = 0 ,
thereefore, it is possible too compare these
t costs with each other. Withhin the classical
fram
mework the first generration pays less and the t future generations
g s pay more than
withiin the multti-decision-m maker frammework. Th his is due too the fact tthat the classical
fram
mework weigghs values in the futurre less than n the multi--decision-m
maker frameework
throuugh the relaatively higheer discount rate.
Figure 4.77. Optimal cover

c veral durations of geneeration τ .
thickkness for sev
-81-
Figure 4.8. Discounted expected costs for each generation (with a duration 25 years).
4.5. Discussion
Figure 4.8 clearly shows the inhomogeneous distribution of costs among the
generations. In particular the first generation pays much more than all following
generations. In order to comply with the first criterion for sustainability,
inter-generational equity, the temporal differences must be compensated by other
means (e.g., by transferring the benefits on capital stocks and natural resources). Such
compensation is beyond the scope of the analysis as presented in this paper, as it
requires that all societal activities must be considered simultaneously within the
multi-decision-maker framework. In this context it is reminded that, although in the
presented example it is the first generation which pays most, many societal activities
have large consequences in the future while only the current generation directly
benefits from them.
The presented framework can be extended to portfolios of structures, which are

distributed over time and space. The optimization of design and maintenance activities
is performed in analogy to the case of the individual structure, but to ensure
inter-generational equity through compensation, it is required to consider the cost
distribution over time for all structures simultaneously.
The analysis presented here ensures that the second criterion of sustainability,
optimality, is fulfilled in such a way that it is consistent with the first criterion. It seems
paradoxical at first that by consideration of multi-decision-makers (which is required
by the inter-generational equity criterion), the optimal design which fulfills the
-82-
optimality criterion leads to an even more inhomogeneous distribution of costs among

generations. For this reason it is crucial that the issue of compensation between the
generations is also addressed.
The presented multi-decision-maker framework provides an analytical approach to the

consideration of the preferences of all generations involved in the life-cycle of
engineering structures. It allows for the assessment of the effect of postponing costs to
the future through the use of large interest rates, which is a common tendency in
societal decision making. In order to be sustainable, the equivalent sustainable discount
rate presented in this paper must be applied.
Finally, it is important to note that whereas the present paper specifically addresses the
problem of sustainable decision making in an inter-generational context the developed
framework also may be valuable for the decision making in intra-generational contexts
involving several decision makers and stakeholders as well as budgets over time. This
is the situation when decision making is considered in organizations which are
responsible for the design, construction and operation of engineering facilities such as
high-way agencies. In such organizations both budgets as well as the persons involved
in the decision making have a substantially shorter life time than the facilities they are
responsible for. The multi-decision-maker framework may serve to set guidelines or
rules for the decision making in such contexts, to help avoid decisions which for the
fulfillment of preferences of individuals may yield a short term benefit but from an
overall life-cycle perspective induce economical losses for the organization.
Furthermore, the framework can be utilized as a rational basis for long term budgeting.
4.6. Conclusions
It is demonstrated how the inter-generational distribution of the life-cycle cost of an
engineering facility can be assessed. This is of importance for ensuring sustainability
of the facility, whereby the considered criteria for sustainability are inter-generational
equity and optimality. It is shown how decisions regarding an engineering facility must
be optimized in order to comply with these criteria and it is outlined that the results of
the optimization may be used as a basis for a broader discussion regarding
inter-generational equity taking into account all kinds of societal activities. Finally, it is
highlighted that the developed framework also may provide a useful basis in any
intra-generational context for organizations involved in decision making concerning
activities with life times significantly exceeding the budgeting periods or the life time
of the individuals responsible for the decision making within the organization.
The developed decision framework is illustrated by the optimization of the design of a

RC structure subject to chloride-induced corrosion and is found to have a significant
effect on the optimal design.
-83-
4.7. Annex A
For easy reference, the applied probabilistic model for deterioration of concrete
structures subject to chloride-induced corrosion is presented in the following. The
modeling corresponds to DuraCrete (2000) and here follows Faber et al. (2005), where
additional details of the models are described.
Corrosion initiates at the reinforcement, when the chloride concentration has reached
the critical chloride concentration CCr . The ingress of chlorides in the concrete is
described by Fick’s second law of diffusion. Based on this model, the random variable
TI representing the time until corrosion initiation is calculated as:
1
⎛ ⎡ −1 ⎛ ⎞⎤
−2
⎞1− n
d2 Ccr
TI = ⎜ ⎢ erf ⎜1 − ⎟⎥ ⎟ (4.15)
⎜ 4k k k D ( t ) n ⎢⎣ ⎜ AC ⋅ ( w / c) + ε C ⎟⎥ ⎟
⎝ e t c 0 0 ⎝ S S ⎠⎦ ⎠
The parameters of the model are given in Table 4.2.
The time until visible corrosion, corresponding to minor cracking and coloring of the
concrete surface, can be determined based on experience. By adding the propagation
time TP to the initiation time TI , the limit state function for visible corrosion is
written as:
gVC ( t ) = X I TI + TP − t (4.16)
The time between visible corrosion and failure is, for illustrative purposes, represented
by the time TP . The limit state function for failure is thus:
g F ( t ) = X I TI + TP + TP 2 − t (4.17)
Note that the model does not account for the dependency between the propagation time
TP 2 and the environmental parameters or the cover thickness.
The values of the distribution parameters for the random variables in Equations (4.15)
to (4.17) can be obtained as functions of indicators, see Faber et al. (2005). For the
considered example, they are stated in Table 4.2. These values are representative for a
concrete with ordinary Portland cement in a splash environment.
The probabilities of the events visible corrosion and failure can be obtained by e.g.
Structural Reliability Analysis (SRA) or simulation techniques.
-84-
Table 4.2. Example parameters for the deterioration model.
Parameter Description Distribution Dimension A B

d Cover Lognormal mm
thickness
ke Environmental Gamma - 0.924 0.155
factor
kc Curing factor Beta - 0.8 0.1 0.4 1.0
kt Test factor Deterministic - 1.0 -
D0 Diffusion Normal mm2/yr 220.9 25.4
coef.
t0 Reference Deterministic yr 0.077 -
period
n Age factor Beta - 0.362 0.245 0 0.98
Ccr Critical Normal * 0.8 0.1
chloride
concentration
w/ c Water/cement Deterministic - 0.40 -
ratio
ACS Chloride Normal * 7.758 1.36
surface
concentration
factor
eCS Chloride Normal * 0 1.105
surface
concentration
factor
XI Model Lognormal - 1.0 0.05
uncertainty
TP Propagation Lognormal yr 7.5 1.9
time
TP 2 Propagation Lognormal yr 10.0 4.0
time
* Mass-% of binder
-85-
A budget management approach for societal infrastructure projects (Paper IV)
5. A budget management approach for societal infrastructure

projects (Paper IV)
Kazuyoshi Nishijima


Structure and Infrastructure Engineering, Vol. 5, Issue 1, pp. 41-47, 2009.
-86-
Abstract
Life cycle costing analysis is broadly applied as a tool for decision support for civil
engineering structures, whereby the expected total cost over the life cycle of the
structure is advocated as the objective function to be minimized. The present paper
takes the new perspective of considering the problem from a budgeting allocation
problem where the aim is to optimize the allocation of budgets for the purpose of
maintaining the operation of the portfolio of structures. Whereas all the consequences
associated to the project must be taken into account in the life cycle costing analysis, it
is important to distinguish the financial costs which must be paid from the user costs
which represent the follow-up consequences, i.e., opportunity losses. This is because
only the costs to be paid are related to the budget. The present paper proposes an
approach to determine the optimal amount of budget and the optimal maintenance
decisions, considering these two types of cost.
Keywords
Objective function, resource allocation, life cycle optimization.
5.1. Introduction
Over the last decade life cycle costing analysis has gained a widespread interest as a
tool for decision support in civil engineering, e.g., Rosenblueth and Mendoza (1971),
as well as in many other engineering fields. It has been appreciated in research and
practice that the efficiency of engineering projects must be assessed with due
consideration of all benefits and costs induced by the projects on time scales
representative for the actual duration of the projects; only when the life-cycle benefits
are larger than the corresponding costs can an engineering project be considered
feasible, e.g. Rackwitz (2000). The feasibility of engineering projects such as societal
infrastructure must thus be assessed considering all phases throughout their life-cycle –
from the concept phase until the decommission.
As opposed to most private business initiatives, infrastructures built for the purpose of
facilitating the development of society serve functions or are in other ways associated
with benefits and/or costs which on the time scale reach well beyond the duration of
the generations who decide to build them. To ensure a sustainable societal
development, i.e., a development which aims to optimize the objectives of not only our
own generation but also those of the future generations, the assessment of life-cycle
costs must take into account the costs implied for future generations. To this end, life
cycle costing analysis together with an appropriately chosen discounting function, see
e.g., Rackwitz et al. (2005), provides a consistent rationale. The objective function to
be minimized is, in many cases, the expected total cost under the assumption that the
benefits from structures are independent of the decision variables, taking the follow-up
consequences, e.g. reduced benefits due to unavailability, into account as user costs.
-87-
Turning our focus to practical situations, however, the decision makers which are
responsible for the maintenance of portfolios of structures request budgets which are in
excess of the expected total costs. The reason for this is obviously in part that their
success as decision makers is measured in terms of whether they are able to meet their
requested budgets and at the same time are able to keep their portfolio of structures in
operation. It may well be that they request more if the lack of budget leads to serious
consequences such as user costs associated with reduced functionality of roadway
systems. Given this practical constraint, an optimal decision which minimizes the
expected total cost does not necessarily lead to an optimal budgeting from a societal
point of view, corresponding to a resource allocation of the society maximizing the
societal net benefit. Thus the optimization of the decision and the total budget by
maximizing the societal benefit becomes an issue in the context of optimal societal
resource allocation.
The present paper proposes an approach to identify optimal decisions related to

maintenance of structures and budget allocation by maximizing the expected net
benefit, where the net benefit is composed of the benefit achieved through the
operation of the considered portfolio, the allocated budget, the financial cost to be
paid, the user cost and additional user cost which arises from the delay of maintenance
activities due to the possible lack of budget. An example of the maintenance of a
portfolio of RC structures subject to chloride-induced corrosion is given to illustrate
how the proposed approach works in practical applications.
5.2. Budget management approach
5.2.1. Resource allocation

Optimal societal resource allocation has gained increased attention since the so-called
Brundtland report (Brundtland (1987)) set focus on sustainability. In principle,
resource allocation in a society is realized through allocation of the total available
budget to the various sectors in the society. Maintenance and operation of civil
engineering infrastructure represents one of the societal activities or sectors which
must be allocated a budget to ensure the continued societal benefit from the structures.
Within this sector, the budget is subdivided into smaller parts for sub-projects,
typically to different groups of structures or e.g. segments of the roadway system.
Despite the fact that a sector-wise or project-wise budgeting system can cause
inefficiency of societal resource allocation, it is assumed in this paper to be a given
constraint. The normative discussion whether or not the sector-wise or project-wise
budget allocation is preferable is beyond the scope of the present paper.
Subject to significant uncertainties related to civil engineering projects, decision

makers must decide on the amount of budget necessary and sufficient for successfully
managing the projects. It has been widely accepted in life cycle costing analysis that
-88-
the objective function to be minimized is the (discounted) expected total cost including
follow-up consequences such as user costs. In this regard, it can be said that the
optimization by the minimization of expected total cost implicitly assumes a “perfectly
flexible budgeting”, namely, a situation where the budget is always available when
needed. For the purpose to assess the optimal amount of budget required for a project
or projects, however, this may not be appropriate. A structure which has reduced
availability due to failure or the need of repair works may not be rehabilitated due to
insufficient budgets, which in turn may lead to additional user costs. The optimal
budget allocation may not correspond to the expected total cost. In order to maximize
the net benefit, the budget allocation, the financial costs and the user cost must be
considered simultaneously.
In a broader sense, the objective function should be an aggregated utility, in which all
the preferences of the decision maker are included, see e.g., Faber and Maes (2003). In
practical situations, a decision maker may be precautious in a sense that he/she
requests more budget than the expected cost in order to ensure a successful
management of projects. In the following section, the net benefit is proposed as a
utility function to represent the preferences of the decision maker.
5.2.2. Net benefit maximization

Life cycle costing analysis for civil engineering structures usually considers only the
cost side, assuming that the benefit B is indifferent to the choice of the decision
variables. The reduced benefits due to the loss of functions caused by adverse events
are included in the cost term as user costs. However, the budget must also be taken into
account in the analysis, since a failed structure or a structure with reduced availability
cannot provide the desired benefits until it has been rehabilitated. If the budget is
insufficient, the operation of such structures cannot be recovered until the budget is
available. This can lead to additional user costs. Thus the evaluation of user costs is
dependent on whether the budget for the recovery of the operation is available or not.
The net benefit NB induced by a structure may be written as:
⎧ B − K − ΔB(e) (C (a, e) ≤ K )
NB = ⎨ (5.1)
⎩ B − C (e, a) − ΔB (e) − ΔB(e, ' C > K ') (C (a, e) > K )
where K is the allocated budget, ΔB(e) is a user cost corresponding to the event e ,
C (a, e) is the financial cost corresponding to (a, e) , a is the decision variable and
ΔB(e, ' C > K ') is a user cost induced by the possibly insufficient budget following the
event e , see Figure 5.1. If the financial cost C (a, e) does not exceed the budget K ,
the net benefit is the difference between the benefit and the sum of the budget and the
user cost associated with the event e . Here it is assumed that the unused part of the
budget within a budgeting period is not transferred to the next budgeting period which
is a commonly known difficulty in the public sector. If the financial cost C (a, e)
-89-
Figure 5.1. Decision event tree including budget.
exceeds the budget K (Budgeting failure), an extra budget must be asked for in order
to reinstate the reduced availability, which will be provided at some later point in time,
e.g. the subsequent budgeting period. Until the extra budget is obtained, the
availability remains reduced, causing the additional user cost ΔB (e, ' C > K ') .
As the amount of budget increases, the probability of budgeting failure P (C > K ) and
the net benefit decreases, and vice versa. The optimal budget K * and the optimal
decision variable a * , e.g. concerning inspection and maintenance activities are
obtained by maximizing the expected net benefit E[ NB ] :
E[ NB ] = ∫ NB ( a, e, K ) dP (e; a ) (5.2)
E
where E is the set of possible events e and P (e; a ) is the probability of the
occurrence of the event e given the decision variable a .
5.3. Example
5.3.1. Maintenance planning for a portfolio of RC structures

