Real Options and Flexibility in Engineering Design:

Emerging Paradigm for Complex Systems Research

Michel-Alexandre CARDIN, PhD

Assistant Professor of Industrial and Systems Engineering
National University of Singapore
 Strategic Engineering Lab
Funding
SGD $2.0M (1.3M) over last 4 years (excluding
current PhD student scholarships)
Projects
8 projects, funded by external (SMART, SEC, NRF-
CREATE) and internal sources (NUS)
Collaboration with local companies/agencies
Manpower
7 post-doc fellows and research associates, 5 PhD
students, > 22 undergraduate theses
CURRICULUM VITAE

PERSONAL INFORMATION
Name: Hu Junfei
Nationality: Chinese (Singapore Permanent Residence)
Contact Number: +65 96165232 (FRS) Future Resilient Systems
Contact Email: hujunfei@nus.edu.sg
Singapore-ETH Centre
Correspondence Address: BLK E1A #05-19, SMALab, ISE, NUS,119260 FRS Project Leader
EDUCATION BACKGROUND 2015 M.-A. Cardin. All rights reserved 1 CREATE Way 2
#06-01 CREATE Tower
Doctor of Philosophy 2008
Singapore present
138602
 Mission
To develop theory of real options and flexibility
in the design, architecture, evaluation, and
management of complex engineering systems

To develop, evaluate, and test systematic

procedures for engineering design and
management under uncertainty

To improve lifecycle performance of complex

engineering systems and products compared to
standard design and project evaluation
approaches

2015 M.-A. Cardin. All rights reserved 3

 What is Flexibility?
Provides right, but not
obligation, to change system
easily in face of uncertainty
Abandon
Defer
Expand/contract
Phase City Group Building, NYC
Switch
Etc.

Also known as Real Option

In system: requires
engineering design
considerations
On system: from managerial
Source: Guma et al., 2009
standpoint

2015 M.-A. Cardin. All rights reserved 4

 Why Flexibility in Systems Matters?
Source: www.comlinks.com
Engineering discipline increasingly Iridium
complex System:
Need socio-technical considerations Demand forecast over
optimistic, too much
capacity deployed at
Uncertainty affects lifecycle performance once filed for
Markets volatile, regulations change, bankruptcy (de Weck
technology evolve et al., 2004)

Flexibility can improve performance by

10%-30% compared to standard design
and project evaluation approaches
Source: en.wikipedia.org
Protects from downsides (e.g. insurance) B-58 Hustler:
Position for upsides (e.g. stock option) No contingency for
Net effect: better expected Soviet surface-to-air
performance! missiles quickly
obsolete, only 10
years of service
Design rigidity may lead to system (Saleh and Hastings,
failure 2000)
Iridium satellite/cell phone system
Convair B-58 Hustler

2015 M.-A. Cardin. All rights reserved 5


Michel-Alexandre Cardin
Department of Industrial and
Systems Engineering,
National University of Singapore,
Block E1A, #06-25,
1 Engineering Drive 2,
117576 Singapore
e-mail: macardin@nus.edu.sg
Enabling Flexibility in
Engineering Systems: A
Taxonomy of Procedures
and a Design Framework
This paper presents a five-phase taxonomy of systematic procedures to enable flexibility
in the design and management of engineering systems operating under uncertainty. The
taxonomy integrates contributions from surveys, individual articles, and books from the
literature on engineering design, manufacturing, product development, and real options
analysis obtained from professional e-index search engines. Thirty design procedures
were classified based on the kind of early conceptual activities they support: baseline
design, uncertainty recognition, concept generation, design space exploration, and pro-
cess management. Each procedure is evaluated based on ease of use to enable flexibility
analysis, whether it can be used directly in collaborative design activities, and has a pro-
ven applicability record in industry and research. The organizing principles integrate the
procedures into a cohesive and systematic design framework. Demonstration applications
on engineering systems case studies show that it helps designers select relevant proce-
dures in different phases of the design process, depending on the context, available ana-
lytical resources, and objectives. In turn, the case studies show that the design framework
helps generate concepts with improved lifecycle performance compared to baseline
concepts. The taxonomy provides guidance to organize ongoing research efforts, and
highlights potential contribution areas in this field of engineering design research.
[DOI: 10.1115/1.4025704]

Keywords: conceptual design, design theory and methodology, systems design, systems
engineering, uncertainty analysis

1 Introduction systems. The literature from real options analysis provides analyt-
ical tools to assess the value of flexibility quantitatively, allowing
This paper presents a five-phase taxonomy of systematic proce-
for objective evaluation of systems design concepts. Combining
dures to enable flexibility in the design and management of engi-
the two literatures provides an extensive and complementary tool-
neering systems operating under uncertainty. It has the dual goal
kit to create better performing systems. The ideas exposed in this
of providing a review of the latest contributions in this field, and
paper are inspired from this unique perspective.
organizing existing procedures into a cohesive design framework.
The paper proposes the notion of a flexible systems design con-
The taxonomy is geared specifically for engineering systems, in
cept to describe a design concept that provides an engineering sys-
particular complex systems in the aerospace, defense, energy,
tem with the ability to adapt, change and be reconfigured, if
housing, telecommunication, and transportation industries. Such
needed, in light of uncertainty realizations. It is different concep-
systems are characterized by a high degree of technical complex-
tually from a robust design concept, which makes systems func-
ity, social intricacy, and elaborate processes fulfilling important
tions more consistent and invariant to changes in the environment,
functions in society [1]. They are long-lived (20 yr), require
manufacturing, deterioration, and customer use patternsinspired
large irreversible investments, will inevitably face much uncer-
from the definition in Ref. [6]. A flexible systems design concept
tainty over their useful lifetime, and have a significantly large
is typically comprised of two components: (1) a strategy, and (2)

