Professional Documents
Culture Documents
Lecture Note #1
ii
Table of Contents
Table of Contents
1
INTRODUCTION ........................................................................................................................ 1
4
THE TASK ENVIRONMENT AND PROBLEM SPACE OF PRELIMINARY SHIP
DESIGN................................................................................................................................................. 51
4.1
PRELIMINARY SHIP DESIGN TASK ENVIRONMENT................................................................ 53
4.1.1
Complex mapping between form and function ............................................................... 53
4.1.2
Multi-dimensional performance evaluation ................................................................... 57
4.1.3
High cost of error ........................................................................................................... 58
4.1.4
Shallow knowledge structure.......................................................................................... 59
4.1.5
Strong domain tradition ................................................................................................. 59
4.1.6
Strict time and resource constraints on the preliminary ship design process ................ 60
4.1.7
Predominantly one-of-a-kind and engineered-to-order solutions ......................... 60
4.2
PRELIMINARY SHIP DESIGN PROBLEM SPACE ...................................................................... 61
4.2.1
Satisficing rather than optimising .................................................................................. 62
Table of Contents
iii
4.2.2
Personalised evaluation functions and stopping rules ................................................... 63
4.2.3
The decomposition of the design problem into leaky modules ....................................... 64
4.2.4
Sequential iterative limited commitment mode design strategy...................................... 65
4.2.5
Extensive use of empirical relations and previous cases................................................ 66
4.2.6
Multiple abstraction levels ............................................................................................. 67
4.3
CRITICAL FACTORS FOR A DESIGN MODEL .......................................................................... 68
4.3.1
Flexibility........................................................................................................................ 70
4.3.2
Expressiveness ................................................................................................................ 71
4.3.3
Perceivability.................................................................................................................. 72
4.3.4
Efficiency ........................................................................................................................ 73
4.4
CHAPTER CONCLUSION ........................................................................................................ 75
5
REFERENCES............................................................................................................................ 77
1
Introduction
The focus of this lecture note is the early stages of the design process. The context is
the design of marine systems, with special emphasis on merchant ship design.
It will centre on the main characteristics of the design process not from a design
analysis perspective, but rather from the perspective of how a design engineer may
understand the design situation at hand in order to select and apply a suitable design
model and its corresponding set of methods.
Different design methods can be grouped together into what we call design
paradigms. These paradigms differ in their fundamental approach to the design
problem. It is important to understand these differences to make proper decisions
about when to use and when not to use a particular model or method. This require
us to take a slight sidestep into basic reasoning processes such as deduction, induction
and abduction, and from these derive a formal theory on elemental reasoning
processes in design.
I run the risk of leaving some of you at a level of abstraction that seems little relevant
to practical design situation. However, we will revisit these elementary processes
throughout the semester. Hopefully you will be patient enough to see how they are
useful to understand the different design paradigms beyond the step-by-step solution
techniques needed to solve text-book problems, but is of little use in most real-world
settings.
In addition, this lecture note contains a brief review of engineering design schools. It
is not comprehensive and it does not go very deeply into each if you want to know
more you should dig into som of the references. Avoid getting fixed on details it is
most important (and most relevant for your final exam) to get the broad picture.
2
The Preliminary Ship Design Process
The term early stages refers to the part of the total ship design process in which the
main features of the ship is determined, taking high-level technical and economic
criteria into consideration. On a task-oriented time-line the early design stages
comprise the activities that starts with an identification of a need and a statement of
the design problem, and ends with a tender or a contractual specification of the design
solution. The features of the vessel to be determined usually include the main
dimensions, powering and propulsion arrangements, overall external and internal
geometry, and cargo access and handling equipment. Sometimes the selection of ship
speed is also covered.
In these early stages some of the most important decisions regarding the vessel are
made. Various sources estimate that between 60% and 80% of the total life cycle cost
is determined during this part of the design process [Dierolf, 1989]. All later design
decisions have to be taken within the frame set by the initial design, with only limited
influence on cost and overall performance.
At the same time the situation at this stage is characterised by a high degree of design
freedom, as illustrated in Figure 21. The design problem is open no decisions have
yet been made that limit the options beyond the bounds given in the problem
definitions. All subsequent decisions will put constraints on this freedom. Since not
making decisions is hardly a viable alternative, the effort must be put on making these
early decisions correct. In [Suh, 1990] the situation is described as:
"Design decisions made at the initial or upstream stage of engineering
affect all subsequent outcomes". Fine-tuning of the later stages of
engineering operations may often have marginal effects on the total
outcome, and certainly cannot rectify the erroneous decisions made at
conception; yet we often relegate the design decisions to the least
experienced or the least educated of engineering professionals. The
reason why this practice has lasted for so long lies in our inability to
Clarificat ion of
t ask
Design
init iat ion
Concept ual
design
Embodiment
design
Cont r act
specificat ion
and events will vary with respect to both the actual design problem and to differences
in design practice, some similarities can be found that together can be used to form a
prototypical preliminary ship design process.
order to start the search for solutions. This includes the identification of promising
market segments, forecasting the development of market supply, demand, prices, cost,
etc., identification of market characteristics (customers, competitors, conferences,
entry and exit barriers, etc.), and the resolution of market constraints and restrictions.
