ARTICLE IN PRESS

Progress in Aerospace Sciences 40 (2004) 487–534

www.elsevier.com/locate/pacrosci

Review of aerospace engineering cost modelling:

The genetic causal approach
R. Curran, S. Raghunathan, M. Price
Centre of Excellence for Integrated Aircraft Technologies, School of Aeronautical Engineering, Queens University Belfast,
David Keir Building, Stranhillis Road, Belfast BT9 5AG, United Kingdom

Abstract

The primary intention of this paper is to review the current state of the art in engineering cost modelling as applied to
aerospace. This is a topic of current interest and in addressing the literature, the presented work also sets out some of
the recognised definitions of cost that relate to the engineering domain. The paper does not attempt to address the
higher-level financial sector but rather focuses on the costing issues directly relevant to the engineering process,
primarily those of design and manufacture. This is of more contemporary interest as there is now a shift towards the
analysis of the influence of cost, as defined in more engineering related terms; in an attempt to link into integrated
product and process development (IPPD) within a concurrent engineering environment. Consequently, the cost
definitions are reviewed in the context of the nature of cost as applicable to the engineering process stages: from bidding
through to design, to manufacture, to procurement and ultimately, to operation. The linkage and integration of design
and manufacture is addressed in some detail. This leads naturally to the concept of engineers influencing and controlling
cost within their own domain rather than trusting this to financers who have little control over the cause of cost. In
terms of influence, the engineer creates the potential for cost and in a concurrent environment this requires models that
integrate cost into the decision making process.
r 2004 Published by Elsevier Ltd.

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 489
1.1. Context . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 489

2. The nature of cost . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 491

2.1. Production. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 491
2.1.1. Customer requirement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 491
2.1.2. Manufacturing practice. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 492
2.1.3. Integrated design and manufacture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 493
2.2. Cost definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 495
2.2.1. Non-recurring and recurring costs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 495

Corresponding author. Tel.: +44 28 90 274190/335; fax: +44 28 90 382701.

E-mail address: t.curran@qub.ac.uk (R. Curran).

0376-0421/$ - see front matter r 2004 Published by Elsevier Ltd.

doi:10.1016/j.paerosci.2004.10.001
 ARTICLE IN PRESS
488 R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534

Nomenclature clframes frame labour coefficient

cm2024 material cost coefficient for 2024 T3 alumi-
E energy (kJ) nium ($/g)
ABC activity-based costing Dfan engine fan diameter
AC acquisition cost FC factor representing complexity
Al artificial intelligence FM factor representing miniaturization
BOM bill of material FP factor representing productivity
CAD computer-aided design fc factor representing cost incurred
CAM computer-aided modelling f Geom factor representing geometric complexity
CEM cost estimating model f Manuf factor representing manufacturing complex-
CER cost estimating relationships ity
CFD computational fluid dynamics fp factor representing performance level
COTS commercial off the shelf achieved
DFA design for assembly f Spec factor representing specification complexity
DFC design for cost ft factor representing time elapsed in reaching
DFM design for manufacture market
DFMA design for manufacture and assembly hf frame height
DFSS design for Six Sigma lf frame flange length
DOC direct operating cost mdata slope of the characteristic
DoD US department of defence nframes number of frames
DTC design to cost p probability of an event occurring
ESDU engineering and science data unit r costing ratios, according to f Geom ; f Manuf and
FB fuel burn f Spec
FEA finite element analysis R the regression coefficient R2 representing
ICT information and communication technology goodness of fit for a data population
IPPD integrated product process development Rr production rate at r production rate curve
KBS knowledge-based systems slope
LCC life cycle cost rlframes the frame labour cost per hour ($/h)
MCR material conversion route tf frame thickness
MDO multidisciplinary design optimisation U unit number
MFC manufacturing cost Vf volume of a ‘C’ shape frame
ROM rough order of magnitude zdata constant of the characteristic
SFC specific fuel consumption r material density
WBS work breakdown structure n factor relating to overhead or mark-up from
b learning curve slope manufacturing cost to unit cost
C cost
CP historical first unit cost Superscript
DC cost differential due to f Geom ; f Manuf and f Spec
DC 0 cost differential from the baseline character- l labour
istic m material

2.2.2. Fixed and variable costs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 495

2.2.3. Direct and indirect costs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 496
2.2.4. Life cycle cost . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 496
2.3. Cost allocation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 496

3. Controlling cost. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 498

3.1. Cost engineering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 498
3.2. Cost estimating . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 499
3.3. Cost-integrated design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 500
3.4. Supply chain cost control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 502
3.5. Knowledge-based systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 504

4. State-of-the-art: cost estimating. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 504

4.1. Classic estimating techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 504
 ARTICLE IN PRESS
R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534 489

4.1.1. Analogous . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 504

4.1.2. Parametric. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 507
4.1.3. Bottom-up. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 510
4.2. Advanced estimating techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 511
4.2.1. Feature-based modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 511
4.2.2. Fuzzy logic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 513
4.2.3. Neural networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 514
4.2.4. Uncertainty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 515
4.2.5. Data mining . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 517

5. State of the science: genetic causal cost theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 517

5.1. State-of-the-art. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 517
5.2. Causation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 519
5.3. Genetic nature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 520
5.4. Relevancy of genetic causal cost modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 520
5.5. Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 521
5.6. Genetic causal case study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 522
5.6.1. Measured costs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 523
5.6.2. Cost prediction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 523
5.6.3. Direct operating cost optimisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 525

6. Conclusions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 527

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 527

1. Introduction relative to the effect of design definition and the

manufacturing processes employed in its causation. A
The paper also reviews the more traditional and case study is presented in detail to illustrate the
advanced methods of cost estimating as the functional applicability of the approach. However, the intention
techniques that are currently available. The final section is not to present a definitive modelling technique but to
reviews the literature in terms of the modelling underwrite the value of having fundamental theoretical
methodology. Cost modelling is a particularly difficult principles to the modelling solution adopted. This is
field to assess in terms of scientific theory as it is not especially relevant to cost modelling where application
normally addressed as a scientific field but rather as an has dominated theory, and where there is a major
attribute of either design and manufacturing decisions influence from environmental factors. There is a lot of
or indeed a product; the latter being further confused current development in the area of systems engineering
with price (as cost plus profit). However, one of the main that is adopting a behavioural approach to integrated
aims of the paper is to consider the basis of the science in technical product development in an attempt to more
some detail in an attempt to establish a consolidating accurately model the performance within the wider
basis for costing methodologies. As there is little customer context, including cost.
literature that addresses the fundamental nature of cost
but rather focuses on establishing, at best, a rationale for
applied relationships and models, this concerns the 1.1. Context
genetic and causal requirements that are a fundamental
requirement of any scientific theory. The resultant The UK aerospace industry is one of the most
theoretical basis of the cost modelling is termed the successful manufacturing sectors with a turnover of
genetic causal approach and underpins the need for an around £20 billion and producing about 10% of UK
analytical foundation that is a platform for applied manufactured exports, with a consistent trade surplus
models that can be adjudged to be appropriate relative since 1980 [1]. The industry, both civil and military,
to both theoretical correctness and application. employs more than 150,000 people and is second only in
Ultimately, the cost modelling domain is reviewed (1) size to the US, with a world market share of 13%. Major
according to its genetic nature: relative to its general companies in the UK aerospace industry include BAE
applicability to engineering products and the concept of Systems and Airbus UK, Rolls Royce, TRW Lucas and
cost being inherited from certain design attributes and Smiths industries, while Shorts of Belfast are now part
manufacturing processes; and (2) its causal nature: of the Bombardier Aerospace group.
 ARTICLE IN PRESS
490 R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534

Already in the late 1980s, the customer was increas- assurance. As production is the result of engineering
ingly being considered more explicitly in the commercial effort, it can be defined by the activities of design,
aircraft design process through their demand for process planning and production planning while the
reduced operating cost and lead-time, whereas technol- associated decision making process is typically driven by
ogy had been the dominant driver in the past [2]. This is technical definition and constraints; although cost is
in the context of the continuing rise in labour rates and being increasingly recognised as an important design
the higher non-recurring costs associated with reduced criterion within the definition process. However, cost is
labour processes. The price of a Boeing 737 is now not known in advance of production and therefore a
approximately 6 times that of 3 decades ago, a rise of cost estimation system is required. Ten Brink [11] has
6.5% per year. Naturally, there have been advances in pointed out that this will rely on the available product
the design and operational capabilities, with both the information at whatever stage of the product develop-
Airbus 380 and Boeing 7E7 being reported to have the ment cycle and relevant information maturity. There is
lowest direct operating costs (DOC) in the large carrier also the possibility of using such a design-oriented
class. With reference to the oil crisis of the mid-1970s, capability to implement product changes that reduce
fuel prices have also fluctuated and air travel is now very cost. For example, concurrent engineering can be used in
cost sensitive [3]. Typical aircraft DOC breakdowns the simultaneous integration of engineering tasks during
show that the aircraft cost contribution to DOC is two the product development cycle but requires the inte-
to four times higher than the contribution made by fuel grated support of a cost estimation capability.
cost [4]. That is reflected in the message from the airlines The cost of manufacturing to produce an output is a
that the paradigm of ‘Better, Quicker to market, and function of resource utilization; including physical
Cheaper’ is replacing the old mantra of ‘Higher, Faster, entities such as: manpower, equipment, facilities, supply,
Farther’ [5]. Aircraft producers now realise that this etc. [12]. The costs are then representative of the
demand to reduce cost and lead-time needs to be tackled resources consumed, such as: machine tools and fixtures,
at the conceptual engineering design phase. Typically, operators and materials, etc. Therefore, it is the
Burt and Doyle [6] report that 70–80% of the total engineering effort that gives rise to cost as decisions
avoidable cost is controllable at the design stage and are made. It is often reported but perhaps not well
indeed many authors agree that conceptual design heeded that conceptual engineering decisions signifi-
wields the greatest cost influence and is often irreversible cantly influence the costs caused by engineering deci-
[7]. Consequently, this results in: (1) a more critical sions later in the engineering cycle, within a reduced
assessment of technology suitability and maturity; (2) a design space. However, although design itself is typically
reassessment of the processes and the establishment of quoted at contributing less than 10% of the product
best practises; and (3) a more rigorous approach to the costs while fixing around 70%, this may be misleading as
issue of cost. product specification has been noted to already commit
The importance of engineering costing within aircraft a significant level of cost. Wierda [13] has noted that
design [8] should have a more directly influential role, design may be responsible for 20–30% of total product
for example as part of an integrated process that is cost, relative to the production environment. This
embedded within multidisciplinary systems modelling unfortunately leads to the cost estimating paradox of
architecture. Differential product evaluation with re- the design process: that product information is not yet
gards to cost, technology, reliability and maintainability, available in detail and consequently, there are varying
along with risk analysis, are all important considerations needs and difficulties in making accurate estimates
in the current aerospace industry. Cost modelling also throughout the duration of the process [14]; leading to
assists in preliminary planning for procurement and the further paradox of confidence being higher after
partnership sourcing. Ultimately, the goal is that aircraft design completion and therefore leading to reengineer-
acquisition is driven by the balanced trade-off between ing and modification.
cost and performance [9]: leading to affordability and The aim of Concurrent Engineering and integrated
sustainability for operators over the product life cycle. product process development (IPPD) is to impose the
The challenge for the industry is to look into all of the simultaneous sharing of task information that originates
aspects of ownership cost and to link these into the design within the individual engineering functions, in order to
decision making process at the conceptual stage on. facilitate and control cooperative decision-making [15].
The recognised need for cost evaluation at the design This is best facilitated with a modular system that has
stage is also intrinsically linked to aircraft production. generic elements, for the enhanced integration of multi-
This is why the principle of Design for Manufacture is so disciplinary analysis and diverse models, along with the
important, addressed in detail in Section 2.2. Chisholm flexibility of extension and system maintenance [16].
[10] has pointed out that manufacture is a series of This is the hard end of concurrent cost engineering and
interrelated activities and operations that involve design, addresses the sharing of information in a holistic
materials selection, planning, production and quality manner through integrating data systems, rather than
 ARTICLE IN PRESS
R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534 491

the fragmented and dated approach of estimating  the availability of the product relative to when
isolated costs with diverse models at disparate levels of needed, and
the product breakdown structure. However, the analysis  the cost of ownership to the customer.
architecture needs to be unified and the linkage between
engineering functions needs to be established, in order to
The first two aspects are associated with the perceived
enable communication and data/knowledge sharing.
performance; the third relates to the timing of the
This is highly dependant on both the availability and
product availability in response to demand; and the
accessibility of coherent information [17] and conse-
fourth relates to cost and robustness. Each aspect can be
quently, engineering databases and systems architectures
related to the ‘‘Better, Faster, Cheaper’’ paradigm,
do play a key role in the continued automation of the
where Murman [19] has proposed that some measure
product development cycle. Billo [18] has pointed out
of Value can be defined with the following functional
that such engineering databases may be populated with
relationship:
geometric, physical, technological or other influential
engineering properties, for example, in an object
oriented framework that also relates to the hierarchical fp
Value ¼ ,
nature of the objects and their attributes being modelled. fc ft

where f p represents performance level achieved, f c

represents cost incurred, and f t represents time elapsed
2. The nature of cost
in reaching market. This formulation highlights the need
for aircraft manufacturing companies to be meeting the
2.1. Production ‘‘Better, Faster, Cheaper’’ requirement by re-assessing
their practises, and improving the existing methods and
2.1.1. Customer requirement work processes with quantitative tools and integration
In aerospace engineering there has always been a wide methodologies. This is a key step towards the new
variety of manufacturing alternatives, whether pro- environment within aviation: the concurrent integrated
cesses, methodologies, or technologies. There are even design of aircraft for the production of a highly
more materials now available. Data management synthesised product that fulfils customer requirement,
systems are continually evolving, and computational whether in terms of performance, cost or availability.
modelling of behaviour is being pursued on all fronts, However, in the shorter term, one can already look
although especially in computational fluid dynamics towards more competitive aircraft for producers, ulti-
(CFD) and finite element modelling (FEM) for aero- mately maximising their profit. Affordability can be
space applications. However, there is still a basic need formalised as relating to a product with a selling price
for tools that help and support engineers in making that has proportional functional worth and which is
reasonable and measured design decisions that are cost- priced within the customer’s range. In addressing
effective and ultimately, more competitive [23]. As affordability, cost can be readily employed as an
mentioned, aircraft engineering is adapting concurrent evaluation criterion at the conceptual design stage in
engineering principles but it is not yet integrated in two ways: (1) design for cost (DFC) and (2) design to
nature as the inter-linkage between key variables and cost (DTC). DFC can be viewed as a feed-forward
parameters has not yet been built into a structured engineering process that makes conscious use of
modelling environment. In addition, there is now a engineering process information during design and is
heightened strategic need between industry and acade- directly aligned with concurrent engineering [24]. Alter-
mia for mutually beneficial research, now much more natively, DTC is a more management driven process
formalised than in times when industry invested more that aims to provide a design that satisfies specification
heavily in its’ own research and development. The requirements for a given cost target [25]. In both cases,
relevance is that cost is now viewed as a metric that can however, cost is used to link design and manufacturing.
facilitate an integrated approach, as engineering variables Consequently, affordability becomes a major design
and parameters can be explicitly linked to cost, whilst also driver that can be measured with cost as the dependent
providing guidelines that directly relate to the quantifiable variable. Ultimately, a more general Cost Integrated
measure of value and competitive advantage. Design approach is advocated, which is less specific
Slack [20] has proposed that value is a measure of than DFC/DTC but encompasses the industrial need
worth for a specific product or service by the customer, for various levels of cost evaluation, for various
and is a function of the following aspects: purposes. This should allow customer defined cost
targets to flow down into the design process and to be
 the product’s usefulness in satisfying customer need, addressed with other engineering requirements and
 the relative importance of the need being satisfied, specifications.
 ARTICLE IN PRESS
492 R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534

2.1.2. Manufacturing practice process plan, determine cycle time and potential bottle-
Concurrent engineering is an important framework necks, and to estimate the product and capital costs.
within which advanced engineering tools and techniques However, specific effort or facilitating software could
can be deployed; with a focus on improving product help capture the knowledge of the manufacturing
definition and development by concentrating life cycle engineer and facilitate the setting of accurate and
issues on the early design process [26]. Such tools should consistent time standards through automated graphical
strengthen the multidisciplinary approach at all phases user interfaces [31]. The need for ease of assembly plays a
of the design process, thus ensuring that the technical dominant role in aircraft production due to the high part
expertise of the participants can be optimally used or at count of an aircraft. Assembly is even more important in
least, successfully utilised to improve the design solu- today’s climate as so much of part manufacture is
tion. Management strategies such as Six Sigma Meth- increasingly being subcontracted to smaller more compe-
odology (a probabilistic approach to process capability titive suppliers. The four main goals of design for
and improvements), Agile Manufacturing (with a focus assembly (DFA) are, as defined by Andreasen [32]:
on flexibility and response), Lean Manufacture (a value
mapping and efficiency approach), and effective human  assembly efficiency,
resource management also need to be taken into  product quality,
consideration if improvements are to be met in the  assembly system profitability, and
areas of manufacture and assembly system profitability.  improved working environment within the assembly
In the Design for Six Sigma [21,27] context, the system.
product design team works with other cross-functional
members from marketing, sales, quality, manufacturing, With reference to the functional relationship of value
procurement and customers. Design for Six Sigma previously described, the first three aspects can be seen
espouses an integrated approach to design, so that the to impact on cost and time, while the fourth influences
product is manufacturable at the highest quality and performance in meeting the challenges of improvements
lowest cost, and satisfies all of the customer require- in the overall product value.
ments. Six Sigma methodology helps identify wastage by Several DFA methodologies exist which concentrate
taking a routine approach to issues that are causing the designers interest on ease of assembly during the
problems. Typically, one key issue addressed within design concept stage, including: the design for manu-
aerospace is the statistical reduction of opportunities for facture and assembly (DFMA) procedure suggested by
defects, scrap and rework. Boothroyd and Dewhurst [33], the Lucas DFA techni-
The concept of Agile Manufacturing [28] is driven by que [34] and the Hitachi Assemblability Evaluation
the need to quickly respond to changing customer and Method (AEM) [35]. The Boothroyd–Dewhurst DFMA
market requirements. Agile manufacturing requires that methodology suggests that the best way to achieve
a manufacturing system is able to efficiently produce a assembly cost reduction is to first reduce the number of
large variety of products and that it can be reconfigured components; standardise where possible; and then
to accommodate changes in both product mix and ensure that the remaining components are easy to
product design. This requires a simple manufacturing assemble. The Lucas DFA technique arose from the
system that is flexible while design for agile assembly is concept of a knowledge-based approach used in
accomplished by considering operational issues of conjunction with a CAD system. This technique uses
assembly systems at the early product design stage. the Boothroyd–Dewhurst principles of reducing compo-
Lean manufacture [29] focuses heavily on the concept nent numbers and analysing the assembly processes. An
of ensuring that value is always added to products and important feature is an emphasis on establishing the
that wasteful practice and processes can therefore be requirements of all customers in the supply chain and
identified and eradicated. The approach has been not limiting the assessment to the immediate business
developed through the Lean Aerospace Initiative (LAI) customer. Hitachi AEM facilitates design improvement
[30] that was born out of the need for affordability as at the concept stage by identifying weak points in the
defence procurement budgets were reduced in the US due design using two key indices:
to increasing costs and military industrial overcapacity
[20]. There is also a UK Lean Aerospace Initiative (UK-
LAI) and a Lean Aircraft Research Program (LARP)
 an assemblability evaluation score that is used to
assess design quality and the difficulty of assembly
based at Linköping University in Sweden.
operations and
Aircraft manufacturing companies are now beginning
to consider commercial software that facilitates assem-
 an assemblability cost ratio that is used to generate
the projected assembly cost.
bly process simulation for the planning and verification
of their operations. Such software can aid the manu- It will be shown in the following section that the
facturing engineer to validate the feasibility of the application of the Boothroyd–Dewhurst methodology
 ARTICLE IN PRESS
R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534 493

can also result in reductions in non-recurring cost. in-process measurement and aids in-service repair
However, tooling engineers are also making a direct operations. It may also be possible to design an
contribution to reducing non-recurring costs through aerospace structure that has sufficient inherent stiffness
approaches such as the jig-less assembly approach to for the assembly tooling to be reduced to simple,
manufacturing. For example, a case study has been reusable and re-configurable (from program to pro-
presented which looks at the redesign of the Airbus gram) supporting structure.
A320 fixed-leading-edge conducted by BAE Systems
[36]. Jig-less assembly aims to reduce cost and to
increase the flexibility of tooling systems for aircraft 2.1.3. Integrated design and manufacture
manufacture through the minimisation of product- There is a substantial amount of general information
specific jigs, fixtures and tooling. During the develop- and case studies available in the realms of DFM and
ment phase, tooling costs are quoted at over a third of DFA, or the more general Design For ‘X’ [38]. However,
the overall cost in the civil sector and nearly a quarter the relative importance and roles of DFM and DFA are
for the military. Consequently, savings in this aspect of not well distinguished, nor is it clear how organised and
aircraft manufacture are significant and they also impact systematic the general approach needs to be to reach its
on the lead time from concept to market. Jig-less full potential, or ultimately, what the quantifiable
assembly does not mean tool-less assembly, rather, the benefits are likely to be relative to the change in design
eradication or at least reduction of jigs. Simple fixtures metrics. In saying this, there is not a conflict of interest
may still be needed to hold the parts during particular between DFM and DFA, as essentially, both work in
operations but other methods are being found to complement to deliver simplified designs, as part of a
correctly locate parts relative to one another, the most concurrent DFMA approach. However, the distinction
advanced systems using lasers for datums. Assembly is correct in terms of either part manufacture or
techniques can be simplified by using precision posi- assembly respectively, and helps simplify the identifica-
tioned holes in panels and other parts of the structure to tion of associated cost drivers and the formulation of
‘‘self-locate’’ the panels. This process, known as rules and guiding metrics.
determinant assembly, uses part-to-part indexing, rather A case study [22] has been presented which illustrates
than the conventional part-to-tool systems used in the the use of the machining process to reduce the number
past. of operations in an assembly, where the baseline design
Within aerospace industry, it is generally recorded involved sheet metal fabrication with fasteners. The
that approximately 10% of the overall manufacturing assembly in question is a Pressure Box that functions as
cost of each airframe can be attributed to the one of two cavities located between the floor beams in
manufacture and maintenance of assembly jigs and the pressurised mid-fuselage section of a regional jet
fixtures. A traditional ‘‘hard tooling’’ philosophy aircraft, where the wing passes through the belly of the
dictates that the desired quality of the finished structure fuselage. The boxes seal the floor for pressurisation at a
is built into the tooling. The tooling must therefore be location where there are two indents that allow flight
regularly calibrated to ensure build-quality through control components to extend beyond the floor-line. The
tolerancing. The alternative philosophy of ‘‘Flyaway baseline and redesign are shown in Fig. 1 while the
Tooling’’ has been conceived with the purpose of process improvement results are presented in Table 1.
reducing tooling costs and improving build quality With regards to the tooling cost, it should be noted that
[37]. This approach envisages that future airframe this non-recurring element was not already spent on the
components will be designed with integral location contract under consideration but relates to the fabri-
features and that they will incorporate positional cated design solution being used on the older aircraft
datum’s that transfer into the assembly. This enables variant. Therefore, the reduced amount refers to the

Fig. 1. Flight control pressure box: baseline design and redesign, respectively.
 ARTICLE IN PRESS
494 R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534

Table 1
DFA results for pressure box

Before After Reduced %

Number of parts 29 1 28 96
Number of fasteners 346 124 222 64
Assembly man-hours 20 3.3 16.7 83
Recurring mfg cost (£) 770 459 311 40
Tooling cost (£) 3863 2847 1016 26

Fig. 3. Initial design of ‘‘I’’ beams (upper) and simpler DFA

redesign (lower).