RESEARCH OVERVIEW
number of design variables and parameters. Dynamic socio-
an enabler in design and management. A strategy is similar con-
technical elements like markets, operational environment, regula-
ceptually to the definition of a real option on systems by Wang
tions, and technology play a significant role in their success and
and de Neufville [7], also referred as real option types by
failure [2]. Crucial decisions have to be made in early conceptual
Mikaelian et al. [8]. These can refer for instance to strategies sug-
design phases, regarding long-term strategic deployment and
gested by Trigeorgis [5]like abandonment, capacity expansion/
operations.
reduction, switching inputs/outputs, deferring investments, etc.
This paper builds upon the definition of flexibility in systems
to provide the system with better flexibility. A strategy represents
engineering and design enabling a system to change easily in the
the aspect of the design concept that captures flexibility, or how
face of uncertainty considering technical and technological
the system is designed to adapt to changing circumstances. The
standpoints [3,4]. It also builds upon the definition of a real
concept of enabler is similar to the definition of real option in
option, which provides the right, but not the obligation, to change
systems by Wang and de Neufville [7], or mechanism by
a system in the face of uncertainty [5]. The literature from
Mikaelian et al. [8]. It represents what is done to the physical
engineering provides tools to help generate flexibility in complex
infrastructure design and management to provide and use the
flexibility in operations. Enablers take a different form for each
Contributed by the Design Theory and Methodology Committee of ASME for
publication in the JOURNAL OF MECHANICAL DESIGN. Manuscript received September
system, depending on the flexibility strategy selected.
24, 2012; final manuscript received October 1, 2013; published online November 7, The following examples provide intuition on why flexibility
2013. Assoc. Editor: Irem Y. Tumer. is a worthwhile design paradigm. The Health Care Service

Journal of Mechanical Design Copyright V

C 2014 by ASME JANUARY 2014, Vol. 136 / 011005-1

Downloaded From: http://mechanicaldesign.asmedigitalcollection.asme.org/ on 11/16/2013 Terms of Use: http://asme.org/terms

M.-A. Cardin, "Enabling Flexibility in Engineering Systems: A

Taxonomy of Procedures and a Design Framework," ASME Journal
of Mechanical Design, vol. 136, 2014. doi: 10.1115/1.4025704

2015 M.-A. Cardin. All rights reserved 6

 Design Framework and Taxonomy
- Axiomatic Design
- Collaboration Engineering
- C-K Theory
- Game Theory
- Concurrent Design
- Serious Gaming
- Function-Based Failure Analysis
- Pahl and Beitz
- Etc.

Quantitative Concept Evaluation

- Binomial Lattice
1. Baseline!
- Decision Analysis Design!
- Simulations

- Bayesian Theory
- Dempster-Shafer Theory
- Possibility Theory
4. Design ! - Probability Theory
5. Process! 2. Uncertainty! - Statistical Analysis
Space!
Management! Recognition! - Binomial Lattice
Exploration! - Decision Tree
- Diffusion Model
- Scenario Planning
Computationally Efficient Search

- Design Catalogs
- Decision-Based Design 3. Concept!
- MATE Generation!
- Screening Methods

Strategy Generation Enabler Identification

- Explicit Training and Prompting - DSM-Based

- Integrated Real Options Framework - Explicit Design Variable Evaluation
- Real Option Strategies - Industry Guidelines

2015 M.-A. Cardin. All rights reserved 7

 Author's personal copy
Res Eng Design (2015) 26:121143
DOI 10.1007/s00163-015-0189-9

ORIGINAL PAPER

Generating flexibility in the design of engineering systems

to enable better sustainability and lifecycle performance
Junfei Hu Michel-Alexandre Cardin

Received: 16 December 2013 / Revised: 12 February 2015 / Accepted: 12 February 2015 / Published online: 5 March 2015
! Springer-Verlag London 2015

Abstract Designing an engineering system that is both application of the proposed methodology through the ana-
environmentally and economically sustainable is a chal- lysis of a waste-to-energy technology in Singapore based on
lenging task. Designers need to cope with socio-technical anaerobic digestion. Results show that the expected net
uncertainties and design systems to provide high perfor- present value of the flexible design concepts provides more
mance during long lifecycles. Flexibility in engineering than 10 % improvement over a fixed benchmark design in
design provides ways to address such challenges by making terms of economic lifecycle performance. This design is
engineering systems changeable in the face of uncertainty. It conducive of better economic sustainability via additional
is difficult, however, to identify suitable system elements for power generation and better use of resources. Results also
designing flexibility, especially when subjected to multiple indicate that the flexible design can reduce downside risks
sources of uncertainty and complex interdependency be- and capitalize on upside opportunities significantly.
tween socio-technical and systems elements. This paper
considers embedding flexibility into the engineering design Keywords Sustainability ! Uncertainty management !
as a mechanism to ensure better sustainability and to im- Flexibility in engineering design ! Waste-to-energy
prove economic performance in long-term lifecycles. The system ! Real options analysis ! Change propagation
main contribution is a novel methodology to identify valu-
able opportunities to embed flexibility as a way to deal pro- Abbreviations
actively with uncertainty in market and environment. The AD Anaerobic digestion
proposed methodology integrates Bayesian network into ADOS Anaerobic digestion of organic slurry
engineering system design to effectively model complex BN Bayesian network
change propagation in the flexibility identification process. CPA Change propagation analysis
It helps structure concept generation activities by identify- CPI Change propagation index
ing candidate areas to embed flexibility in the system. It CPM Change prediction method
compares favorably to other concept generation methods DCF Discounted cash flow
(e.g., prompting, brainstorming) that require modeling and DSM Design structure matrix