The maintenance planning for a portfolio of RC structures subject to deterioration due
to corrosion is considered. The portfolio consists of 50 structures, each of which is
composed of 100 elements. For illustrational purposes, they are all assumed to be 10
years old, operable but subject to deterioration. The objective of the maintenance
planning is to find the optimal inspection interval and the optimal budget for each year
so that the net benefit is maximized. The benefits induced by the portfolio are assumed
to be independent of the inspection. Therefore, by letting CT = B − NB , the objective
function to be minimized is written as:
E[CT ] = ∫ CT ( Δt I , e, K ) dP (e; Δt I ) (5.3)

E
-90-
where Δt I is the inspection interval (corresponding to the decision variable a in

Section 5.2.2) and:
⎧ K + ΔB ( e ) (C ( Δ t I , e ) ≤ K )
CT = ⎨ (5.4)
⎩C (e, Δt I ) + ΔB (e) + ΔB (e, ' C > K ') (C ( Δt I , e) > K )
CT can be considered the total cost including all consequences and is thus referred to
as the “total cost” in the subsequent. It should, however, be noted that the total cost
represented by Equation (5.4) differs from the typical definition in commonly applied
life cycle analysis in the sense that it includes the budget and possible additional user
cost due to insufficient budget. The term ΔB (e, ' C > K ') accounts for the effect of
insufficient budget to the reduction of the net benefit. Still, the net benefit defined by
Equation (5.2) is maximized by minimizing the ‘total cost’ defined by Equation (5.3).
5.3.2. Inspection, repair and failure

Two states of visually observable corrosion for an element of a structure are
considered, i.e., the state which will induce a repair eR , and the state which
corresponds to failure eF . The former state requires the need of repair, e.g.,
replacement of concrete cover, while the latter state needs more serious action, e.g.,
replacement of reinforcement. Thus, the set of events E in this example is expressed
as:
E = {(e0 , eR , eF ); for all years and all elements in all structures} (5.5)
where e0 is the state when no action is required.
For the purpose of the illustration but with no effect on generality, it is assumed that
the inspections are made visually and the probability of detection of corrosion is
assumed to be equal to one, i.e., perfect inspections. As long as the budget is sufficient
for performing the necessary repairs of the corroded elements, those are assumed
performed in connection with the performed inspections. It is further assumed that the
repaired elements are brought back to their original states, i.e., described using the
same probabilistic models as those for new elements and the realization of the new
elements are independent of the previous elements. The basic characteristics of the
probabilistic modelling of deterioration are provided in the next section.
The number of structures to be inspected is uniformly distributed over time in

accordance with the inspection planning – for instance, when the inspection period is 4
years, the number of structures to be inspected during 4 years is 13, 13, 12 and 12,
respectively.
-91-
Figure 5.2. Probability of repair (left) and probability of failure (right) for a given
realization of element.
5.3.3. Probabilistic corrosion model

The probabilistic model adopted here corresponds to DuraCrete (2000) and follows
Faber et al. (2005). All the uncertain parameters are assumed to be independent
between different elements. Two limit state functions are explicitly considered: One is
related to the time until the realization of visual corrosion, which corresponds to the
event eR and the other is related to the time until the element fails, which corresponds
to the event eF .
An element is repaired if visual corrosion is observed at the time of inspection as long

as the budget is sufficient. An element fails if and only if the degradation reaches the
failure limit state between two subsequent inspections. Thus the probability of repair,
qR (t ) , and the probability of failure, qF (t ) , at time t for a given realization of an
element both depend on the inspection interval. In this example, the design or nominal
cover thickness is assumed to be equal to 50mm.
Figure 5.2 shows the probability of repair and failure at time t after construction,
repair or recovery due to failure for a given realization of an element in the case of
ΔtI = 5 . Both the probability of repair and failure are calculated by Monte Carlo
simulation. The probability of repair is different from zero only at iΔtI , ( i = 1, 2,3,... ).
This is because the repair is made only if visual corrosion is observed at the inspection.
On the other hand, failure can occur at any point in time. The probability of failure
over time varies significantly as the inspection interval changes. When the inspection
period is small, e.g., Δt I = 1 , the probability of failure is low, since more elements are
already repaired. In contrast, if the inspection period is large, elements may fail more
frequently before repair due to the few inspections. Thus, the inspection period affects
both the probability of repair and the probability of failure. It should be mentioned that
the time axis in the figure does not necessarily represent the structure age after the first
-92-
installation, since a structure may have been repaired or replaced. When a structure
performs poorly (i.e., the realizations of the random variables are unfavorable), it will
be repaired or reconstructed relatively early. After each repair or replacement, the new
structures are identical but stochastically independent of the old ones. A structure with
an initially bad performance will thus eventually be replaced by one with a good
performance. This is why the probabilities of failure and repair decrease after their
peak.
5.3.4. Cost model

The financial maintenance cost C , consists of inspection cost CI , repair cost CR ,
and failure cost CF .
C = CI + CR + CF (5.6)
These costs do not include any user costs associated with the repair actions. The user
costs are considered separately in terms of the reduced benefits ΔB(e) . It is assumed
that the reduced benefits ΔB(e) are additive and proportional to the number of
repaired elements. Due to the uncertainties associated with the physical process of the
deterioration, the maintenance costs can be considered as random variables. As the
inspection interval decreases, the repair cost increases, while the probability of failure
decreases, and vise versa. The additional user cost due to the lack of budget is assumed
to be proportional to the user cost for repair:
ΔB (e, ' C > K ') = g ΔB (e) (5.7)

where g is a multiplying factor. In order to see the significance of the additional user
cost to the optimal inspection interval, g is set to 2, 10 and 100.
The total life cycle period considered for the maintenance planning is 200 years and
budgeting is assumed to be made annually. The discount rate is assumed to be 2% per
year equivalent to the economic growth per capita. The discount rate by time
preference is neglected in this paper for simplicity. This may be justified for short
budgeting periods, i.e., 1 year, by the result in Nishijima et al. (2007). The cost
parameters assumed in this example are summarized in Table 5.1 together with other
parameters.
-93-
Table 5.1. Parameters assumed in the example.
Number of structures 50
Number of elements in each 100
structure
Total life cycle time to be 200 years
considered
Discount rate by economic growth 2% per year
Inspection cost for each structure 1
Repair cost for each element 1
Failure cost for each element 10
User cost for each repair 1
User cost for each failure 10
Multiplying factor g 2,10 and 100
5.3.5. Numerical results

The optimization of the amount of budget and inspection interval is made based on
Monte Carlo simulations in accordance with the probabilistic model for corrosion and
cost model. Figure 5.3 shows the probabilities of the number of elements to be repaired
each year. For the purpose of simplicity it is assumed that repairs are made
immediately after inspections if necessary whether or not the budget is available. The
differences between the case where the budget is available for repair and the case
where the budget is not available are considered through the additional user cost. This
assumption significantly simplifies the analysis, while there is little difference in the
assessment of probabilities of the number of elements to be repaired each year. The
probabilities of the number of failed elements each year are also simulated. The
deviation of the numbers of elements to be repaired and the number of failed elements
is of relevance for the optimization of the budget for each year. If the budget is
insufficient additional user costs may be implied. On the other hand, if the requested
budget is too large, the net benefit decreases.
First, the optimization is made for the optimal budget for each year, for a given
inspection interval. Figure 5.4 (left) shows an example of the optimization of the
budget. The expected total cost E[CT ] , which is defined in Equation (5.4), becomes
large as the multiplying factor, g , becomes large. Accordingly, the optimal budget
which minimizes the expected total cost becomes large as g becomes large. After the
budget for each year is optimized, the expected total costs for all years are summed up
weighted with the corresponding discounting factors. In Figure 5.4 (right), the
discounted expected total costs are shown for each inspection interval. The optimal
inspection interval is obtained as the one which minimizes the discounted expected
total cost. As the multiplying factor increases, the corresponding discounted expected
-94-
total cost increases and the optimal inspection interval decreases. Since the optimal
budget for each year for a given inspection interval is already obtained, the optimal
combination of budget and inspection time is derived, see Figure 5.5. The higher
“penalty” due to the lack of budget, which is represented by the multiplying factor g ,
is reflected in the optimal amount of budget in Figure 5.5. In both cases of g = 10 and
g = 100 , the expected financial costs remain the same, while the optimal budget is
higher in the case of g = 100 than in the case of g = 10 , reflecting the precautionary
attitude toward larger consequences due to the lack of budget. It should be mentioned
that the periodic fluctuations of the expected financial cost and the optimal budget
come from the different number of structures to be inspected as mentioned in Section
5.3.2.
Figure 5.3. Probability of number of elements to be repaired each year (inspection

interval: 5 years).
Figure 5.4. Optimal amount of budget at 20th year in the case of inspection interval: 5
years (left) and optimal inspection intervals (right).
-95-
Figure 5.5. Optimal budget and expected financial cost at each year for g = 10 (left)
and g = 100 (right), (not discounted).
5.4. Discussions
In the example the features and advantages of the proposed approach are shown
considering maintenance planning for RC structures. The approach works especially
well in the case of relatively high probability of occurrence of adverse events and
relatively low consequences. For the case where the occurrence probability of adverse
event is relatively small and the consequence is relatively large, e.g., floods or
earthquakes, the annual budget approach may not work well. In such situations,
establishment of a fund shared by projects, which corresponds to the integration of
projects into one portfolio, may be a good strategy. However, the basic idea in the
proposed approach, namely, optimization of budgeting by maximization of the net
benefit still works in these situations. In the present example, optimal budget
distribution over time has a sharp peak, which is inconvenient in practical budgeting.
However, the budget distribution will be averaged out by considering a portfolio which
is composed of structures of different ages. Thus, the budget distribution over time
shown in the example is due to the fact that only structures whose ages are identical are
considered; not indicating a limitation of the present approach.
In regard to the net benefit induced by a project, it is assumed that the unused portion
of the budget in the case where the cost does not exceed the budget is lost. However,
this can underestimate the net benefit, since the unused portion of the budget can be
spent for relevant activities: in the case of the example, for instance, it could be used
for additional (unplanned) inspections. In applications, this aspect should be properly
taken into account.
Finally, the assumption made in the simulation that repairs are made immediately after
inspections if necessary whether or not the budget is available, may not be suitable if
the repair time is crucial. The repair time is, in general, dependent on when the budget
is available, therefore, the budget for one year does affect the time for repair, which
-96-
must be reflected in the simulation of deterioration. The repair time also affects the
user cost associated with the delay of repair due to a possible insufficient budget. As
the delay increases, the user cost increases.
5.5. Conclusions
Optimal decision making for maintenance of structures is addressed from a societal
perspective as an optimal budget allocation problem. An approach to find the optimal
budget to be allocated and the corresponding optimal inspection and maintenance
strategy is proposed. Thereby the expected net benefit is adopted as the objective
function to be maximized. In addition to the user costs associated with repair activities
the user cost which might result from postponed repair and consequential reduced
availability due to insufficient budget is taken into account.
The proposed approach provides a rational framework for decision makers responsible
for the budgeting and planning of maintenance activities for portfolios of structures
and leads to optimal budgets which are consistent with the adverse consequences of
possible insufficient budgets. For the purpose of illustrating the application of the
proposed approach the problem of maintenance planning for a portfolio of RC
structures subject to chloride-induced deterioration is considered. The example clearly
shows that the optimal budgets differ from the commonly applied expected total costs
and this also has an effect on the optimal choice of inspection plans.
-97-
Societal performance of infrastructure subject to natural hazards (Paper V)
6. Societal performance of infrastructure subject to natural

hazards (Paper V)
Kazuyoshi Nishijima


Australian Journal of Structural Engineering, Vol. 9, No.1, pp. 9-16, 2009.
-98-
Abstract
The present paper proposes a methodology for assessing the effect of different design
and maintenance policies for infrastructure on societal economic growth. The approach
adopted takes basis in the general economic theories and economic models, and
provides an interface between economics and civil engineering with which the
engineering knowledge can be reflected in the economic models. The proposed
methodology can be utilized by societal decision makers to identify the optimal
investments into infrastructure for ensuring sustainable societal development. An
illustrative example is provided considering sustainable decision making in regard to
design and maintenance of infrastructure subject to natural hazards. Thereby the
advantage of the proposed methodology is shown; it enables one to analyze the
economic growth and the associated uncertainties corresponding to different design
and maintenance policies for infrastructure.
Keywords
Sustainability, societal decision making, reliability theory, economic theory.
6.1. Introduction
Sustainable societal development has become an issue of increased and wide spread
societal attention especially during the last two decades. The tremendous economic
developments of former third world nations such as China and India and the general
impact of globalization have put even larger pressures on our limited natural resources
and fragile environment. Faced with an ever increasing amount of evidence that the
activities of our own generation might actually impair the possibilities for future
generations to meet their needs it has become a political concern that societal
development must be sustainable. The issuing of the famous Brundtland report “Our
Common Future” (Brundtland (1987)) forms a milestone on the political arena. This
important event has enhanced the public awareness that substantial changes of
consumption patterns are called for and has further significantly influenced the
research agendas worldwide; it is fair to state that “sustainable development” has
strongly influenced the consciousness and the moral setting in society.
Recent disasters caused by natural hazard events, e.g. the tsunamis in Southeast Asia in
2004 and the flood induced by the hurricanes in the United States of America in 2005,
have proven the importance of infrastructure in society and revealed how societies in
both developing countries and developed countries supported by infrastructure are
vulnerable to natural hazards. Recognizing the lesson learned from these recent
disasters it is necessary to reconsider the framework for identifying the optimal level of
reliability of infrastructure in regard to the performance with due consideration of the
role that the infrastructure plays for societies.
-99-
Infrastructure such as road networks, water and electricity distribution systems assists
economic growth. Aschauer (1989) has reinforced this perception by showing that
investment into infrastructure has a strong explanatory power for societal productivity
taking up the case of the United States of America. A number of studies have
confirmed and generalized this observation; some of these studies, however, claim that
the estimated return rates of investment into infrastructure might be biased, see the
review paper by Gramlich (1994).
In the field of civil engineering, the life cycle cost (LCC) optimization concept has
gained a reputation as being a means for identifying optimal designs as well as
maintenance strategies for infrastructure with due consideration of possible
consequences and proper discounting for future expenses. More recently, the LCC
optimization concept often has been applied in the context of sustainable decision
making for infrastructure projects. However, the application of the LCC optimization
concept in this context may not be appropriate since it tends to focus on the marginal
analysis of the benefits and the costs of projects. For instance, the LCC optimization
concept implicitly assumes that the necessary budget is available whenever it is
needed, which in practice is not necessarily true. Nishijima and Faber (2006) discuss
this issue taking into account the opportunity costs that the lack of budget may incur.
Furthermore, the LCC optimization concept does not aim to identify how to optimally
allocate limited resources into different projects; it primarily addresses the
optimization of each individual project or a portfolio of projects assuming these
projects are in any case undertaken. This is especially problematic in the context of
sustainable decision making, since sustainability fundamentally concerns the issue of
allocation of limited resource in different societal sectors and projects. From this
perspective, the optimization problem in the context of sustainable decision making
should be formulated as: 1) given the amount of investment into the civil engineering
sector, how much of the investment should be directed to new construction and
maintenance works respectively and then 2) at the level of societal decision making
how much of the investment should be allocated to the civil engineering sector.
Whereas the latter optimization is conducted from the perspective of societal decision
makers, the former optimization is a civil engineering issue. However, these two
optimizations have never been discussed jointly due to the lack of the interface
between civil engineering and economics.
Economics plays the central role in analyzing the development of society in the most
aggregated way. It considers not only economic development but also environmental
issues, societal preferences regarding e.g. issues of human safety and inter- and intra-
generational equity etc. The general discussion on the implications of sustainability is
also ongoing in the field of economics, although no agreement is yet established, see
e.g. Perman et al. (2003); the present paper assumes that the agreement on implications
of sustainability should be made in the general economics. Therefore, the present paper
does not aim at defining the objective function and constraints concerning sustainable
-100-
societal development but at providing a methodology with which economic output,

which is one of the relevant indicators concerning sustainable societal development,
can be evaluated as a function of the amount of investments allocated to the civil
engineering sector.
The main problem in employing general economic models in sustainable decision

making in connection with civil engineering is that they do not account for the
performance of infrastructure based on scientific and engineering knowledge; mostly
they are based on aggregated statistical analysis using historical data. Therefore, it is
difficult to study the effects of different design and maintenance policies on societal
economic growth.
With this background the present paper proposes an interface between the general
economic theories and civil engineering. The proposed methodology takes basis in the
methodology proposed by Nishijima and Faber (2007c). However, an extension is
made such that the losses of infrastructure capital due to natural hazards can be
considered in an explicit probabilistic manner. After formulating the methodology this
is applied for the investigation of the effect of different target reliabilities for
infrastructure facilities on the economic growth and the degree of uncertainties
associated with the economic growth.
6.2. Problem setting