Table 2
DFA results for fire bulkhead

Before After Delta %

Fig. 2. Tailcone forward firewall.
Raw material (kg) 143 96 47 32
Machine time (h) 138 90 48 34
relative savings to be made upfront at the conceptual Weight (kg) 10.6 9.9 0.7 6
design stage. Recurring mfg cost (£) 17,827 8413 9414 52
A quite different example of DFM implementation is
on the firewall bulkhead of a tail-cone on a Lear 45
business jet as shown in Fig. 2. The purpose of the quality and timeliness while inflating cost. Therefore,
Firewall is to resist the excessive heat that may emanate cost becomes an integral part of the design process and
from a malfunctioning of the auxiliary power unit an explicit design driver [41,42].
(APU) and in extreme conditions, to withstand a fire A methodology for integrating competitive manufac-
situation. Typically, for such thermal applications, a turability is also highlighted in Fig. 4, with specific
stiffened Titanium structure is used consisting of sheet reference to the three underlying design principles that
metal sub-assemblies and a number of ‘‘I’’ shaped underwrite the genetic causal approach later advocated
beams, as shown in Fig. 3. A cross-functional engineer- in the paper; namely, DFM, DFA, and DFC. An
ing team was established which promptly identified the example of this is to first simplify the assembly concept
five ‘‘I’’ beams in the baseline as a major manufacturing through DFA application; then to match the materials
cost driver, and focused on these as part of the DFM and process through the combined use of DFA and
process. The redesign then focused on the minimisation DFM; and finally, to simplify the part design through
of material usage and reduced machining time, achieving DFM. Cost-integrated design is applied at all stages in
a cost reduction of around 50% and a weight reduction order to provide quantitative information regarding the
of 6%, as shown in Table 2. cost impact of the decisions being made [43]. Supporting
Commercially, an aircraft’s specifications list is methodologies such as statistical process control (SPC)
drafted when considering the market niche and asso- can also be exploited concurrently with cost-integrated
ciated requirements. From that point, the design concept design so that as a consequence, the competitive
is established and at this juncture the manufacturability manufacturability of the product can be maximised in
of the aircraft already needs to be integrated into the measures of cost, quality and time. This can all be
early design process in a concurrent engineering context. carried out in the wider context of Six Sigma for
Fig. 4 refers to an aircraft as composed of a number of example, following the principle of: Definition; Mea-
interrelated multidisciplinary systems [39,40]. The key surement; Analysis; Improvement and Control [27]. The
design parameters that characterise these systems must Six Sigma methodology is not restrictive and advocates
be optimally, or at least satisfactorily, integrated to the use of whatever supporting tools that can be used to
reduce the negative manufacturing implications that improve processes and products, such as DFA, tolerance
compromise competitive advantage and value: reducing design, robust design, pareto analysis, etc.
 ARTICLE IN PRESS
R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534 495

Integrated Aircraft
New
Advanced
Large Technologies
Number
Of
Interdependent
Systems
Integration of
disciplines
within each
system

Output =
performance Integration
cost of Systems

Fig. 4. Aircraft systems design integration.

Ultimately, it is necessary to couple all of the key decision-making processes whereas all other costs are
design parameters at the early definition stage so that the then termed irrelevant [48].
aircraft can be integrated as a whole system [44].
Methodologies need to be formulated into models that
2.2.1. Non-recurring and recurring costs
can provide the linkage between performance models
A non-recurring cost refers typically to a capital
and production realities. In this context, DFM, DFA
expenditure that is incurred prior to the first unit of
and cost-integrated design are not only approaches that
production and is an element of the development and
support the principle of designing with a view towards
investment costs that generally occurs only once in the
the implications on manufacture, assembly and cost,
life cycle of a work activity or work output [12]. It may
respectively. Rather, these should be embodied into
be broadly defined as a one-time cost per programme or
models that drive the process by providing a quantita-
narrowly as per contract. Typically, this would include:
tive predicted outcome that can then be analytically
initial engineering effort in design; jigs and tooling
linked into an integrated engineering system that may
acquisition and/or upgrade; system testing and certifica-
include more traditional engineering models such as
tion; pre-production manufacturing costs such as plan-
computational fluid dynamics (CFD) and finite element
ning, etc. On the other hand, capital expenditures
analysis (FEA).
allocated to prepaid materials, supplies and parts used
to produce a unit of output are designated as recurring
2.2. Cost definitions costs. Recurring costs are ongoing costs that are
proportionally incurred from the production of the first
unit of output then on but are also required in order to
This section includes a brief explanation of the
maintain and update the manufacturing set-up as a
various cost categories recognised as being incurred by
whole. These costs occur throughout a programme’s life
an aircraft producer. The following categorisations are
and arise due to the repetitive nature of: commercial
well documented in the literature [45,46] and are
procurement costs; production overhead costs; materials
included primarily for clarity and fullness. A product’s
procurement costs; technical upgrade costs; labour and
costs can be arranged into a cost breakdown structure,
personnel costs; consumables; utility costs, etc. These
such as presented by Fabrycky and Blanchard [47] or
are similar to variable costs, explained in the following
Liebers [48]. This cost breakdown structure is driven by
section, as they vary as a function of quantity acquired.
the design of the particular product and must include all
It is important to note that both non-recurring and
costs only once. Some useful classifications that facilitate
recurring costs are important when modelling learning
this process are: (1) non-recurring or recurring; (2) direct
and improvement curves, especially as the recurring
or indirect costs, and (3) variable/fixed costs. Another
estimates should decrease over the production run.
distinction sometimes made is to relevant and irrelevant
costs [52] where relevant costs are treated as those that
are in one of several design alternatives but absent in 2.2.2. Fixed and variable costs
other alternatives, and therefore can be treated as Recurring and non-recurring costs can be incorrectly
differential costs. These costs play a specific role in the confused with variable and fixed costs respectively. The
 ARTICLE IN PRESS
496 R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534

terms variable cost and fixed cost are often associated dated by labelling such costs as overheads or a burden
with higher-level financial studies and with break-even that is summed and then spread over the enterprise
analysis over investment decisions. Typical examples as a whole, typically being added as a portion of
would be the cost of telecommunication, executive the direct labour cost. This may typically include the
board salaries, leasing, etc. Consequently, when a cost of electrical power, cleaning, building works,
company is being assessed financially the fixed costs pilfering, etc.
are often investigated in order to see whether the
company’s profits are superior, that being a sign of
2.2.4. Life cycle cost
general economic health. Schiller [49] defines fixed costs
It is worthwhile to introduce the concept of life cycle
as the ‘‘costs of production that do not change when the
cost (LCC). This is a customer driven cost assessment
rate of output is altered’’. Therefore, the association
that is concerned with the overall LCC of the product,
with non-recurring is clear. However, recurring over-
facility, system, service, or other. This is of interest when
heads could be fixed while non-recurring costs could be
making acquisition decisions but aircraft producers are
programme or contract specific. In general, fixed costs
also using it increasingly to assess the competitiveness of
remain unchanged on the global level and are indepen-
their product’s design. For instance, an LCC analysis
dent of the enterprise performance. They are therefore
might be useful when the estimate is to be used in a
treated as a general production cost incurred in keeping
performance trade-off study of a process or activity
the company operational. Conversely, variable costs are
within a company or enterprise. LCC is typically
costs of production that change when the rate of output
associated with the estimation of total acquisition cost,
is altered. Typical examples include many recurring
from ‘womb to tomb’ or ‘cradle to grave’. LCC
elements such as labour and material costs, machining
components can be defined in many different ways
expenditure, etc. Stewart [46] has defined variable costs
but, nevertheless, all classifications tend to start with
as those which change with the rate of production or the
either product development or acquisition, and continue
performance of services, whereas fixed costs are those
through to product disposal or retirement. Asiedu and
that do not vary with the volume of business. Company
Gu [47] divide the total product cost or life cycle cost
financiers like to have a good understanding of the
into four distinctive phases: (1) research and develop-
general variable cost expenditure so that they can put a
ment costs; (2) production and construction costs; (3)
case forward for reducing it as a way of increasing
operations and maintenance costs; and (4) retirement
profit. However, this can be counter-productive as
and disposal costs (as illustrated in Fig. 5).
variable costs must be incurred in the production of
Notwithstanding the LCC breakdown shown in Fig.
good quality products that satisfy customer expecta-
5, commercial airlines tend to focus in on several aspects
tions; quality as well as quantity. In addition, semi-
of this and in particular DOC. This is addressed later in
variable costs can be considered as varying in relation to
the paper and will be presented through Fig. 30, which
volume although the percentage change is not equal to
shows a DOC breakdown for a regional transport jet
that of the volume change [52]. Finally, stepped fixed
and incorporates the key cost elements incurred by the
costs can be considered to be fixed costs that alter as the
company. In particular, there is the cost of: ownership,
activity level moves from one level to another [52].
which is a function of price and borrowing; and of
operation, which is a function of fuel burn and the cost
of aviation fuel and maintenance, the latter being a
2.2.3. Direct and indirect costs
function of quality, complexity and spares. One might
A direct cost is an expenditure that can be broken
consider the fact that the cost of maintenance for the
down and allocated to specific items or causes. Conse-
airline industry is some $36billion in comparison with
quently, they are more easily identified and associated
the industry’s fuel cost at only $8 billion, while the cost
with an end result such as a product, service, pro-
of food on flights is $12 billion.
gramme, function, or project. These costs are typically
charged directly to a given contract in the way that
procured items can be easily associated with the bill of 2.3. Cost allocation
material (BOM) for a particular aircraft unit. On the
other hand indirect costs cannot be identified specifically Cost allocation refers to the interpretation of cost and
and consistently with an end objective [52]. Conse- its categorisation in order to arrive at a reasonable
quently, indirect costs are the opposite of direct, and distribution of those costs [12]. As mentioned in Section
where direct costs can be allocated directly as the 2.2.3, direct costs can be readily allocated according to
allocation base is known, the allocation base for indirect their nature whereas indirect costs need to have their
cost has to be defined [51]. These costs may be difficult allocation base pre-defined. The definition can be based
either to identify in the first instance or to be associated on historic information or from prognoses or a
with a given operation or outcome. This is accommo- combination of both. The traditional approach is to
 ARTICLE IN PRESS
R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534 497

Total Product Costs

Research and Production and Operations and Retirement and

Development Construction maintenance Disposal Costs
Cost Costs Costs

Product Manufacturing Operation Disposal of Non-

Management Management Management repairable

Product Industrial Engineering Product Product

Planning and operations Operation Retirement
Analysis

Product Product Documentation

Research Manufacturing Distribution

Design Design
Documentation Construction Maintenance

Product Inventory
Software Quality Control

Product Test and Operator and

Evaluation Initial Logistic Maintenance
Support Training

Technical Data

Product
Modification

Fig. 5. Cost breakdown structure.

allocate the overhead using volume-based allocation A more detailed method that meets this requirement is
bases such as labour hours and machine hours. activity-based costing (ABC), which assumes [51,52] that
However, this can lead to incorrect conclusions if the costs are caused by activities and that products consume
allocation base is chosen incorrectly. This is evident those activities. The implementation procedure is as follows:
when indirect costs are calculated with the direct cost
burden rate, which incorrectly implies that every
product with high direct costs also has high indirect  Determine the activity centres that relate to certain
costs. cost aspects of the product development cycle, as
It has been noted that the actual ratio between direct monitored individually by management. These ad-
and indirect costs has significantly changed due to the ministrative units are basic units of control in cost
increased use of automation [50]. Half a century ago, the accounting with managerial responsibility.
indirect cost was a small fraction of the total product  Determine the activity pools that relate to sets of
cost in comparison to direct labour. Consequently, it activities which are carried out by the functions.
was not important to have extremely accurate estimates  Determine the allocation base per activity pool as the
of the indirect costs and the traditional estimating cost driver that is a measure directly related to the
method was appropriate for overheads. That paradigm amount of an activity used.
has changed significantly and now overheads constitute  Determine the overhead costs per activity pool, which
the major share of total product cost, with direct labour are typically based on the adjusted overhead costs
costs being only a small component and material costs from the previous year.
remaining relatively unchanged. Therefore, there is now  Calculate the overhead costs per cost driver (rate),
a need to accurately calculate overheads by some other which are divided by the budgeted quantity for the
allocation base that is more realistic. allocation base.
 ARTICLE IN PRESS
498 R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534

3. Controlling cost can be distinguished in the architecture: (1) engineering

and planning; (2) order acceptance; (3) production; and
3.1. Cost engineering (4) accounting. The engineering and planning loop
provides decision makers with both qualitative and
Humphreys [53] has defined cost engineering as ‘‘the quantitative cost information for the various design
application of scientific and engineering principles and alternatives. The order acceptance loop provides cost
techniques to problems of cost estimation, cost control, information to the decision maker about the total cost
business planning and management science’’. In the consequences and needs to be of a quantitative nature.
wider sense this includes aspects of profitability analysis, In the production loop, information from the actual
project management, and the planning scheduling of production of a product is fed back in order to compare
major engineering projects, in general, ensuring that the cost estimate with the actuals, again facilitating
technically feasible engineering projects are economic- improvement of the cost models. Finally the accounting
ally attractive [54]. Firstly, the function of cost control loop feeds back information from production over a
includes the detection of cost values and the causes of given period of time in terms of the comparison of
those costs in order to keep cost within a pre-determined estimates with the actuals during that period, in order to
range or to identify opportunities for cost reduction. improve the rates.
Secondly, cost control must be able to compare and The modelling of cost as a means of enhancing cost
contrast cost estimates with actual values in order to control can be traced back to some very specific
feed findings back into the process and improve under- equations that were formulated to estimate the cost of
standing. This is facilitated for complex systems, such as aircraft over long production runs; later to become
in manufacturing, by the decomposition of the system known as learning curve theory [57]. This early work
into a reference model. Such models represent the was later developed into the parametric cost modelling
system as an structured organisation of relatively technique that was fully established in the 1950s by the
independent, interacting components, and their globally Rand Corporation. The Rand Corporation is credited
defined tasks [55]. Many reference models for the with the development of cost estimating relations
manufacturing system have been developed [17]. One (CERs) for different classes of aircraft and various
possible representation of decomposition is by an operational parameters, developed at that time to help
architecture that defines the functions within a given the department of defence (DOD) estimate the cost of
framework, each function’s input and output being new military aircraft [58].
required to perform the task of the system [56]. There are three well-recognised methods that are
The reference model of Liebers [48] was developed to currently employed in evaluating cost in aerospace
clarify the relation between cost control and manufac- engineering. The bottom-up method is associated with
turing so that when the position of cost control in the collecting all of the product cost values that are
manufacturing system is known, the cost control available; the analogous method is associated with
component can also be decomposed. Ten Brink [11] comparative costing according to the similarity and
explains that the hierarchical model consists of three differentiation of like products; and the parametric
main components in planning, execution and control, method is associated with the use of probabilistic
which are then sub-divided into sub-components. The relations between appropriate product features and cost
four planning and control levels are the strategic, (the CERs). Rush [59] has pointed out that cost
tactical, operational and production levels. The cost modelling is knowledge intensive and that it requires
control component can be decomposed into four the skills and knowledge capture from a number of
functions: (1) cost estimation; (2) production monitor- disparate disciplines. It relies on an accurate under-
ing; (3) cost calculation and evaluation; and (4) cost standing of the company’s and supplier’s product
modelling [48]. The cost estimation function generates development capabilities, which ensures that the models
cost estimates that are based on the specification of a are provided with the appropriate data. Hammaker [60]
solution by a decision maker, in conjunction with a cost has noted that the reasoning and logic that an estimator
model with defined cost rates. The production monitor- develops, is not readily evident because the knowledge
ing function provides the relevant information and data required is complex while its sources are varied, as
from the execution of the production plan, to the cost depicted in Fig. 6.
calculation, evaluation and accounting. The manufac- Cost estimators are required to apply a combination
turing input data is used to generate the actual costs, of logic, common sense, skill, experience, and judge-
which are then compared with the cost estimates and ment, in order to generate a final estimate that is timely,
their underlying assumptions. This then becomes the relevant, and meaningful [61]. Normally, engineers are
basis of the cost modelling that learns, while the cost more required to do this when interpreting predictions
accounting generates the cost rates based on the and modelling results, not within the actual modelling
manufacturing data. There are four feedback loops that process itself. They interpret and manipulate data from
 ARTICLE IN PRESS
R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534 499

Fig. 6. Skills and knowledge of cost estimating.

all of the functions that contribute throughout the cost details, such as the number of operations, time per
product development cycle in order to provide the operation, labour cost, material cost and overhead costs,
platform on which cost estimating and project planning etc. However, Boehm [67] offers a more detailed
may be built [62]. This view is confirmed by the Society classification of estimating methods that includes the
of Cost Estimating Analysts (SCES) that defines a cost following:
model as: ‘‘a compilation of cost estimating logic that
aggregates cost estimating details into a total cost  Parametric: using cost drivers that represent and
estimate... an ordered arrangement of data, assump- model certain characteristics of the target system and
tions, and equations that permits translation of physical the implementation environment.
resources or characteristics into costs’’. In general, a cost  Expert judgement: the advice of knowledgeable staff
model can be said to consist of a set of equations, logic, is solicited.
programs and input formats that specify the problem.  Analogy: a similar, completed, project is identified
Some formulation or framework of these can be and its recorded costs are used as a basis.
supplied with input program information of a descrip-  Parkinson: the premise that work expands to fill the
tive nature in order to produce an output format. This time available and uses the available resource level to
also highlights the fact that the origin of cost modelling drive the estimate.
is always with data analysis or data mining (see Section  Price to win: a figure that is sufficiently low to win the
4.2.5), which serves as the basis for the development of contract.
analytical models [63].  Top down: an overall estimate of effort for the whole
project that can be broken down into the effort
3.2. Cost estimating required for individual component tasks.
 Bottom-up: component tasks are identified and sized
Cost estimating is the process of predicting or and the individual estimates are aggregated to
forecasting the cost of a work activity or output [64] produce an overall estimate.
by interpreting historical data. Rush [65] has noted that
traditionally there are two main estimates: (1) a first- Boehm [67] refers to all seven entries in his list as
sight estimate early on in the design process; and (2) a ‘software cost estimation techniques’, although Hughes
detailed estimate that is associated with precision [68] correctly points out that the ‘Parkinson’ method is
costing. First-sight estimates are useful for what is often not an effort prediction method but a way of setting the
referred to as a rough order magnitude (ROM) estimate scope of a project. Similarly, ‘Price to win’ is a pricing
[66] and provide useful information at an early stage of tactic and not a prediction method, although both are
product definition but are not suitable for decisions recognised management techniques. However, Boehm’s
regarding product detail. On the other hand, detailed or list can be further distilled [58] to leave the three
bottom-up cost estimates are based on specific recorded most basic and inclusive classifications of bottom-up,
 ARTICLE IN PRESS
500 R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534

analogy, and parametric. These three are addressed in Dilts [78] have presented an automated design-to-cost
Section 4.1 while some of the more advanced techniques tool which can be linked to a CAD package in order to
that are currently being developed are presented in provide the estimated cost of machined parts from a
Section 4.2. Bode [14] uses a similar logic to define two particular material. Within aerospace, this would be
basic approaches for cost estimation: generative cost most relevant to the detailed design process for a range
estimation and variant-based cost estimation. Genera- of parts from smaller complex machinings but could be
tive cost estimation is seen as the composition of costs extended to larger fuselage frames of machined-finish
from the key constituents while variant-based cost aluminium forging for example. The cost tool interprets
estimation uses similar products that have been manu- the machined part using Feature-Based Modelling [79]
factured in the past. The various techniques for cost and classes it accordingly using Group Technology [80].
estimating are presented in detail in Section 4. Various costing modules then plan and cost the
machining process using a mixture of activity-based
costing [81,82] and analogous costing in a comparative
3.3. Cost-integrated design manner. Taylor [83] has also advocated a feature-based
approach to aerospace cost estimating and this is often
It is well documented in the literature that cost is an used in traditional aircraft cost estimating, although in a
important attribute of any product and highly relevant less formalised manner.
to the engineering design process [69,70]. Sheldon [71] Analogous costing is also a traditional costing
has stated that customer affordability, product quality technique that uses the cost of a similar product to gain
and market timeliness are the three key elements of a first baseline estimate. Deviations in the design or
competitiveness. He also points out that there are two manufacture of the new product are then used to
fundamental engineering approaches to controlling cost: account for alterations in the initial cost estimate [83].
namely, (1) designing for cost and (2) costing for design. Apart from the analogous, ABC and feature-based
Within aerospace, Dean [72] is well known for promot- techniques, there are a range of other methods for
ing such considerations within NASA. Although Shel- generating the actual cost estimates from input data and
don defines the DFC methodology as being driven by constraints [85] including: regression-oriented para-
management imposed cost targets, this is usually metrics [86], bottom-up costing, fuzzy logic [87] and
referred to specifically as DTC [73]; implying that a neural networks [88]. It is the level of input data and the
cost target has to be met and adhered to. DFC is range of constraints, as well as the technique itself,
generally taken to mean that the design process is which tend to differentiate these techniques and to make
mindful of cost. Many authors now believe that them more or less suitable to a given application,
imposing strict target costing on engineering design, as especially according to the level of product and process
for DTC, is not effective as it tends to result in inferior definition available. The parametric estimating techni-
design that still overshoots the poor cost estimates used que [89,90] is widespread within aerospace and varies
as the initial guidance [73] . Rather, it seems to be more greatly from being based on purely statistical signifi-
important to give designers supportive costing tools that cance, to being more causal in nature; being either
facilitate the product definition process by linking design linear, exponential (logarithmic linearity) or polynomial
decisions to estimated cost impact. in form.
Fig. 7 shows a typical generic model of a cost It is also well documented in the literature that the
estimating tool which can be used within the design impact of cost needs to be introduced upfront at the
domain [74]. However, most of these DFC/DTC tools concept design stage. Pugh [91] has advised that a top-
are application specific and highly customised within the down cost estimation should carried out even before the
aerospace industry [75–77]. For example, Geider and aircraft development process begins. Thurston [92]

Production standards Material costs Labour rates

Part features / geometry COMMERCIAL FACTORS

Cost by part, assembly, material, etc.
INPUTS OUTPUTS
Feature attribute Design guidance

Planned process
Cost Algorithms
Inputs to risk

Material/BO details Producibility guidance

Design rules Producibility rules Process decision models

Fig. 7. Typical generic model of a design-oriented cost model.