PHASE 3:
evaluation of a large number of concepts generated in order ENPV Expected net present value
to identify the ones offering better performance. It differs ESM Engineering system matrix
from other flexibility enabler identification methods by GBM Geometric Brownian motion
considering indirect as well as direct dependencies, in ad- IRF Integrated real options framework
dition to the probabilistic nature and risk resulting from Logical-MDM Logical multiple domain matrix
possible changes. Another contribution is the demonstration NPV Net present value
ROA Real option analysis
J. Hu ! M.-A. Cardin (&) RSI Risk susceptibility index

CONCEPT GENERATION
Department of Industrial and Systems Engineering, National
TPD Ton per day
University of Singapore, Block E1A, #06-25, 1 Engineering
Drive 2, Singapore 117576, Singapore VOF Value of flexibility
e-mail: macardin@nus.edu.sg WTE Waste-to-energy

123

J. Hu and M.-A. Cardin*, "Generating Flexibility in the Design of

Engineering Systems to Enable Better Sustainability and Lifecycle
Performance," Research in Engineering Design, vol. 26, pp.
121-143, 2015. doi: 10.1007/s00163-015-0189-9

2015 M.-A. Cardin. All rights reserved 8

How to Generate Flexibility in Systems Design?
Iridium satellite cell-phone communication system
90s award-winning technology, $4BN development cost
Optimized deployment to demand forecast 66 satellites
Successful land-based cell phones wrong demand forecast
over-capacity design!
Sold for a < 1% in bankruptcy

Proposed flexibility strategy (de Weck et al., 2004)

Phased deployment
Different design: reconfigurable constellation
Reduce downsides; capture upsides
20% expected lifecycle cost savings
Source: h*p://comlinks.com

How to make flexible design concept generation more

accessible, systematic?
2015 M.-A. Cardin. All rights reserved 9
analyze the system from multiple domains standpoint. In ibility from an ex
addition, multiple uncertainties that affect system perfor- baseline by conside
mance are modeled. Using the identified information, such subsequent phases.

Systematic Methodology
as system-level dependencies and the cost of changing
from one state of design to another, the potential flexibility
scratch for flexibil
large, with many p
become easily intr
lution and trying
determine the initia
lows designers to s
know and have exp
from the design of
performance follow
Once a baseline
techniques can be u
discounted cash fl
simulation and com
for example: It is c
cost and revenue c
DCF model is ana
point forecasts of
demands and requi
sent value (NPV) li
Fig. 1 Summary of proposed methodology to generate flexibility in design concept is c
2015 M.-A. Cardin. All rights reserved 10
engineering systems design better lifecycle pe
 to these child nodes downstream due to a change in system
element si upstream and finally avoid this risk in the initial
design phase. The risk generated by system element si can
be calculated as:
RGenerated
si
X
Risk Susceptibility Index
Psj jsi ;8uj 2U # Psj j8uj 2U Csj
sj 2Dsi
3
Here, sj represents a child node of system element si,
si = sj, Dsi is a set of system elements that contains all of
the child nodes of system element si , Psj jsi ;8uj 2U is the
posterior conditional probability of a change in system
element sj given a change in system element si ; and
Psjj 8uj 2U is the posterior conditional probability of a change
node B is a fixed design for V6 engine initially (e.g., BV6 ).
Given node A changes, node B has a high probability to
change to accommodate a V8 engine (e.g., BV6 changes to
BV8 ). If the cost of change for node B (i.e., cost for
changing BV6 to BV8 ) is also high, this means node B has
high RReceived . It motivates designers to consider an ad-
vanced design (e.g., flexibility or robustness) for node B in
the initial stage, with the goal of reducing cost of change in
the future. To further determine whether node B is suitable
for flexibility or robustness, risk susceptibility RGenerated
should be analyzed and node B is considered as an internal
source of uncertainty.

Reveived
in system element sj only subject to the external uncer-
tainties. The subtraction here represents the increased Robustness System elements

R
Flexibility
probability of child node (sj) changing due to the effect of High
changing the parent node si . A large difference between the
probabilities means that given the new information from
parent node si (e.g., the parent node si will be changed with Low Fixed Design
Robustness
100 % probability), the probability for changing the child
node sj has been significantly increased and the child node
sj becomes more likely to change. Variable RGenerated si Low High RGenerated
indicates the degree of risk generated by a change in sys-
RSI
temSi element si under uncertainty U.
Fig. 3 Risk susceptibility of system element

123
2015 M.-A. Cardin. All rights reserved 11
Application: Anaerobic Digestion
Waste-to-Energy (WTE) System

Source: www.keppelseghers.com
2015 M.-A. Cardin. All rights reserved 12
The impurity portion is then sent to incineration plants. The
separated wastes are mixed with circulate material from 4.1 Step 1: Initial design
digesters, which are then pumped into the digesters. During
the digestion process, bio-gas is produced and stored in Based on the procedure described in Sect. 3, a model is