Often this activity is performed by the owner or brokers, and not by the shipyard.
Ideally, the result of this activity is a set of pure functional requirements and
objectives that the final design solution should meet, that is, the formulation of the
design task should not presuppose any particular solution. This is important for
maintaining a creative element in the design process, hopefully leading to a new and
innovative solution. For instance, formulations like design a bulker with less than 10
m draft should be avoided, since important aspects of the solution will be implicitly
defined. A better alternative is the formulation: design a transport system that can
unload bulk cargoes in port X where the maximum permissible draught is 10 m
(example from [Erichsen, 1989]). In the latter formulation the emphasis is on the
problem, keeping the choice of solution open. However, in practice the openness of
the design problem is restricted due to factors such as risk avoidance, the adaptation
to existing production technologies and facilities, and sometimes an inherent
conservatism in the maritime industry as such.
selection of the optimal solution taking both monetary and non-monetary goals into
consideration.
An example illustrating these characteristics for the outline specification is given in
Box 2-1. In paragraph (i) the intention of the outline specification of stating the
minimum requirements for the vessel to be designed. In (ii) the top-level preferences
of the potential customer is given, thus providing the basis for forming an evaluation
function that can be used to determine the overall performance of a particular design.
Paragraph (iii) in Box 2-1 states the selected conceptual solution, while paragraph (iv)
states constraints on the design solution by giving (minimum) performance and
operational requirements.
Box 2-1: Excerpts from an outline specification from [Statoil, 1991] illustrating
typical characteristics. The outline specification serve as a starting point for the
embodiment design phase
(i) This outline specification contains design and minimum performance and
operational requirements of a tanker for cargoes as listed in DnV Rules ().
(ii) On a competitive basis, the Charterer will give preference to vessels offering the
best design and operational characteristics for the following high-lighted items: ()
(iii) The vessel shall be a single screw fully welded motor tanker with bulb,
forecastle, pump room aft, and one semi-spade rudder. The vessel shall be built with
double bottom and double skin as ballast tanks, and preferably arranged with centre
longitudinal bulkhead in the cargo area. ()
(iv) The vessel shall be constructed and outfitted for world-wide trade. Dimensions
for length overall, draft, and extreme beam will be subject for due consideration with
respect to limitations at major European and US terminals. Max. draft = 7.5 m. The
vessels deadweight shall be about 6000-8000 tonnes on design draft in SW. The tank
capacity is to be sufficient in order to carry a full deadweight with a cargo of specific
gravity 0.8 t/m3. The mean service speed for design loaded/ballast shall not be less
than 12.5 knots with main engine running at 90% MCR, and at head wind speed and
sea state up to and including Beaufort Number 5. ()
General arrangement
Preliminary lines
Price estimates
A principal difference between the conceptual and embodiment design phase is the
dominant type of decisions made. According to [Mistree, 1991] there are two types of
primary decisions: selection and compromise. The selection decision involves a
choice between a finite number of given alternatives, based on a trade-off between a
set of attributes. In the compromise decision the task is to find the value of a set of
design variables that best satisfies some given objective within the system constraints
and bounds. Other kinds of decisions can be represented as a combination of these.
In the conceptual design phase selection is the dominant decision type, in making a
choice between a number of solution principles. In the embodiment the dominant
decision type is compromise, in finding the values of the ship main characteristics in
order to best satisfy mission requirements. This results in a principally different
decision-making process in the conceptual and embodiment design phase, making a
common approach with respect to decision support inappropriate.
3
A Review of Engineering Design
It is said that design is in a state of change, and that we see a development towards a
"science of design". The notion of a design science was first put forth by H. A. Simon
in his book The Sciences of the Artificial [Simon 1982]. Following this, the design
community has put considerable effort into developing a theoretical fundament on
which such a science can be built, though so far no general consensus has been
reached on the actual content of this fundament.
An excellent survey of the current status of design science is given by [Finger and
Dixon 1989]. Particular research areas comprise the formulation of a consistent design
taxonomy [Dixon, et al. 1988, Ullman 1992], the identification of fundamental design
axioms and invariants [Mistree, et al. 1990, Suh 1990, p 12], the development of
procedures for hypothesis formation and testing in design [Antonsson, 1987, Dixon
1987], and the establishment of a theory for understanding design processes [Mistree,
et al. 1990, Hubka 1982, Hubka and Eder 1987, Coyne, et al. 1990, Rzevski 1980,
Yoshikawa and Koyama 1982].