 ARTICLE IN PRESS
R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534 501

advocates a holistic approach to the design process that The basic principle of relevance to LCC is still as
is appropriate at the concept stage where a product is prevalent today as shown by Murman et al. [104] who
defined in terms of a measure of its utility value to the defines better–faster–cheaper life cycle needs in terms of
customer. This includes cost in a multi-attribute analysis value-oriented cost, performance and time functions.
[93] of the design that can then be mathematically The process technology aspects are addressed by
optimised [94]. Another form of this design methodol- considering ‘Lean’ practises for design, engineering
ogy has also been applied by Collopy [95] to satisfy the and manufacturing. Marx et al. [98] have presented a
more holistic design requirements of an unmanned arial parametric solution for linking life cycle needs back to
vehicle (UAV). A high-level objective function that design. They use the case study of a high-speed
reflected the wider design requirements of both cost and commercial transporter to investigate the best structural
performance is at the core of the method, providing a layout for the wing in terms of life cycle requirements;
trade-off mechanism that through maximisation pro- including chord-wise stiffened, span-wise stiffened and
motes the optimal choice of design parameters within bi-axially stiffened structural layouts. On the other
stipulated ranges of constraint. hand, a much more detailed analysis platform for
This type of approach can be traced back to much of manufacturing cost drivers has been developed by
the classic research within the aerospace industry into Rais-Rohani [105], where he incorporates many of the
parametric optimisation [96]: the identification of key relevant manufacturing issues in terms of parametrically
design parameters that drive performance and which can defined complexity factors; including; compatibility,
be optimised when combined in mathematical formula. complexity, quality, efficiency and coupling. Rais-
Much of the current mainstream research is focused on Rohani’s work is integrated into the aircraft design
multi-disciplinary optimisation (MDO), whether at a process using a three-tier MDO methodology [106]. For
high level or a lower level that links discrete computa- example, with respect to the three alternate structural
tional models [97]. Marx and Mavris [98] have linked designs of a wing box (thin heavily stiffened skin; thick
MDO to Life Cycle Analysis by defining high level lightly stiffened skin; multispar), the authors advocate
objective functions that encompass the life cycle needs of firstly setting out the structural design configuration, as
aircraft, supported by necessary disciplinary models well as defining materials, part manufacture and
which facilitate the optimisation process through a assembly method. Secondly, a single or multiple
linkage that is defined by an objective function. In optimisation procedure is carried out according to some
aerospace, life cycle analysis tends to be associated with objective function with a multidisciplinary set of design
military applications while the commercial sector and manufacturing constraints. Thirdly, the design is
focuses on DOC; the latter being more associated with validated and the cost estimates improved to allow for
the cost of transporting a person a number of air-miles trade-off, sensitivity studies and optimisation of the
at as cheap a cost as possible. There are various DOC airframe structures.
models available, which tend to be of a parametric With regards to the aircraft fuselage panel case study
nature [99,100], which allow the trade-off of design considered later in this paper, the need to understand the
parameters and which can be linked to manufacturing linkage between material and process selection, structur-
models to couple the impact on production [73,101]. al design needs and LCC was driven by industrial need;
It has been shown from the literature that aerospace in the face of ever-tighter competition and demanding
design is a key fundamental driver of the overall cost of passenger requirements. Sandoz [107], a chief engineer
aircraft, whether considering high-level cost control on the Boeing 747, was projecting a value-oriented
methodologies such as DFC/DTC or cost integrated approach to the integration of these needs for aircraft
design methodologies; for higher level concept stages or structures already in the early 1970s. Other authors have
at the lower level preliminary scheming and detailed continued to address the impact on manufacturing by
stages. The impact of the work of Boothroyd and characterising the various manufacturing processes for
Dewhurst [101] in highlighting the need for a methodol- fuselage panel parts [108], along with the associated
ogy that links the impact of design decisions on assembly processes [109]; with respect to key design
manufacture is well referenced. The major contribution drivers and cost. Much of the work has again been
in addition to firmly establishing the DFMA principle industrial-oriented and focuses on assessing the trade-off
was in providing an analytical technique that introduced between technologies or materials [110]. However, there
quantitative analysis when comparing a given design has been very limited published work carried out in the
with a theoretical baseline in terms of design complexity; linkage and simulation of accurate cost estimation and
classically with regard to part count and fastener count. detailed structural requirements.
Stoll [102] has also addressed many of the organisational Consequently, this paper sets out a methodology in
and implementation aspects of DFMA while other Section 5 for the integration of cost into the airframe
authors were also reporting the important linkage design process, at the performance analysis stage so that
between DFMA and LCC [103]. a proper trade-off of design solutions can be carried out
 ARTICLE IN PRESS
502 R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534

through the explicit optimisation procedures, involving

both structural performance and manufacturing cost in
this case.

3.4. Supply chain cost control

Within the procurement and logistics function, accurate

costing information is required to drive market strategy,
design and manufacture, and ultimately, to ensure
competitive advantage. The procurement function tends
to be characterised as exploiting the aerospace supply
chain in order to develop opportunities for increased
profitability. It has been noted [111] that this is envisaged
through manipulating the areas that directly affect asset
and resource utilisation, and profit margins, including:
production decisions, supplier relationships, outsourcing
verses in-house management, and inventory turnover. Fig. 8. Make/buy flow chart.
Humphreys [112] states that organisations traditionally
buy on the basis of lowest price, only sometimes taking
other factors into account such as quality and delivery.
Other authors have also noted this very limited apprecia-  internal and external analysis for major part families,
tion of the wider issues of delivery reliability, technical manufacturing process categories, cost allocations
capability, cost capability and financial stability [113]. and the alignment of parts and technologies within a
Williamson’s [114] theory of Transaction Cost Ana- competitiveness/importance matrix;
lysis provides a conceptual basis for the make/buy  strategic evaluation of sourcing options now identi-
decision making process. The analysis can be likened to fied in conjunction with the business data;
Activity Based Modelling (see Section 4) as it considers  final selection based on current and future projections
the transfer of goods and services across technologically through application of financial decision support
separate units as they move from one stage of distinct models.
activity to another. The transactions, rather than the
commodity, are the basis of the analysis, which focuses
Typically, it is recommended to formalise best practise
attention on the cost of planning, adapting and
monitoring activities under alternative governance procedures for all of the activities that describe the
structures. Williamson [115] has further noted the need procurement function. For example, Fig. 9 [116] high-
to understand and control the factors that make lights the degree of risk associated with the degree to
transactions simple or difficult to mediate, and especially which an item is interrelated to other items or activities.
to establish monitoring and governance structures that The best practise principles that have been identified as
can be matched to the transactions. He also combines procedurally correct need to be supported by facilitating
economic theory with management theory in order to tools that provide quantitative measures of cost, time,
lay the foundations for a purchasing discipline that risk, quality, etc. In particular cost-modelling tools can
respects both internal and external boundaries in both be easily related to the following procurement needs as
described in the literature [119,120]:
the short and long term [112], whereas design and
manufacture is traditionally ineffective in even appre-
ciating their in-house cost base.  eliciting active support from top management,
Fig. 8 [116] shows the most important ‘exit points’ in  integrating and modelling the supply chain,
the process at which a company can opt to buy rather  understanding cost drivers in appropriate detail,
than make, including several stages of product design  measuring the performance of suppliers, systems, and
and process design, rather than just basing outsourcing employees,
decisions on the reduction of immediate overheads [117].  developing cooperative supplier relations,
However, many of the generic aspects in Fig. 8 are  delivering and establishing a culture of continuous
shared in a more collaborative relationship. Alterna- improvement,
tively, Probert [118] has proposed a 4-stage process  facilitating a cross-functional approach linked
characterised by the following methodological steps: through cost,
 managing and reducing cost across the whole
 preliminary business and strategic appraisal based on business structure,
the company’s, competitors’ and supplier’s data;  developing integrated data management systems and
 ARTICLE IN PRESS
R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534 503

Fig. 9. Matrix of dependency and decomposability.

 justifying investment in procurement/supply tooling and Brand [124]. They define four levels of development
and management. for the purchasing function:

(1) traditional: vendor selection for the lowest possible

Supply chain management is driven by the need for
price;
aerospace companies to reduce cost, shorten product
(2) partnership-relational: close supplier relations for
development time and manage risk, all in an effort to
reduced total cost and risk in an atmosphere of trust;
maximise value added [121]. The transactions between
(3) operational (material logistics management): coor-
companies in supply chains or the extended enterprise
dinating material and information flows to improve
can be conceptualised by the adding of value up through
quality, inventory levels, and overall cost;
the chain and consequent payment in return back down
(4) strategic (integrated value added): flexible business
the chain. This is the integration of key business
processes for speed, flexibility and advantage in the
processes from the end user through to the original
market place.
supplier, in providing products, services, and informa-
tion that add value. On the other hand, it has been noted
that lack of cohesion destroys value in the supply chain Narasimhan [126] has noted that the key concept that
[122], and therefore collaboration is the process that distinguishes a supply chain from its constituent entities
results in the opportunity to create value. Lockamy and is the integration of operations across the extended
Smith [123] have characterised the supply chain with enterprise. The management of the supply chain goes
three common components: suppliers, producers and beyond the simple interface coordination which sees
customers. The components must interact in a coordi- firms optimise local objectives. It explicitly recognises
nated manner in order to ensure the efficient delivery of interdependencies and the wider need for adequate
goods and services, rather than the more typical supply within the global market, while protecting profit
management of each as a separate independent entity margins under such global competition [125]. With the
with localised objectives [122]. rise of global opportunities, the outsourcing of manu-
The changing nature of purchasing towards supply facturing activities has been followed by the outsourcing
chain management has been investigated by Giunipero of design and development work. To an increasing
 ARTICLE IN PRESS
504 R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534

extent, suppliers are contributing to the technical experts in order to formalise that into similarity
development of the end products and are therefore functions and analogy rules [143,144]. However, such a
increasing the importance of supply chain management formalised knowledge-based tool can be complex and
and intelligence as part of their strategic approach [127]. always entails the use of subjectivity to some degree.
Therefore, its development, underlying rules and
3.5. Knowledge-based systems assumptions, and its repeatable utilisation are difficult
and subject to the expertise and understanding of the
user [145,146].
It has been pointed out by Rush and Roy [7] that
The analogous costing methodology is characterised
Knowledge-Based Systems (KBS) help to formalise
by adjusting the cost of a similar product relative to
specialised knowledge so that it can be reused. This
differences between it and the target product. As stated,
overcomes many of the human fallibilities that have
the principle is widely used within aerospace costing and
negative impact, such as poor memories, bias, incon-
there is a similarly wide range of implementation
sistency, retirement, job change, and illness, etc. There-
techniques, ranging from subjective expert opinion
fore, this aspect of KBS is intrinsic to the approach of
[146,147] to objective use of calculated differentials [83]
capturing human expertise in order to be able to make it
according to percentage of unit cost or even from
available when required [128], However, such expert
bottom-up variations in the BOM. The effectiveness of
knowledge needs to be captured and formalised in a
this method depends heavily upon the ability to identify
meaningful way so that it can be reused, although the
correctly the differences between the two cases [47].
capture and embedding of knowledge is not easy and has
Analogous estimates can utilise a single historical data
been viewed as a key weakness of KBS design [129].
point as the basis for the estimate or a programme cost
These difficulties are exacerbated when trying to identify
estimate may use a number of analogous estimates
representative experts and then interpreting their multi-
relative to a number of cost elements that make up the
ple views [130].
programme. There is an obvious risk in basing a single
Kingsman and de Souza [130] have presented a
point estimate on one historical instance and in addition,
methodology in support of a knowledge-based decision
the technique usually involves a high degree of expert
support system for made-to-order companies. The
judgment. However, it is a reasonable approach for
method included identifying when most judgments were
estimating the unit cost of a new product that does not
made and then examining both the cost estimating and
incorporate very different design features or utilise new
pricing processes. The identified judgments are then
processes for that company. The FAA Life Cycle Cost
taken to represent the expert knowledge capture and are
Estimating Handbook [148] recommends its use for a
formalised through the use of ‘‘If (Condition)...Then
new product or system that is primarily a combination
(Action)’’ rules. The research method included the use of
of existing sub-systems, equipment or components for
expert interviews to facilitate the capture and develop-
which recent and complete historical cost data is
ment of the rules. It is reported that managers found the
available. They also point out that analogy methods
end result to be useful as an aid to their decision-making
are less likely to overlook the impact of rapid technology
but it was also noted that one of the limitations of the
changes, whereas it may be less obvious that a
approach is that it is more suited to companies that have
parametric cost model database is no longer valid and
a similar project base as KBS tends to be domain
needs updating. The recommended practice for generat-
specific.
ing analogous estimates is lengthy but the standardisa-
tion helps to ensure that the process is as rigorous as
possible, as presented in Fig. 10:
4. State-of-the-art: cost estimating (1) The first stage is one of definition. This includes
the general features of the estimate, including its type
4.1. Classic estimating techniques and accuracy, and the assumptions made in terms of
inflation, quantities, scheduling, etc. The product must
4.1.1. Analogous also be defined in terms of its physical design
Analogous costing is one of the best-established and parameters; performance characteristics such as relia-
applied methods of costing [131–139]. In industry, it is bility and maintainability; training and operational/
still deployed in an ad hoc and expert oriented manner support issues; test and certification requirements;
but the term is also synonymous with case-based technology maturity levels, etc. This then allows the
reasoning tools [140–142]. Typically, a CBR tool will estimate breakdown structure to be identified in terms of
store and organise past projects with a view to later the hardware and activity components whose estimates
retrieving these projects in order to help identify a costed are to be incorporated in a cumulative estimate.
solution for a new project. Consequently, the develop- (2) The second stage is one of practical preparation in
ment entails capturing the knowledge from domain assessing the availability of data downstream in the
 ARTICLE IN PRESS
R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534 505

Fig. 10. Best practices for generating analogous cost estimates.

process. This includes data relating to the quantity, these relative factors map across to the cost ratios
design and performance characteristics of the product through their performance and design ratios, and are
components for both the historical and new cases, and not influenced by productivity improvement differences
the cost data for the historical case. The components for between the cases. Miniaturisation factors are also used
the new case also need to be described in relative terms as typically in aerospace the smaller the subsystem is for
to the most comparable historical cases that are most a given level of performance; the more costly it is likely
likely to reflect the cost differentials. to be to produce. These factors may relate to weight or
(3) The third stage is actual data collection, which space constraints and again are assessed initially by
includes both quantitative and qualitative data for as technical specialists. Productivity improvement factors
many historical cases as possible that are current and are used to map the cost reduction expected from
comparable with the new specification. The historical significant productivity improvements between the
cost data should be as well defined as possible and historic and new cases, being anticipated from improved
distinguish between prototype, full-scale development design for manufacturability, more effective manufac-
and production costs, and between non-recurring and turing technology and reduced material costs.
recurring costs. All historical data also needs to be (5) The fifth step is the generation of the actual cost
normalised relative to time and a baseline year, as well estimates. It is recommended to initially estimate the
as ascertaining the first unit recurring costs and the first-unit cost from the historic cost C P for first-unit
improvement slopes. This then provides the necessary value in conjunction with the three ratios generated for
factors, etc. based on historic costs, including those from complexity F C ; miniaturisation F M and productivity F P :
the extrapolation of historic cost elements to the new Therefore, the analogous cost estimate is calculated
case or adopting existing factors that have been according to C N ¼ C P F C F M F P : Typically, these factors
reconciled for any major differences. These ratios, are estimated by expert opinion within the companies
factors and improvement curve should then be reviewed but could be more rigorously defined on an analytical
with input also from technical specialists who are basis from historical data, e.g. miniaturisation being
familiar with the historical and new design cases. modelled according to the recorded impact of reduced
(4) The fourth step is to generate a range of factors part size on cost for components with a like functional
that characterise the product in terms of design features, value. Following on from that, the first unit values
etc. that influence cost and manufacturing capabilities. estimated are combined with the cost improvement
Complexity factors are recommended by the technical curve slope values developed to generate the total
specialists relative to cost. There is an assumption that recurring costs for each component. The non-recurring
 ARTICLE IN PRESS
506 R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534

costs are developed similarly or are based on recurring 5

R2 = 0.955
to non-recurring ratios. The relevant full-scale develop- 4.5
2
R2 = 0.8716

Normalised manufacturing cost

R = 0.8625
ment estimates must be aggregated separately for the 4
R2 = 0.8971
specified production amount, unless one is to be 3.5
developed based on the other. In addition, other costs 3
can be generated using determined factors; for systems 2.5
engineering, programme management, spares, support Fan Diameter
2 Weight
equipment, training and IT.
1.5 Airwash Area
(6) The final step is the development of the total Thrust
1
programme cost estimates. This includes the addition of
0.5
agreed profit levels according to market research and
company policy and any additional factors such as, 0
0 2 4 6 8 10 12
mission support, testing facilities (if external), contrac-
Normalised cost driver
tors costs, etc. Ultimately, the final estimate is to be
reviewed in terms of the results and the balance of Fig. 11. Plot showing high-level design cost drivers.
complexity value judgements. It is important that the
documentation should not only list the total costs but
also the main complexity judgments applied, the Table 3
historical cases used, and the qualification of the Attributes used to characterise cost drivers
technical specialists who set the criterion.
Specification Manufacturability Geometry
An example of analogous costing taken from the
literature [83] details one methodology that was devel- Functionality Part count Cylindricity
oped for the costing of nose-cowls on engine nacelles. Certification Process capability Circularity
The cost of a nose-cowl is driven fundamentally by the Aerodynamic Assembly Concentricity
various design requirements that meet aerodynamic, smoothness philosophy
thermodynamic, and structural needs, with some addi- Structural efficiency Manufacturing Curvature
tional functionality such as thermal anti-icing and Full tolerances
Authority Digital Electronic Control (FADEC) systems,
and engine integration. However, with regard to
production, this assembled component is relatively
generic in form and nature, the key function being to the baseline cost driver and then generate measures of
direct airflow cleanly into the engine fan. Consequently, differential cost driver to refine the cost estimate, such as
this commonality reduces the complexity of the costing those listed in Table 3. The presented model was based
as it is less likely that there will be major design on the premise that recurring manufacturing cost is a
differences that make analogous costing more difficult in function of four parameters, including size (rather than
terms of accuracy. The clear symbols in Fig. 11 show the weight) for the baseline relation and three specific design
unit recurring costs of a number of nose-cowls plotted and manufacturing drivers. The component size is given
against component size or engine fan diameter. It can be by the engine fan diameter Dfan being also linked to
seen from the trend line that the characteristic is linear design specification through engine fan size.
and there is a statistical significance of R2 ¼ 0:9; where Having determined the key cost drivers; the next step
approximately 90% of the scatter in the points is being was to gather data that would quantify these. These
modelled by the linear regression trending. In terms of were largely determined through knowledge capture
analogous costing, there is an assumption that there is a based on expert opinion. For example, a rating of ‘1’
linear baseline relationship between the two variables was assigned to a baseline level and a rating of ‘4’ to the
and that it is the complexity factors which give rise to the most extreme deviation from that baseline. This
cost differentials from that baseline characteristic or cost qualitative approach can be easily replaced by a more
floor. quantitative approach, which should be developed
Three categories of cost driver were identified as relative to the product definition available in terms of
relevant to characterising the cost variance and are pre-concept bid, preliminary design or detailed design
typified in Table 3 as: geometric complexity factor for example. The approach presented identified two
f Geom ; manufacturing complexity factor f Manuf ; and Nacelles with one of the complexity ratings ( f Geom ;
specification complexity factor f Spec : However, there are f Manuf or f Spec ) being equal and the third with a different
also other higher-level cost drivers that can be used to value in order to ‘calibrate’ the cost differential. In a
develop a rough order of magnitude (ROM) for the principle similar to the solution of simultaneous
baseline prediction. For example, many commercially equations, the difference in the dependant variable, i.e.
available cost estimating packages [150] use weight as the cost differential between the Nacelles, was equal to
 ARTICLE IN PRESS
R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534 507

the rating differential for that complexity factor.