Step 1: Initial Design

gasholders. The biogas can be continuously fed into gas
engines for generating electricity. The generated power is
pumped to the grid and sold in the wholesale electricity
developed in the first step to quantify the lifecycle per-
formance of a WTE system design. The design configura-
tion (i.e., combination of design variables) producing the

Major Tankage
Food Waste
Tipping Floor Electricity
Gas Holder Gas Engine Power Grid
Waste Collector

Gas
Bag Breaker Waste
Digester Flaring and Odor
Control
Fresh Circulate Digested
Waste Material Waste
Screener

Pre-processing Residue
Equipments Mixer Screw Press Compactor and
Containers
Impurity
Post-processing
Pumps Equipments Residues

Main Equipments
Incineration
Plants
Anaerobic Digestion general process flow
Fig. 5 General process flow for anaerobic digestion plants

2015 M.-A. Cardin. All rights reserved 13

123
 Author's personal co

Discounted Cash Flow Model Res Eng Design (2015) 26:121143

Table 2 List of assumptions for the economic model Table 3 Capit

national Ltd (2
Parameter Assumption and source
Main system e
Cdis S$ 77/tona (this is the disposal cost which should be paid
by the AD plant to dispose of residues. The
transportation cost for disposing of residues is not General site w
considered in this case) utility conne
Pto S$70/ton (Pto is assumed to be slightly lower than Cdis to Tip floor and e
encourage organic waste separation) Major tankage
Pe S$0.267/kwhb gasholders, s
Nt 30 workers for capacity of 400 tpdc (based on economies Pre-processing
of scale, 40 workers are assumed for a capacity of 600 equipment)
tpd, and 50 workers are assumed for capacity of 800 Main equipme
tpd) units, convey
W S$3000/monthe Post-processin
Mtc Approximately 2 % of capital costc,d presses)
Cope S$ 10/tonc (this cost includes the fuel and raw material Flaring and od
for digestion) Electrical gene
q 30 % of the total amount of disposed wastes at year td,g engine)
(i.e., Qtr = 30 % 9 QtTot) Residue comp
RGbio 100 m3/t (3531 ft3/t) for a general casec. It can achieve Total initial ca
150 m3/t (5297 ft3/t) by using advanced technologyf
Ven 522.1 BTU/ft3 c

RCe 35 %c (the generated heat will be channeled to the and analyzed

digester to maintain its operating temperature and the available inf
electricity will be sold out) prior conditi
Rnea 80 %c (20 % of the generated electricity is used to expert elicita
maintain its operations) moderate and
Ra 90 % for a general casec in Fig. 6 rep
a
National Environmental Agency (2013) ships: the hi
2015 M.-A. Cardin. All rights
b reserved
Singapore Power (2013) 14 between the
c
RIS International Ltd (2005) (T1) strongly
d by issuing
 Step 2: Dependency
Author's personal copy and
132
Uncertainty Analysis Res Eng Design (2015) 26:121143

System Stake-
Objective Functions Objects
EngineeringESM
System Matrix drivers holders
S1 S2 S3 T1 T2 T3 O1 O2 O3 F1 F2 F3 F4 B1 B2 B3 B4 B5 B6 B7 B8 B9
Total amount of organic waste (S1)
System drivers Organic waste recycling rate (S2)
Organic waste moisture content (S3)
Government's strategy (T1) 0.6 0.6 0.6 0.3
Stakeholders Public behavior (T2) 0.6
Waste-to-energy company (T3) 0.9
Eliminate land use (O1) 0.6 0.6 0.9 0.6
Objectives Maximize waste to energy generation rate (O2) 0.9 0.9
Achieve a clear and safety environment (O3) 0.9 0.3 0.9
Waste sorting mechanism (F1) 0.6 0.6
Waste volume reduction (F2) 0.6
Functions
Power generation(F3) 0.6
Emission control (F4) 0.6
General site works (B1)
Tipping floor and new buildings (B2) 0.6 0.6
Main equipment (B3) 0.6 0.6 0.9
Pre-processing equipment (B4) 0.6 0.6 0.6 0.9
Objects Post processing equipment (B5) 0.9 0.9
Major tankage (B6) 0.6 0.6 0.9
Power generation control (B7) 0.6 0.6
Residue compactor and containers (B8) 0.9
Flaring and odor control (B9) 0.6 0.6

Fig. 6 ESM representation with triggering probability for an AD plant

2015 M.-A. Cardin. All rights reserved 15

best performance is then selected and referred as the where Qte is the total amount of electricity (KWh) available
 Author's personal copy
Stochastic Process
Res Eng Design (2015) 26:121143

probability of each
Bayesian network (
about the cost show
can be quantitativel
the system compo
example, the poste
and the normalized
Eq. (2), the RReceiv
lated as 0.33. Since
not impact other sy
tionships as seen in
0 according to Eq.
process equipment
summarizes the ra
component. It sugg
floor and equipmen
Fig. 7 GBM simulation
2015of future
M.-A. generated
Cardin. organic wastes (5 out of
All rights reserved
tankage16
have the t
2000 runs) selected as valu
 Main Equipment (B3) Major Tankage (B6)

Step 3: Flexible Design Opportunity

P(B5|B3)=0.9 P(B5|B6)=0.9
Post Processing
Equipment (B5)