Public infrastructure is the primary concern in the present paper, e.g., road networks,
water and electricity distribution systems, for which societal decision makers, to a
large extent, can decide the amount of investment for design and maintenance policies.
The methodology presented in Section 6.4 can be partly applied to private
infrastructure, e.g. machinery and residential houses. The question, however, remains
whether sustainable decision making can be expected from private stakeholders;
societal policy measures, e.g. imposing taxes and giving subsidies, may be required in
order to lead decision makers in private sectors to societal optimal actions.
Two issues are addressed in the present paper. The first concerns the reliability of
infrastructure facilities. Economic models must be able to account for the different
reliabilities of infrastructure facilities resulting from different design and maintenance
policies. In general, the deterioration rate of infrastructure depends on the target
reliability in regard to any type of reduction of performance of the infrastructure and
thus depends on the policy in regard to design and maintenance. Usually, in the field of
economics the deterioration rate is estimated directly or indirectly based on historical
data, see e.g. Aschauer (1989), Gramlich (1994) and Greenwood et al. (2000). Using
historical data as a basis, however, there is no possibility to reflect the effect of new
policies on the future deterioration of infrastructure facilities. The proposed idea in the
present paper is that the reliabilities of infrastructure facilities in the engineering sense
-101-
can be related to the deterioration rates in an economic sense. Secondly, two different
types of investments into infrastructure should be differentiated; 1) the investment into
new construction of infrastructure facilities, which will increase the economic output
through the increased infrastructure capital stock, and 2) the investment for achieving
higher reliability of infrastructure facilities, which does not directly increase the
economic productivity but improves the durability of the structures and prolongs their
lifetime. The distinction between these two different types of investments is realized by
assessing the infrastructure capital stock by physical units, as opposed to monetary
units.
The necessity of increased investments into infrastructure in terms of maintenance

works has been appreciated for both developing and developed countries for different
reasons. In developed countries the investments into maintenance are considered as an
urgent issue in the light of the severe deterioration of aged infrastructure. For
developing counties on the other hand the necessary investments into maintenance
works have been considered as a potential opportunity for increasing the investment
efficiency of expenditures into the built environment; the investment for deteriorating
infrastructure into maintenance works may be more efficient than the investment into
construction of new infrastructure, see e.g. World Bank (1994), Rioja (2003) and
Kalaitzidakis and Kalyvitis (2004). The present methodology is formulated allowing
for considering both types of investments.
6.3. Role of infrastructure in economic context

The role of infrastructure in an economic context is illustrated in Figure 6.1. The
performance of infrastructure must reflect societal needs in regard to productivity and
societal preferences, for instance, concerning life safety and damages to the qualities of
the environment. Societal preferences in regard to life safety have been discussed in
the context of economic output and consumption through the recently developed
concept of the Life Quality Index, see e.g. Nathwani et al. (1997) and Rackwitz (2002).
These considerations fall into the category of how to define a utility function and/or
constraints in the context of decision making. The present paper, in contrast, focuses
on the relation between the productivity of societal infrastructure versus investments
into new and existing infrastructure.
-102-
Figure 6.1. Focused role of infrastructure in an economic context, after Nishijima and
Faber (2007c).
The technology currently available determines the level of economic output given the
amounts of different types of capitals, e.g. human capital, physical capital etc. This
relation is in the field of macroeconomics often represented by a production function
as:
Y = f ( K (1) , K (2) ,...) (6.1)
where Y is the economic output in a given period, K (i ) ( i = 1, 2,... ) represents the

amounts of different types of capitals. The level of differentiation of capitals depends
on the level of analysis and data available. For instance, K ( n ) may represent the
amount of the aggregated capital of infrastructure including different types of
infrastructure or may represent one specific type of infrastructure. It is also possible
that K ( m ) represents the amount of a capital differentiated according to its relevance
that, however, belongs to the same type of infrastructure, e.g. road networks that
connect large cities versus road networks in remote areas. In what follows, however,
only one type of the capitals is considered and it is abbreviated as K (1) = K for the
purpose to make clear the concept of the proposed methodology, and it is not the
limitation of the proposed methodology.
The production function can be estimated using historical data by time-series analysis
and/or cross-sectional studies. Thereby the capitals are measured in physical terms e.g.
kilowatts of electricity generating capacity or length of road or in monetary terms (by
multiplying the amount measured in physical units with the corresponding prices).
However, for the present purpose it is important to measure the infrastructure capital in
-103-
physical terms since otherwise the investment for achieving higher reliability and the
investment for increasing the amount of infrastructure cannot be distinguished. Several
datasets and estimated production functions in regard to several types of infrastructures
are available, e.g. Canning (1998) and Canning and Bennathan (2000).
The equation of capital accumulation is often written in the following form:
ΔKt = Kt new − δ Kt (6.2)
where the subscript t in Kt represents that the amount of capital K is evaluated at

time t , ΔKt is the net increment of the amount of the capital between time t and
t + Δt , Δt is the increment of time, Kt new is the amount of infrastructure capital
constructed between time t and t + Δt , and δ is the deterioration rate. Note that the
amount of the capital is measured in physical units and Δt is often chosen as Δt = 1
year. As mentioned previously, the deterioration rate δ is usually estimated using
historical data and it is often represented as a deterministic value. The exceptions for
this are Bulow and Summers (1984) and Zeira (1987), who consider the uncertainty in
depreciation of capital 9 . From a civil engineering point of view the amount of
deteriorated capital between time t and t + Δt represented by δ Kt in Equation
(6.2) indeed is a function of design and maintenance policies. In general, the amount of
deteriorated capital should be considered as a random variable unless it may be
assumed to converge to its expected value; this may not be the case for infrastructure
facilities subject to natural hazards when the geographical sizes of the hazard events
are relatively large compared to the sizes of societies. In the following section a
methodology for solving these issues is proposed.
6.4. Proposed methodology
6.4.1. Definition of infrastructure failure

Let Rt denote a set of states that represent the performance of an infrastructure
facility at time t . Rt may consist of not only physical states of infrastructure
facilities but also societal states of relevance that are related to the use of the
infrastructure facilities. The infrastructure facility is considered to have failed if the
performance of the infrastructure facility does not satisfy the societal requirements.
The failure of an infrastructure facility may occur e.g. due to natural hazards, physical
deterioration and societal obsolescence. The societal requirements to the infrastructure
facility are assumed expressed through the failure domain ΩF ,t . The failure domain
ΩF ,t can be a composite set of single-failure events, each of which relates to different
9 They consider the uncertainty of capital depreciation in terms of uncertain changes of the monetary
value of capitals. Thus, the depreciation therein does not concern the change of the amount of
physical capital due to e.g. natural hazards.
-104-
societal requirements. Examples hereof include collapse of a structure, severe

deterioration where repair actions are not feasible as well as situations where the safety
of a structure does not fulfill given acceptance criteria and must be demolished and/or
replaced. Then failure may be defined as:
Rt ∈ Ω F ,t (6.3)
The conditional probability pF ,t of failure of infrastructure in time period (t , t + Δt ]

is defined as:
pF ,t = P ⎡⎣ Rt ∈ Ω F ,t ; (t , t + Δt ] | Rt ∉ Ω F ,t ;[0, t ]⎤⎦ (6.4)
In cases where the failure domain consists of m independent failure event sets
Ω F ,t (i ) ( i = 1, 2,..., m ) and Ω F ,t = ∪ i Ω F ,t ( i ) , the conditional probability of failure can
be written as:
⎡ ⎤
pF ,t = P ⎢ Rt ∈ ∪ Ω F ,t (i ) ;(t , t + Δt ] | Rt ∉ Ω F ,t ;[0, t ]⎥
⎣ i ⎦ (6.5)
(
= 1 − ∏ 1 − P ⎡⎣ Rt ∈ Ω F ,t ;(t , t + Δt ] | Rt ∉ Ω F ,t ;[0, t ]⎤⎦
i
(i )
)
In this way, the conditional probability of failure defined in terms of Equation (6.4) can
be regarded as a generalized measure of capital deterioration. The advantage of the
definition of infrastructure failure in this manner is that it enables the use of the
reliability theory for the calculation of the probabilities corresponding to the structural
design and maintenance policies for infrastructure facilities whenever probabilistic
models are available. Otherwise the probabilities estimated by expert judgments can be
partly integrated into probabilistic terms in Equation (6.5), hence, it is possible to
combine objective and subjective evaluations in order to quantify the conditional
probability of failure.
6.4.2. Equation of capital accumulation

The increment ΔKt of the infrastructure capital from time t to t + Δt can be
generally written taking basis in Equation (6.2) as:
ΔKt = Ktnew (at ) − X t (6.6)
where Ktnew (⋅) is the amount of new infrastructure constructed at time t and X t is
the amount of failed infrastructure. In general, X t should be considered as a random
variable. Note that applying expectation operation Equation (6.6) is reduced to
-105-
Equation (6.2). Ktnew (at ) is a function of the design policy at at time t and is
written as:
It
Ktnew (at ) = (6.7)
qt (at )
where It is the budget allocated to construction of new infrastructure at time t and

qt (at ) is the unit cost of the construction corresponding to the design policy at . The
probability distribution of the amount of failed infrastructure X t is characterized by
the amount of the capital Kt at time t , the sequence of design policies
{ai }i =1 = {a1 , a2 ,..., at } and the sequence of maintenance policies {bi }i =1 = {b1 , b2 ,..., bt }
t t
for the infrastructure until time t . In cases where large-scale hazards, e.g. earthquakes
and hurricanes, are of concern the geographical distribution of the infrastructure is also
a relevant factor. Finally, since the budget allocated to infrastructure is divided into the
investments into new construction and maintenance works the following equation must
hold:
Gt = I t + M t ({ai }i =1 , {bi }i =1 )
t t
(6.8)
where Gt is the allocated budget for the civil engineering sector at time t and M t
is the budget necessary for maintenance works. M t is a function of {ai }i =1 and
t
policies {ai* }
t
{bi }i =1 . With these settings it ist possible to identify the optimal design
t
and maintenance policies {bi*} given the budget sequence {Gi }i =1 = {G1 , G2 ,..., Gt } .
t i =1
i =1
The methodology proposed above requires as an input parameter the amount of

investment into infrastructure, considers the design policy and the maintenance policy
to be decision variables to be controlled and provides as outputs the sequence of the
amount of capital Kt and the corresponding economic growth Yt .
6.5. Illustrative example

The economic model in the following example assumes that infrastructure is the only
capital that affects the economic production in society. The production function is
assumed to be written as:
Yt = AKtα (6.9)
which is a special form of Equation (6.1). Therein A is the factor that represents the
technology in the society, which is assumed constant. The exponent α represents the
marginal increase of the economic output with respect to the infrastructure capital. It is
assumed that the infrastructure capital is exposed to natural hazards and that the
infrastructure capital can be geographically divided into n segments within which the
-106-
failures of infrastructure facilities are perfectly correlated and between which the
failures are independent. Namely, the parameter n represents the relative
geographically affected size of the natural hazards compared with the size of the
society. Furthermore, the occurrence of natural hazards is assumed temporary
independent. Under these assumptions the amount of capital which is lost at time t
can be expressed as:
Nt
Xt = Kt (6.10)
n
where Nt represents the number of failed segments among n independent segments
with the probability of failure p f and follows the binomial distribution with n n
trials and the probability of failure being equal to p f . p f is the probability of failure
within the duration Δt = 1 year. Note that as n becomes large X t converges to its
expected value, E[ Nt ] / n ⋅ Kt = p f Kt thus the equation of capital accumulation is
reduced to the form of Equation (6.2). By substituting Equation (6.10) into Equation
(6.6), the equation of capital accumulation is written as:
It N
ΔK t = − t Kt (6.11)
qt (at ) n
The values of the parameters assumed in this example are shown in Table 6.1. These
values are postulated for illustrative purposes, however, in practice these values can
and should be determined by economic as well as engineering analyses.
Table 6.1. Assumed parameters in the example.
Investment ratio into infrastructure λ = 0.05