However, this was calculated as a ratio of fan diameter Original data
Dfan in order to allow for the baseline influence of
Baseline prediction
component size. This procedure yields three costing

Manufacturing cost
ratios r of the form described in Eq. (1) for geometry:
R2 = 0.8971 2
R = 0.9372
ðC 2 =Dfan2 Þ  ðC 1 =Dfan1 Þ
rGeom ¼ . (1)
ð f Geom2  f Geom1 Þ

Consequently, the total cost impact of each of the

complexity ratings could be calculated, relative to some
baseline cost that is purely a function of size. For
example, Eq. (2) shows the form cost differential
associated with each of the geometric complexity factor: Fan diameter

DC Geom ¼ rGeom Dfan ð f Geom  1Þ. (2) Fig. 12. Evidence of the baseline concept used in analogous
costing.
To establish the linear baseline equation as a function of
size, the trend identified for the original data points is
shifted vertically downwards by the cost differential DC 0
between it and the baseline Nacelle. This gives an
equation of the form described in Eq. (3), where z is the Prediction
linear constant from the original data: Original data
Manufacturing cost

0
C ¼ mdata Dfan þ zdata  DC . (3)

Subsequently, the predicted cost C Pred of any new

Nacelle with a given engine fan diameter Dfan and
complexity factors of f Geom ; f Manuf and f Spec is
calculated as shown below:

C Pred ¼ mdata Dfan þ zdata  DC 0 þ DC Complexity . (4)

The diamond symbols in Fig. 12 denote the original cost

of each Nacelle while the circles represent that actual Fan diameter
cost minus the predicted cost differential arising from Fig. 13. Comparison of analogous cost estimates.
the complexity. The latter should represent the baseline
cost or cost floor that is a function of size only and is
important in suggesting whether the methodology is
improving the regularity and predictability of the cost. formalised process to cost knowledge capture and
In support of this, it can be seen that the linearity is utilisation. The results of pair-wise comparisons have
further improved to R2 ¼ 0:94: Finally, Fig. 13 plots the been recorded to outperformed experts who do not use a
original data against the predicted values, with reason- structured approach [152]. However, the user requires
ably good correlation. It is interesting to note that the knowledge or data relating to both the functional and
trend line for the original regression analysis of the data, the technological aspects of the product [134] and
shown in Fig. 10, had an average absolute error of 14% therefore, there is an implicit requirement for a
in predicting the cost of each nose-cowl while the structured technique for capturing such input data in
proposed complexity method delivered a reduced order to provide better results.
average error of 10%.
Yet another technique within analogous cost model-
ing is the pair-wise comparison approach. It has been 4.1.2. Parametric
used for various estimating tasks such as software sizing According to the Parametric Cost Estimating Hand-
and manufacturing design effort [132–134,150]. Given a book of the Department of Defence [90]: ‘‘A parametric
number of reference projects, the comparative analysis is cost estimate is one that uses CERs and associated
carried out by quantitatively rating how similar in size mathematical algorithms (or logic) to establish cost
or attribute the various projects are [151]. This estimates’’. This is a commonly used technique within
quantitative approach utilises statistical analysis to aerospace which typically utilises linear regression for
normalise and order the ratings and provides a more CER development [153–155]. The CER is developed by
 ARTICLE IN PRESS
508 R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534

establishing a relationship between one or more para- period by Stanford Research Institute, establishing the
meters that are observed to change as cost changes. relationship as a function of the cost of the first set and
These parameters are typically referred to as cost the total unit number to be investigated. In the
drivers, as they are known to be highly influential in formulation, there is typically an exponential term that
effecting a change in cost or at least, to vary similarly determines the slope of the characteristic and which is
with cost. Using historical data, a correlation between associated with several influencing factors. Most im-
cost as the dependant variable and the cost driving portantly, the learning exponent would be function of
parameters as independent variables, establishes the the efficiency of the company’s processes in general, the
statistical accuracy of the relationship. An example of a use of new technology and the design complexity of the
simple CER would be the relationship or correlation aircraft. It should be noted that in the time domain
between the number of design drawings and the cost of analysis, such cost data needs to be normalised
the design process for a large aircraft assembly. The according to financial rates and inflation index so that
rationale behind the choice of drawing number (as the the analysis is fair and true. This is especially true and
cost driver) is that one would expect the number of relevant for unstable periods in history when rates
drawings to increase with the complexity and part count fluctuate more widely [156].
of the assembly, which is linked therefore to design Typically, learning is factored into the estimating
effort and time, and ultimately to the design cost [93], process through some deviant of the following formula-
The latter part of the DOD definition quoted above tion:
refers to the way in which the CERs are used to arrive at
Hours=unit ¼ U b Rr
a cost estimate for a product. In a sense this is driven by
the perceived costing architecture that is used to describe or
all the relevant costs and how they are combined to
account for the product’s total cost. Typically, this is ðFixed_year_costÞ=unit ¼ ðFirst_unit_costÞU b Rr ,
referred to as a Cost Estimating Model (CEM) and for where U is unit number, b is learning curve slope, R is
the above example of an aircraft sub-assembly might production rate, r is production rate curve slope. The
include additional CERs that are required to generate an slope of the curve can be estimated or derived from
estimate of unit cost. For example, in addition to the historical data from particular programs but then would
design cost, this might include CERs for estimating the have a specific range of application, similar to a CER.
cost of: materials and treatments, fabrication and The slope should be determined while holding the
assembly, support and inspection, overheads, contin- learning curve constant as the rate effect can vary
gency, etc. [45–47]. From these, the estimator is able to considerably with changes in plant facilities, manpower
generate a cost estimate for a similar product that and redeployment, and overtime.
accounts for all of the perceived costs, with the accuracy The main period of fast development for parametric
being dependant on the combined correlation accuracies methods started in the 1950s with the establishment of
of all the individual CERs. The resulting parametric the Rand Corporation [151] by the military, which was
models can be used easily and speedily by engineers of to be an independent civil forum for discussion and
varying experience and at a very early stage in the design analysis. The main concern of the DoD and the United
process when there is little product definition. States Air Force in particular, was to have the capability
The birth of parametric cost estimating is often traced to analyse future scenarios in terms of technology and
back to the work of Wright when he first proposed the cost. In terms of current technology utilisation there was
learning curve [57]. That early work was a forerunner of no established methodology for estimating the first unit
parametric techniques to come as it specifically con- cost, also being the required input value for the learning
sidered the relation between the unit cost of aircraft as a curve formulation. In addition, although the learning
function of the number of aircraft produced, i.e. linked curve addressed recurring cost, there were no methods of
cost to an observed cost driver. His theory was used estimating the early non-recurring costs such as
extensively during World War II when there was an research, development, testing and evaluation. During
exponential increase in the production of military the 1950s the Rand Corporation established parametric
aircraft but little knowledge of how the unit cost would ways of both estimating first unit cost and the non-
decrease with the benefits of production scale and recurring costs [151]. It has been noted that even then
learning. A typical learning curve in its class, for these techniques were being utilised for all phases of
example for the high production C47 aircraft, would aircraft systems during the 1960s.
record the unit cost after 10,000 aircraft have been made Due to the potential for fast and easy estimating
decreasing to approximately a quarter of that of the first capabilities based on company practise, the world of
aircraft. However, the major point of interest is that the parametric costing has grown and spread into other
unit cost was already close to that level after some 3000 fields and the civil sector. In the same way that certain
units. Wright’s work was validated in the post-war drivers can be chosen to relate to aircraft cost or weight
 ARTICLE IN PRESS
R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534 509

Fig. 14. Methodology for developing parametric models.

[151], any dependant cost or performance variable has and the Space Systems Cost Analysis Group (SSCAG)
the propensity to be related statistically to a product’s in 1977.
attributes. One of the strongest deployment areas for The basic methodology for developing parametric
parametric technology is within the construction estimating models was developed in the 1950s by the
industry [159]. Typically, they relate costs to size, Rand Corporation, illustrated in Fig. 14, who are
assuming that statistically this provides a reasonable accredited with the following key developments [90]:
estimate based on historical data, regardless of unfore-
seen wastage, build problems or other variations in cost.
In aeronautics, it is substantially used at the bidding and  Developing the most basic tool of the cost estimating
cost-targeting stage. However, manufacturing also use discipline, the Cost Estimating Relationship (CER).
parametric relations as an experience-based guide when  Merging the CER with the learning curve to form the
facilitating ultimate Estimated At Completion (EACs) foundation of parametric aerospace estimating.
cost estimates; although with the advent of design for  Deriving CERs for aircraft cost as a function of such
manufacture (DFM), there is a growing recognition of variables as speed, range, and altitude.
the additional potential as a DFM enabler. The growth  Observing acceptable statistical correlations in check-
in this method has been a commensurate with the ing the CERs.
appearance of supporting organisations such as Inter-  Developing families of curves data segregated by
national Society of Parametric Analyst (ISPA) in 1978, aircraft types, e.g., fighters, bombers, cargo aircraft,
the Society of Cost Estimating and Analysis (SCEA), etc.
 ARTICLE IN PRESS
510 R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534

 Developing curves corresponding to different levels accuracy of the model is subject to the relevancy of
of product or program complexity. application. The Parametric Handbook [90] notes the
following pitfalls to avoid:
The three categories of parameters central to the
development of parametric relationships as defined  using the parametric model outside the database
within RAND [158] are: range,
 using a parametric model not researched or validated,
 Performance and physical parameters are measures of  using a parametric model without adjustment when
technical capability and may be further divided into new system requirements are not reflected in the
parameters that are scale dependent and independent. database,
 Technical risk and design maturity parameters  using a parametric model without access to realistic
measure or quantify the relative difficulty of devel- estimates of the independent variables and
oping and producing a particular system.  requesting impossible or impractical point estimates
 Programmatic parameters address issues related to for independent variable values over a required range.
the way in which programs are operated.
Beltramo [162] has stated that with parametric model-
ling development there is often a poor correlation
Ideally, parameters from all three groups would be
between the data analysis and the actual product
included when developing parametric relationships
breakdowns and therefore the modelers need to carefully
however there are some limitations to including all
document their assumptions in order to help the users to
parameters. Parameters should be selected based upon
put the models to appropriate use [163–166]. For
the availability of appropriate information while a
example, Kitchenham [165] has reported that in the
rationale must exist as to why a particular parameter
case of the COCOMO parametric cost model, many of
correlates with the dependent variable, i.e. a causal link.
the underlying assumptions were not valid while
It is well documented that parametric relations are
Shepperd and Cartwright [167] reported that much of
extremely sensitive to range of use due to their inability
the cost input data was inaccurate and captured from
to estimate for differences in the product definition not
people with a poor recollection of projects that were
evident in the historical data. The other fundamental
completed a long time ago. Ultimately, this is a highly
aspect is the choice of data, its gathering and manipula-
speculative process and is subject to both technology
tion. In this respect, one must first determine the input
and organisational process changes. Nonetheless, poor
variables to be related. The independent variables are
quality data is often all that is available and therefore
the cost drivers that are (thought to be) related to a
requires extensive use of expert judgment [168] in
change in cost while the dependant variable is the actual
formulating models that do aid in providing a formal
cost data. Some form of regression analysis can then be
method of generating cost estimates [166,169]. Pengelly
formed on the two sets of data, e.g. linear, multiple
[170] agrees that subjective measures and assumptions,
linear, or curvilinear. However, it is very important that
which are often embodied in ratings within the models,
the various cost data is well understood in terms of
are a necessary requirement during the analysis and
auditing and is of a similar makeup. This ensures that
input of data. This raises another question of the quality
the data points are comparative in terms of what they
and adequacy of the data collection [171,172]. This is
represent and how they arose in the first place. To this
exacerbated by the inability of models to predict the cost
end, normalisation is often necessary to account for
of a technology that is not a part of the underlying
variations in the inflation rate. Other factors include the
database [162,90]. Within aerospace, the design of new
learning curve already mentioned and also the produc-
aircraft often entails a step increment in the technology
tion rate. It is recognised that the production rate is
exploited on previous products, which necessitates
related to the speed at which learning [161] can be
expert judgment and knowledge in adjusting costs
established, with faster production rates leading to a
relative to these changes. This judgment must guide in
steeper gradient in the learning curve.
whether a particular parametric CER can be used and
In a similar way to factoring the basic CER with
whether this is feasible [173], and whether the result
production information that adjusts the cost, the CERs
reflects the cost of new technologies and if the outputs
can also be calibrated to give an improved estimate at
are relevant.
current expectations. Calibration is also important to
commercial CEMs that use more universal data and
therefore, require tailoring to a given company database 4.1.3. Bottom-up
[150]. Furthermore, this brings in validation and the As the name suggests the bottom-up or engineering
comparison of estimates with actuals for any parametric build-up method [174] identifies and sizes the component
model. The validation process and the estimating parts and tasks, and then estimates these to be
 ARTICLE IN PRESS
R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534 511

Table 4
Matrix of comparative assessment for tradition methods

Approach Advantages Disadvantages

Bottom-up Cause and effect understood Difficult to develop and implement

Substantial, detailed expert data are required
Very detailed estimate Requires expert knowledge

Estimate by analogy Cause and effect understood Appropriate baseline must exist
Substantial, detailed data are required
More easily applied than the bottom-up method Requires expert knowledge

Parametric Easiest to implement Can be difficult to develop

Non-technical experts can apply method Factors might be associative but not
causative (i.e. lack of direct cause-and-effect
relationships)
Uncertainty of the forecast is generated Extrapolation of existing data to forecast the
future, which might include radical
Allows scope for quantifying risk technological changes, might not be properly
forecast

aggregated in order to produce the overall estimate. The In addition, Table 4 summarises the advantages and
bottom-up approach relies on detailed engineering disadvantages associated with each of the three approaches
analysis and calculation to determine an estimate. To [175]. It appears that the bottom-up method is strong in
apply this approach to any system manufacture, the detail and causation but difficult to implement while
analyst would need the detailed design and configura- inversely the parametric method is too associative in
tion information for the various system components and generating relationships but is easy to implement. The
accounting information for all material, equipment, and analogous method is somewhere between the two and
labour [175]. Within the software industry, the bottom- perhaps is seen as the compromise. However, apart from
up approach is also used [170]. The result of either is a finding a comparable program, it is very difficult generally
detailed estimate and breakdown of costs. within aerospace to gain access to well documented and
Some of the characteristics of the method are as understood costing data. In addition, all three methods
follows: rely heavily on that historic data and relate well to new
materials, technology or design features.
 It is performed at a detailed level within the Work
Breakdown Structure (WBS). 4.2. Advanced estimating techniques
 Cost is estimated for basic tasks such as engineering
design, tooling, fabrication of parts, manufacturing 4.2.1. Feature-based modelling
engineering, and quality control. Design features are often used as relational drivers of
 The cost of materials is estimated or obtained from cost for two reasons as set out by Wierda [176]: (1) cost
the supplier. functions can be derived for classes of similar objects
 The approach requires detailed and accurate data and that serve as key drivers of global cost estimation and
should be undertaken by an experienced engineer. are linked to the engineering domain; and (2) the
designer wants to know the causes of cost so that when
Consequently, it can be seen that relative to the bottom- linked to design features, they are able to influence
up method, the parametric method can be used at the committed cost directly.
early stage of a program when limited data and technical Wierda [176] has also identified three components of
definition is available. Similarly, the analogous method cost that relate to design features and which are valid for
also does not require highly detailed definition as it uses any class of similar objects to which the costs are related
the actual cost from a comparable program although the [177,178]. The difference between the allocation of direct
adjustments to cost require information regarding and indirect costs is also illustrated through:
differences in the program’s complexity as well as the
technical and physical differences to the baseline chosen  Costs assigned directly to individual design features:
as comparable. at a feature or assembly level,
 ARTICLE IN PRESS
512 R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534

 Costs incurred for a collection of design features: at a

component or batch level and
 Costs assigned to design features: at an order or
facility level.

In simple implementation, feature interrelationships can

be ignored so that resource selection is allocated
according to each separate feature. For example,
production times and resultant costs can be calculated
from a simple time formulation for each of the standard
design features [179], which may include feature para- Fig. 15. Design feature-based definitions.
meters and machining rates. Kiritsis [180] has assigned
machining operations to surfaces in calculating cost
while Schaal [181] has used only a rough process plan for Table 5
each of the features as a gauge of manufacturability and Definition of feature drivers
cost. These plans include some information relating to
Feature Examples
the next level of aggregation in the hierarchy, which can type
be used in conjunction with manufacturing rules. A
rough process plan is sufficient at the early design stage Geometric Length, width, depth, perimeter, volume, area
when the cost estimate does not need to be so accurate. Attribute Tolerance, finish, density, mass, material,
As more detailed production information becomes composition
available, the complexity of the cost estimation can be Physical Hole, pocket, skin, core, PC board, cable, spar,
increased as necessary relative to accuracy. wing
Typically, material costs can be related directly to the Process Drill, lay, weld, machine, form, chemi-mill, SPF
Assembly Interconnect, insert, align, engage, attach
material blank with some additional design features
Activity Design engineering, structural analysis, quality
being incorporated if they further influence material assurance
costs. Wierda [176] presents one approach to this
procedure: (1) material costs can be directly assigned
directly to a feature if it implies a positive volume; and
(2) negative material costs (waste revenue) can be and order level costs are associated with certain product
directly assigned if a feature implies a negative volume. levels within the Work Breakdown Structure but costs
For the latter, however, a negative volume may also associated at the facility and other product levels not
be created directly by casting or injection moulding, being evident.
which does not entail material removal. Further It has been noted by Rush and Roy [65] that the
complication arises when both positive and negative growth of CAD/CAM technology and 3D modelling has
volumes of different features overlap, or when parts of a probably played a significant part in the development of
blank lie outside the final product envelope and are not feature-based costing. Most manufacturers do have a
described by any design feature. An inherent anomaly good supply of historical geometric data (if not direct
with feature-based cost attribution is that most opera- cost) that can be related to features and therefore can be
tions are carried out for groups of inter-related features linked to technical specification through functionality
[182] making allocation difficult, however, the approach and performance, and manufacturing capability. Con-
demands that the costs involved with the operations sequently, many researchers are using the feature-based
must be assigned to the number of features identified. approach in costing studies looking at the integration of
This can be confusing when cost does not fall as a design, process planning and manufacturing [183–185].
feature is removed, due to the fact that the cost is This is driven by the ability of a feature-based
incurred regardless, and actually results in an increase in methodology to describe the product as a number of
cost for the other features still included in the opera- associated features that the designer and manufacturer
tions. In terms of the ultimate usefulness of feature- both relate to, i.e. to holes, faces, edges, folds, etc. (see
based costing, another fundamental difficulty is that the Fig. 15). A key observation is that typically, the more
preoccupation with the costs of individual features mat features a product has, the more designing, manufactur-
not lead to the global reduction in cost. For this reason, ing, planning it will require [186]; leading to an increase
Wierda [176] has suggested the use of high-level features in committed cost downstream in the life cycle.
which include both the component and assembly levels With respect to this problem, companies are faced
at which the costs occur. However, there is a clear with producing their own feature definitions. Table 5
problem with the cost allocation, as the assembly, batch shows an example of how one cost engineering group
 ARTICLE IN PRESS
R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534 513

categorised features for the purpose of costing [187]. It is transition from member to non-member being gradual;
evident that on one level of feature definition however as illustrated in Fig. 16.
there are several levels of feature’s definitions. For Fig. 16 highlights that the membership function is
example, a feature of an aircraft could be a wing, yet this described by a characteristic that defines how each
wing contains many parts, each of which consists of instance within the design space is mapped to a degree of
many lower level features. Therefore companies are membership between 0 and 1 [195]. However, the key
also left to decide on how to cope with the changing contribution of the fuzzy methodology is that these
product definition and the application of an appropriate membership functions can be of any characteristic
feature-based CER. The feature-based costing approach shape, within the known boundaries assigned to 0 and
is not yet well established and its application is not yet 1. The characteristic is dependent on the relationship
fully understood although companies do seem to being modelled and is usually described by the simplest
appreciate the concept, features apparently being one function that represents the relational behaviour.
way in which engineers decompose or define a design Typically, these include the: piece-wise linear function,
concept. Gaussian distribution, sigmoid curve, and quadratic or
cubic polynomial curves [196]. These are often described
by straight line characteristics to give the triangular or
4.2.2. Fuzzy logic trapezoidal functions illustrated in Fig. 17. Fuzzy
Ting [188] has stated that most traditional cost modelling is a formulaic representation of a knowl-
modelling tools are crisp, deterministic, and precise in edge-based approach that consists of a collection of n
character. However, in the actual industrial aerospace rules of the form: If V 1 is Li1 and V 2 is Li2 and . . . V p is
environment there are many parameters that are Lip then U is M i ; where Lij and M i are linguistic values
uncertain in nature. Fuzzy logic addresses this char- associated with the corresponding variables. As such,
acteristic and is a mathematical discipline that was the linguistic variables are controlling rules within a
originally created to bridge the gap between the binary fuzzy inference mechanism, as distinguished by the
world of digital computing and that of continuous appropriate use of inputs and outputs. Ultimately, this is
intervals, as displayed in nature [189]. Fuzzy theory was
first introduced in 1965 by Lotfi Zadeh to deal
quantitatively with imprecision and uncertainty
[190,191]. The literature agrees that the major contribu-
tion of fuzzy set theory is in its inherent capability of
representing vague and imprecise knowledge, as applied
to classification, modelling and control [192]. Cross [193]
states that since its inception, fuzzy set theory has been
advocated as a formal and quantitative method of
specifying vagueness in human knowledge. Typically,
the fuzzy approach provides a methodology in which
algorithms for the prediction or control of a system are
arrived at through qualitative expressions that link
linguistic variables [194].
It is of special interest to cost modelling to consider
that the theory states that fuzzy sets are the basis of the
logic, this being the collective name given to the set of
conditions that a fuzzy variable can belong to. A fuzzy
set F is defined as a set of ordered pairs ðx; mðxÞÞ: The
membership function f establishes the relationship:
f : x ! mðxÞ; where x is the value of an element in the
domain of function f (mðxÞ being the value of f at x) and
mðxÞ has values in the interval [0,1]. For a given value
x; mðxÞ ¼ 0 denotes x with null membership within F
while mðxÞ ¼ 1 denotes x having full membership.
Therefore, the membership function mðxÞ consists of
real numbers within the interval [0,1] and represents the
degree of membership that an object exhibits within a
fuzzy set. Kishk [190] points out that the fuzzy set
introduces vagueness by eliminating the sharp boundary Fig. 16. Membership functions for (a) crisps and (b) fuzzy sets.
dividing members of the set from non-members, the
 ARTICLE IN PRESS
514 R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534

mathematical concepts utilised; (2) the ability to match

any set of input–output behavioural data; and (3) the
integration with traditional techniques and experiences.
In terms of knowledge utilization, the fuzzy logic
approach can be viewed as a form of Artificial
Intelligence (Al) that formulates the human thought
process [197], similarly as for neural networks. Conse-
quently, it is appropriate to be developed and applied to
the realm of aerospace cost estimating [198–200]. A
number of authors have explored the use of captured
and coded fuzzy logic within cost estimating [201–203].
However, the technique is not well established and it can
be said that these ‘models only know what the expert has
told the model builder’ [204]. As a consequence, fuzzy
cost estimating is subject to the domain rule of being
limited through limited scope and application.