Identification (a) (b)

Fig. 8 a Dependent relationship between system components, b the conditional probability table for constructing the Bayesian network
Bayesian Network
AmontOfOrganicWaste OrganicWasteRecyclingRate WasteMoistureContent
LargerThanExpectation 100 ReachCertainLevel 100 ReachCertainLevel 100
WithinCurrentRange 0 WithinCurrentRange 0 WithinCurrentRange 0

PublicBehavior GovernmentStrategy OperationCompany

ChangeBehavior 56.2 ChangePolicy 93.6 ChangeStrategy 84.2
Stable 43.8 Stable 6.40 Stable 15.8

EliminateLandUse
HigherRequirement 99.2 MaxEnergyGenerationRate AchieveClearEnvironment EmissionControl
CurrentRequirement 0.79 HigerRequirement 91.8
HigherRequirement 92.1 HigherRequirement 55.3
CurrentReuirement 8.18 CurrentRequirement 7.86 CurrentRequirement 44.7

WasteSortingMechanism
WasteVolumeReduction PowerGenerationService
HigherRequirement 81.8
CurrentRequirement 18.2 HigherRequirement 59.5 HigherRequirement 55.1
CurrentRequirement 40.5 CurrentRequirement 44.9

TippingFloorNewBuildings PreprocessEquipment
MajorTankage
CapacityExpansion 84.0 AdvancedFacilityCapacityExp 98.2
CapacityExpansion 96.1
CurrentCapacity 16.0 CurrentState 1.77
CurrentCapacity 3.90

PowerGenerationControl
FlaringAndOdorControl
AdvancedFacilityCapacityExp 68.3
AdvancedFacility 68.6
MainEquipment CurrentState 31.7 CurrentFacility 31.4
AdvancedFacilityCapacityExp 96.5 PostProcessingEquipment
CurrentState 3.47 ResidueCompactor
CapacityExpansion 97.5
CurrentCapacity 2.48 CapacityExpansion 86.5
CurrentCapacity 13.5

Fig. 9 Bayesian network model of AD plant with uncertainties occur

2015 M.-A. Cardin. All rights reserved 17

123
Residue Compactor and Containers 0.24 The first sour
enabled in the tip
ponent. This sou
Risk Susceptibility Index sion for the tippi
and quickly deli
Table 5 RSI for each system component under uncertainty scenario cessing equipme
System components RReceived RGenerated RSI recycling rate inc
value exposure to losse
Tip floor and equipment building 0.84 0.03 0.81
initial phase (i.e.
capacity). In add
Main equipment (e.g., feed pumps, 0.76 0.01 0.75
mixing units, conveyors) to capitalize on t
Major tankage (e.g., digestion tank, 0.71 0.03 0.68 more feedstock b
gasholders, storage tank) needed. This stra
Electrical generation (e.g., generator, 0.57 0.00 0.57 the operation pr
engine) equipment requir
Post-processing equipment (e.g., 0.51 0.00 0.51 components (see
screw presses) changing the tipp
Pre-processing equipment (e.g., 0.33 0.00 0.33 very high. Addin
screen equipment)
change more ea
Residue compactor and containers 0.21 0.00 0.21
needed in the f
Flaring and odor control 0.07 0.00 0.07
structure is const
floors can be adde
2015 M.-A. Cardin. All rights reserved 18
be expanded. In a
 Flexible Systems Design
IEEE TRANSACTIONS ON ENGINEERING MANAGEMENT, VOL. 58, NO. 3, AUGUST 2011

Concept Generation

ms and types.

resources
onal funds
s view).
s multiple 2015 M.-A. Cardin. All rights reserved
Source: Mikaelian et al. (2011)
19
employees Fig. 7. Examples of mechanisms and types across the enterprise views.
 Real Options
Capacity expansion (both for tipping floor and
major tankage)
Start with 400 tpd
Expand by 200 tpd organic waste in previous year
> installed capacity, up to 800 tpd

Switching to technology for different water

contents
Switch to advanced technology moisture rate >
80%

2015 M.-A. Cardin. All rights reserved 20

 137

Step 4: Value Flexibility

dstock Table 6 Design of experiments for flexible design concepts
is the Flexible Expansion Expansion Switching
ocess, design option in option in option in
al Pte concepts tipping floor major tankage main equipment
ecially
Design concept 1 Y N N
ch can
Design concept 2 N Y N
y and
Design concept 3 N N Y
esters.
Design concept 4 Y Y N
m that
Design concept 5 Y N Y
ss and
Design concept 6 N Y Y
oisture
Design concept 7 Y Y Y
Benchmark design N N N
from
xpand 2015 M.-A. Cardin. All rights reserved 21
r than
 Design concept 5 Y N Y
the dry process and
Design concept 6 N Y Y
the waste moisture
Design concept 7 Y Y Y

Cumulative Distribution Functions

n options starts from
on rule is to expand
Benchmark design N N N

astes are larger than

year. This rule ex-
h time and until the
k design is reached
n with the switching
e as the benchmark
a dry process in the
option is exercised
n 80 %.
exibility gives rise to
zed in Table 6. The
o three factors. Each
ibility turned on and
The flexible design
ce of flexibility em-
design concepts 47
en flexible strategies
zed in Sect. 4.4.2. Fig. 10 Cumulative probability of AllNPV
2015 M.-A. Cardin. for flexible strategies and
rights reserved 22
benchmark designs
 Flexible design concept 6 165.03 115.90
Flexible design concept 7 167.17 117.41
Benchmark design 150.81 92.08