The pursuit of a theoretical basis for design aimed at linking design method and
scientific method has also been criticised. Cross [Cross 1980] points to the
fundamental difference in the goals of these two activities: While science aims at
describing the nature of what exists, design is concerned with inventing artefacts that
are to be built. As a consequence of this basic difference, a common theoretical
foundation for these two activities is not possible, nor desirable. He further indicates
that the underlying reason for the science approach to design is not based on the
nature of design as such, but rather motivated by the attractiveness of scientific
qualities, like rationality, neutrality, and universality, primarily aimed towards giving
design a higher status both in academia and in society in general [Cross, 1980].
A similar criticism has been put forth by Sargent [Sargent 1994]. He claims that
design requires the integration of techniques from several fundamentally incompatible
views of the world. From this he concludes that we can have no unitary science of
design at best we have a number of partial theories that are useful. Thus, an
essential aspect of design is to reconcile the incommensurate requirements and goals
created by the partial theories.
It is not within the scope of this lecture note to discuss the ontological1 properties
the essence of design. Rather, a more utilitarian view is taken, as follows: If we
by "science" understand any methodological activity towards identifying, describing,
investigating, and explaining a certain class of phenomena, then a scientific approach
towards design is both useful and necessary to conduct research and development
work in design.
Box 3-1: On science and design, from [Suh 1990, pp. 5-6]:
"One of the major causes of the dismal state of design is simply a mental block: the
notion that design, unlike the natural sciences, cannot stand on a scientific basis. This basic
hypothesis is both unnecessary and incorrect. (...) Due either to lack of sufficient effort or to
the lack of truly successful paradigms for dealing with creative processes on a systematic
basis based on "principles" and "laws", it has been assumed that these topics cannot be treated
as subject for scientific discourse. (...) In the absence of a scientific basis, human intellectual
endeavours ranging from fine arts to engineering are performed subjectively in the realm of
the "creative" activity. Since the output of such activities cannot be understood rationally in
the absence of commonly accepted criteria, they are treated as such. What this really means is
that we can appreciate the outcome of the intellectual endeavour but do not understand the
process that produces the outcome, and cannot quantify the results. (...) When "know-how"
and knowledge are not codified, each generation must gain similar experience all over, again
and again, and develop its own intuition. These are typical characteristics of a field that has
not yet matured into a "science". A truly scientific discipline is powerful because governing
principles or laws describe the underlying thought process and reduce a seemingly complex
arrays of facts and observations into consistent and explicitly stated knowledge.
As support of this view, two development trends are of special importance: First, we
have seen design evolve from mainly being a one-man invention process into a multiperson/multi-team core activity in the industrial organisation. Where before the
designer could to a large extent rely on his/her own skills and experience in
organising and carrying out the design task, it is now common that a large number of
people with different backgrounds, sometimes located at geographically distant
places, work closely together on the same design in parallel. From this follows a
requirement for both a systematisation and a common taxonomy of design and design
processes in order to co-ordinate the activity.
ontology The branch of philosophy that deals with being [American Heritage Dictionary, 1991]
Secondly, and most important for this thesis, is the central role that computers have
come to play in design. Until recently, their main function in the ship design process
was to automate the often tedious and complex task of design analysis, and to serve as
an advanced drawing aid in graphical CAD tools. However, in the last few years we
have seen a development towards the computer increasingly acting as a partner to the
human designer. Examples of technologies enabling this are large scale integrated
design systems, knowledge-based systems, and product models. Today, it is no
exaggeration to say that the presence of computers has become a prerequisite for realworld ship design. As a consequence, there is a need for a systematic and formalised
treatment of all aspects of the design process, in order to satisfy the computers
demand for logic and rigour.
Box 3-2: The top-level design task can be described as a mapping between a
functional and a descriptive [Chandrasekaran 1989 p. 77]:
The design problem is specified by a set of functions to be delivered by an artefact, a set
of constraints to be satisfied by the artefact, a repertoire of components assumed to be
available, and a vocabulary of relations between components. The solution to the design
problem consists of a complete specification of the set of components and their relation which
together describe an instance of the artefact delivering the functions and satisfying the
constraints.