4.2.3. Neural networks

In a similar way to fuzzy logic, neural networks have
also been developed with a view to simulating the
human thought process, and as a method of linking
historic costing information with design stimuli [205]. As
Fig. 17. Characteristic relationship membership functions such, this can be viewed as a form of artificial
(distributions). intelligence that can be used to develop links between
cost as the effect and certain cost drivers as the cause
[206–209]. The method is based on the concept of a
used to map an input space to an output space through
system that learns to predict the effect on cost when
the use of a set of rules that take the form of a series of
presented with a range of product-related attributes.
‘if-then’ statements. As such, antecedent clauses im-
This in turn is derived from the analogy of a number and
mediately follow ‘if’ statements but precede ‘then’
hierarchy of neurons as logic gates being able to
statements of the rule and contribute to the logic
simulate various procedural permutations and combina-
process by implementing evaluation measures that
tions as it trains itself in being able to repeatedly arrive
control progress through fuzzified rules [194]. The
at a logical conclusion, given input data available from
consequence of the process defines the action to be
historic case studies. Once trained, the attribute values
taken as the antecedent is satisfied, relative to the degree
can be supplied to the network of neurons in order for it
of membership of the input to the antecedent. The three
to apply the approximated functional steps in comput-
main procedures within fuzzy logic are as follows
ing an expected resultant cost. The technique does not
[189,195,194]:
simplify any of the analysis but does transfer much of
the logic and rules to the coded neural network process.
1. Recognise one or more assigned physical conditions However, the analyst must still define the problem
that require analysis or control. domain and apparent cost drivers, and also must supply
2. Process these as inputs according to fuzzy ‘if-then’ the relevant cost data perceived to be important.
rules that are expressed linguistically. Bode [210] states that under certain conditions, neural
3. Average and weight the outputs from all of the networks can produce better-cost predictions than more
individual rules into a single defuzzified output that conventional parametric regression costing methods.
results in the decisions and/or actions required of the However, it is also made clear that in certain cases there
system. are disadvantages in terms of accuracy, variability,
model creation and model examination [211]. Notwith-
Kishk [190] proposes that fuzzy logic is appropriate in standing, one of the key advantages is that a neural
two kinds of situations: firstly, very complex models network can detect obscure relationships within the
where understanding is limited or judgmental, and database. These would not be evident if the user had to
secondly, processes where human reasoning, human provide the complete input assumptions [212]. One of
perception, or human decision making are inextricably the defining aspects of neural networks is that they
linked. This results in a number of key advantages in the require a large historic data bank from which to learn
costing sphere: (1) the simplicity and transparency of the and that the data base needs to be comprised of similar
 ARTICLE IN PRESS
R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534 515

information and form to the new products that are to be risk is highest. Rather than focusing on the actual
analysed. Consequently, their prediction accuracy is as management of risk, the focus is more on combining
poor as the quality, quantity and relevancy of the input statistical analysis with cost estimation in order to
learning data is. Neural networks are not applicable to predict the cost estimation uncertainty to be attributed.
novel or innovative product developments that deviate It is more realistic to have a range of cost estimates
significantly from the historic precedent or where the rather than a discrete value, and this is more likely to be
environmental aspects have changed [213]. Furthermore, accurate in modelling the effect of cost variance, which
one must trust to the ‘‘black box’’ nature of the process is a reality for any product. At a more detailed level,
whereas the regression approach assumed with para- such an analysis facilitates the mitigation of risk in
metric analysis does have a more transparent audit trail reducing uncertainty through avoidance, adjustment
for the estimating procedure. It has been said that the and contingency. At a higher level, risk analysis
neural network solution often does not appear to be facilitates go/no-go decisions that need to be made
logical [65], even if one were to extract it by examining regarding exit criteria when moving from each stage
the weights, architecture, and neuron functions that within the Integrated Product Process Development life
were adopted by the final trained model. Consequently, cycle. It can also be used to rate all of the potential
the ‘‘black box’’ nature of the costing relationship is less design solutions between the range of scenarios envi-
appropriate CER for users that need a transparent audit sioned at the concept stage: when the majority of the
of the reasons and assumptions behind the cost estimate, aircraft’s life cycle costs are committed. This shows
which also impacts on the use of additional analysis which variables and parameters have the most impact on
tools such as risk and uncertainty. Naturally, this is a the design and therefore, highlight where most of the
fundament requirement of the designer who wants to be effort should be targeted in making decisions that
able to learn from the estimating procedure in order to influence the cost and viability of the product. In terms
be able to influence the design process in arriving at a of the benefits of risk management, Edmonds [215] has
more optimal solution [214]. noted that the use of risk analysis provides under-
standing with regards to the consequences of risks to
programme cost and scheduling. However, risk analysis
4.2.4. Uncertainty needs to be first employed during the commercial
The aerospace industry poses substantial difficulties bidding and planning stages when a programme’s price
for the financiers and directors who are trying to develop and duration are being estimated, a range of probability
sustainable products with established in-house capabil- level being attributed to each cost estimate required of
ities and a stable extended supplier base. Changeable the project definition process.
markets and global issues through shifts in emphasis In context, the majority of research carried out into
regarding development, politics, commerce and military risk analysis has been concerned with the combined
action exacerbate this. There is also the continued need effect of an accumulation of uncertainties associated
for product differentiation, cost rationalisation and with the estimates required to estimate a product’s cost.
increased competitiveness, with regard to lead-time, cost This provides a better understanding of the potential
and customer defined quality. This is embodied in the correlation between itemised cost variations and the
European Vision 2020, which sets out cost and efficiency combined effect on the overall distribution [216,217]. As
goals such as a 20–50% reduction of aircraft operating a consequence, risk analysis is being used to alter the
costs in the short to long term, respectively, and 20–50% normal cost/price estimate at an early stage in order to
reduction of aircraft development costs in the short to raise awareness of the sensitivity of the product cost to
long term, respectively; along with substantial reduc- the cost breakdown. This contingency range of values is
tions in lead time. That is set against technological quantified and rated relative to uncertainty and can be
progress and policy, such as reduced impact on the used to guide bidding and planning and ultimately, the
environment through quantitative reductions in emis- product development process. There are a number of
sions and noise, and the requirement for improved statistical methods that are suited to performing this
safety margins and air transport network flexibility and function and software tailored towards risk assessment
service. This drives the industry into higher risk areas of is now more readily available. However, much of the
research and development, forcing them to manage and actual risk assessment within a company is more of a
mitigate that risk accordingly. In addition, the aerospace procedural exercise that is qualitative and not bench-
industry is characterised often as having lengthy project marked.
time scales and extremely high initial investment up One form of a risk model is described by the
front. Stochastic Aggregation Model (SAM) that is based on
This section looks predominately at some of the key a Monte Carlo analysis [218]. The model is essentially a
costing issues to be addressed during the early stages of simulation program that quantifies the uncertainty
product development and definition, where potential associated with parametric cost estimates and it
 ARTICLE IN PRESS
516 R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534

the bidding or concept stage: the more likely the change,

the less accurate the cost estimate.
Assessment: The relationship between the likelihood
of a change in the products cost drivers (the perceived
risk) and the impact on cost needs to be formalised and
quantified. Consequently, the risk, or more accurately
the cost-impact of risk, is quantified by calculating the
probability of an event occurring p (a change in a cost
driver) and the impact that will have on cost c; as
described in Eq. (2). It is important to note that this
Fig. 18. Risk management process. incorporates both the probability of the risk occurring
and its impact on cost:
specifically addresses: (1) uncertainty related to the
magnitude of the independent variable used as a cost Risk ¼ p  c. (2)
driver; (2) uncertainty in rating the complexity factor
used to increase the accuracy and (3) the actual Fig. 19 shows the likelihood of variance, on the y-axis,
statistical uncertainty of the relationship developed. characterised by a triangular probability distribution to
The model has generic application but requires a give the range from minimum variation, to most likely
formalised equation relating inputs that are statistically variation, to maximum variation. Consequently, the
relevant to the output cost estimate. Another example is likelihood of the design variable being within a
given by the RiTo (Risk Tool) model has been designated range needs to also be provided. The
developed by Crossland et al. [219], which was especially magnitude of the variation in cost being represented
developed to deal with the uncertainty experienced along the x-axis is given by multiplying by the coefficient
during the early stages of design. It was based on an associated with that independent variable (as in Eq. (1)),
object-oriented approach and again incorporated a while Eq. (2) can be used to calculate the actual risk by
number of features to model and assesses various risks multiplying this cost variation with the value from the
associated with the estimations. The model was oriented risk assessment. Consequently, the final value for risk
towards decision tooling that could be used by designers grows with variation in cost but also occurrence.
in evaluating the cost impact of conceptual and detailed Analysis: With reference to the previous section, if
definitions; relative to the design space and constraints there is a 50% likelihood of a mass increase (see Eq. (1))
that drives the product’s likely cost base and price range. then Fig. 19 can be used to give the cumulative
Turner [220] has noted five key steps as part of a likelihood of a variation of a given magnitude not
methodology for managing and mitigating risk. These occurring from 50% to 100%. This is shown typically in
are concerned with controlling risk so that the final Fig. 20. This can be completed using a Monte Carlo or
product is as it was envisaged and at a realistic cost and Latin Hypercube simulation. This type of a risk analysis
schedule that were targeted. The procedure is illustrated provides a range of costs and probabilities rather than a
in Fig. 18. Roy et al. [216] has used the core of Turner’s single value as normal. Therefore, one can assume with
model in suggesting a number of ways in which each step some level of confidence that the cost will not exceed a
can be facilitated to produce a number of model types: specific value, typically a threshold of 85% probability
Identification: Risk is driven by the uncertainty being used.
introduced by the inclusion of the independent variables The above methodology was also extended to the risk
selected through the statistical analysis. For example, a analysis for the prediction of the range of the actual
parametric CER could relate cost as a function of both CER. This is particularly relevant when there is limited
weight and surface area, as a result of the statistical data available for the statistical analysis of new high-risk
analysis: product developments. The aim is to provide a predic-
tion of the maximum value of the cost estimate with its
Y ¼ C 0 þ C 1 ðMassÞ þ C 2 ðsurface areaÞ, (1) associated probability occurrence. The initial two stages
where Y is the estimate of the dependent variable; C 0 is
a constant; C 1 is a coefficient associated with mass; and
C 2 is a coefficient associated with surface area. The two
independent variables can be assumed to be potential
sources of uncertainty due to their highly influential
relation with that estimate. Both are likely to change as
the product definition develops and the statistical
Minimum Most Likely Maximum
analysis infers that this will have a significant influence
on the accuracy of the initial cost estimate generated at Fig. 19. Triangular relation adopted.
 ARTICLE IN PRESS
R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534 517

in some pre-determined databank. He points out that

some of these requirements may be conflicting where, for
example, data security issues can often conflict with
interactive mining of multiple-level knowledge from
different angles. A methodology for the mining of cost
data can be defined as follows [221–223]:

1. Data cleaning: to manage noisy, erroneous, missing

or irrelevant entries.
2. Data integration: for the integration of multiple,
heterogeneous data sources.
3. Data selection: to retrieve data that is relevant to the
analysis task.
Fig. 20. Cumulative likelihood of occurrence. 4. Data transformation: for consolidation through
summing or aggregation.
are implemented as before using the data that was used 5. Data mining: where intelligent methods are applied to
for the development of the CER, as shown in Fig. 20. extract data patterns.
6. Data pattern evaluation: to identify the significant
patterns that constitute knowledge.
4.2.5. Data mining 7. Knowledge presentation: visualisation and represen-
All cost modelling is ultimately based on the mining tation for the user.
and analysis of data [63] which is a form of Knowledge
Management. Liao [221] has classified other forms of According to Sorensen [223], two general types of data
knowledge management technology as: knowledge mining approaches exist: (1) knowledge and (2) predic-
management frameworks, knowledge-based systems tion discovery. Prediction discovery identifies causal
(KBS), information and communication technology relationships between certain fields (parameters) in the
(ICT), artificial intelligence (AI), expert systems, data- database. These relationships are established by finding
base technology, and modelling. Data mining is an predictor variables that model the variation of other
interdisciplinary field that Chen [222] describes as a independent variables. If a causal relationship has been
process of non-trivial extraction from databases of established, action can be undertaken to reach a specific
implicit, previously unknown and potentially useful goal such as cost reduction. Knowledge discovery
information, such as rules, constraints, and regularities. problems are usually associated with the stage prior to
This therefore can be used to facilitate decision-making, prediction, where information is insufficient for predic-
problem solving, analysis, planning, diagnosis, detec- tion. Sorensen [223] states that data mining techniques
tion, integration, prevention, learning, and innovation. can be characterised according to the kind of knowledge
Liao [221] notes that quantitative methods for exploring to be mined, which for costing includes: association
the issues of knowledge discovery, knowledge classifica- rules, characteristic rules, classification rules, discrimi-
tion, knowledge acquisition, learning, pattern recogni- nate rules, clustering, evolution, and deviation analysis.
tion, artificial intelligence algorithms, and decision In particular, data classification is a process that finds
support are the modelling technology of knowledge the common properties among a set of objects in a
management. database and then classifies them into different classes,
In conducting effective data mining, Chen [222] has also referred to as clustering [222], whether for the
highlighted the need to first examine what kind of
grouping of physical or abstract objects into classes of
features an applied knowledge discovery system is similarity. This was a technique that was already being
expected to have and what kind of challenges one may advocated in the 1950s by the Rand Corporation when
face at the development of data mining techniques. This they grouped aircraft into clusters of a similar type in
includes: the handling of different types of data; the order to increase the predictability of the CERs [58].
efficiency and scalability of data mining algorithms; the
usefulness, certainty and expressiveness of data mining
results; the expression of different kinds of data mining
results; the interactive mining knowledge at various 5. State of the science: genetic causal cost theory
levels of abstraction; the mining of information from
different sources of data; and the protection of privacy 5.1. State-of-the-art
and data security. The term data mining is increasingly
being used to describe the process of extracting From the assessment of cost modelling techniques in
probabilistic characteristics from a mass of data held Section 4 it is evident that there is no consolidating
 ARTICLE IN PRESS
518 R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534

theory on which the models are based and indeed there has been observed and recorded that the cost architec-
seems to be many different types of model. Anthologies ture, or cost breakdown structure, can be organised
often include many recognised but yet inconsistent according to factors that give rise to cost whether due to:
classifications. For current purposes, it is helpful to first activity performed, resources utilised, parts assembly,
separate costing methodologies into two specific func- product life cycle stages, or part design features.
tional classifications: (1) compilational costing: aggre- However, these are all types of compilation methods
gating various identified costs; and (2) relational costing: and each framework requires additional techniques to
comparative relation of product defining parameters. supply the actual cost estimates they refer to. This is true
The first category represents the compilational method also of Feature-Based Modelling, which is more often
of modelling cost within a designated cost breakdown associated with the second category yet requires some
structure and includes: functional technique that enables it with the capability
to estimate the actual costs it requires for each feature.
 Activity-based costing (ABC): assigning costs to each On the other hand, scenario-based costing is more
activity performed. ambiguous and undefined in terms of which aspect of
 Absorption costing: assigning cost according to the life cycle is being considered, and to what aspect it
resources utilised. refers. Notwithstanding, all those listed have factual
 Bottom-up costing: accumulating cost from the BOM relevancy and address costs that arise due to some
and Work Breakdown Structure (WBS). element of causation relevant to the application. They are
 LCC: attributing costs to all stages of the life cycle all functional driven and have a technological nature.
from ‘womb to tomb’ The distinction between causal and non-causal
Scenario-based reasoning: a subset projecting and foundation becomes much more acute when applied to
forecasting future product scenarios, inclusive of the second category: relational costing techniques. The
market. only relational method that is intrinsically founded on a
 Feature-based costing: attributing cost to geometric causal basis is physical process modelling. An example
part features. of this would be a cost model for a machining process
that is based on cutter speed, feed rate, etc., and
The second category represents the relational method of therefore, may be based on the modelled usage of
linking cost to one or more attributes to form discrete material and time. However, the other methods listed do
associations and includes: have varying degrees of causality, although in all cases it
must be explicitly enforced. For example, parametric
 Physical process modelling: focusing on the time models do not intrinsically require that causal cost
required to carry out work. drivers (as independent variables) be used but that
 Parametrics: stochastic relations within product explicit distinction could be used as a desirable attribute
classes. when identifying the cost drivers for the parametric cost
 Neural nets: learnt mapping of attributes to cost. estimating relations. Neural network models seem to be
 Analogous costing: using precedent at product level. the least causal as the technique operates to a large
Case-based reasoning: a subset using precedent at degree as black box, the neurons learning how to map
detailed level. cost to independent variables given a databank of
 Fuzzy logic: interpolating along established cost historical data, in order to replicate the result. The
functions. network can be designed to a degree while the
 Financial modelling: using mathematical series for independent variables can be chosen for their causal
cost variance. relation to cost, even although it is likely that there will
be very little insight that can be used to facilitate the
The above provides a categorisation that is based on the choice-dilemma of engineering decision making.
basic nature of the method and as such, the distinction is Looking at the current state of the art in cost
more technological and relates to their industrial use. modelling in general the following observations can be
However, science is concerned with the causal founda- made:
tions for each. The importance of the scientific basis of
the modelling method will be expanded in the following
section because of its role as a key differentiator in  The major effort is directed towards estimating costs
assessing the engineering understanding on which each rather than first developing a causal understanding
method relies. Understanding greatly increases the that is a basis for that modelling:
flexibility, usefulness, robustness, and accuracy of any function over foundation!
engineering model.  Modelling is directed towards a particular element of
It is clear that most of the compilational costing cost but is not mindful of the holistic cost architecture:
methods do have a strong causal basis. In each case it micro over macro!
 ARTICLE IN PRESS
R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534 519

 Modelling is directed towards a particular stage in the understood and bounded. Unfortunately, that under-
cost life cycle and is not mindful of the holistic cost standing is the very quality that is often suppressed in
structure: following a stochastic technique that expresses a casual
inhibiting over inheriting! relationship rather than a causal one. Non-causal
 Methodologies are based primarily on a mechanistic relations can be used out of context with unclear
approach and not a causal truth: boundary limits being set and there can be little
casual over causal! appreciation for their total inability to deal with
 Modelling is product specific rather than generic: anything that has not been instrumental in their
generative over genetic! formulation.
 Costing is experience based rather than scientific: This issue of application and relevancy is the main
experience over experiment. functional limitation of non-causal models. A second
more fundamental limitation is the near total lack, or at
Many of the above points highlight the negative best incidental inclusion, of understanding regarding the
scientific basis that surrounds engineering cost model- reason for cost behaviour. Consequently, such relations
ling and the lack of a consolidating theory that can are severely limited and cannot be used readily in
establish its fundamental basis. The discipline is made making decisions regarding product definition and
more difficult to define because it has such a breadth of development; remembering that the weight-cost relation
relevancy and application and has both qualitative and would encourage the designer to always choose the
quantitative aspects. Notwithstanding, the basis of all lightest option. This states that such an approach will
scientific thought and theory is built on the principle of always result in the lowest cost, regardless of the certain
understanding and the modelling of cause and effect direct costs, perhaps having to remove more of a
relations. Consequently, the following section will look material that is likely to be more expensive in its raw
at the importance of causality in this respect. form due to its higher structural efficiency. Although
Bertrand Russell once stated that in terms of the
philosophy of science, ‘‘[the] law of causality... is a relic
5.2. Causation of a bygone age’’ [226], the physical world and its
relation to cost can only be understood truly in terms of
It is easier to first begin with non-causal modelling causal understanding [227,228].
and to say that good examples of such models can be The need for a causal approach to modelling is
extremely useful in estimating the likely behaviour of founded on a few basic intentions that are summarised
cost as the dependant variable. These models should according to Cowan and Rizzo [229]: ‘those that render
conform to the covering-law of Hempel and Oppenheim the overall explanatory structure complete, and those
[224], which gives validation to the explanation of a that make it more nearly correct’. Primarily, complete-
phenomenon if that phenomenon is subsumed under ness helps show that which drives outcomes and
some general formulation of regularity [225]. The ideal secondly, it also helps formulate guiding principles and
gas law is an example of this, where pressure, volume, useful rules. These are linked in providing a more full
temperature and quantity of matter are all incorporated explanation that can be developed into a predictive
into an expression of repeatable consistency. Non-causal model for engineering purposes. On the other hand
models highlight general trends at a higher level with correctness is also a necessary attribute that will provide
little thought to abstraction and therefore, are suited to greater insight and detail. This will lead to more robust
an appreciation of the likely systems’ behaviour, or in modelling that is based on the correct causal relations
this instance, the cost of complex products. Such models and which gives a more useful understanding of the
tend to be of simple formulation and are therefore easy influence certain parameters wield. Correctness will
and quick to deploy and maintain, thereby facilitating distinguish between a coincidence (possibly statistical)
the immediate engineering task at hand of estimating and result (causal). A more thorough understanding of
cost. A potent example is the infamous relation within causation will be based on completeness and correctness
aerospace of product cost as a function of weight. The and will therefore result in an improved predictive
weight and unit cost relation do show a remarkable capacity.
degree of statistical significance and indeed there is a Cowan and Rizzo [229] have also noted that the
partial truth in the proposition that heavier things tend existence of causation is also highlighted by: (1)
to be larger in size and in turn cost more. However, the purposeful endeavour; and (2) the time span between
aerospace industry has always been striving at great cost cause and effect. The purposefulness is an obvious but
and effort to reduce weight in order to reduce the area important aspect as it points to doing something to
required of lifting surfaces and ultimately, the fuel burn. instigate change and produce something new. With
The scientific proposition is disproved although it has a purpose is associated worth and therefore, this has given
range of limited usefulness that needs to be well rise to the monetary value attributed to such products.
 ARTICLE IN PRESS
520 R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534