Exponent in production function α = 0.2
Factor in production function Α = 10
Independent segments of infrastructure n = 5,50
Policy1 Policy2
Probability of failure per year p f = 0.01 p f = 0.001
Construction cost per unit qt = 1 qt = 2
Maintenance cost 0.01Kt 0.01Kt
The probability of failure p f is a function of the policy in regard to design and

maintenance. Here, two policies are considered, each of which targets the probability
of failure shown in Table 6.1. The corresponding construction costs and maintenance
costs are also shown in the table. In practical situations, the probability of failure and
the associated costs can be identified using the definition of infrastructure failure
represented by Equation (6.3) employing the civil engineering knowledge and the
structural reliability theory for the calculation of Equation (6.4).
-107-
The analyzed economic output paths as the function of the policies and different
numbers of independent segments are shown in Figure 6.2. The figure shows the
median, 5% and 95% of the economic output as a function of time. Figure 6.2 (left)
shows the economic output paths when the number of independent segment is
relatively small ( n = 5 ). The economic growth is faster when policy 1 (lower reliability
associated with lower construction cost) is adopted. However, in a long run the
economy grows more when policy 2 (higher reliability associated with higher
construction cost) is adopted. It should be mentioned that the economic growth path
under policy 1 is associated with larger uncertainty, i.e. results in a less stable
economic growth, compared to the economic growth path under policy 2. The
economic growth paths are more stable in a sense that the uncertainty on the economic
output is smaller when the number of independent segments is larger ( n = 50 ), see
Figure 6.2 (right). The results shown in Figure 6.2 and the interpretation of the results
stated above are coherent with engineering understanding. Furthermore, it is possible
with the proposed methodology to evaluate in a quantitative manner the effects of
different policies on the economic growth within a general economic model
framework.
Figure 6.2. Economic output paths for different policies and different numbers of
independent segments n = 5 (left) and n = 50 (right).
6.6. Discussion
In practical applications, the amount of infrastructure capital losses due to natural
hazards can be readily assessed by risk analysis together with Geographical
Information Systems (GIS), see Bayraktarli and Faber (2007). The design costs and
maintenance costs for infrastructure facilities corresponding to different policies can be
optimally identified using the framework proposed by Nishijima et al. (2008). This
framework models the infrastructure using hierarchical Bayesian networks and
formulates the problem as a constrained optimization problem where the expected
costs are considered as the objective function and the requirements to the performance
of infrastructure, e.g. target reliabilities and acceptance criteria for fatalities, are
-108-
accounted for by constraints. These techniques can be incorporated into the proposed
methodology. The assumption that failures of structures are independent generally does
not hold even if the hazard events that affect each structure are independent. This is
because of the presence of modeling uncertainties, e.g. on the resistance of structures,
that may commonly affect all considered structures, see Faber et al. (2007a). Thus the
proposed methodology and the analysis in the example of this paper should be
considered as being conditional on the modeling uncertainties. In general the
integration with respect to the modeling uncertainties is necessary in the analyses of
losses of infrastructure facilities and societal economic growth.
6.7. Conclusion
The present paper proposes a methodology for assessing the effect of different design
and maintenance policies for infrastructure on societal economic growth. The proposed
methodology can serve as a component of a general decision making framework for
optimal resource allocation in the context of sustainable societal development. The
proposed methodology requires the amount of investments into infrastructure as an
input parameter. It incorporates the design policy and the maintenance policy as
decision variables. It provides the sequence of the amount of capital together with the
corresponding economic growth as outputs. In an example the advantage of the
proposed methodology is illustrated; it enables one to analyze in a quantitative manner
the economic growth and the economic stability corresponding to different design and
maintenance policies for infrastructure.
-109-
Optimal design and maintenance policy for infrastructure from a macroeconomic
perspective
7. Optimal design and maintenance policy on infrastructure

from a macroeconomic perspective
7.1. Introduction
In Chapters 3, 4 and 5, the optimization problem of the reliability of individual
structures or groups of structures is addressed. In these chapters, the reliability or
decision variables related to structural performance are optimized based on the
life-cycle cost optimization concept. Strictly speaking, the life-cycle cost optimization
concept can be applied only if the benefit and cost of the project concerned are
assumed marginal in the economy; that is, the economic growth is not affected by
whether or not or how the project is undertaken. Thus, the life-cycle cost optimization
concept may not be appropriate as the philosophical principle for decision making in
cases where the consequences of the decisions are considered as non-marginal.
In practice, there are many situations where the consequences of the decisions are
considered as non-marginal. Such decision situations include, for example, code
making in which the acceptable reliability of structures is controlled, and design and
maintenance strategies on nationwide infrastructure projects. These decisions affect the
capital accumulation of infrastructure and thus, in turn, the long-term development of
the economy. Therefore, in these decision situations a non-marginal economic
framework has to be adopted.
As a first step to develop a general decision framework for facilitating these decision
situations, this chapter examines how the optimal reliability of infrastructure may be
identified within the economic growth theoretical framework. For this, a simplistic
economic model is developed, employing the approach proposed in the previous
chapter for incorporating the reliability of infrastructure in economic models. Using
the developed economic model, it is investigated how the reliability of infrastructure
affects the economic growth and the optimal reliability at each point in time depends
on the economic level. The aim of this chapter is to show the potential that such a
general framework can provide the optimization principle for non-marginal decision
analysis.
The structure of this chapter is as follows. First, the principle of the life-cycle cost
optimization concept is reviewed. Then, the assumptions and limitations of the concept
are pointed out. Second, previous research work on the role of civil infrastructure
within the economic growth theoretic framework are introduced briefly, followed by
some critical reviews on the assumptions made in these works. Third, a simplistic
-110-
perspective
economic model is presented. Finally, the optimal reliability of infrastructure is

examined within the model, and the results are discussed.
7.2. Principle of life-cycle cost optimization concept

The life-cycle cost optimization concept is considered as an extension of cost-benefit
analysis. Thus, before deriving the life-cycle cost optimization concept, the derivation
of the principle of cost-benefit analysis is introduced. The derivation introduced here is
based on Stern (2006)10.
7.2.1. Derivation11
A project is socially profitable if the social welfare is increased through the project.
This is expressed as:
ΔW = W 1 − W 0 > 0 (7.1)
where W is the social welfare function, and W 0 and W 1 are the social welfares
when the project is not undertaken and the project is undertaken respectively. In
general, the social welfare function is a function of many variables that concern the
utilities of all members in the society. However, here it is assumed that the social
welfare function consists of the utility function of a representative household and a
discount factor, and the utility is a function only of the consumption of the household.
Under these assumptions, the social welfare function can be written as:
∞
W = ∫ u (ct )e − ρ t dt (7.2)
0
where u(ct ) is the utility function of the representative household, ct is the

consumption at time t , and ρ is the discount rate for pure-time preference.
Assuming that the change of the consumption in Equation (7.2) is small, and
substituting Equation (7.2) into Equation (7.1), ΔW can be written as:
∞ ∂u (ct ) ∞
ΔW = ∫ Δct e − ρ t dt = ∫ λt Δct dt (7.3)
0 ∂ct 0
where Δct is the perturbation of the consumption from a baseline consumption path,
and λt is the discount factor and is written as:
10 Other derivations can be found in e.g. Ramsey (1928) and Koopmans (1965) in the context of the
economic growth theory.
11 For simplicity, here it is assumed that a representative individual lives for an infinite time. However,
the derivation can be extended for the case where many generations live for finite lifetimes, which is
the situation assumed in the generation-adjusted discounting concept introduced in Chapter 4.
Furthermore, it is assumed that the population is constant over time.
-111-
perspective
∂u (ct ) − ρ t
λt = e (7.4)
∂ct
Here, the increase/decrease of consumption at each point in time corresponds to the

benefit/cost from the project. The rate of the temporal change of the discount factor
λt / λt is obtained as12:
λt {u ''(ct )ct − ρ u '(ct )} e − ρt

=
λt u '(ct )e − ρt
ct u ''(ct ) ct
= −ρ (7.5)
u '(ct ) ct
= −(ηδ t + ρ )
where u '(ct ) = ∂u / ∂ct , u ''(ct ) = ∂ 2u / ∂ct 2 , η = ct u ''(ct ) / u '(ct ) and δ t = ct / ct . η is
the elasticity of the marginal utility of consumption. δt is the growth rate of
consumption, which is assumed to be exogenously given.
If the growth rate of consumption is assumed constant and given as δ t = δ , then the
discount factor is obtained as13:
λt = e− (ηδ + ρ )t (7.6)
By substituting Equation (7.6) into Equation (7.3), the criterion for the project
appraisal is finally obtained as:
∞
ΔW = ∫ Δct e − (ηδ + ρ )t dt > 0 (7.7)
0
In cases where several decision alternatives are available for the project, the above
criterion should be applied for the decision alternative that maximizes ΔW .
The life-cycle cost optimization concept typically employed in civil engineering

decision analysis is derived by further assuming that the perturbation of the
consumption Δct (a) is a function of the decision variable a regarding structural
performance and this is equal to the benefit Bt less the cost Ct (a) , which is also a
function of the decision variable a as:
Δct (a) = Bt − Ct (a) (7.8)
12 The dot " ⋅ " on the top of symbols represents the derivative with respect to time.
13 The choice of the constant λ0 is arbitrary, thus here it is chosen as λ0 = 1 .
-112-
perspective
Note that the benefit Bt may vary as a function of time but is often assumed to be
independent of the decision variable a . By substituting Equation (7.8) into Equation
(7.7), neglecting the constant benefit term, and taking the negative sign, the objective
function, i.e. the life-cycle cost CT (a) is obtained as a function of the decision
variable a as:
∞
CT (a ) = ∫ Ct (a )e − (ηδ + ρ ) t dt (7.9)14
0
Whenever uncertainty is involved in the cost term, the expectation should be taken as:
∞
C T (a ) = ∫ E[Ct (a )]e − (ηδ + ρ )t dt (7.10)
0
where C T (a) is the expected life-cycle cost as a function of the decision variable a ,
and this should be employed as the objective function in the optimization.
7.2.2. Assumption and limitation

The fundamental assumption in the derivation of the life-cycle cost optimization
concept shown in the above is that the growth rate δt of the consumption is
exogenously given and is not affected by the benefits and costs from the project;
namely, the benefit from the project at each point in time is consumed at the time (i.e.
not invested for capital accumulation), and costs incurred by the project at each point
in time are compensated by the decrease of consumption at the time (thus the amount
of investment remains unchanged). Note that the stock losses of the infrastructure
capital due to failure (direct consequence of failure) and the economic flow losses
associated with the capital losses (indirect consequence of failure) in case of failures
should be interpreted as reduced benefits, which are also assumed not to affect the
growth rate of consumption. The application of the life-cycle cost optimization concept
should be limited to the extent that the assumption can be considered as reasonable.
As is clear from the above derivation, the growth rate δt of consumption does not
need to be constant, although in practice it is often assumed constant. It may be also
worth mentioning that whenever uncertainty is involved in the discount rates δ and
ρ , the expectation should be taken as15:
∞
C T (a ) = ∫ E[Ct (a )]E[e − (ηδ + ρ ) t ]dt (7.11)
0
14 If u (ct ) = ln ct , then η = 1 , and it coincides with the formulation of the objective function in
Chapter 4.
15 Here, the cost term is assumed to be independent of the term of the discount factor. However, if they
are not considered as being independent, the expectation operator should be only applied to the
product of the two terms; this may be the case when some of the costs included in the cost term,
which are measured in real terms, may change in accordance with the economic growth.
-113-
perspective
Note that the expectation operator is applied to the discount factor, not to the discount
rates, see e.g. Newell and Pizer (2004) for more discussion.
7.3. Available economic models for infrastructure

In order to describe the role of infrastructure within the economic growth theoretical
framework, two types of component economic models are required; one for describing
the contribution of the infrastructure capital to the economic productivity, and the other
for describing the accumulation of the infrastructure capital. The former is represented
in terms of a production function, and the latter is represented in terms of so-called
"equation of motion," which describes how the capital is accumulated as a function of
the investment into new construction of infrastructure and the deterioration rate of the
infrastructure capital.
Concerning the production function that incorporates the infrastructure capital, there
are a number of research works available both theoretically (e.g. Glomm and
Ravikumar (1994) and Duggal et al. (1999)) and empirically (e.g. Aschauer (1989),
Easterly and Rebelo (1993) and Canning and Bennathan (2000) ). There are also some
research works on the estimation of the deterioration rate of infrastructure capital, see
e.g. Gramlich (1994) and Greenwood et al. (2000). However, only a few research
works are available that explicitly treat the deterioration rate of infrastructure capital as
a variable which can be controlled in terms of maintenance policy on infrastructure,
e.g. Rioja (2003) and Kalaitzidakis and Kalyvitis (2004).
For example, Rioja (2003) considers the amount of investment in maintenance work
for the infrastructure (relative to economic output) as a control variable, and the
optimal investment ratio in maintenance work is derived. Kalaitzidakis and Kalyvitis
(2004) extend Rioja's economic model by endogenizing the decision of budget
allocation into both investment in the construction of new infrastructure and
investment in maintenance work for existing infrastructure.
These pioneering works are remarkable in the sense that the deterioration rate is
considered as a variable and can be optimized through the investment ratio into
maintenance work. However, the relations between the deterioration rate and the
investment ratio assumed in the models are not realistic. One of the drawbacks of these
assumptions is that the deterioration rate at any time is dependent only on the current
investment ratio in maintenance work; the current deterioration rate is not a function of
past maintenance policies, and the current maintenance work does not affect the future
deterioration rate. Furthermore, the effect of differing design policies on the
deterioration of infrastructure is not considered.
However, in civil engineering it is commonly agreed that a slight increase of initial

cost for the purpose of increasing the durability of infrastructure would significantly
-114-
perspective
reduce the future costs for maintenance work. Similarly, undertaking maintenance
work at an earlier stage of deterioration would reduce additional maintenance costs in
the future. Thus, the investment in construction and maintenance works for reducing
the deterioration rate can be considered at least partly as an investment into the future.
However, the economic models proposed by those pioneering works may fail to
capture this nature of the investment.
In order to overcome these drawbacks of the economic models proposed previously, in

the following section, a simplistic economic model that enables one to capture this type
of investment is developed, and the economic model is examined.
7.4. Analysis with simplistic economic model
7.4.1. Economic model

The aggregated output Y (t ) is assumed to be produced by means of capital K (t )
and labor L(t ) at time t . This relation is assumed to be represented by the
neoclassical production function16:
Y (t ) = F ( K (t ), L(t )) (7.12)
Herein, it is furthermore assumed that the capital K (t ) consists only of infrastructure
capital. Assuming that the production function exhibits constant return to scale, the
production function can be reformulated in terms of variables per capita17 as:
Y (t ) F ( K (t ), L(t ))
y (t ) = = = F ( K (t ) / L (t ),1) = f (k (t )) (7.13)
L (t ) L (t )
where y (t ) and k (t ) denote the output and capital per capita at time t , and f (⋅)
represents the production function in terms of the variables per capita. In addition to
these assumptions, it is assumed that the saving rate of the household is exogenously
given as e ( 0 < e < 1 ) and the amount of labor is constant over time18.
The important difference in the economic model assumed here from the models
employed in Rioja (2003) and Kalaitzidakis and Kalyvitis (2004) appears in the
equation of motion for the capital accumulation, especially on the way of modelling
infrastructure deterioration.
Consider the infrastructure constructed at time s . The expected service life time T s
of the infrastructure and associated costs q s for construction and maintenance work
16 Namely, ∂F / ∂K > 0, ∂F / ∂L > 0, ∂ 2 F / ∂K 2 < 0, ∂ 2 F / ∂L2 < 0 .