Risk vs. Return Tradeoff Best design Flexible design 7 Flexible d

portfolio theory in
versus return trade
comes is used as a
As illustrated in
be grouped into th
return and high-ris
3 and the benchm
are designed with
stage. Compared w
cept 3 has a slight
indicates that only
with the uncertain
provement, but th
cludes mid-range
(i.e., design conce
proved ENPV sinc
Fig. 11 ENPV versus 2015
standard deviation plot for the eight design
M.-A. Cardin. All rights reserved mitigated
23 by only e
concepts with design concep
er; a zero CPI indi- Pre-processing equipment (e.g., screen 1 0 -1 A
equipment)
a negative number
Residue compactor and containers 1 0 -1 A
The multipliers that

Sensitivity Analysis
Post- processing equipment 2 0 -2 A
ived are suggested (e.g., screw presses)

the CPA method for

stem component are
own in the objects
n between two ele-
ity in the matrix
1, since CPA does
nd only account for
in a direct manner.
n are displayed in
er and C is carrier).
g floor and major
e flexibility, which
Bayesian approach.
method and the pro-
method is calculated
proposed method is
n " RGenerated
si =out)],
s. The reason is that Fig. 12 Value of flexibility forCardin.
2015 M.-A. different
All rightsdiscount
reserved rate i 24
out for these two
 Sensitivity Analysis
Author's personal copy
Res Eng Design (2015) 26:121143

10 runs of 2000 scenarios). provides guidance to designers and decision-makers,

sitivity of VOF with respect to especially when resources are limited.
summarizes the key inputs/out- The 2015
second objective
M.-A. Cardin. is to demonstrate application of25
All rights reserved

VOF changes as each volatility the proposed tool in the analysis of a WTE infrastructure
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS: SYSTEMS
Training Design and Management of Flexible
Engineering Systems: An Empirical Study
Using Simulation Games
Michel-Alexandre Cardin, Jiang Yixin, Howard K.-H. Yue, and Fu Haidong
AbstractThis paper presents the results of an empirical study
of training procedures enabling flexibility in the design and man-
agement of large-scale engineering systems. The work relies on
the development and use of a simulation game environment
to study decision-making dynamics under different treatment
conditions. Evaluation of short-term, long-term, and in-game
training is completed to assess the main and interaction effects
on quantitative lifecycle performance indicators, and qualitative
user impressions. Forty-six graduate engineering students par-
ticipated in controlled experiments where they worked on the
design and management of a flexible emergency medical services
system. Results show that in-game training produces a statis-
tically significant improvement on lifecycle performance score,
while long-term flexibility training significantly reduces decision-
making time. In-game training improves process satisfaction.
Both short-term and in-game training lead to improved satis-
faction with the results, and contribute to improved anticipated
quality of the results. Correlation studies suggest that participants
taking more time for decision-making may improve lifecycle per-
formance scores. Lifecycle score improvement also increases as
satisfaction with the process and anticipated quality of results
increase. Experimental results show that different training pro-
cedures produce different effects on design and management
decision-making for flexible engineering systems operating under
uncertainty. They provide further insights to support the devel-
opment and evaluation of novel training approaches useful for
systems engineering practice and education.
Index TermsComputer simulation, decision making, large-
scale systems, risk analysis, systems analysis and design, systems
engineering education.
I. I NTRODUCTION
O N May 1997, the first five satellites of the Iridium
constellation were launched successfully in space. This
large-scale engineering system was meant to revolutionize
PHASE 5: PROCESS
wireless communications by offering satellite-based phone ser-
vices almost anywhere on the planet. By September 1998, the
66 satellite infrastructure was fully launched. Rapid deploy-
ment was needed to accommodate an anticipated customer
Manuscript received July 14, 2014; revised October 28, 2014;
accepted December 14, 2014. This work was supported by the National
University of Singapore Faculty Research Committee via MOE AcRF
Tier 1 under Grant WBS R-266-000-067-112, Grant R-266-000-061-133, and
Grant R-266-000-079-112. This paper was recommended by Associate Editor
W.-K. V. Chan.
MANAGEMENT
The authors are with the Department of Industrial and System
Engineering, National University of Singapore, Singapore 117576 (e-mail:
macardin@nus.edu.sg).
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TSMC.2015.2392072
2168-2216 c 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
M.-A. Cardin*, Y. Jiang, H. K. H. Yue, and H. Fu, "Training Design
and Management of Flexible Engineering Systems: An Empirical
Study Using Simulation Games," Accepted for publication in IEEE
Transactions on Systems, Man, and Cybernetics: Systems, 2015.
doi: 10.1109/TSMC.2015.2392072
2015 M.-A. Cardin. All rights reserved
1
base of three million subscribers. Unfortunately, demand grew
much slower than anticipated, and the company was soon
unable to honor debt payments on the US$4 billion devel-
opment costs. By the early 2000s, the company had to file for
bankruptcy [1].
de Weck et al. [2] showed later that flexibility in design
and management of Iridiumdefined as the ability to change
the system easily in the face of uncertainty [3]could have
saved up to 20% in expected lifecycle cost, perhaps even sav-
ing the technological venture from bankruptcy. The idea was
to design each satellite so it can be redeployed in orbit as
required coverage increases, and stage capacity deployment of
the constellation gradually as demand grows (i.e., start with
fewer satellites, and add more as demand grows while recon-
figuring the constellation in space to cover changing demand
areas). This contrasts with a strategy of optimizing design
and capacity deployment in view of deterministic (and per-
haps optimistic) demand forecasts, which may lead to a more
rigid design solution.
The Iridium case is an extreme illustration of a tension in
standard design and management practice for large-scale engi-
neering systems, explored for some time in [4][6]: is it best
to invest in design flexibility early to provide better adapt-
ability in the view of an uncertain future, or to design the
system optimally for a particular view of the future? The for-
mer approach may require additional costs upfront, which may
be lost if the flexibility is not used. The latter may reduce
upfront cost, but will expose the system to sub-optimal per-
formance if forecast conditions do not materialize, and may
require more costs to adapt.
The Iridium system is an example of engineering sys-
tems, defined broadly as socio-technical artifacts fulfilling
important functions for society for healthcare, power gener-
ation and supply, telecommunications, transportation, etc. [7].
Such large-scale systems typically operate for a long-time,
and will inevitably face a wide range of changing condi-
tions over their useful life in terms of market environment,
regulations, and technology. Yet, standard approaches to sys-
tems analysis and design often focus on optimizing design
and management under deterministic conditions. These may
not fully account for the impact of uncertainty on lifecycle
performance, and the potential value of flexibility. As seen in
the Iridium case, such approaches may give rise to engineer-
ing systems that are rigid and cannot cope well with changing
conditions.
26
 Motivation
Assuming flexibility exists, how to best manage in operations?