The second aspect, of time, is also of fundamental (g) Cost will be inherited by a derivative version of
significance as it introduces the concept of a process that parent product.
sees something converted from material state A to
material state B. Menger noted this already back in 5.4. Relevancy of genetic causal cost modelling
1871: ‘‘The idea of causality... is inseparable from the
idea of time. A process of change involves a beginning It has been established that there is no recognised
and a becoming, and these are only conceivable as scientific method of cost modelling and little common-
processes in time’’ [230]. The consequence of this is that ality between the various models. Technological cost
time is a fundamental characteristic of a cause and a models are based on a wide range of principles and
process and therefore, anything that has a time-span methodologies and have been devised for a wide range
associated with it has a causal nature. This is especially of applications. The review of current modelling
relevant to labour costs and it can be concluded that cost techniques raised the issue of causality. It was proposed
can indeed have a scientific basis. that this is fundamental in terms of establishing a
To summarise there is evidence of: the shortcomings scientific understanding that is both more complete and
of non-causal models; a fundamental scientific nature to more correct. This will then provide a better basis for
costing that causal models should exhibit; the need for engineering models that are more robust and accurate.
causal models that encompass our current experience The additional concept of adopting a genetic scientific
and understanding; the need for recognition of the key basis was then addressed. This is especially relevant to
attributes of completeness and correctness. All these modelling as it provides a potential scientific basis and
form the basic tenements on which the genetic causal generic framework for developing any analysis. Specifi-
approach to cost modelling is based, as suggested in this cally, and with respect to the previous points, it identifies
paper. that:

5.3. Genetic nature  Cost should be primarily viewed as a design attribute

of a product, i.e. it is a variable or parameter that is
It has been noted that there is evidence of a genetic designed into a product.
nature within economics, which is described as the  Cost originates primarily within the product defini-
tendency for economic processes to be unidirectional: tion and therefore is primarily determined at the
the outcome of which is the effect [231]. The importance design stage.
of the descriptor ‘genetic’ relates also to the causal  Cost is also influenced by secondary environmental
nature and the observation that there is origination, the factors such as economics (supply and demand) and
process being unidirectional from some start-point. technology.
Genetic nature would be more appropriately defined as  Cost can be broken down into hierarchical groupings
the evidence of the same recurring prime drivers that can that have their own distinct influence or nature.
be assigned as the causal advent of cost. This is a  Fundamentally, cost is caused by a small number of
powerful proposition that is underscores the concept of base cost drivers: materials, time, and energy.
a Cost Gene; like the analogous genealogy within the  It is the manner in which these base cost drivers are
natural world. This implies that product cost is a formulated which dictates the cost categorizations.
function of certain building blocks that determine the  The genetic nature of cost gives rise to the concept of
resultant cost make-up. These can be viewed as universal inheritance, where cost can be passed down to
cost drivers that have some absolute nature that does derivative versions (or derivative features) of the
not change. There are a number of observations that one parent product.
can make regarding the analogy between natural Materials are converted by human endeavour from a
genetics and the cause of cost: raw state to a manufactured form through the use of
devised processes. This ability is facilitated by techno-
(a) Cost is an attribute of a product. logical know-how that results in a primary cost that is
(b) Cost has physical causes. determined by the product definition. This primary cost
(c) Cost is not fixed but is influenced by the economic then becomes some marketed product that is influenced
scenario. by its environmental. However, underlying all of these
(d) Cost can be broken down into a number of distinct are the fundamental base cost drivers of material
categories. availability, labour time and energy utilisation, although
(e) There is a small number of discrete primary cost it is equally important to establish the hierarchical
drivers that are building blocks for all the higher structure of cost in order to structure this theory into
level cost groupings. a useful framework that can be used as a scientific
(f) The sequencing of these quantities gives rise to cost. basis for cost modelling. Production costs are often
 ARTICLE IN PRESS
R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534 521

categorised according to the procurement, labour, and gineering design, there are two key aspects that are seen
capital costs, and investment associated with producing to consistently relate to cost: form (or geometric
engineering designs. However, in the development of a definition) and the relation of production processes to
science of cost, the causal and genetic principles of materials. It is also evident that there are a number of
origination can be used to formulate some basic rules ways in which to quantitatively formulate relations but
that underwrite the subject matter. Cost has been that statistical significance is a fitting manner in which to
explicitly linked to product definition and therefore, formulate relations that are sensitive to environmental
design-oriented rules might include: noise but yet characterised by certain generic aspects,
typically relating to design information. The genetic-
1. The required material characteristics affect produc- causal approach is proposed as a valid scientific
ability: part cost increases as the amount of design approach to the modelling of manufacturing cost, as
information increases, for constant process capabil- arising from the work done in converting a raw material,
ity. through a number of stages, into a part that may then be
2. Assembly cost increases as part complexity and part assembled into a product.
number increase, for constant process capability. It is proposed that manufacturing cost is modelled
3. Production cost increases as tolerance is tightened. using a new methodology referred to as the genetic-
4. The design process results in a non-recurring cost. causal method. This is achieved by

Secondly, complimentary rules would be more oriented 1. Classifying the generic cost elements that are linked
towards the production of the designs and could include: to particular genetic indicators, according to product,
life cycle phase or process.
5. Materials cause cost through labour, capital equip- 2. Developing parametric relations that link the manu-
ment and market ‘supply and demand’. facturing cost to design attributes within each of the
6. Part forming processes cause cost through labour, identified genetic families.
capital equipment and wastage.
7. Assembly processes cause cost through labour,
This is illustrated conceptually in Fig. 21. In proceeding
materials (gigs and tools), capital equipment and
with a hierarchical design-oriented classification there
wastage.
are three key aspects that can be considered as genetic,
8. Unit production cost depends on both recurring and
cost being a result of design definition. The relevant
non-recurring costs.
information from these three aspects can be thought of
9. Unit production cost decreases with number of
as bits of genetic information that are coded into the
units, learning and ‘economies of scale’.
design and which give rise to cost. The actual cost
Finally, the overriding law of economics applies: however, is only fixed if all things remain equal.
Otherwise, environmental factors such as rates, interest
10. All costs are adjusted by environmental equilibrium and technology vary while process cycle indexes will
through the law of ‘supply and demand’. vary relative to Company efficiency. Therefore, any
scientific cost prediction really is truly termed an
estimate as the prediction is the most likely potential
In summary, there is a hierarchical framework:
cost given (1) the nature of the pure design and (2) the
environmental factors that could influence in the
 The basic resources of materials, labour time and production domain.
energy are the fundamental building blocks of cost. The aerospace application presented in the following
 The product definition is the primary cost driver and section is for stringer-skin panels that make up the
imbues cost into a design. aircraft fuselage. With this application in mind, the
 The production process is the subsequent cost driver genetic-causal method utilises the following drivers and
and actualises that propensity to have cost. hierarchy:
 The environmental market scenario will drive design 1. Form—the required shape: the classification accord-
effort towards an equilibrium that is dictated by ing to form or geometric similarity is crucial for linking
supply and demand. manufacturing cost into the design definition process.
This may also include additional form definition in
terms of identified features or increased fidelity ratings
5.5. Application relating to detailed design information; such as through
complexity factors. It will be seen in the case study
Although there is not an established theory to the presented in the following section that a first-order
scientific modelling of manufacturing cost within en- classification is imposed to identify: skin, stringer,
 ARTICLE IN PRESS
522 R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534

Design
attribute
Part Geometric
Cost families1-n => count shape

• Product
• Phase Cost family 1
Fastener
• Process Weight
count

Size Features

Fig. 21. Conceptual illustration of the genetic causal cost modelling approach.

frame, cleat and rivet as forms within the skin-stringer It can be seen from the above three aspects that design
application, while a second-order classification of light- information is absolutely fundamental to the under-
ening-hole is used in conjunction with the Frame Form standing of manufacturing cost, according to the genetic
to improve the resolution of design information. causal cost coding imposed by the designer through the
2. Material—relative to the required behaviour: The impact of their decisions on form, process and material.
choice of material is associated with the required The impact of environmental noise has also been
behaviour of the parts but is strongly coupled to process included in tempering the casual impact of form, process
selection. Producers may preference a process and then and material. This justifies these causal relations being
work to satisfy material requirements; for example, modelled using statistical significance with appropriate
developing stringer alloys that can be welded; although normalisation for the environmental factors. This results
it is recognised that the material categorisation con- in scientifically based relations that formally link cost to
tributes to both the raw material and treatments costs. their causal sources embedded in the design definition.
This is a function of the material quantities required by Apart from being a highly generic cost modelling
the design Form but it is also coupled to the process type technique, the genetic-casual technique is also inherently
in terms of material addition or material removal. A suited to use within an integrated design platform as
further complication with materials procurement is the changes to the design for performance benefit are
degree of pre-processing, such as rolling, forming or mapped to cost. Such interactions can now be directly
the extrusion of the stringer lengths. This need not affect traded off relative to some global objective function, as
the costing accuracy significantly but does impact on the exemplified in the following section with a case study.
practical implementation of the trade studies, within the
context of the design process. However, the addition of
bought-out and subcontracted items does require a 5.6. Genetic causal case study
procurement factor.
3. Process—the available material conversion route A preliminary case study of the application of the
(MCR): the classification of physical form can then be genetic causal cost modelling approach has been carried
matched to potential available processes that can out [232], the study being based on an empirical case
achieve the Form identified. There are two aspects to carried out in conjunction with Bombardier Aerospace
this: (1) understanding the various process stages, (2) Shorts. The main aim was to provide a manufacturing
understanding each of those processes. The significant cost model based on the theory, and then to link this to a
stages in the production cycle are identified through the structural analysis in order to show that detailed
definition of a material conversion route (MCR), after engineering design can be driven by such a modelling
which individual process models can be assigned to each technique to minimise the Direct Operating Cost to the
stage. At this stage, cycle time factors and established customer. Therefore, it explicitly links customer require-
rates need to be introduced to characterise the processes ment and affordability to the design process. The
relative to influential geometric information. For exam- application focused on the design of a traditional
ple, it will be seen that the form: stringer and feature: T- metallic fuselage panel but could be applied to more
shape is first used to classify the stringer riveting, after advanced processes such as laser welding of stringers or
which the cost is predicted using the design length of friction stir welding of panels, or to different materials
stringers in conjunction with a process performance rate such as carbon composites or metal fibre laminates such
and its cost rate. as GLARE. A semi-empirical numerical analysis using
 ARTICLE IN PRESS
R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534 523

ESDU reference data [233,234] was coupled to model

the structural integrity of these thin-walled metal Assembly
Fabrication
35%
structures with regard to material failure and buckling: 37%
the latter including skin buckling, stringer buckling,
flexural buckling and inter-rivet buckling. The optimisa-
tion process focuses on the minimisation of DOC as the
objective, being a function of acquisition cost and fuel
burn.
Material
28%
5.6.1. Measured costs
The genetic causal cost modelling methodology Fig. 22. Total cost breakdown.
imposes a breakdown of the cost into a number of cost
elements, including material cost, fabrication cost and
assembly cost; so that cost can be formulated into semi- Frames sub-
assembly Rivets
empirical equations to be linked to the same design 12% 3%
variables as considered in the structural analysis. The
generic product families used on a typical stringer-skin Final riveting
panel are: the panel, which forms the skin of the aircraft; 12%
the stringers and the frames that support it in the
Lay-off Drilling and set
longitudinal and lateral directions respectively; the cleats operations up
that are present at every stringer-frame junction; and the 7% 49%
rivets that fasten the assembly together. The overall Automatic
breakdown in the manufacturing cost analysis is riveting
summarised through Eq. (1), expressed in term of the 13%
identified product families (skin, stringers, frames, cleats Manual riveting
4%
and rivets):
Fig. 23. Actual assembly cost.
X
5
C Panel ¼ C i ¼ C Skin þ C Stringers þ C Frames
i¼1
as they are not part of the structural configuration but
þ C Cleats þ C Rivets , ð1Þ
are added at the end of the estimation process as a fixed
where C Panel is the total cost of the panel and C i the total cost for accuracy. The cost of such supply and
cost for the family i: commercial off-the-shelf (COTS) items is a function of
According to the empirical data provided for the different cost drivers and would require a different
stringer-skin panel, the repartition of material costs, implementation of the genetic causal methodology, for
fabrication costs and assembly costs is shown in Fig. 22. example, relative to manufacturing quality of supply,
It is worth noting that the fabrication costs only include quantity ordered and performance specification. The
the in-house labour costs. This means that for several part family cost breakdown is given in Fig. 24, showing
parts the material costs also include fabrication costs a rivet (33%), then skin (30%), then stringer (18%)
while the rivets are part of the material cost. The total hierarchy.
cost breakdown illustrated in Fig. 22 shows that the
repartition of the three cost elements are almost
equivalent. The assembly or riveting cost has been 5.6.2. Cost prediction
further divided into various causal processes as shown in For each part family identified in Eq. (1) there are two
Fig. 23. The major contribution is from the drilling cost, causal cost components that are modelled as genetic
which also includes the cost linked to the set-up and contributors: the material cost C m i and the labour cost
preparation of the parts. It is interesting to note that the C li ; the latter being subdivided into either fabrication or
cost of the rivets is insignificant relative to the later assembly, where assembly is all the remaining costs after
assembly cost associated with them, thereby highlighting fabrication repartition:
the need for a causal breakdown rather than using
Ci ¼ Cm l
i þ Ci , (2)
higher-level parametrics. The remaining costs account
for the sub-assembly of the frames, the manual and where superscripts m and l denote material and labour,
automatic riveting, the final riveting and the lay-off respectively. The associated cost coefficients were
operations such as cleaning and inspection. Additional determined empirically from the data supplied from
parts such as antennas, lighting or electrical provisions the industrial partner. Each coefficient is computed, for
(totalling 8% of the all-up cost) have not been included each family part and cost element, as an average of the
 ARTICLE IN PRESS
524 R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534

Additional parts
8%
Skin
30%

Rivets
33%

Cleats Stringers
2% 18%
Frames
9%

Fig. 24. Panel cost breakdown. Fig. 25. Section of the panel.

actual cost data found in the WBS spreadsheets. Three

types of coefficients are employed in the equations: the
material coefficient cm i ($/[unit]) and two labour
coefficients for the time factor cli (h/[unit]), which
includes learning, etc., and the wage rate per hour rli ($/
h). The drawing in Fig. 25 illustrates a section of the
panel from which the geometrical data are issued,
including: panel length, width, and thickness; frame
pitch, rivet pitch, and cross-section dimensions; stringer
pitch, rivet pitch and cross-section dimensions.
Fig. 26. Frame design.
It is useful to illustrate the modelling implementation,
for example, to the frames exemplified in Fig. 26. The
frames were manufactured from 2024 T3 aluminium 0.25
alloy and investigations showed that the material cost Data
for the frames should be computed as a function of the Estimates
0.20
Normalised contribution

volume. For tf being the frame thickness, hf the frame

height, l f ; the frame flange length, the volume V f of one
‘C’ shape frame is given by 0.15

V f ¼ ðð2l f þ hf Þtf  2ðtf Þ2 ÞW . (7)

0.10
Given nframes as the number of frames, r the material
density and cm2024 ($/g), the material cost coefficient for 0.05
the 2024 T3 aluminium, the material cost for the frames
is computed as
0.00
Cm m
frames ¼ nframes V f rc2024 . (8) Skin Stringers Frames Cleats Rivets

The frame labour coefficient clframes (h/hole) was found Fig. 27. Comparison of material costs.
to be a function of the number of lightening holes in the
frames nholes : For rlframes as the frame labour cost per panel is the most significant expenditure. Fig. 28 shows
hour ($/h), the total frames labour holes cost was the breakdown of labour costs for the various product
calculated as families that constitute the stringer-skin panel. It can be
seen that the labour cost associated with the rivets is
C lframes ¼ nframes nholes rlframes clframes . (9)
now significant, as for the stringers. Finally, the overall
Using all of the derived cost relations, the comparison of breakdown of the total manufacturing costs is shown in
the actual and predicted costs for the complete skin- Fig. 29 being the aggregate of Figs. 27 and 28. It is
stringer panel is shown in Figs. 27–29. The cost data and evident that the greatest expenditure is caused by the
estimates have been normalised for proprietary reasons riveting process, the assembly process and the skin being
relative to the total actual cost. Fig. 27 shows the almost 35% of the total cost, while the stringers
breakdown of material costs and highlights that the contribute 20%.
 ARTICLE IN PRESS
R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534 525

0.40 again be simplified and stripped of overheads, con-

Data
0.35
tingency, etc. to be a function of the cost of manufacture
Estimates
for design trade-off purposes. Fuel burn is a function of
Normalised contribution

0.30 the specific fuel consumption (SFC) and the cost of fuel
0.25
and therefore can be said to be a function of weight in
the current context.
0.20 For the purposes of structural optimisation relative to
0.15 DOC, it is simple to use some estimate of the cost of
transporting each unit weight of structure over the life
0.10 span of the aircraft: effectively being a cost per unit
0.05 mass-distance with units of either d/kg km or $/lb m for
example. With respect to the isolation of manufacturing
0.00
cost and structural weight being the key DOC drivers, it
Skin Stringers Frames Cleats Rivets
can be seen that manufacturing cost has a direct relation
Fig. 28. Comparison of labour costs. to the magnitude of DOC/unit mass-distance while
weight is its multiplier. Therefore, one cannot assume to
0.40 use a fixed figure for the DOC estimate within the
Data optimisation process but a more correct weighted
0.35 Estimates
formula that includes the direct relation of manufactur-
Normalised contribution

0.30 ing cost as well as the more obvious one of weight.

Essentially, to optimise according to an objective
0.25
function that only includes a fixed DOC/unit weight-
0.20 distance would lead to the improper assessment of the
minimum manufacturing cost condition; as occurring at
0.15
that point at which the weight corresponds to minimum
0.10 DOC rather than the minimum manufacturing cost
being the decider. This is consistent with literature that
0.05
has stated that minimum manufacturing cost does not
0.00 necessarily correspond to minimum weight [237]. There-
Skin Stringers Frames Cleats Rivets fore, a change in manufacturing cost must be linked
through the impact on both acquisition cost (AC) and
Fig. 29. Comparison of total costs.
fuel burn (FB) (at that associated weight) while a change
in weight is linked through fuel burn alone. The pie
5.6.3. Direct operating cost optimisation chart shown in Fig. 30 shows that a 50% weighting for
The study was ultimately concerned with linking and acquisition cost and 15% weighting for fuel burn is
trading off structural efficiency with manufacturing cost. reasonable for the DOC split for an aircraft of the
Structural efficiency is already a trade-off between regional type; in keeping also with the panel sizing used
maximising material strength utilisation and reducing in the paper. It was found from this basic correlation
weight [235], while manufacturing cost is a trade-off that the manufacturing cost (MFC) needs to be multi-
between specified design requirements (within tolerance) plied by a weighting factor n that truly reflects the cost
and process capability. Eq. (29) highlights that the trade- penalty. This is relative to the factory and company
off can be achieved through the minimisation of DOC: overheads, etc. and would be typically from 2 to 4 times

DOC ¼ fn ðacquisition; fuel burn; maintenance,

crew and navigation; ground servicesÞ. ð29Þ Crew
13% Fuel
However, for the purposes of the structural design trade- 15%
off, all DOC drivers can be said to be fixed apart from Landing fee
the acquisition cost and fuel burn. The neglected 2%
elements can be said to be of much less importance to
Insurance
the structural airframe designer where for example even
3%
airframe maintenance has been estimated by Russell
[236] to be of the order of only 6%; relative to Ownership
Maintenance
subsystems and the power plant. Acquisition cost is 54%
13%
driven by the cost of financing the acquisition cost of the
aircraft, plus a 15% profit margin for example, and can Fig. 30. Life cycle cost breakdown for regional jets.
 ARTICLE IN PRESS
526 R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534

the basic value: The panel was loaded in compression-shear, not

introducing tensile loading and crack propagation, at a
DOC ¼ FB þ AC ¼ FB þ n MFC. (30) structural index value p=LF ¼ 0:5 N=mm2 : This will
result in a relatively low stress level that is appropriate to
The above approach to DOC optimisation has been the design of panel studied. The panel was first
investigated more rigorously for the optimal trade-off optimised for maximum theoretical efficiency Z; which
between aerodynamic tolerances and manufacturing is equivalent to minimising the cross-sectional area of
cost [4,8,96,85,238–242]. In that study, DOC was the skin and stringers to give a maximum efficiency Z ¼
calculated using industry PIANO software while the 0:693: The cost and the total weight of this optimised
manufacturing cost was stochastically modelled as a panel was used as the reference datum for decrease or
function of tolerance allocation and process capability. increase in cost or weight in the subsequent optimisation
Incidentally, it was found that the introduction of approaches. The optimisation was then repeated for
manufacturing cost into the objective design function minimum total weight, minimum material cost, mini-
changed the design definition, and that is again reflected mum total manufacturing cost and minimum direct
in the following results. operating cost. The marginal change in direct operating
For cost-weight optimisation of the panel, a marginal cost with different choice of objective function is
saving in the direct operating cost of the aircraft (i.e. illustrated in the bar chart in Fig. 31 and detailed in
saving directly attributable to the design of the panel) is Table 3; the first column showing the quantity mini-
assumed to be made up of a saving in manufacturing mised, and the other columns the relative change.
cost offset against a fixed cost penalty for any increase in Positive values denote reductions relative to the refer-
structure weight. The fixed cost penalty is a function of ence panel while negative values indicating increase.
the fuel burn with normal utilisation over the useful life It is evident from Table 6 that substantial reductions
of the aircraft, expressed in terms of its all-up weight. A in both weight and direct operating cost are obtained
reduction in manufacturing cost through design itera- when the panel is optimised for minimum total weight,
tion of the panel will result in weight increase and
implies an increase in the cost of fuel consumed.
saving in direct operating cost US $ / m2