17 Here, it is assumed that the population is equal to the amount of labor capital.
18 Thus, the analysis can be made only in terms of variables per capita. In what follows, the small
symbols for the corresponding variables represent the variables per capita.
-115-
perspective
are assumed to be a function of the design and maintenance policy a s at time s , i.e.
T s = T (a s ) and q s = q(a s ) . Herein, the associated costs refer to all the costs that are
required in order to realize the target expected service life T (a s ) . Whereas the service
life time of infrastructure is in general a random variable, it is assumed here for
simplicity that the service life time is deterministically represented by its expected
value, T s . Furthermore, it is assumed that the infrastructure provides full functionality
until it exceeds the expected service life time and does not provide any functionality
when it exceeds the expected service life time. Note that for the assessment of the
expected service life time the approach presented in Section 6.4.1 is useful. In this
setting the failure of the infrastructure should be interpreted in a broader sense; the
relevant failure modes include not only physical collapse but also unavailability of
required functionality for any reasons, e.g. severe deterioration, failure to satisfy given
acceptable safety level, and even societal obsolescence19. The expected service life T s
thus defined can be interpreted to represent the reliability of infrastructure; the longer
the expected service life of infrastructure, the higher the reliability of the
infrastructure.
To consider these properties in the economic model, the following function is

introduced;
⎧0 (t < s, s + T s < t )
g (t ; s, T s ) = ⎨ s
(7.14)
⎩1 ( s ≤ t ≤ s + T )
Then the contribution κ s (t ) of the infrastructure constructed at time s to the capital

accumulation is written as:
κ s (t ) = k s ⋅ g (t; s, T s ) (7.15)
where k s represents the per-capita amount of the infrastructure constructed at time

s , see Figure 7.1. The effect of the design and maintenance policy a s at time s on
the deterioration of the infrastructure in the future can be represented through the
function g (t; s, T s ) in terms of the expected service life time T s of infrastructure.
Assume that all the current and future costs for construction and maintenance work for
the infrastructure constructed at time s are invested at time s , and denote the overall
cost per unit capital by q s 20. Since the amount of investment is assumed given
exogenously as i ( s ) = ey ( s ) , the increment k s of the capital due to the investment at
time s is given as:
19 See Section 6.4.1 for the definition of the generalized capital deterioration.
20 If the variables in the economic model are measured in terms of a physical unit, this cost per unit
capital should be interpreted as the multiplying factor for adjusting the difference of the required
amount of resources for different design and maintenance policies.
-116-
perspective
Figure 7.1. Increment of capital due to investment into infrastructure at time s .
i ( s ) ey ( s )
ks = = s (7.16)
qs q
Finally, the amount k (t ) of capital at any given time t is represented as:
t t
k (t ) = ∫ κ s (t ) ds + k0 (t ) = ∫ k s g (t ; s, T s )ds + k0 (t ) (7.17)
0 0
where k0 (t ) represents the amount of the initial capital remaining at time t .
The objective function of the dynamic optimization problem here is the social welfare
function, which is defined as:
∞
W = ∫ U (c(t ))e − ρ t dt (7.18)
0
where U (c (t )) is the utility function of the representative household in the economy

and ρ is the discount rate for pure-time preference. Note that the consumption
c(t ) = (1 − e) y (t ) in the utility function is implicitly a function of the set of the decision
variable {as }s = 0 until time t . The dynamic optimization problem for the design and
t
maintenance policy on infrastructure is thus formulated as:

∞
max W = ∫ U (c(t ))e − ρ t dt (7.19)
{a }
∞
s 0
s =0
subject to:
c(t ) = (1 − e) f (k (t )) (7.20)
ef ( k ( s ))
ks = (7.16)'
q (a s )
-117-
perspective
t
k (t ) = ∫ k s f (t ; s, T (a s ))ds + k0 (t ) (7.17)'
0
with the initial conditions: the initial amount k0 (0) = k0 of the infrastructure capital,
and the expected service life time of the infrastructure initially available. The set of
decision variables that should be optimized by societal decision-makers is the set of
design and maintenance policies {as }s = 0 .
∞
7.4.2. Steady state analysis

First, the steady state is analyzed where the amount of capital is constant, i.e. the state
of no economic growth, which is characterized by k = 0, (k > 0) . For this state, the
increment of the capital due to the investment in the infrastructure capital should
exactly compensate the decrease of capital due to deterioration. This is represented as,
see also Figure 7.2:
ef (k * ) k *
= * (7.21)
q* T
where the superscript " * " on the symbols signifies that the quantities are the quantities
at the steady state. The left hand side comes from Equation (7.16), and the right hand
side is obtained from the assumptions made on the deterioration of the infrastructure.
Figure 7.2. Steady state where the increment of the capital exactly compensates the
depreciation.
-118-
perspective
Reformulating Equation (7.21),
q* *
ef ( k * ) = k (7.22)
T*
From the assumed properties of the production function ( df / dk > 0, d 2 f / dk 2 < 0 ), k *
is maximized when q* / T * is minimized. Since the highest production level leads to
the highest consumption level for a given saving rate, the optimal policy at the steady
state is the policy a * that minimizes q* / T * .
Note, however, that this steady state does not necessarily correspond to the optimal
state in the sense that the consumption is maximized. This is because of the assumption
that the saving rate e is exogenously given. Since the saving rate e corresponds
one-to-one with the amount k * of the capital at the steady state through the relation
given by Equation (7.22), the optimal saving rate that maximizes consumption at the
steady state is characterized by k * as:
* * q* *
*
max c = (1 − e) f (k ) = f ( k ) − * k (7.23)
k* T
Thus, the optimal amount kopt of the capital that maximizes the consumption at the
steady state is obtained as the amount that satisfies the following equation:
q*
f '( kopt ) = (7.24)
T*
where f ' represents the derivative with respect to k . This corresponds to the golden
rule of accumulation for the Solow-Swan model, see Phelps (1961).
The optimization principle obtained from Equation (7.22) for the design and
maintenance policy on infrastructure shows that the sum of the initial cost and
maintenance cost of infrastructure divided by the service life time, i.e. average cost per
unit time, should be minimized. This is intuitively appealing. In order to investigate
this principle further, an illustrative relationship between the expected service life time
T and the average cost q(T ) / T 21 per unit time is shown in Figure 7.3. For a smaller
expected service life time, the average cost per unit time is higher. This is because the
overall cost divided by a shorter expected life time is disproportionately large. On the
other hand, infrastructure with a very long expected service life may be very costly due
to technical reasons and/or may not even be feasible for other reasons, e.g. societal
obsolescence. This is why in Figure 7.3 the average cost per unit time increases sharply
21
Both the cost q(a) and the expected service life T ( a ) are functions of the decision variable a .
However, since the expected service life corresponds to the decision variable one-to-one, the cost is
considered as a function of the expected service life time, i.e. there exists such a function as
q(T ) = q (a) . The variables without the superscript s represent the variables at any arbitrary time.
-119-
perspective
for a very long expected life time. Between these two extremes, the average cost per
unit time moderately decreases as the expected service life time increases.
One of the most relevant differences between the optimization principle obtained here
and the life-cycle cost optimization commonly utilized in engineering decision-making
is that the principle obtained here does not involve failure cost terms, which in the
life-cycle cost optimization play an important role. The explanation for this is: first, in
the economic model considered here (also in most economic models), the loss of
infrastructure due to failure is considered in the deterioration terms in the equation of
the capital accumulation (see Equation (6.2) or Equation (7.17), although in Equation
(7.17) the deterioration term is implicit); second, the reduction of the economic output
associated with the loss of capital is considered through the production function by
substituting a smaller amount of capital due to the loss of capital. Namely, possible
consequences due to the loss of infrastructure are already taken into account. However,
note that although the objective function in the optimization principle obtained above
and the objective function in the life-cycle cost optimization principle are not the
same22, this is not contradictory. In fact, the contexts in which these two principles are
assumed to be applied are different; the life-cycle cost optimization principle is
suitable for marginal decision analysis, and the principle obtained above is suitable for
non-marginal decision analysis.
Figure 7.3. Relationship between expected service life time T (a s ) and average cost
per unit time q(a s ) / T (a s ) at any given time s .
22 Since the objective function derived in this section is the objective function at the steady state, the
objective function in the life-cycle cost optimization should assume zero discount rate for economic
growth, δ = c / c = 0 , in order for the comparison to be meaningful.
-120-
perspective
7.4.3. Transition state analysis

In the previous section, the optimal design and maintenance policy on infrastructure is
considered at the steady state. However, the application of this optimal policy in a
transition state (i.e. economy is under development, y > 0 ) may not be optimal in the
sense that the overall social welfare for all relevant generations defined in Equation
(7.19) may not be maximized. In order to answer this question, the dynamic
optimization problem defined by Equations (7.16) to (7.20) in Section 7.4.1 is
considered.
The dynamic optimization problem is solved numerically here, since it appears

difficult to apply commonly available algorithms for the analytical solution of dynamic
optimization problems, e.g. the variation methods or the maximum principle (see
Chiang (1999)). For this reason, the parameters required for solving the problem are
postulated as shown in Table 7.1. The functional forms of the utility function,
production function, cost function, and the function that represents the deterioration of
the initial capital are also shown. Note that the values of the parameters are assumed
only for performing the numerical calculation, thus the values themselves are not
relevant.
In the dynamic optimization problem, the equations are discretized on a multi-annual

basis. It is assumed that the design and maintenance policy can be changed every 10
years for the first 100 years, and the same policy is taken after 100 years. The reasons
for this assumption are 1) that a more frequent change of the policy, e.g. every year,
may be feasible but not realistic in practice, 2) that a more frequent change of the
policy increases the number of the variables to be optimized in the optimization
problem, which makes the optimization cumbersome, and 3) the optimization of the
policies in the distant future is computationally more demanding because the
contribution of the change of the policies in the distant future to the objective function
is much less due to discounting. Thus, the optimization variables are effectively eleven
expected service life times for the infrastructure that is constructed in each respective
period.
It should be mentioned that this optimization problem is reduced to identify the

optimal balance between the quality of infrastructure (measured in terms of the
expected service life time T t ) and the quantity of infrastructure (the amount k t of
new construction), since the size of budget available is limited to ef (k (t )) ; the
constraint ef (k (t )) = q(T t )k t must be satisfied each time.
Note that in this assumption the optimal policy at the steady state corresponds to
T * = 100 years, because the annual average cost, q(T ) / T = 1/ T + T /1002 (see Table
7.1), is minimized at T * = 100 .
-121-
perspective
Table 7.1. Functional forms and parameters postulated in the optimization problem.
Utility function u (c (t )) = ln c(t )

Discount rate for pure-time preference ρ = 0.02 [1/year]
Production function y(k (t )) = Ak (t )α , A = 10 , α = 0.2
Design and maintenance cost q(T ) = 1 + (T /100)2
Amount of initial capital k0 (0) = 10
Deterioration of initial capital k0 (t ) = k0 (0) (1 − t / 30 ) , ( 0 < t ≤ 30 )
Saving rate e = 0.2
The optimized23 service life time in each period and corresponding economic growth
path (denoted by a dynamically optimized policy) are shown in Figure 7.4. It is seen
that the optimal policy, which maximizes the social welfare, is to choose a shorter
expected service life time at an earlier stage of the economy and then to switch to a
longer expected service life time later. It should be mentioned that the optimized
expected service life time after 100 years is not T * = 100 years, which could lead to
the highest steady state. This is because the contribution of the utility of future
generations to the social welfare is small so that higher consumption of earlier
generations is more important to reach a higher social welfare. For comparison
purposes, two other economic paths for different policies are calculated; the economic
growth path in the case where the expected service life time is fixed at 100 years for all
periods (denoted by " T s = 100 years, fixed" in the figure) and the economic growth
path in the case where the expected service life time is incrementally increased from 40
years to 100 years (denoted by "step" in the figure). It is clearly seen that with the
"fixed" policy the economy suffers lower economic output in earlier years, although in
the long run the economic output can reach the highest value. Under the "step" policy
the economy can grow as fast as the economy under the dynamically optimized policy
in the earlier years. However, the economic growth becomes slower in later years
because of higher design and maintenance costs for the infrastructure with the longer
expected service life time. The calculated social welfare is highest in the case of the
dynamically optimized policy, the second highest in the case of the "step" policy and
the lowest in the case of the "fixed" policy. That is, from the viewpoint of the social
welfare maximization the application of the optimal policy at the steady state in the
transition state is suboptimal.
23 Note that this dynamic optimization result is only approximate. The reason is that the time horizon is
truncated at a finite time (in this calculation, 200 years). This might be problematic because with this
truncation a sound strategy for a future generation living just before the 200 years time limit is to
construct infrastructure with a very short expected service life time to increase the economic output
for a short term, without considering severe deterioration of the infrastructure which could occur after
200 years. However, the main conclusion in this section is valid, i.e. that the application of the
optimal policy at the steady state in the transition state is suboptimal, because the social welfare that
corresponds to the "dynamically optimized" policy is calculated using the obtained expected service
life times, and this is larger than the social welfare that corresponds to the policy whereby the
expected service life time of T s = 100 [year] is taken for the whole period of time.
-122-
perspective
Figure 7.4. Economic growth paths (top) and expected service life times (bottom).
7.4.4. Discussion and conclusion