Emergency Medical Services (EMS) systems very flexible:

Station allocation and timing
Resource allocation/reallocation
Abandonment of unused capacity

Singapore collaborating agency relying on simple, deterministic

heuristics for design, planning and management

Can training help better manage flexible engineering systems?

What procedures are best?
What is the impact on quantitative lifecycle performance, and qualitative
user impressions?

2015 M.-A. Cardin. All rights reserved 27

 Experimental Methodology (Generic)
Step 1:
1. Design Problem Description Design
Problem
Description

Step 2:
2. Computer Model Computer
Model

Step 3:
3. Simulation Game Simulation
Game

Step 4:
4. Data Collection Data
Collection

Step 5:
5. Statistical Analysis Statistical
Analysis

2015 M.-A. Cardin. All rights reserved 28

 topics of uncertainty, flexibil
and management.2 Current tr
and assumes that participants
own educational background.

Preliminary Setup baseline/control in the statistic

TABLE II B. Step 2: Computer Model

23 factorial design S ETUP F OR 23 DOE A computer model is devise
quantitatively after a sequen
Cardin et al. [8] used a discou
with Monte Carlo simulations
mance of different flexible d
problem. Discrete event simul
modeling techniques can be u
Here the model is develo
quantitative lifecycle perform
(Appendix A provides full de
tions). The original incident lis
ments are based on documen
provider. The model is develo
Participants
training (L), and in-game training (G)) with two levels each in DES and urban OR [38]. U
(Table II). Participants are randomly assigned to each treat- operational, economic, and ser
46 NUSment,graduate
with seven students
participants in treatment 1, five participants tion systems, and logistically
7 (Treatment 1), 24,
in treatments 5 (Treatments 2-4), 6 in
and six participants (Treatments 5-8)Upon
treatments 58. The DES focuses on simulatin
57% > completing
25 yearstheir old,treatment,
85% have participants
Bachelor are notonly,
told what
48%to > 1 yearmedical
handle workincidents that
do or how to do it during the game. It is left to them to depending on the spatial confi
experience
decide,inupon
industry
receiving their assigned training, how to apply ity allocations. The DES eng
the concepts during the game. stochasticallyreferred here
The factor L consists of aCardin.
2015 M.-A. shortAll 1520 min lecture focusing
rights reserved tured by key parameters,
29 like
explicitly on flexibility using a narrated PowerPoint presen- rates, travel times subject to r
tation (L = +1), or current training (L = 1), relying on
 Step 1: Design Problem: Emergency Services
Simplified version of realistic EMS system in Singapore
Focused on medical (i.e. hospitals, station/fire posts, ambulances)
Model developed in collaboration with Singapore Civil Defence Force
Quantitative performance-based metric (response time, lost calls)
Described benchmark design (initial station/ambulance deployment)
Explained task to improve existing benchmark design

Source: therealsingapore.com

Source: livinginsingaporetoday.com

Source: Ong et al., 2009

2015 M.-A. Cardin. All rights reserved 30

Step 2: Computer Model:
Discrete Event Simulation

Activation Station to Scene Scene to Hospital Hospital to

Time scene time Time hospital time Time station time

Response time
2015 M.-A. Cardin. All rights reserved 31
Step 3: Simulation Game

2015 M.-A. Cardin. All rights reserved 32

Performance Measurements

2015 M.-A. Cardin. All rights reserved 33

 Step 4: Data Collection
Factorial experiments designed to measure pretest-posttest
response improvement between Sessions 1 and 2: y = y2 y1
n n n
y(x1, x 2 ,, x n ) = 0 + i x i + ij x i x j + Design
Problem
i=1 i=1 j=1 Description &
j>i GUI Training