Minimisation of this total cost (i.e. manufacturing 4000

cost + fixed cost penalty) is the basis of the optimisation
3500
performed. It should be noted that additional fuel costs
are paid for over the life of the aircraft, whereas 3000
manufacturing costs are met at the outset. A fixed cost 2500
penalty (often referred to as the economic value of
2000
weight saving) of 300 US $/kg was been adopted, this
figure having been adjusted to reflect interest on the 1500
initial investment. 1000
In the optimisation process the structural analysis
500
simply ensures that the panel continues to withstand the
applied loads. Due to the explicit nature of both 0
min min min min
the genetic causal manufacturing cost model and the
W mat mfc doc
structural modelling, it was possible to employ the
simple ‘Solver’ optimisation routine within MS Excel, Fig. 31. Saving in direct operating cost.
which uses a generalised reduced gradient method. The
formulations for the various modes of failure from the Table 6
structural modelling act as constraints in the cost Savings according to the choice of objective
optimisation, together with constraints arising from
the limits of validity for the buckling data and further Panel optimised for Saving in
constraints imposed to reflect practical limits of spacing, W MAT MFC DOC
etc. The weight of the panel, its bare material cost, the
total manufacturing cost (i.e. including material cost) Minimum W 1.60 11 807 2898
and the marginal saving in direct operating cost were Minimum MAT 0.99 36 680 2335
assessed by the objective function. The active design Minimum MFC 2.29 108 1186 2872
variables, which also are genetic links to cost, were Minimum DOC 0.58 26 1122 3539
chosen to be: stringer pitch b; stringer height h; skin
All cost savings are in US $ per m2 of panel, weight in kg per m2
thickness t; stringer thickness ts and rivet pitch rp : The (values in italics indicate an increase). W ¼ total weight;
last was chosen primarily as it is causal in being a major MAT ¼ bare material cost; MFC ¼ total manufacturing cost;
contributor to assembly cost of manufacture. DOC ¼ direct operating cost:
 ARTICLE IN PRESS
R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534 527

Table 7 has been illustrated. Furthermore, this has been shown

Panel dimensions after optimisation to be an appropriate basis for the assessment of the
scientific relevance of the methods presented in the
Panel optimised for H b h t ts rp
literature. Therefore, although no different from the
Efficiency Z 0.693 42.8 27.6 0.85 1.61 31.1 proper basis that would be adopted on a scientific basis,
Minimum W 0.632 71.5 31.0 1.60 1.60 61.6 the genetic causal theory of cost modelling can now be
Minimum MAT 0.628 65.5 27.0 1.09 2.53 41.9 referenced in assessing the balance of modelling applic-
Minimum MFC 0.383 192.3 38.64 2.52 6.07 124.7 ability and fundamental basis.
Minimum DOC 0.517 125.1 28.2 1.97 3.73 83.7 Finally, it is concluded that engineering can be
scientifically modelled and that a consequence of this
All dimensions in mm (values in italics indicate that the limits of
is that it can be integrated into the engineering design
validity of the local buckling data have been reached).
process and promoted to the status of a key design
rather than for maximum theoretical efficiency. This variable. This is a contentious issue for many design
only emphasises the causal importance of including the purists who still adhere to the performance and technical
weight of connections and similar items in the optimisa- specification paradigm but who will be increasingly
tion. Minimisation of material cost and total manufac- marginalised by the age old need for the engineering
turing cost both show improvements with regard to profession to be called on to apply science in meeting the
direct operating cost, even though they induce a weight perceived market need. That now most definitely
penalty. Optimisation for minimum direct operating includes both value and initial cost, as demonstrated
cost rather than for minimal weight shows a further through the culminating case study presented.
improvement of 10% for total DOC. This is a significant
result as much structural optimization is performed
according to a minimum weight goal, it being implicitly References
assumed that this also reduces cost. When optimised for
minimum direct operating cost, it was found that the [1] DTI. Aerospace & Defence Technology Report, 2001/
2002.
ratio of acquisition cost to fuel burn was typically 4:3,
[2] Rubbert PE. CFD and changing world 006Ff aircraft
although a different impact might be expected for other
design. AIAA Wright Brothers Lecture. New York:
panel cases. Finally, it can be seen in Table 7 that the AIAA Publications; 1994.
various criteria for optimisation lead to widely differing [3] Bekker J, Prentice E. The contribution of commercial
panel dimensions. Minimisation of direct operating cost aviation to the economy of manitoba: an economic
leads to a stringer pitch almost triple that of the impact assessment. Asper School of Business, University
theoretical optimum, at the same time more than of Manitoba, 2002.
doubling the skin and stringer thickness and the rivet [4] Kundu AK, Watterson J, Raghunathan S, McFadden R.
pitch. Increased stringer pitch implies a reduced number Parametric optimisation of manufacturing tolerances at
of connecting cleats, and this together with increased the aircraft surface. J Aircraft 2002;39(2).
[5] Murman E, Walton M, Rebentisch E. Challenges in the
rivet pitch leads to substantial cost savings in assembly.
better, faster, cheaper era of aeronautical design, en-
Again, this is significant in the cost modelling proving to
gineering and manufacturing. Aeronaut J (October) 2000.
be a very influential factor in the design process. The [6] Burt DN, Doyle MF. The American Keiretsu, Business
genetic nature of the costing method is of relevance to One. Homewood IL: Irwin; 1993.
the general applicability while the causal nature ensures [7] Rush C, Roy R. Expert judgement in cost estimating:
that design relevance is inherent and quantitatively modelling the reasoning process. Concu Eng: Res Appl
linked for use in decision making facilitated with more (CERA) J 2001;9(4).
recognised modelling. [8] Kundu A, Raghunathan S, Curran R. Cost modelling as
a holistic tool in the multi-disciplinary systems architec-
ture of aircraft design—the next step ‘design for
customer’. 41st AIAA Aerospace Sciences Metting and
6. Conclusions Exhibit, Reno, USA, 6–9 January 2003.
[9] Hart-Smith LJ. On the adverse consequences of cost-
The paper has presented the more established performance metrics usurping the role of goals they were
supposed to support. 21st ICAS congress, Melbourne,
techniques presented in the modelling of aerospace
Australia, September 1998.
costing, while also presented the recognised definition of
[10] Chisholm AWJ. Nomenclature and definitions for
these costs. It has been established that there is no manufacturing systems. Ann. CIRP 1990;39/2:735–42.
consolidating theoretical approach to the domain and [11] Ten Brink E. Costing support and cost control in
consequently, this has been proposed using the so-called manufacturing: a cost estimation tool applied in the sheet
genetic causal approach. Through this proposition, the metal domain. Ph.D. thesis. The Netherlands: Print
inherent genetic and causal nature of aerospace costing Partners Ipskamp, Enschede; 2002.
 ARTICLE IN PRESS
528 R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534

[12] Stewart RD, Wyskida RM, Johannes JD, editors. Cost international society for parametric analysis (ISPA),
estimator’s reference manual, 2nd ed. New York: Wiley; Washington, DC, 2001.
1995. [32] Andreasen M, Kahler S, Lund L. Design for
[13] Wierda LS. Cost information tools for designers, a survey assembly, 2nd ed. Berlin: IFS Publications/Springer;
of problems and possibilities with an emphasis on mass 1998.
produced sheet metal parts. Ph.D. thesis, University of [33] Boothroyd G, Dewhurst P, Knight W. Product design for
Delft, Delft, The Netherlands, 1990. manufacture and assembly, 2nd ed. New York: Marcel
[14] Bode J. Decision support with neural networks in the Dekker; 2001.
management of research and development: concepts and [34] Tibbetts K. An introduction to teamsetTM. CSC manu-
application to cost estimation. Inf Manage 1998;34: facturing. Birmingham, England: Computer Sciences Ltd;
33–40. 1995.
[15] Sohlenius G. Concurrent engineering. Ann CIRP [35] Miyakawa S, Ohashi T. The Hitachi assemblability
1992;41(2):645–55. evaluation method (AEM). In: Newport, Rl Proceedings
[16] Leung ACK, Wainwright CER, Leonard R. The devel- international conference on product design for assembly;
opment of an integrated cost estimation system. Int J April 1986. p. 15–7.
Comput Integrated Manuf 1996;9(3):190–204. [36] Naing S, Burley G, Odi R, Williamson A, Corbett J.
[17] Lutters D, Streppel AH, Kroeze B, Kals HJJ. Adaptive Design for tooling to enable jigless assembly—an
press-brake control in air bending. In: Belfast Proceed- integrated methodology for jigless assembly. Soc Auto-
ings of the fifth international conference on sheet metal: motive Eng 2000;2000-01-1765.
SheMet’97; 1997. p. 471–80. [37] Burley G, Corbett J. Flyaway tooling for higher quality,
[18] Billo RE, Rucker R, Shunck DL. Integration of a group more cost-effective aerostructure. Soc Automotive Eng
technology classification and coding system with an 1998;981843.
engineering database. J Manufact Syst 1987;6(1):37–45. [38] Huang GQ, editor. Design for X: concurrent engineering
[19] Murman E, Walton M, Rebentisch E. Challenges in the imperatives. London: Chapman & Hall; 1996.
better, faster, cheaper era of aeronautical design, en- [39] Krammer J, Sensburg O, Vilsmeier, Journaland Berch-
gineering and manufacturing. Aeronaut J 2000; 481–9. told G. Concurrent engineering in design of aircraft
[20] Slack R. The application of lean principles to the military structures. AIAA J Aircraft 1995;32(2).
aerospace product development process. MIT SM thesis, [40] Dean EB. Parametric cost analysis: a design function.
December 1998. Transactions of the American association of cost
[21] Eakin DJ. Design for six sigma (DFSS). Proceedings of engineers, 33rd annual meeting, San Diego, CA, June
the time-compression technologies conference, Cardiff, 25–28, 1990.
UK, October 10–11, 2000. [41] Meisl CJ. The future of design integrated cost modelling.
[22] Eakin DJ. Aerospace DFMA. In: RI Proceedings Proceedings AIAA/AHS/ASEE aerospace design confer-
international forum on design for manufacture and ence. California: Irvine; 1992, AIAA 92-1056.
assembly; June 8–9, 1998. [42] Marx WJ, Mavis DN, Schrage DP. Cost/time analysis for
[23] Broade J, Pfoertner H, OLMOS in GAF MRCA theoretical aircraft production. J Aircraft 1998;35(4):
Tornado—10 years of experience with on-board life 637–46.
usage monitoring. Proceedings of the 33rd AIAA/ [43] Marx WJ, Mavis DN, Schrage DP. A knowledge-based
ASME/SAE/ASEE joint propulsion conference and system integrated with numerical analysis tools for
exhibit, Seattle, 1997. aircraft life-cycle design. Artif Intell Eng Des Anal Manuf
[24] Unal R, Dean EB. Design for cost and quality: the robust J 1998;12:211–29.
design approach. J Parametr 1991;11(1):73–93. [44] Rais-Rohani M. Manufacturing and cost consideration in
[25] Hirt RJ. Air force design-to-cost methodology develop- multi-disciplinary aircraft design. NASA Grant NAG-1-
ment. In: Los Angeles, CA Proceedings of the 21st 1716, NASA Langley, 1996.
national SAMPE symposium and exhibition; April 6–8, [45] Ostwald P. Engineering cost estimating. Englewood
1976. Cliffs, NJ: Prentice-Hall; 1992, 576pp, ISBN 0-13-
[26] Wu T, O’Grady P. A concurrent engineering approach to 276627-2.
design for assembly. Concu Eng Res Appl September [46] Stewart R, Wyskidsa R, Johannes J. Cost estimator’s
1999;7(3). reference manual, 2nd ed. New York: Wiley Interscience;
[27] Chambers I. Lean design and cost of quality: voids in six 1995.
sigma deployment efforts. Soc Automotive Eng [47] Asiedu Y, Gu P. Product life cycle cost analysis:
2000;2000-01-1730. state of the art review. Int J Prod Res 1998;36(4):
[28] Kusiak A, He D. Design for agile assembly: an 883–908.
operational perspective. Int J Prod Res 1997;35(1): [48] Liebers A. An architecture for cost control, the use of cost
157–78. information in order-related decisions. Ph.D. thesis,
[29] Womack J, Jones D, Roos D. The machine that changed University of Twente, Enschede, The Netherlands, 1998.
the world. New York: MacMillan; 1990 ISBN 0 89256 [49] Schiller B. Essentials of economics, 4th ed. New York:
350 8. McGraw-Hill/lrwin; 2001 ISBN 0072374071.
[30] http://lean.mit.edu/ (last accessed 01/10/2004). [50] Thompson F. Cost measurement and analysis. In: Meyers
[31] DeMarco, Anthony A, Geiser, Todd A. True planning: R, editor. Handbook of government budgeting. San
the next generation of estimating tools. Proceedings of Francisco: Jossey-Bass; 1998. p. 381–411.
 ARTICLE IN PRESS
R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534 529

[51] Cooper R, Kaplan RS. The design of cost management [68] Hughes RT. Expert judgement as an estimating method.
systems; text, cases and readings. New Jersey: Prentice- Inf Software Technol 1996;38:67–75.
Hall; 1991. [69] Wierda L. Design-oriented cost information: the need
[52] Shuford Jr RH. Activity-based costing and traditional and the possibilities. J Eng Des 1(2):146–67.
cost allocation structures. In: Stewart RD, Wyskida RM, [70] Hoult DP, Meador CL, Deyst J, Dennis M. Cost
Johannes JD, editors. Cost estimator’s reference manual. awareness in design: the role of data commonality. SAE
2nd ed. New York: Wiley; 1995. p. 41–94. Technical Paper, No. 960008, 1996.
[53] Humphreys K, Wellman P. Basic cost engineering, 3rd ed. [71] Sheldon DF, Huang GQ, Perks K. Design for cost: past
(Revised and expanded), 1996. experience and recent development. J Eng Des
[54] Park WR. Cost engineering analysis. A guide to the 1991;2(2):127–39.
economic evaluation of engineering projects, 1973. [72] Dean EB, Unal R. Elements of designing for cost. In:
[55] Biemans. A reference model for manufacturing planning February 1992. California, CA: Irvine Proceedings of
and control. Ph.D. thesis, University of Twente, En- AIAA 1992 aerospace design conference; 1992.
schede, 1989. [73] Wood, Michael J. Design to cost. New York: Wiley
[56] Arentsen AL. A generic architecture for factory activity Interscience; 1989.
control. Ph.D. thesis, University of Twente, Enschede, [74] Curran A, Kundu A, Ray A, Woods K, Crosby S,
The Netherlands, 1995. Raghunathan S, Shields P. Aerospace cost estimating for
[57] Wright TP. Factors affecting the cost of airplanes. competitive design for manufacture. Proc Concu Eng
J Aeronaut Sci 1936;3(November 2). 2002; 885–93.
[58] Rand Corporation. Military jet acquisition: technology [75] Curran R, Rush C, Roy R, Raghunathan S. Current cost
basics & cost-estimating methodology. MR-1596, estimating practice in aerospace. Proc Concu Eng Res
2002. Appl (CE2002) 2002; 894–902.
[59] Rush C, Roy R. Capturing quantitative & qualitative [76] Roy R, Jones P. Developing an integrated approach to
knowledge for cost modelling within a CE environment. design and manufacturing cost modelling. In: Lyon,
ISPE international conference on concurrent engineering: France Proceedings of CE2000 conference; 17–21 July,
research and applications, Anaheim, Los Angeles, 2001. 2000. p. 31–9.
Pennsylvania, USA: CETEAM; 2001. p. 209–18. [77] Heinmuller B, Dilts DM. Automated design-to-cost:
[60] Hammaker J. The faster, better, cheaper approach to application in the aerospace industry. Annual Meeting
space missions: a cost analysis perspective. SSCAG’s 69th of the Decision-Science-Institute, San Diego, CA, vol.
meeting, European space agency, Noordwijk, The Neth- 1–3, November 22–25, 1997. p. 1227–9 [chapter 569].
erlands, May 11–12, 2000. [78] Gieger TS, Dilts DM. Automated design-to-cost: inte-
[61] Rush C, Roy R. Expert judgement in cost estimating: grating costing into the design decision. Comput Aid Des
modelling the reasoning process. Concu Eng Res Appl 1996;28(6/7):423–38.
(CERA) J 2001;9(4). [79] Brimson JA, Downey PJ. Feature technology: a key to
[62] Roy R, Bendall D, Taylor JP, Jones P, Madariaga AP, manufacturing integration. CIM Rev 1986;2(3):21–7.
Crossland J, Hamel J, Taylor IM. Development of [80] Bashir HA, Thompson V. Estimating design complexity.
airframe engineering CER’s for aerostructures. Proceed- J Eng Des 1999;10(3):248–57.
ings of the second world manufacturing congress [81] Cooper R, Kaplan RS. Measure cost right: make the right
(WMC’99), 27–30 September, Durham, UK, 1999, decisions. Harvard Bus Rev 1988;66(5):96–103.
p. 838–44. [82] Cokins G. ABC can spell a simpler, coherent view of
[63] Wahl MG, Ambler T, Maab C, Rahman M. From DFT costs. Computing Canada, September 1, 1998.
to systems test—a model based cost optimisation tool. [83] Taylor IM. Cost engineering—a feature based approach.
Texas at Austin and Southwest Texas State: Universities 85th Meeting of the AGARD Structures and Material
of Siegen; 2000. Panel, Aalborg, Denmark, October 13–14, No. 14, 1997.
[64] Roy R, Palacio A. Cost estimating and risk analysis in p. 1–9.
manufacturing processes. Proceedings of MATADOR [85] Curran R, Kundu A, Raghunathan S, Eakin D. Costing
2000 conference, 13–14 July. Manchester: UMIST; 2000, tools for decision making within integrated aerospace
p. 177–82, ISBN 1-85233-323-5. design. J Concu Eng Res 2002;9(4):327–38.
[65] Rush C, Roy R. Analysis of cost estimating processes [86] Smith AE, Mason AK. Cost estimation predictive
used within a concurrent engineering environment modelling: regression versus neural network. Eng Econ
throughout a product life cycle. Seventh ISPE interna- 1997;42(2):137–62.
tional conference on concurrent engineering: research and [87] Villarreal JA, Lea RN, Savely RT. Fuzzy logic and neural
applications, Lyon, France, July 17–20. Pennsylvania, network technologies. In: Houston, TX 30th aerospace
USA: Technomic; 2000. p. 58–67. sciences meeting and exhibit; January 6–9, 1992.
[66] Curran R, Watson P, Cowan S, Mahwinney J, Raghu- [88] Bode J. Neural networks for cost estimation. Amer Assoc
nathan S. Development of an aircraft cost estimating Cost Eng 1998;40(1):25–30.
model for program cost rationalisation. In: Montreal [89] Roy R, Bendall D, Taylor JP, Jones P, Madariaga AP,
Proceedings of the Canadian aeronautics and space Crossland J, Hamel J, Taylor IM. Development of
institute (CASI); April 2003. airframe engineering CERs for military aerostructures.
[67] Boehm BW. Software engineering economics. IEEE In: Durham, UK Second world manufacturing congress
Trans Software Eng 1984;10(1):7–19. (WMC’99); 27–30 September 1999.
 ARTICLE IN PRESS
530 R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534