In this chapter, first the derivation of the life-cycle cost optimization concept from the
more general principle, i.e. the social welfare maximization concept, is introduced.
Then, the assumptions and limitations of the life-cycle cost optimization concept are
pointed out. Thereafter, an optimization principle for the design and maintenance
policy on infrastructure is presented in a macroeconomic context based on the
economic growth theory. The optimization principle derived here can only be applied
at steady states of the economy. Finally, the dynamic optimization problem is
considered where the policy on the target expected service life time of infrastructure is
optimized. With the assumptions made in Sections 7.4.1 and 7.4.3, it is shown that a
better policy in the sense that it leads to a higher social welfare is to choose a shorter
expected service life time when the economic output level is relatively low, and then to
shift to a longer expected service life time when the economy grows enough to afford
-123-
perspective
the high cost but highly reliable infrastructure; the optimal policy on the reliability of
civil infrastructure at each time depends on the current economic output level.
The economic model considered in this chapter is simplistic. There are many
possibilities to extend the economic model; including private capitals and other types
of capitals in the economy; considering technological development, or employing
endogenous economic models; modeling the deterioration of infrastructure in different
probabilistic ways; using a more realistic budget framework especially for maintenance
costs. These extensions are addressed as future research tasks.
-124-
Conclusions and outlook
8. Conclusions and outlook
8.1. Conclusions
In the present thesis, the issues of sustainable decision-making in civil engineering,
especially design and maintenance strategies for structures, are addressed. These issues
are examined from two perspectives, i.e. marginal decision analysis and non-marginal
decision analysis. Within the context of marginal decision analysis, sustainable
decision problems can be formulated as constrained optimization problems. Therein,
the objective function is the expected discounted life-cycle cost associated with the
projects concerned, and the constraints correspond to the societal preferences with
respect to different aspects of sustainability, which are usually represented in terms of
acceptance criteria. In the context of non-marginal decision analysis, sustainable
policy-making on the design and maintenance of civil infrastructure can be discussed
within a macroeconomic framework. Focusing on individual issues in marginal
decision analysis as well as non-marginal decision analysis, the present thesis proposes
methods useful to formulate and solve the constrained optimization problems, and a
methodological approach for implementing the structural performance of infrastructure
in terms of reliability in the economic growth theoretical framework.
In the context of marginal decision analysis, the main constituents of the objective
function are: probability of failure; discount factors; cost terms such as initial cost,
maintenance cost, cost of failure and indirect cost beyond the direct cost associated
with structural failure. In Chapters 2, 4 and 5, these constituents are individually
addressed and investigated from a sustainability perspective. On the other hand, in
Chapter 3 a computational method is presented for formulating and solving the
constrained optimization problems integrating these constituents.
Chapter 2 considers the treatment of aleatory and epistemic uncertainties in the

probabilistic assessments of events. The motivation for this chapter is to emphasize the
importance of the consistent treatment of these types of uncertainty in probabilistic
assessment in general, and in the probabilistic assessment of extreme events during
longer periods which usually requires the extrapolation of knowledge concerning e.g.
the probabilistic characteristics of events during shorter periods in particular. Its
importance is emphasized by introducing three practical examples in which the
uncertainties are integrated in an inconsistent manner, and then by showing that such
inconsistent treatments can lead to highly biased estimates of the probabilistic
characteristics of extreme events. The principle presented in this chapter serves as a
philosophical basis for the treatment of uncertainties in sustainable decision analysis
for civil infrastructure.
-125-
Chapter 3 presents a platform on which the constituents of the constrained optimization

problems are fully represented, and the calculations required for optimizations can be
performed in a generic manner. For this, the Bayesian probabilistic networks and
influence diagrams are adopted as the probabilistic representation platforms. Such
representation directly allows for calculating any conditional probabilities and
conditional expected values of variables of interest by use of the generic algorithms
developed for such calculations as a function of any given decision alternative.
Furthermore, by linking the networks/diagrams to the generic algorithms available for
solving constrained optimization problems, the constrained optimization problems of
interest can be solved quasi automatically once the networks/diagrams corresponding
to the problems are established. The decision-makers can thus focus on the
development of such networks/diagrams, which is highly useful in practical
applications. Another practical advantage of employing the Bayesian probabilistic
networks or influence diagrams is that they can be facilitated as communication tools
among experts as well as between experts and non-experts. The use of the
networks/diagrams as a communication tool is especially useful in decision analysis
for civil infrastructure, since civil infrastructure is in general a complex system
composed of components at different levels and therefore the modelling of these
components and their possible consequences require the collaboration of experts from
different disciplines.
Chapter 4 addresses the issue of discounting. The motivation of this chapter is to

reconsider the formulation of the life-cycle cost optimization problem from an
intergenerational-equity perspective. The focus is on discounting for pure-time
preference. Because discounting for pure-time preference reflects the myopic nature of
individuals, the application of discounting for pure-time preference can be logically
justified only within individual generations. However, often in life-cycle optimization
problems, discounting for pure-time preference has been applied without considering
the finite duration of the generations (or as if one generation lives for ever). In this
chapter, based on the consideration similar but independent from the
generation-adjusted discounting concept proposed by Bayer and Cansier (1999), a
formula is proposed to calculate an equivalent discount rate. The equivalent discount
rate is the rate which, if applied to a decision problem with the classical perspective in
which one generation that is assumed to have an infinite lifetime, yields the same total
expected utility as when the decision problem is analyzed in accordance with the
consistent consideration of discounting over generations. The use of the formula can
thus avoid tedious calculations required for the assessment of the discounted life-cycle
costs if the generation-adjusted discounting concept was applied in a straightforward
manner. Furthermore, from the formula it directly turns out that the classical
perspective tends to put more burden on future generations by applying a higher
discount rate than the rate that is logically consistent with the implication of
discounting for pure-time preference and the finite duration of individual generations.
-126-
Chapter 5 reformulates the life-cycle cost optimization concept from a budget

allocation perspective. The background of this reformulation is that the life-cycle cost
optimization concept implicitly assumes that the necessary budget is available
whenever it is needed, which is not realistic in practice. Since the failure to acquire the
necessary budget on time incurs additional indirect costs due to the delay of actions, an
explicit consideration of such costs is required to formulate the objective function in
the life-cycle cost optimization concept. Thereby, because the probability of
occurrence of the non-availability of the budget is a function of the size of the budget
allocated to the project, the size of the budget to be allocated should be one of the
decision variables in the optimization problem. This shift of perspective is especially
useful for societal decision-makers who have to decide how to optimally allocate the
limited budget to different projects.
In Chapters 3, 4 and 5, it is implicitly assumed that the decision analyses are marginal.
That is, these analyses are only valid if the consequences of the decisions are
reasonably assumed not to influence long-term economic growth. However, there have
been some cases where the life-cycle cost optimization concept, which is the typical
concept for the marginal decision analysis, seems to have been applied beyond its
limitation24. Therefore, in order to clarify its underlying assumption and limitation, the
derivation of the life-cycle cost optimization concept from a broader decision principle
is introduced in Section 7.2.
In contrast to Chapters 3, 4 and 5, the non-marginal decision analysis is addressed

within the framework of economic growth theory in Chapters 6 and 7. The original
contribution here is that a methodological approach is proposed in regard to how the
reliability of structures defined in terms of limit state functions, and initial cost and
maintenance cost can be integrated into the framework. The direct benefit of the
proposed approach is that sustainable design and maintenance policies on
infrastructure can be discussed in the macroeconomic context, thereby enabling one to
assess the effects of the policies on long-term economic development. In Chapter 7, an
optimization principle for the design and maintenance policy on infrastructure is
derived in a macroeconomic context, which can apply at steady states of the economy.
Its objective function takes a different form from that of the objective function in the
life-cycle cost optimization concept. However, this is not contradictory; the
optimization principle derived in Chapter 7 should be applied for non-marginal
decision analysis, and the life-cycle cost optimization concept should be applied for
marginal decision analysis. It is also shown that the presented methodological
approach enables one to identify the optimal reliability (represented by the expected
service life time) of infrastructure as a function of the economic states, which could be
difficult within the marginal decision analysis framework.
24 In reality, it is often very difficult to check the marginality. Hence, the marginality is often merely an
assumption for the decision analysis. Still in such cases, it is important that the interpretation and
application of the analysis results should be consistent with the assumption.
-127-
The methods proposed for the marginal decision analysis are directly useful in present
practical decision situations where due consideration of sustainability is required. On
the other hand, the proposed approach for non-marginal decision analysis will serve as
the first step for a further development of the general framework for sustainable
policy-making on civil infrastructure.
8.2. Scientific achievements and limitations

The scientific achievements of the present thesis work are: (1) introduction of the
concept of marginality of engineering decision-making; (2) adaptation of the classical
life-cycle cost optimization concept to sustainable decision-making in the context of a
marginal decision analysis; (3) development of a non-marginal engineering decision
analysis framework.
However, these scientific achievements are limited. Regarding (1), the marginality of
decisions introduced in the present thesis is difficult to assess in practice; strictly
speaking, any engineering decisions can affect economic growth. Hence, by definition
any engineering decision can be non-marginal. Thus, marginal decision analysis can be
considered as an approximation. Therefore, the concept of marginality should be used
in practice to check if the assumption of the marginality is reasonable in given decision
situations; only when the assumption is reasonable can the marginal decision analysis
be performed. Otherwise, a non-marginal decision analysis should be undertaken.
Concerning (2), it is assumed that the objective function in the decision problems can
be represented by or otherwise converted to a monetary term. However, it is not clear if
the objective function can be fully described by the monetary term. Even if it were
possible, it is still not obvious how the values of different actions and consequences are
objectively quantified in the monetary term. The present thesis does not provide any
justification for this assumption and ways to quantify them. Furthermore, the boundary
conditions in the decision-making process are assumed to be given. However, since the
choice of the boundary conditions affects the optimization of the decision-making,
these boundary conditions should be carefully assessed and chosen. The way in which
they are assessed and chosen is addressed as a future task. Finally, in regard to (3), the
proposed framework is under development in the sense that the role of civil
infrastructure in the economy is considered only in terms of productivity; civil
infrastructure plays other important roles in society such as amenity for leisure and
safety measures to mitigate natural hazards. At the same time, the operation of civil
infrastructure impacts on the quality of the environment. These aspects are not
considered in the proposed framework and the consideration of these aspects is
addressed as an additional future task.
-128-
8.3. Outlook
8.3.1. Assessment of the boundary conditions in marginal decision analysis

Throughout the present thesis, the research focus is mainly on theoretical aspects of the
issues mentioned in the introduction. That is, whereas the method for solving the
constrained optimization problems given the constraints is presented (Chapter 3), the
constraints are assumed to be given in terms of the acceptance criteria related to the
aspects of sustainability, e.g. human safety and environmental impact; whereas the
formula for obtaining sustainable equivalent discount rates is presented (Chapter 4), it
is not investigated which values should be assumed for the discount rates for economic
growth and pure-time preference. In practice, these may be equally or even more
relevant in decision-making.
Concerning the acceptance criteria for human safety in the context of engineering
project appraisal, a number of approaches have been proposed and utilized in practice.
One common approach in practice is the use of the Farmer diagram, often called the
F-N curve 25 . In this approach, the F-N curve concerning a considered project is
compared with the criterion F-N curve, which is usually provided by regulatory
authorities; the considered project is acceptable if the F-N curve concerning the project
is below the criterion F-N curve. However, several inconsistencies in the use of F-N
curves for project appraisal concerning human safety have been pointed out. Among
others, it is possible that a project that associates higher expected fatality due to
possible accidents in a given time period is accepted, whereas another project that
associates lower expected fatality is rejected. This is because the F-N curve-based
project appraisal essentially concentrates on one extreme feature of the distribution of
fatalities due to different possible accidents disregarding the overall characteristics of
the distribution of the fatalities, see Evans and Verlander (1997). In Evans and
Verlander (1997), it is also shown that the F-N curve-based project appraisal fails to
pass a logical test for a prescriptive criterion. Recently, a promising approach has been
developed based on the life quality index (LQI) proposed by Nathwani et al. (1997).
The LQI is a social indicator that is composed of the gross domestic product per capita,
the life expectancy and the fraction of lifetime spent in working for a living. In this
approach, the LQI is considered to represent the indifference between the
increase/decrease of life expectancy and the decrease/increase of consumption per
capita. Thus, the willingness to pay for life-saving measures can be derived from this
index, see e.g. Skjong and Ronold (1998) and Rackwitz (2003). Further development
of this approach is necessary and is on-going, see e.g. Ditlevsen (2004), Kübler and
Faber (2005), Pandey et al. (2006), and Ditlevsen and Friis-Hansen (2008).
25 An F-N curve represents, for different n , the mean absolute frequency F (n) of the accidents in a
reference time period which associate n or more fatalities in a considered project. Normally, in the
diagram the horizontal axis corresponds to the number n of fatalities and the vertical axis
corresponds to the mean absolute frequency F (n) .
-129-
The acceptance criteria for environmental impacts, e.g. targets in pollution control, use
of non-renewable resources, and recycling of partly-renewable materials, have been
intensively discussed in environmental sciences and economics, see Perman et al.
(2003) for an overview. Among others, what is relevant to the design and maintenance
on civil infrastructure are: the recycling of the construction materials, e.g. cements,
aggregates in concrete and steel; emissions of carbon dioxide in the construction and
operation of the infrastructure. These should be addressed in the context of life-cycle
optimization problems as well as the life-cycle cost. Thereby, a controversial issue
arises; how to identify the decision alternatives among a set of the Pareto-optimal
solutions in the optimization problem if formulated as a multi-objective optimization
problem, or otherwise which attribute should be taken as the (scalar) objective function
and which should be considered as constraints in constraint optimization problems.
This should be addressed as a challenging research task.
Different choices of the values of discount rates often lead to different conclusions.
One of the examples can be seen in the debate between Nordhaus (2007) and Stern and
Taylor (2007) on the necessity for urgent countermeasures to global warming.
Nordhaus (2007) criticizes the choice of the values of the discount rates in the Stern
Report (2006) (the discount rate for pure-time preference: ρ = 0.001/ yr , the discount
rate for consumption growth: δ = 0.013 / yr , and the elasticity of the marginal utility
of consumption: η = 1 ) by arguing that the resulting real return rate,
r = ηδ + ρ = 0.014 / yr , is far smaller than the real return rate observed in the capital
market. For this criticism, Stern and Taylor (2007) justify their choice by claiming that
1) the discount rate for pure-time preference should be significantly smaller when the
consequences of the decision affect both current and future generations26, and 2) the
capital market is imperfect in the sense that those who do not or cannot participate in
the market (i.e. the young, the poor and the future generations) have little or no
influence on current market behavior. The underlying issue in this debate is the choice
of the perspective to be followed in societal decision-making; normative or descriptive.
The normative perspective seems reasonable for societal decision-making. However,
with this approach it is difficult to directly obtain the value of the discount rate for
pure-time preferences without relying on statistics that may be affected by the capital
market. Furthermore, the justification to assume a positive discount rate for pure-time
preference is a controversial issue, see e.g. Price (1993); the choice of the value of the
discount rate can be subjective.
Decision-making in civil engineering often encounters the similar situation in which

the consequences of the decisions affect current and future generations, e.g.
construction of nuclear power plants and nationwide infrastructure projects. In these
decision situations, whereas it is difficult to choose the commonly agreed discount
rates, it is important that the process leading to the choice is clear and transparent, and
26 For this, the discussions in Chapter 4 may provide a rationale.
-130-
consistent with the state of the art of the philosophical discussions on discounting;
continuous literature reviews and dissemination of the review results to
decision-makers is important.
8.3.2. Further development of non-marginal decision framework