Here y = S, can be others Session 1 +

(e.g. time, satisfaction with process, results, quality) Survey

Controls for inherent creativity levels and prior Session 2 +

Survey
knowledge of design procedure within-groups
Debrief
Improves between-groups vs. within-groups
variability comparison, internal validity of results
(Campbell and Stanley, 1966)

2015 M.-A. Cardin. All rights reserved 34

 Step 5: Effects on Lifecycle
Performance Score (S)
GLM response
S = 49.7 1.51M + 7.9L + 14.6G 2.0ML 5.5MG + 6.1LG 2.8MLG

Significant main effects for in-game training (G = +1)

G = 14.6, t = -2.0, p 0.05
Interpretation
In-game training had main effect on S vs. benchmark

2015 M.-A. Cardin. All rights reserved 35

 scenarios are used as the system performance converged to a steady state value with a negligible variation.
Figure 9Manuscript to be submitted to
demonstrates the ENPV and standard deviation of the simulation response using the enumeration

Other Ongoing Work

decision-making; and 4)method with optimum
sensitivity analysis. The proposed flexible
framework isdesign shown
then applied in anred
to analyze dots.
on-shore
real-world LNG production system as a case study. Figure 1 shows the process. The goal is to quantify the
potential value improvements not recognized by standard design and evaluation approaches while benefiting
from the efficient and effective design space exploration to find promising flexible design and management
strategies for the system. While this paper focuses on on-shore LNG systems, the proposed frame work could be
adapted to measuring the value of flexibility in many other engineering systems as well.

Phase 1: Problem modeling Phase 2: Screening procedure

1. Baseline
(1): Deterministic quantitative
performance model
Synthesis flexible design by
sampling and running
simulation model
Build/update the response
Design
(2): Quantitative performance No
model under uncertainty
Meta-model is
(3): Quantitative performance adequate?
model for flexibility

Go to the next phase 4. Design

5. Process 2. Uncertainty
Yes

Phase 4: Sensitivity analysis Phase 3: Multi-criteria decision-making Space

Management Recognition
Local and global
(a)
(1): Multi-objective exploration
Preference
Exploration
(b)
sensitivity analysis informatio
(2): Find best tradeoff flexible design
Figure 9: ENPV and standard deviation of simulation using exhaustive enumeration, 2000 scenarios

Figure 1: Proposed multi-criteria screening framework for flexibility in engineering design

3.1 Phase 1: Problem modelling 3. Concept
Problem modeling is the starting phase of the framework that includes the three following steps, as illustrated in
Figure 2.
Generation
Systems modeling inputs such as market study data, design parameters and variables

Define the problem scope; identify Identify key sources of Identify appropriate
! !!!
Stochastic and the robust optimization for real decision rule is
main cost and revenue drivers; uncertainty flexible strategies
!
define design variables and input
parameters !!! !!! !! !! ! ! , !! denote total profit. The problem of choosing an optimal
options analysis in energy, emergency
Generate different Embed corresponding
uncertainty scenarios decision rules
Step 1:
Develop a deterministic then:
quantitative performance model
Step 2:
Develop a quantitative
services,
(b)
and transportation systems
(a)
performance model under
uncertainty
! !
Computational framework for simulation- max
Figure 10: Kriging meta-model for ENPV and standard deviation 2000
Step 3:

! r !! ! , ! = !!
Develop a quantitative
performance model for !!! !! !!! , !!! (6)
!scenarios
based optimization in oil and gas and
flexibility analysis
!
!!! !!!
Figure 2: Problem modeling phase for flexibility and uncertainty analysis (Phase 1)

transportation systems September 06, 2014

s. t. !! !!! !! , !, !
Page 9 of 37
(7)
September 06, 2014 Page 25 of 37
! !
!! !!! = ! !! !!! ,!!!if!!!! = !!! , !, ! ! , ! (8)

!! ! (9)
2015 M.-A. Cardin. All rights reserved 36
where! is a subset of!!, the variables with superscript ! correspond to the variables in scenario !.
 Conclusions
Standard design and architecture practice may not account well for
uncertainty and flexibility in complex engineering systems

Explicit considerations of uncertainty and flexibility show to improve

lifecycle performance significantly

Enabling/using flexibility challenging; need R&D for systematic design and

training procedures

Need new quantitative analytical tools to assess lifecycle performance

impact on decision-making

Need empirical work to determine which procedures are most suitable for
real-world use

2015 M.-A. Cardin. All rights reserved 37

Acknowledgments and Contacts
Thanks to team members
Post-doc fellows: Drs. Mark De Lessio, Hu Junfei, Simon Ng, Chang Sun, Aakil Caunhye
Research Associates: Jiang Yixin, Howard Ka-Ho Yue
PhD students: Mehdi Ranjbar-Bourani, Yinghan Deng, Zhang Sizhe, Xie Qihui, Ashwani Kumar, Sixiang
Zhao

Thanks for financial and other support provided by

NUS Faculty Research Committee via MoE AcRF Tier 1 grant
Singapore-MIT Alliance for Research and Technology (SMART)
Singapore-ETH Centre (SEC)
National Research Foundation Campus for Research Excellence And Technological Enterprise (NRF-
CREATE)
Keppel Offshore and Marine Technology Centre (KOMTech)
Singapore Civil Defence Force (SCDF)

More details at https://www.ise.nus.edu.sg/research/strategic/index.html

PI contact: Dr. Michel-Alexandre CARDIN

Email: macardin@nus.edu.sg
Phone: +65 6516 5387
2015 M.-A. Cardin. All rights reserved 38