[90] Department of Defence (DoD). Parametric estimating [108] Noton BR. Cost drivers in design and manufacture of
handbook, 2nd ed. DoD, 1999, http://www.ispa-cost.org/ composite structures. Compos Struct Anal Des 1987,
PEIWeb/cover.htm (last accessed 01/10/04). 419–28.
[91] Pugh P. Working top–down: cost estimating before [109] Noton BR. Cost drivers and design methodology for
development begins. J Aerospace Eng G 1992;206: automated airframe assembly. In: Proceedings of 31st
143–51. international SAMPE symposium; April 7–10, 1986.
[92] Thurston DL, Essington SK. A tool for optimal p. 1441–55.
manufacturing design decisions. Manufact Rev [110] Ermanni P, Ziegmann G. Cost-efficiency of highly
1993;6(1):48–59. integrated fuselage structures-comparison between metals
[93] Keeney RL, Raiffa H. Decisions with multiple objectives: and composites. In: Proceedings of SAMPE advanced
preferences and value trade-offs. New York: Wiley; 1976. materials: cost effectiveness, quality control, health and
[94] Vanderplaats GN. Numerical optimization techniques for environment; 1991. p. 347–59.
engineering design: with applications. New York: [111] Hicks C, McGovern T, Earl CF. Supply chain manage-
McGraw-Hill; 1984. ment: a strategic issue in engineer to order manufactur-
[95] Collopy PD, Eames DJH. Aerospace manufacturing cost ing. Internat J Product Econ 2000;65(2):179–90.
prediction from a measure of part definition information. [112] Humphreys P, Mclvor R, Huang G. An expert system for
Warrendale, PA: SAE Publications; 2001. evaluating the make or buy decision. Comput Ind Eng
[96] Kundu AK, Raghunathan S, Curran R, Cather G. Cost 2002;42(2–4):567–85.
modelling as a holistic tool in the multidisciplinary [113] Dooley K. Purchasing and supply—an opportunity for
systems architecture of aircraft design—the next step OR? Or Insight 1995;8(3):21–5.
‘design for customer’. 41st AIAA Aerospace Sciences [114] Williamson OE. Markets and hierarchies. New York:
Meeting and Exhibit, USA, 6–9 January 2003, Reno, NV. Free Press; 1975.
[97] Butterfield J, Yao H, Curran R, Price M, Armstrong CG, [115] Williamson O. The economics of organization: the
Raghunathan S. Integration of aerodynamic, ‘structural, transaction cost approach. Am J Sociol 1981;87:548–77.
cost and manufacturing considerations during the con- [116] Fine C. Is the make-buy decision process a core
ceptual design of a thrust reverser cascade’. AIAA Paper competence? MIT Centre for Technology, Policy and
2003. 42nd AIAA Aerospace Sciences Meeting and Industrial Development, 1996.
Exhibit, 5–8 January 2004. [117] Blaxill M, Hout T. The fallacy of the overhead quick fix.
[98] Marx WJ, Mavris DN, Schrage DP. Effects of alternative Harvard Bus Rev 1991; July–August:93–101.
wing structural concepts on high speed civil transport life [118] Probert D. The practical development of a make or buy
cycle costs. Conference proceedings of 37th AIAA/ strategy: the issue of process positioning. Integr Manuf
ASME/AHS/ASC structures, structural dynamics, and Syst 1996;7(2):44–51.
materials conference, Salt Lake City, UT, April 15–17, [119] Fitzgerald KR. Best practices in procurement. Ascet, vol.
1996, Paper No. AIAA-96-1381. 4, Ascet—Achieving Supply Chain Excellence Through
[99] Roskam J. Airplane design. Roskam aviation/engineering Technology, 2002
company, vol. 8, 1985 [120] Handfield RB. Avoid the pitfalls in supplier management.
[100] Westphal R, Scholz D. A method for predicting direct Sloan Manage Rev January 2002.
operating costs during aircraft system design. Cost Eng [121] Hicks C, McGovern T, Earl CF. Supply chain manage-
1997;39(6):35–9. ment: a strategic issue in engineer to order manufactur-
[101] Boothroyd G, Dewhurst P. Product design for assembly. ing. Int J Prod Econ 2000;65(2):179–90.
Wakefield, Rl, USA: Boothroyd Dewhurst; 1990 (first [122] Fu Y, Piplani R. Supply-side collaboration and its value
edition published in 1983). in supply chains. Eur J Oper Res 2002;152(2004):
[102] Stoll HW. Design for manufacture: an overview. Appl 281–8.
Mech Rev 1998;39(9):1356–64. [123] Lockamy A, Smith W. Target Costing for supply chain
[103] Bloom HM. Design for manufacturing and the life cycle. management: criteria and selection. Ind Manag Data Syst
In: Proceeding of the NSF design theory 88 conference; 2000;100/5:210–5.
1998. p. 302–12. [124] Giunipero L, Brand R. Purchasing’s role in supply chain
[104] Murman EM, Walton M, Rebentisch E. Challenges in the management. Int J Log Manag 1996;7(1):29–38.
better, faster, cheaper, era of aeronautical design, [125] Marquez A, Blanchar C. The procurement of strategic
engineering and manufacturing. Aeronaut J 2000; 481–8. parts. Analysis of a portfolio of contracts with suppliers
[105] Rais-Rohani M. A framework for preliminary design of using a system dynamics simulation model. Int J Prod
aircraft structures based on process information. NASA Econ, in press. (Corrected Proof, available online 23 July
Grant NAG-1-1716, 1998. 2003.)
[106] Rais-Rohani M, Greenwood AG. Product and process [126] Narasimhan R, Mahapatra S. Decision models in global
coupling in multidisciplinary design of flight vehicle supply chain management. J Ind Market Manag
structures. Seventh AIAA/NASA/USAF/ISSMO sympo- 2003;5598.
sium on multidisciplinary analysis and optimisation, [127] Dubois A. Strategic cost management across boundaries
September 2–4, St. Louis, MO, USA, Paper No. AIAA of firms. J Ind Market Manag 2003;32:365–74.
98-4820, 1998. [128] Mills P, Jones R, Sumiga J. The integration of cost
[107] Sandoz PL. Structural design of future commercial knowledge and uncertainty into expert systems for
transports. Paper No. AIAA 73-20, 1973. engineering design. In: Fourth Eur conference on
 ARTICLE IN PRESS
R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534 531

automated design. UK: Kempston Publications; 1987. Ireland. In: Cranfleld University, July 28–31. Ninth ISPE
p. 443–52. international conference on concurrent engineering:
[129] Jackson P. Introduction to expert systems. Reading, MA: research and applications. Netherlands: A.A. Balkema
Addison-Wesley Publishing Company; 1986 ISBN: 0-201- Publishers: 2002. p. 849–59.
14223-6. [148] FAST. FAA pricing handbook. The federal aviation
[130] Kingsman B, Souza A. A knowledge-based decision administration acquisition system toolset, 1999, accessible
support system for cost estimation and pricing decisions from: http://www.fast.faa.gov/index.htm (last accessed
in versatile manufacturing companies. Int J Prod Econ 01/10/04).
1997;53(2):119–39. [150] Ferens V. Parametric estimating: past, present, future. In:
[131] Hughes R. Expert judgement as an estimating method. Cambridge, England PRICE systems 19th European
Inf Software Tech 1996;38:67–75. users symposium, price across the enterprise; 20–21
[132] Shepperd M, Schofield C, Kitchenham B. Effort estima- October 1999.
tion using analogy. In: Berlin Proceedings of the 18th [151] Crawford G, Williams C. The analysis of subjective
international conference on software engineering; 1996. judgement matrices. USA: The Rand Corporation; 1985
p. 170–8. ISBN 0-8330-0639-8.
[133] Shepperd M, Schofield C. Estimating software project [152] Miranda E. Improving subjective estimates using paired
effort using analogies. IEEE Trans Software Eng comparisons. IEEE Software 2001;18(1):87–91.
1997;23(12):736–43. [153] Tuer G. ICM (Integrated Cost Modelling) business case.
[134] Bashir H, Thompson V. An analogy-based model for BAE SYSTEMS, Warton, Preston, UK, Internal pre-
estimating design effort. Des Stud 2001;22(2):157–67. sentation, 2002, unpublished.
[135] Cowderoy A, Jenkins J. Cost-estimation by analogy as a [154] Tuer G. Integrated cost modelling at BAE SYSTEMS.
good management practice. Software Engineering 88, SSCAG’s 69th meeting, European Space Agency, Noord-
Second IEE/BCS conference, 1988. p. 80–4. wijk, Netherlands, 11–12 May 2000.
[136] Tessem B, Modeling S. Analogy and complex software [155] Taylor I. Cost engineering: a feature based approach.
modelling. Comp Human Behav 1997;14(4):465–86. 85th Meeting of the AGARD structures and material
[137] Klein G. Applications of analogical reasoning. Metaphor panel, Aalborg, Denmark, 1998. p. 1–9.
Symbol Activity 1987;2(3):201–18. [156] Robert S. Inflation conversion factors for dollars 1700 to
[138] Klein G. Sources of power: how people make decisions. estimated 2010, Oregan State University.
Massachusetts Institute of Technology, 1998, ISBN: [158] The Rand Corporation. USA, ISBN 0-8330-0639-8.
0-262-11227-2. [159] Akintoye A, Fitzgerald E. A survey of current cost
[139] Boehm B. Software engineering economics. Englewood estimating practices in the UK. Constr Manag Econ
Cliffs, NJ: Prentice-Hall; 1981. 2000;18(2):161–72.
[140] Kadoda G, Cartwright M, Chen L, Shepperd M. [161] Gammack J, Jenkins D. Learning from design histories in
Experiences using casebased reasoning to predict soft- concurrent engineering. Comput Ind 1997;33(1):83–90.
ware project effort. Conference on empirical assessment [162] Beltramo M. Beyond parametrics: the role of subjectivity
in software engineering (EASE), Keele University, 2000. in cost models. Elsevier: Eng Costs Prod Econ: Int J
p. 1–23. Industry 1988;14(2):131–6.
[141] Duverlie P, Castelain J. Cost estimation during design [163] Kitchenham B, Pfleeger S, McColl B, Eagan S. A case
step: parametric method versus case based reasoning. Int study of maintenance estimation accuracy. J Syst Soft-
J Adv Manuf Tech 1999;15:895–906. ware 2002.
[142] Rehman S, Guenov M. A methodology for modelling [164] Kitchenham B. The certainty of uncertainty. Keynote
manufacturing costs at conceptual design. Comput Ind address. Business improvement through software mea-
Eng 1998;35(3–4):623–6. surement. European Software Measurement Conference,
[143] Briand L, Emam K, Surmann D, Maxwell K, Wieczorek FESMA 98, Antwerp, Belgium, 1988.
I. An assessment and comparison of common software [165] Kitchenham B. Empirical studies of assumptions that
cost estimation modeling techniques. International Soft- underlie software cost estimation models. Inf Software
ware Engineering Research Network. Technical Report: Tech 1992;34(4):211–8.
ISERN-98-27, 1998. [166] Stensrud E, Myrtveit I. Human performance esti-
[144] Briand L, Wieczorek I. Resource estimation in software mating with analogy and regression models: an em-
engineering. International Software Engineering Re- pirical validation. IEEE proceedings from the fifth
search Network. Technical Report: ISERN-00-05, 2000. international software metrics symposium, 1998;
[145] Mukhopadhyay T, Vicinanza S, Prietula M. Examining p. 205–13.
the feasibility of a case-based reasoning model for [167] Shepperd M, Cartwright M. Predicting with sparse data.
software effort estimation. MIS Quart 1992;(June): Software Eng 2001;27(11):987–98.
155–71. [168] Gray A, MacDonell S, Shepperd M. Factors system-
[146] Myrtveit I, Stensrud E. A controlled experiment to assess atically associated with errors in subjective estimates of
the benefits of estimating with analogy and re- software development effort: the stability of expert
gression models. IEEE Trans Software Eng 1999;25(4): judgment. Sixth international software metrics sympo-
510–25. sium, 1999, p. 216–27.
[147] Curran R, Rush C, Roy R, Raghunathan S. Cost [169] Mukhopadhyay T, Vicinanza S, Prietula M. Examining
estimating practice in aerospace: England and Northern the feasibility of a case-based reasoning model for
 ARTICLE IN PRESS
532 R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534

software effort estimation. MIS Quarterly, 1992. [188] Ting P-K, Zhang C, Wang B, Deshmukh A. Product and
p. 155–71. process cost estimation with fuzzy multi-attribute utility
[170] Pengelly A. Performance of estimating techniques in theory. Eng Econ 1999;44(4).
current development environments. Software Eng J 1995; [189] Gerla G. Fuzzy logic mathematical tools for approximate
162–70. reasoning. Dordrecht: Kluwer Academic Publishers; 2001
[171] Briand L, Emam K, Surmann D, Maxwell K, Wieczorek ISBN 0-7923-6941-6.
I. An assessment and comparison of common software [190] Kishk M, Al-Hajj A. An integrated framework for life
cost estimation modeling techniques. International Soft- cycle costings in buildings. RICS Research Foundation;
ware Engineering Research Network. Technical Report: 1999 ISBN 0-85406-968-2.
ISERN-98-27,1998. [191] Nachtmann H, Needy K. Fuzzy activity based costing: a
[172] Hughes R. Expert judgement as an estimating method. methodology for handling uncertainty in activity based
Inf Software Tech 1996;38:67–75. costing systems. Eng Econ 2001;46(4):245.
[173] DOE. Cost estimating guide. US Department of Energy: [192] Cordon O, Gomicide F, Herrera F. Ten years of gentic
Associate Deputy Secretary for Field Management: DOE fuzzy systems: current framework and new trenads. Fuzzy
G 430.1-1, 1997. Sets Syst 2003.
[174] Goodman PA. Application of cost-estimation techniques: [193] Cross V. Defining fuzzy relationships in object models:
industrial perspective. Inf Software Tech 1992;34(6): abstraction and interpretation. Fuzzy Sets Syst
379–82. 2003;140:5–27.
[175] Rand Corporation. Military jet acquisition: technology [194] Mamdani EH, Gaines BR. Fuzzy reasoning and its
basics & cost-estimating methodology. MR-1596, 2002. applications. New York: Academic Press; QA
[176] Wierda LS. Linking design, process planning and cost 248(MAMD), 1981.
information by feature-based modelling. J Eng Des [195] Klir GJ, Ruan DA. Fuzzy logic foundations and
1991;2(1):3–19. industrial applications, International series in intelligent
[177] Nee AYC, Kumar AS, Prombanpong S, Puah KY. A Tech. Dordrecht: Kluwer Academic Publishers;
feature based classification scheme for fixtures. Ann 1996.
CIRP 1992;41(1):189–92. [196] Klir GJ. Fuzzy set theory: foundations and applications.
[178] Zhang YF, Fuh JYH, Chan WT. Feature-based cost Englewood Cliff, NJ: Prentice-Hall; 1997 ISBN 0-13-
estimation for packaging products using neural networks. 341058-7.
Comput Ind 1996;32:95–113. [197] Villarreal J, Lea R, Savely R. Fuzzy logic and neural
[179] Geiger TS, Dilts DM. Automated design-to-cost: inte- network technologies. In: 30th Aerospace sciences meet-
grating costing into the design decision. Comput Aid Des ing and exhibit, Houston, TX; 6–9 January 1992.
1996;28(6–7):423–38. [198] Smith A, Mason A. Cost estimation predictive modelling:
[180] Kiritsis D, Xirouchakis P. A software prototype for cost regression versus neural network. Eng Econ
estimation of process plans of machined parts. In: 1997;42(2):137–62.
Proceedings of the international symposium on auto- [199] Hornik K, Stinchcombe M, White H. Multilayer feed-
motive technology and automation (29th ISATA), forward networks are universal approximators. Neural
Florence; 1996. p. 19–26. networks, 2:359–66. In: Smith A, Mason A, editors.
[181] Schaal S, Ehrlenspiel K. Design concurrent calculation: a (1997). Cost estimation predictive modelling: regression
CAD- and data-integrated approach. J Eng Des versus neural network. Engineering Economist 1998,
1993;4(2):75–89. 42(2): 137–62.
[182] Srikantappa AB, Crawford RH. Automatic part coding [200] Office for Naval Research (ONR). A comprehensive,
based on inter-feature relationships. In: Shah JJ, Mäntylä robust design simulation approach to the evaluation/
M, Nau DS, editors. Advances in feature based selection of affordable technologies and systems. ONR
manufacturing. Amsterdam: Elsevier Science; 1994. Affordability Program Review, Presentation made
p. 215–37. 21–22nd July 1999, Washington DC, Grant No.
[183] Bronsvoort WF, Jansen FW. Multi-view feature model- N00014-97-1-0783.
ling for design and assembly. In: Advances in [201] Pedrycz W, Peters J, Ramanna S. A fuzzy set approach to
feature based manufacturing; 1994. p. 315–29 [chapter cost estimation of software projects. In: Shaw conference
14]. centre, Edmonton, Alta. Canada Proceedings, IEEE,
[184] Catania G. Form-features for mechanical design and Canadian conference, electrical and computer engineer-
manufacturing. J Eng Des 1991;2(1):21–43. ing; 9–12 May 1999.
[185] Ou-Yanang C, Lin TS. Developing an integrated [202] Joumier H. Fuzzy numbers as a tool for measuring
framework for feature based early manufacturing imprecision in cost estimating: a pragmatic implementa-
cost estimation. Int J Adv Manuf Tech 1997;13: tion example. In: Proceedings of second joint
618–29. ISPA-SCEA conference San-Antonio TX; 8–11th June
[186] Brimson JA. Feature costing: beyond ABC. J Cost 1999.
Manag 1998; 6–12. [203] Office for Naval Research (ONR). Development and
[187] Taylor IM. Cost engineering—a feature based approach. implementation of an IPPD approach to system afford-
In: 85th Meeting of the AGARD Structures and Material ability. ONR affordability program review. Presentation
Panel, Aalborg, Denmark, vol. 14, October 13–14, 1997, made 3rd June 1998, Washington DC, Grant No.
p. 1–9. N00014-97-1-0783.
 ARTICLE IN PRESS
R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534 533

[204] Blattberg R, Hoch S. Database models and managerial [220] Turner RJ. The handbook of project-based management.
intuition: 50% model þ50% manager. Manage Sci New York, UK: McGraw-Hill International Limited;
1990;36(8):887–99. 1993.
[205] Villarreal JA, Lea RN, Savely RT. Fuzzy logic and neural [221] Liao S. Knowledge management technologies and appli-
network technologies. In: 30th Aerospace sciences cations—literature review 1995 to 2002. Expert Syst Appl
meeting and exhibit, Houston, TX; January 6–9, 2003;25:155–64.
1992. [222] Chen M-S, Han J, Yu PS. Data mining: an overview from
[206] Harding A, Lowe D, Hickson A, Emsley M, Duff R. The a database perspective. IEEE Trans Knowledge Data Eng
cost of procurement: a neural network approach. In: 1996;8:866–83.
Proceedings international conference in construction [223] Sorensen K, Janssens G. Data mining with genetic
information. Technology Reykjavik, Iceland; 28–30 June algorithms on binary trees. Eur J Oper Res 2003;151:
2000. 253–64.
[207] Harding AM, Lowe DJ, Hickson A, Emsley MW, Duff [224] Hempel CG, Oppenheim P. Studies in the logic of
AR. Implementation of a neural network model explanation. Philos Sci 1948;15:75–135.
for the comparison of the cost of different procure- [225] Silberberg E. The structure of economics: a mathematical
ment approaches. In: 15th annual ARCOM con- analysis, 2nd ed. New York: McGraw-Hill Publishing
ference, Liverpool John Moores University; 1999. Company; 1990.
p. 763–71. [226] Russell B. On the notion of cause. Proc Aristotelian Soc
[208] Geiger M, Knoblach J. Cost estimation of sheet metal 1912–23;13:1–26.
parts with neural networks. In: Proceedings of the fifth [227] Suppes P. A probabilistic theory of causality. Amster-
international conference on sheet metal: SheMet’97, dam: North Holland; 1970.
Belfast; 1997. p. 69–78. [228] Cartwright N. How the laws of physics lie. Oxford:
[209] Zhang YF, Fuh JYH, Chan WT. Feature-based cost Clarendon Press; 1983.
estimation for packaging products using neural networks. [229] Cowan R, Rizzo MJ. The genetic-causal tradition and
Comput Ind 1996;32:95–113. modern economic theory. 1997.
[210] Bode J. Neural networks for cost estimation. Cost Eng [230] Menger C. Principles of economics (trans James Dingwall
1998;40(1):25–30. and Bert F. Hoselitz). New York: New York University
[211] Smith AE, Mason AK. Cost estimation predictive Press; 1981 [1871].
modelling: regression versus neural network. Eng Econ [231] Cowan R. Tortoises and hares: choice among technolo-
1997;42(2):137–62. gies of unknown merit. Econ J 1991;101:14–801.
[212] Hornik K, Stinchcombe M, White H. Multilayer feed- [232] Curran R, Rothwell A, Castagne S. A numerical method
forward networks are universal approximators. Neural for the structural cost optimisation of stringer skin
Networks 1998;2:359–66. panels. AIAA structure conference, Palm springs,
[213] Elhag TMS, Boussabaine AH. Statistical analysis and 2004.
cost models development. EPSRC Research Grant [233] ESDU International. Buckling in compression of sheet
Report, University of Liverpool, 1998. between rivets. Engineering sciences data item 02.01.08
[214] Bode J. Decision support with neural networks in the (Structures series), 1962.
management of research and development: concepts and [234] ESDU International. Local buckling of compression
application to cost estimation. Inf Manage 1998;34: panels with unflanged integral stiffeners. Engineering
33–40. Sciences Data Item 70003 (Structures series), 1970.
[215] Edmonds RJ. A case study illustrating the risk assessment [235] Rothwell A. Structural optimisation. Lecture notes.
and risk analysis process at the bid phase of a project. Faculty of Aerospace Engineering, Delft University of
British Aerospace Defence Limited, Dynamics division. Technology, The Netherlands, 2004.
2000 [chapter 2]. [236] Sandoz PL. Structural design of future commercial
[216] Roy R, Bendall D, Taylor JP, Jones P, Madariaga AP, transports. Paper AIAA-73-20, 1973
Crossland J, Hamel J, Taylor IM. Identifying and [237] Rais-Rohani M, Greenwood AG. Product and process
capturing the qualitative cost drivers within a concurrent coupling in multidisciplinary design of flight vehicle
engineering environment. In: Chawdhry PK, Ghodous P, structures. Proceedings of the seventh AIAA/NASA/
Vandorpe D, editors. Advances in concurrent engineer- USAF/ISSMO symposium on multidisciplinary analysis
ing. Pennsylvania, USA: Technomic Publishing; 1999. and optimisation, St. Louis, MO, USA, September 2–4,
p. 39–50. 1998, Paper AIAA 98-4820.
[217] Hull K. AEPS/ETG, Ministry of defence, procurement [238] Curran R, Kundu A, Raghunathan S, McFadden R.
executive. Risk analysis techniques in defence procure- Impact of aerodynamic surface tolerance on aircraft cost
ment, 1991, unpublished. driver. J Aerospace Eng Proc IMecE 2002;216(G1):
[218] Hamaker JW. SAM user manual. In: Huntsville AL, 29–39.
Stewart RD, Wyskida RM, Johannes JD, editors Cost [239] Curran R, Kundu A, Raghunathan S, McFadden R.
estimator’s reference manual. 1980 [chapter 8]. Influence of manufacturing tolerance on aircraft direct
[219] Crossland R, Sims Williams JH, McMahon CA. An operating cost (DOC). J Mater Process Tech 2003, ISSN:
object—oriented design model incorporating uncertainty 0924-0136.
for early risk assessment. In: Boston, MA International [240] Kundu AK, Watterson JK, Raghunathan S. A multi-
Computers in Engineering Conference; 1995. disciplinary study of aircraft aerodynamic smoothness
 ARTICLE IN PRESS
534 R. Curran et al. / Progress in Aerospace Sciences 40 (2004) 487–534

requirements to reduce operating costs, Seventh AIAA/ tolerance on aerodynamic surface features on turbo fan
USAF/NASA/ISSMO symposium on multidisciplinary nacelles. Int J Adv Manufact Technol 1999;14:
analysis and optimisation. Paper 98-4874, September 894–900.
1998, Missouri. [242] Kundu A, Raghunathan S, Cooper RK. Effect of aircraft
[241] Sanchez M, Kundu AK, Hinds BK, Raghunathan SA. A smoothness requirements on cost. Aeronaut J 2000;
methodology for assessing manufacturing cost due to 104(1039):415–20 (paper 2389).