The proposed macroeconomic decision framework is a promising platform, on which
the effects of policies in different sectors concerning the long-term development of the
economy can be examined individually as well as jointly. This is possible because the
framework defines the generic format for the component models based on the
economic growth theory: the social welfare function, the production function, and the
equation of motion that governs the changes in the amount of capitals. Thereby, it is
readily possible that the quality of different types of capitals can be modeled based on
the corresponding engineering knowledge through the limit state representation, the
reliability in the generalized sense is calculated using the structural reliability theory
and it is implemented into economic models in similar manners as illustrated for civil
infrastructure in Chapters 6 and 7. Here, it is mentioned that the policy making
concerning the global warming is one of such possible applications of the proposed
framework.
However, as is pointed out in Chapter 7 the equation of motion typically employed in

the economic growth theory is too simplistic for some types of capitals, including
infrastructure capital. On the other hand, realistic equations of motion as presented in
Chapter 7 significantly complicate dynamic optimization problems, so that the
application of the maximum principle is not feasible. Therefore, either a simplification
of the model that does not lose the relevant characteristics or an efficient algorithm for
solving the complicated dynamic optimization problem is required.
At the same time, more sophisticated, realistic economic models need to be developed
to fully capture the interaction between infrastructure capital and other capital, and the
socio-economic roles of infrastructure. It is also required to extend the framework in
order to enable one to take in account environmental aspects such as exploitation,
recycling and reuse of non-renewable resources and protection of the biodiversity. The
goal in the development of a non-marginal decision framework is to obtain a
framework that can identify the sustainable policies on infrastructure taking into
account all relevant aspects of sustainability including economic growth, the
socio-economic role and environmental issues jointly.
-131-
References
References
Arrow, K. J. (1962). The economic implications of learning by doing. Review of Economic
Studies, 29, 155-173.
ASCE7-98. (2000). Minimum design loads for buildings and other structures, Revision of
ANSI/ASCE.
ASCE. (2005). Report card for America's infrastructure.
Aschauer, D. A. (1989). Is Public Expenditure Productive? Journal of Monetary Economics,
23, 177-200.
Ayres, R. U., van den Bergh, J. C. J. M., and Gowdy, J. M. (1998). Viewpoint: Weak versus
strong sustainability. Tinbergen Institute Discussion papers.
Baker, J. W., Schubert, M., and Faber, M. H. (2007). On the assessment of robustness.
Structural Safety, In Press, Corrected Proof.
Bangso, O., Flores, M. J., and Jensen, F. V. (2003). Plug & Play OOBNs. Lecture Notes in
Artificial Intelligence, 3040, 457-467.
Bangso, O., and Olesen, K. G. (2003). Applying Object Oriented Bayesian Networks to Large
(Medical) Decision Support Systems. Proceedings of 8th Scandinavian Conference on
Artificial Intelligence, SCAI'03, Bergen, Norway.
Barro, R. J., and Sala-i-Martin, X. (2004). Economic Growth Second Edition, The MIT Press,
Cambridge, Massachusetts.
Bayer, S. (2003). Generation-adjusted discounting in long-term decision-making. International
Journal of sustainable development, 6(1), 133-149.
Bayer, S., and Cansier, D. (1999). Intergenerational Discounting: A New Approach. Journal of
International Planning Literature, 14(3), 301-325.
Bayraktarli, Y., and Faber, M. H. (2007). Bayesian network approach for managing earthquake
risks. International Forum on Engineering Decision Making, IFED3, Shoal Bay, Australia.
Benjamin, J. R., and Cornell, C. A. (1970). Probability, Statistics and Decision for Civil
Engineers, McGraw-Hill, New York.
Bobbio, A., Ciancamerla, E., Franceschinis, G., Gaeta, R., Minichino, M., and Portinale, L.
(2003). Sequential application of heterogeneous models for the safety analysis of a control
system: a case study. Reliability Engineering & System Safety, 81(3), 269-280.
Bobbio, A., Portinale, L., Minichino, M., and Ciancamerla, E. (2001). Improving the analysis
of dependable systems by mapping fault trees into Bayesian networks. Reliability
Engineering and System Safety, 71(3), 249-260.
Brundtland, G. H. (1987). Our Common Future, Oxford University Press.
Bulow, J. I., and Summers, L. H. (1984). The Taxation of Risky Assets. The Journal of
Political Economy, 92(1), 20-39.
Canning, D. (1998). A Database of World Infrastructure Stocks, 1950-95. World Bank Policy
Research Working Paper, 1929, World Bank.
Canning, D. (1999). Infrastructure's contribution to aggregate output. World Bank Policy
Research Working Paper, World Bank.
Canning, D., and Bennathan, E. (2000). The Social Rate of Return on Infrastructure
Investments. World Bank Policy Research Working Paper, 2390, World Bank.
Cass, D. (1965). Optimum Growth in an Aggregative Models of Capital Accumulation. Review
of Economic Studies, 32, 233-240.
Clark, J. S., and Gelfand, A. E. (2006). Hierarchical Modelling for the Environmental
Sciences, Oxford University Press, New York.
Coles, S. (2001). An Introduction to Statistical Modeling of Extreme Values, Springer-Verlag,
London.
Coles, S., Pericchi, L. R., and Sisson, S. (2003). A fully probabilistic approach to extreme
rainfall modeling. Journal of Hydrology, 273, 35-50.
Cornell, A. (1969). A Probability-Based Structural Code. ACI Journal, 974-985.
-132-
References
Dasgupta, P. S., and Heal, G. M. (1974). The optimal depletion of exhaustible resources.
Review of Economic Studies, 41, 3-28.
De Groot, H. (1970). Optimal Statistical Decisions, John Wiley&Sons, Inc.
Der Kiureghian, A., and Ditlevsen, O. (2007). Aleatory or epistemic? Does it matter? Special
workshop on risk acceptance and risk communication, Stanford University, California,
USA.
Der Kiureghian, A., and Ditlevsen, O. (2008). Aleatory or epistemic? Does it matter?
Structural Safety, available online.
Der Kiureghian, A., Haukaas, T., and Fujimura, K. (2006). Structural reliability software at the
University of California, Berkeley. Structural Safety, 28(1-2), 44-67.
Der Kiureghian, A., and Moghtaderi-Zadeh, M. (1982). An integrated approach to the
reliability of engineering systems. Nuclear Engineering and Design, 71(3), 349-354.
Der Kiureghian, A., and Song, J. (2008). Multi-scale reliability analysis and updating of
complex systems by use of linear programming. Reliability Engineering & System Safety,
93(2), 288-297.
Ditlevsen, O. (2004). Life quality index revisited. Structural Safety, 26(4), 443-451.
Ditlevsen, O., and Bjerager, P. (1986). Methods of structural systems reliability. Structural
Safety, 3(3-4), 195-229.
Ditlevsen, O., and Friis-Hansen, P. (2008). Cost and benefit including value of life, health and
environmental damage measured in time units. Structural Safety, In Press, Corrected Proof.
Ditlevsen, O., and Madsen, H. O. (2005). Structural Reliability Methods.
Duffy-Deno, K. T., and Eberts, R. W. (1991). Public infrastructure and regional economic
development: A simultaneous equations approach. Journal of Urban Economics, 30,
329-343.
Duggal, V. G., Saltzman, C., and Klein, L. R. (1999). Infrastructure and productivity: a
nonlinear approach. Journal of Econometrics, 92, 47-74.
DuraCrete. (2000). Statistical Quantification of the Variables in the Limit State Functions.
Final DuraCrete report, European Union.
Easterly, W., and Rebelo, S. (1993). Fiscal policy and economic growth. Journal of Monetary
Economics, 32(3), 417-458.
Engle, R. F., and Granger, C. W. J. (1987). Co-integration and error correction: Representation,
estimation, and testing. Econometrica, 55(2), 251-276.
EUROCONSTRUCT. (2007). European construction market trends to 2010, Country report.
Evans, A. W., and Verlander, N. Q. (1997). What Is Wrong with Criterion FN-Lines for
Judging the Tolerability of Risk? Risk Analysis, 17(2), 157-168.
Faber, M. H. (2003). Uncertainty Modeling and Probabilities in Engineering Decision
Analysis. Proceedings of the 22nd International Conference on Offshore Mechanics and
Arctic Engineering, OMAE2003, Cancun, Mexico.
Faber, M. H., Bayraktarli, Y., and Nishijima, K. (2007a). Recent Developments in the
Management of Risks Due to Large Scale Natural Hazards. XVI Congreso Nacional
Ingenieria Sismica, Ixtapa-Zihuatanejo, Mexico.
Faber, M. H., Engelund, S., Sørensen, J. D., and Bloch, A. (2000). Simplified and Generic Risk
Based Inspection Planning. Proceedings OMAE2000, 19th Conference on Offshore
Mechanics and Arctic Engineering, New Orleans, Louisiana, USA,
[OMAE2000/S&R6143].
Faber, M. H., Maes, M., Baker, J., Vrouwenvelder, T., and Takada, T. (2007b). Principles of
risk assessment of engineered systems. ICASP 10, Tokyo, Japan.
Faber, M. H., and Maes, M. A. (2003). Modeling of Risk Perception in Engineering Decision
Analysis. Proceedings of 11th IFIP WG7.5 Working Conference on Reliability and
Optimization of Structural Systems, 113-122.
Faber, M. H., and Maes, M. A. (2005). Epistemic Uncertainties in Decision Making.
Proceedings of the 24th International Conference on Offshore Mechanics and Arctic
Engineering, OMAE2005, Halkidiki, Greece.
-133-
References
Faber, M. H., and Nishijima, K. (2004). Aspects of Sustainability in Engineering Decision

Analysis. Proceedings 9th ASCE Specialty Conference on Probabilistic Mechanics and
Structural Reliability, Albuquerque, New Mexico, USA.
Faber, M. H., and Rackwitz, R. (2004). Sustainable decision making in civil engineering.
Structural Engineering International, 14(3), 237-242.
Faber, M. H., Straub, D., and Maes, M. A. (2005). A Computational Framework for Risk
Assessment of RC structures Using Indicators. Computer-Aided Civil and Infrastructure
Engineering, accepted for publication.
Feller, W. (1966). An Introduction to Probability Theory and Its Applications, John Wiley &
Sons Inc., New York.
Ferry-Borges, J., and Castanheta, M. (1971). Structural Safety, 2nd Edition. National Civil
Engineering Laboratory, Lisboa, Portugal.
Freudenthal, A. M. (1947). The safety of structures. Transactions of American Society of Civil
Engineering, 112, 125-159.
Freudenthal, A. M. (1954). Safety and Probability of Structural Failure. Transactions of
American Society of Civil Engineering, 1337-1397.
George, D., and Hawkins, J. (2005). A Hierarchical Bayesian Model of Invariant Pattern
Recognition in the Visual Cortex. Neural Networks, 3, 1812-1817.
Glomm, G., and Ravikumar, B. (1994). Public investment in infrastructure in a simple growth
model. Journal of Economic Dynamics and Control, 18(6), 1173-1187.
Gramlich, E. M. (1994). Infrastructure Investment: A Review Essay. Journal of Economic
Literature, 32(3), 1176-1196.
Greenwood, J., Hercowitz, Z., and Krusell, P. (2000). The role of investment-specific
technological change in the business cycle. European Economic Review, 44, 91-115.
Guha-Sapir, D., Hargitt, D., and Hoyois, P. (2004). Thirty years of natural disasters 1974-2003:
The numbers. Centre for Research on the epidemiology of Disasters.
Guikema, S. D., and Pate-Cornell, M. E. (2002). Component choice for managing risk in
engineered systems with generalized risk/cost functions. Reliability Engineering & System
Safety, 78(3), 227-238.
Hartwick, J. M. (1977). Intergenerational equity and the investing of rents from exhaustible
resources. American Economic Review, 67, 972-974.
Hartwick, J. M. (1978). Substitution among exhaustible resources and intergenerational equity.
Review of Economic Studies, 45(347-354).
Hasofer, A. M., and Lind, N. C. (1974). Exact and Invariant Second-Moment Code Format.
Journal of Engineering Mechanics Division, Proceedings of the American Society of Civil
Engineers, 100, 111-121.
Hohenbichler, M., and Rackwitz, R. (1982). First-order concepts in system reliability.
Structural Safety, 1(3), 177-188.
Holtz-Eakin, D., and Schwartz, A. E. (1995). Infrastructure in a structural model of economic
growth. Regional Science and Urban Economics, 25(2), 131-151.
JCSS. (2001a). Probabilistic Assessment of Existing Structures, A publication of the Joint
Committee on Structural Safety (JCSS), RILEM Publications S.A.R.L The publishing
Company of RILEM.
JCSS. (2001b). Probabilistic Model Code. The Joint Committee on Structural Safety.
Jensen, F. V. (2001). Bayesian Networks and Decision Graphs, Springer, New York.
Johnson, V. E., Graves, T. L., Hamada, M., and Reese, C. S. (2002). A Hierarchical Model for
Estimating the Reliability of Complex Systems. Bayesian Statistics, 7, 199-214.
JSCE. (2008). わが国におけるインフラの現状と評価インフラ国勢調査 2007 -体力測
定と健康診断-.
Kübler, O., and Faber, M. H. (2005). LQI: On the correlation between life expectanca and the
gross domestic product per capita. Proceedings ICOSSAR2005, 9th International
Conference on Structural Safety and Reliability, Rome, Italy, 3513-3517.
Kalaitzidakis, P., and Kalyvitis, S. (2004). On the macroeconomic implications of maintenance
in public capital. Journal of Public Economics, 88, 695-712.
-134-
References
Kjaerulff, U. (1995). dHugin: A computational system for dynamic time-sliced Bayesian

networks. International Journal of Forecasting.
Koopmans, T. C. (1965). On the concept of optimal economic growth, Amsterdam: North
Holland.
Korb, K. B., and Nicholson, A. E. (2004). Bayesian Artificial Intelligence, Champman &
Hall/CRC.
Krautkraemer, J. A. (1999). On sustainability and intergenerational transfers with a renewable
resource. Land Economics, 75, 167-184.
Leadbetter, M. R., Lindgren, G., and Rootzén, H. (1983). Extremes and Related Properties of
Random Sequences and Series, Springer Verlag, New York.
Li, F.-F., and Pietro, P. (2005). A Bayesian Hierarchical Model for Learning Natural Scene
Categories. IEEE Conference on Computer Vision and Pattern Recognition San Diego,
USA.
Lin, Y. K. (1967). Probabilistic theory of structural dynamics, McGraw-Hill.
Lindley, D. V. (1965). Introduction to probability and statistics : from a Bayesian viewpoint,
Cambridge University Press, Cambridge.
Lindley, D. V. (1980). Introduction to Probability & Statistics - from a Bayesian viewpoint,
Part.1-Probability, Cambridge University Press, Cambridge.
Lindley, D. V., and Smith, A. F. M. (1972). Bayes Estimates for the Linear Model. Journal of
the Royal Statistical Society. Series B (Methodological), 34(1), 1-41.
Lucas, R. E. J. (1988). On the mechanics of economic development. Journal of Monetary
Economics, 22, 3-42.
Maes, M. A. (1990). The Influence of Uncertainties on the Selection of Extreme Values of
Environmental Loads and Events. Civil Engineering Systems, 7(2), 115-124.
Maes, M. A., and Jordaan, I. J. (1985). Extremal Analysis of Loads Using Exchangeability.
Proceedings of the 4th International Conference on Structural Safety and Reliability,
ICOSSAR 1985, Tokyo, Japan, 579-589.
Mayer, M. (1926). Die Sicherheit der Bauwerke, Springer, Berlin.
Munich Re. (2005). Annual review: Natural catastrophes 2005. Topics Geo.
Munnell, A. H. (1992). Policy Watch: Infrastructure Investment and Economic Growth. The
Journal of Economic Perspectives, 6(4), 189-198.
Nathwani, J. S., Lind, N. C., and Pandey, M. D. (1997). Affordable Safety by Choice: The Life
Quality Method, University of Waterloo, Waterloo.
Newell, R. G., and Pizer, W. A. (2004). Uncertain discount rates in climate policy analysis.
Energy Policy, 32(4), 519-529.
Nijkamp, P., and Poot, J. (2004). Meta-analysis of the effect of fiscal policies on long-run
growth. European Journal of Political Economy, 20(1), 91-124.
Nishijima, K., and Faber, M. H. (2006). A Budget Management Approach for Societal
Infrastructure Projects. IABMAS'06, 3rd International Conference on Bridge Maintenance,
Safety and Management, Porto, Portugal.
Nishijima, K., and Faber, M. H. (2007a). Bayesian approach to proof loading of quasi-identical
multi-components structural systems. Civil Engineering and Environmental Systems, 24(2),
111-121.
Nishijima, K., and Faber, M. H. (2007b). A Bayesian framework for typhoon risk management.
12th International Conference on Wind Engineering, 12ICWE, Cairns, Australia.
Nishijima, K., and Faber, M. H. (2007c). On Structural Performance vs. Societal Economic
Growth. 10th International Conference on Applications of Statistics and Probability in Civil
Engineering, ICASP10, Kashiwa, Japan.
Nishijima, K., Maes, M. A., Goyet, J., and Faber, M. H. (2008). Constrained optimization of
component reliabilities in complex systems. Structural Safety.
Nishijima, K., Straub, D., and Faber, M. H. (2004). Sustainable decisions for Life-Cycle Based
Design and Maintenance. First Forum on Engineering Decision Making, IFED, Stoos,
Switzerland.
-135-
References
Nishijima, K., Straub, D., and Faber, M. H. (2005). The Effect of Changing Decision Makers
on the Optimal Service Life Design of Concrete Structures. Proceedings of the 4th
International Workshop on Life-Cycle Cost Analysis and Design of Civil Infrastructures
Systems, Cocoa Beach, Florida, 325-333.
Nishijima, K., Straub, D., and Faber, M. H. (2007). Inter-generational distribution of the
life-cycle cost of an engineering facility. Journal of Reliability of Structures and Materials,
3(1), 33-46.
Nordhaus, W. (2007). Critical Assumptions in the Stern Review on Climate Change. Science,
317, 201-202.
O'Hagan, A., and Oakley, J. E. (2004). Probability is perfect, but we can't elicit it perfectly.
Reliability Engineering & System Safety, 85, 239-248.
Pandey, M. D., Nathwani, J. S., and Lind, N. C. (2006). The derivation and calibration of the
life-quality index (LQI) from economic principles. Structural Safety, 28(4), 341-360.
Pate-Cornell, M. E. (1996). Uncertainties in risk analysis: Six levels of treatment. Reliability
Engineering & System Safety, 54, 95-111.
Perman, R., Ma, Y., McGilvray, J., and Common, M. (2003). Natural Resource and
Environmental Economics (3rd edition), Pearson Education Limited, Harlow Essex UK.
Pezzey, J. C. V. (1992). Sustainability: an interdisciplinary guide. Environmental Values, 1,
321-362.
Pezzey, J. C. V. (1997). Sustainability constraints versus optimality versus intertemporal
concern, and axioms versus data. Land Economics, 73(4), 448-466.
Pezzey, J. C. V., and Withagen, C. A. (1998). The rise, fall and sustainability of
capital-resource economics. Scandinavian Journal of Economics, 100(2), 513-527.
Press, W. H., Flannery, B. P., Teukolsky, S. A., and Vetterling, W. T. (1988). Numerical Recipes
in C, Cambridge University Press.
Price, C. (1993). Time, Discounting & Values, Blackwell Publishers, Oxford, United Kingdom.
Rackwitz, R. (2000). Optimization - the basis of code-making and reliability verification.
Rackwitz, R. (2002). Optimization and Risk Acceptability Based on the Life Quality Index.
Structural Safety, 24(2-4), 297-332.
Rackwitz, R. (2003). Acceptable Risks and Affordable Risk Control for Technical Facilities
and Optimization. Reliability Engineering and Systems Safety.
Rackwitz, R., Lentz, A., and Faber, M. (2005). Socio-economically sustainable civil
engineering infrastructures by optimization. Structural Safety, 27(3), 187-229.
Raiffa, H., and Schlaifer, R. (1961). Applied Statistical Decision Theory, Cambridge
University Press, Cambridge.
Ramsey, F. (1928). A mathematical theory for saving. Economic Journal, 38, 543-559.
Raudenbush, S., and Bryk, A. S. (1986). A Hierarchical Model for Studying School Effects.
Sociology of Education, 59(1), 1-17.
Rioja, F. K. (2003). Filling potholes: macroeconomic effects of maintenance versus new
investments in public infrastructure. Journal of Public Economics, 87, 2281-2304.
Romer, P. M. (1986). Increasing returns and long-run growth. Journal of Political Economy,
94, 1002-1037.
Rosenblueth, E., and Mendoza, E. (1971). Reliability Optimization in Isostatic Structures.
Journal of the Engineering Mechanics Division, 97(6), 1625-1648.
Royset, J. O., Der Kiureghian, A., and Polak, E. (2003). Reliability-based optional design:
problem formulations, algorithms and application. Proceedings of the 11th IFIP WG7.5
Working conference on reliability and optimization of structural systems, Banff, Canada,
1-12.
Salazar, D., Rocco, C. M., and Galvan, B. J. (2006). Optimization of constrained
multiple-objective reliability problems using evolutionary algorithms. Reliability
Engineering & System Safety, 91(9), 1057-1070.
Skjong, R., and Ronold, K. (1998). Societal Indicators and Risk Acceptance. 17th
International Conference on Offshore Mechanics and Arctic Engineering, Lisbon, Portugal.
-136-
References
Smyth, P. (1997). Belief networks, hidden Markov models, and Markov random fields: A
unifying view. Pattern Recognition Letters, 18(11-13), 1261-1268.
Solow, R. M. (1956). A contribution to the theory of economic growth. Quarterly Journal of
Economics, 70, 65-94.
Solow, R. M. (1974). Intergenerational equity and exhaustible resources. Review of Economic
Studies, 41(22-46).
Solow, R. M. (1986). On the intergenerational allocation of natural resources. Scandinavian
Journal of Economics, 88(1), 141-149.
Song, J., and Kang, W.-H. (2008). System reliability and sensitivity under statistical
dependence by matrix-based system reliability method. Structural Safety, In Press,
Corrected Proof.
Stern, N. (2006). Stern Review: The Economics of Climate Change. HM Treasury.
Stern, N., and Taylor, C. (2007). Climate Change: Risk, Ethics, and the Stern Review. Science,
317, 203-204.
Stiglitz, J. E. (1974). Growth with exhaustible natural resources: efficient and optimal growth
path. Review of Economic Studies, 41, 123-137.
Straub, D. (2004). Generic approaches to risk based inspection planning for steel structures,
PhD thesis, ETH Zurich, Zurich.
Straub, D., and Der Kiureghian, A. (2008). Improved seismic fragility modeling from
empirical data. Structural Safety, 30(4), 320-336.
Straub, D., and Faber, M. H. (2005). Risk based inspection planning for structural systems.
Straub, D., and Faber, M. H. (2006). Computational Aspects of Risk-Based Inspection
Planning. Computer-Aided Civil and Infrastructure Engineering, 21(3), 179-192.
Swan, T. W. (1956). Economic growth and capital accumulation. Economic Record, 32,
334-361.
Tang, W. H. (1973). Probabilistic Updating of Flaw Information. Journal oft Testing and
Evaluation, 1, 459-467.
Thoft-Christensen, P., and Sørensen, J. D. (1987). Optimal Strategy for Inspection and Repair
of Structural Systems. Civil Engineering Systems, 4, 94-100.
Turner, R. K. (1992). Speculations on Weak and Strong Sustainability. Centre for Social and
Economic Research on the Global Environment (CSERGE).
USNRC. (1975). Reactor Safety Study - An Assessment of Accident Risks in U.S. Commercial
Nuclear Power Plants, WASH-1400 (NUREG-75/014). U.S. Nuclear Regulatory
Commission.
USNRC. (1990). Severe Accident Risks: An Assessment for Five U.S. Nuclear Power Plants
(NUREG-1150). U.S. Nuclear Regulatory Commission.
Valente, S. (2005). Sustainable development, renewable resources and technological progress.
Environmental and Resource Economics, 30(1), 1573-1502.
Vanmarcke, E. (1983). Random Fields: Analysis and Synthesis, MIT Press, Cambridge,
Massachusetts.
Vesely, W. E., Goldberg, F. F., Roberts, N. H., and Haasl, D. F. (1981). Fault Tree Handbook
(NUREG-0492). U.S. Nuclear Regulatory Commission.
Volovoi, V. (2004). Modeling of system reliability Petri nets with aging tokens. Reliability
Engineering & System Safety, 84(2), 149-161.
Wen, Y. K., Ellingwood, B. R., Veneziano, D., and Bracci, J. (2003). Uncertainty Modeling in
Earthquake Engineering. MAE Center Project FD-2 Report.
Wierzbicky, W. (1936). La sécurité des constructions comme un problème de probabilité.
Annales de l'academie des sciences techniques, 7, 63-74.
Withagen, C. A. A. M. (1996). Sustainability and investment rules. Economic Letters, 53, 1-6.
World Bank. (1994). World development report 1994: infrastructure for development. World
Bank.
Zeira, J. (1987). Risk and Capital Accumulation in a Small Open Economy. The Quarterly
Journal of Economics, 102(2), 265-280.
-137-
Curriculum vitae
Curriculum vitae
PERSONAL DETAILS
Name Kazuyoshi Nishijima
Address Hofwiesenstr. 200, 8057 Zurich, Switzerland
Telephone 044 633 43 16
E-mail nishijima@ibk.baug.ethz.ch
Date of birth 13 August, 1978
Citizenship Japan
EDUCATION
2004- ETH Zurich, Institute of Structural Engineering (IBK), D-BAUG

Ph.D. student in the Group of Risk and Safety
2003-2004 University of Tokyo, Institute of Environmental Studies

Ph.D. student
2001-2003 University of Tokyo, Institute of Environmental Studies

Master of Environmental Studies
1997-2001 University of Tokyo, Faculty of Engineering (Architecture)

Bachelor of Engineering
-138-
Curriculum vitae
PROFILE
Awards
2005 Japan Association for Wind Engineering Award (shourei-sho)

from the Japan Association for Wind Engineering
2003 Award for excellent master thesis from department of

environmental studies, courses of socio-cultural and
socio-physical environment, the University of Tokyo
Scholarship
2003-2004 Research fellowship for young scientists of the Japan Society for
the Promotion of Science (JSCS), DC1
1997-2001 Scholarship student of the Kinoshita Scholarship Foundation
Language
Japanese Native
English Very good
German Good
-139-

Eth 41460 02

Uploaded by

Copyright:

Available Formats

Eth 41460 02

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Eth 41460 02

Uploaded by

Copyright:

Available Formats

What is the thesis about?

What is the thesis about?

What challenges does the author face during their research?

What challenges does the author face during their research?

Research Collection

Issues of sustainability in engineering decision analysis

Issues of sustainability in engineering decision analysis

for the degree of

Master of Environmental Studies, University of Tokyo

accepted on the recommendation of

Prof. Dr. Michael Havbro Faber, examiner

The realization of a sustainable development of society necessitates that a holistic

Whereas the development of a general framework for sustainable decision-making is

In the reformulation of the classical life-cycle cost optimization, its practical

Die Umsetzung einer nachhaltigen gesellschaftlichen Entwicklung erfordert eine

Zwei Typen von Entscheidungsanalysen im Ingenieurwesen wurden in dieser Arbeit

In der Neuformulierung des klassischen Lebenszyklusansatzes wird die praktische

2. PROBABILISTIC ASSESSMENT OF EXTREME EVENTS SUBJECT TO EPISTEMIC

3. CONSTRAINED OPTIMIZATION OF COMPONENT RELIABILITY IN COMPLEX

4. INTER-GENERATIONAL DISTRIBUTION OF THE LIFE-CYCLE COST OF AN

5. A BUDGET MANAGEMENT APPROACH FOR SOCIETAL INFRASTRUCTURE

7. OPTIMAL DESIGN AND MAINTENANCE POLICY ON INFRASTRUCTURE FROM A

7.1. INTRODUCTION .............................................................................................................. 110

8. CONCLUSIONS AND OUTLOOK.......................................................................................... 125

8.1. CONCLUSIONS ................................................................................................................ 125

REFERENCES ................................................................................................................................... 132

CURRICULUM VITAE ..................................................................................................................... 138

In some developed countries, a considerable amount of economic resources is allocated

The statistics on economic losses due to natural hazards are summarized by

Deterioration of infrastructure and losses of infrastructure due to natural hazards are

Today, due consideration of sustainability is required in almost all civil engineering

1.2. Aim of the thesis

are regarded as best practices for the realization of a sustainable development of

Issue 2: Adaptation of optimization problems to sustainable decision-making

resources, degree of homogeneity of welfare between members in society, and human

Issue 3: Inter-generational equity

Issue 4: Balance between quality and quantity

1.3. Scope of the thesis

In principle, any engineering decision-making may affect economic growth. Hence,

1.4. State of the art in relevant research topics

1.4.1. Structural performance of civil infrastructure

general framework for sustainable decision-making on civil infrastructure in a

1.4.2. Socio-economic role of civil infrastructure

A pioneering work on the effectiveness of investment in civil infrastructure is that of

1.4.3. Implication and formulation of sustainability

The first aspect is the substitutability of different types of capital in production

The second aspect concerns the economic concepts of sustainability; opportunity-based

• A sustainable state is one in which utility (or consumption) is non-declining

Other concepts, which originate in ecology, are stated as:

• A sustainable state is one in which resources are managed so as to maintain a

1.5. Outline of the thesis

Chapter 2 (Paper I) focuses on the treatment of aleatory and epistemic uncertainties in

6 From Table 4.2 in Perman et al. (2003).

Chapter 4 (Paper III) considers the issue of discounting in the context of

Chapter 5 (Paper IV) reformulates optimization problems of civil infrastructure

In Chapters 3 to 5, the primary objective is to optimize individual civil infrastructure

exogenous, independent of the decisions regarding individual projects. However,

2. Probabilistic assessment of extreme events subject to

Institute of Structural Engineering, ETH Zurich, ETH Hönggerberg, HIL E 22.3,