In the following I will look into the foundation for developing such a formalised
model of the ship design process. The focus will be on both the theoretical aspects of
design science and design methodology, and on new methods and the techniques of
decision support in designing complex, large-scale, engineering systems. In the first
part of the chapter I will briefly discuss elementary design inference processes, since
they represent the building blocks on which more complex design processes will be
based in later chapters. Further, a number of different design paradigms will be
described, and their usefulness for supporting the preliminary ship design process will
be evaluated. In the last part of the chapter, I will survey some relevant new
technologies that is expected to influence the design process in the years to come.
10
Figure 31: The top-level design process as a mapping between a decision space and
a performance space
In the ship domain, the decision space might be spanned by a numerical description of
the ship. This description may comprise a set of attribute-value pairs corresponding to
for instance the most important design characteristics, a geometric description of the
hull form using NURBS surfaces, or a topological description of the main ship
system. Most common in a preliminary design setting is a decision space spanned by
a numerical representation of the ship main characteristics, thus forming a ndimensional space of real numbers, n.
The performance space specifies the functions delivered by the design object. This
may be both independent performance functions, such as steel weight, production
cost, resistance, etc., or some compound criteria.
Using these representations, the baseline design process may be described as the
mapping of points from the performance space to the corresponding point in the
decision space, thus arriving at a design solution. If this mapping had been ideal it
would bring us directly to the optimal design solution. However, for real-world
practical design problems we are forced to take a number of iterative steps in order to
arrive at a satisfying design solution. In the following I will review different types of
such iterative steps, or inference processes. The discussion is to a large extent based
on [Coyne, 1990]. Later in this chapter these basis processes will be used in the
discussion of different design paradigms.
11
12
syntax 1. GRAMMAR a. The way in which words are put together to form phrases and
sentences () 2. COMPUTER SCIENCE The rules governing the construction of a machine
language. [American Heritage Dictionary, 1991]
13
Induction is the inference process of reasoning from the specific to the general,
also termed the logic of scientific discovery. Induction can be represented by
the following reasoning pattern:
(1,1),(2,2),()
Using these terms, we can now describe a number of basic inference processes that
are part of the total design process Mapping these generic types into the design
domain, the following basic inference steps can be identified [Coyne, et al. 1990]:
(2.1)
14
Ki:
D:
I:
Resistance(ShipA) = 1200
The first statement says that if we know the deadweight and speed of any ship (within
the relevant Universe of Discourse) we can determine its resistance by using the
function f. Since this is a mapping from a description (DW and V) to an interpretation
(the performance parameter resistance), the statement is classified as interpretive
knowledge.
The second statement simply says that we are given a specific instance, ShipA, of
which we actually know the deadweight and the speed. On the basis of these two
premises, the corresponding resistance can be correctly interpreted as 1200, and,
given that the two premises are true, the interpretation is also true by logical
necessity.
The inference step just described is the core process in most existing preliminary ship
design tools. Referring back to Figure 31, it represents a mapping from the design
space to the performance space. Thus, it represent a necessary, but not sufficient,
inference step to support the top-level design process. In addition we need to support
the mapping the opposite way, in going from a specification of function to a
specification of form in the design process. For the aforementioned design tools this
has the consequence that the actual design process becomes sequential and iterative,
often described as a DESIGN-EVALUATE-REDESIGN cycle. Here, the computer
is primarily responsible for the EVALUATE part based on the interpretation
presented here, while the human designer is responsible both for proposing possible
design solutions in the DESIGN step, and to interpret the outcome of the
EVALUATE step in order to redesign the solution.
arbitrary generalisation, for instance a rule of thumb, a physical law, or an empirical relation
15
From a design process perspective, this represents the ideal inference step, since it
brings us directly from a functional description in the performance space to the
corresponding description of a design solution in the decision space. As an example,
we may have:
Ki:
I:
Resistance(ShipA) = 1200
D:
The example is the same as in the previous section, with the exception that the
interpretation, I, is now one of the premises, and the description, D, is the conclusion.
The inference pattern of the example above corresponds to that of abduction, or
inference to the best explanation. The problem with abduction is that it does not
represent a valid argumentation in the same way as for deduction, that is, the truth of
the conclusion, D, does not follow logically from the truth of the premises, I and Ki.
On the contrary, there is most likely an infinite number of possible solutions that are
consistent with the two premises. Still, the abductive inference step represented by the
transformation 2 above is both useful and necessary in the production of design
descriptions. In particular, it is useful to reduce the search space by generating
reasonable hypotheses, that later can be validated by deductive interpretations, such
as the transformation 1 in Equation 1.1.
16
(1.4)
(2.5a)
(2.5b)
or alternatively as
(2.6)
The inference process in the transformations above are equivalent to induction, where
a general knowledge statement is concluded from a set of observations or test cases.
For example, we may have:
17
D1:
D2:
D3:
Ks:
One typical example of the transformations 5 and 6 is the use of statistical inference
techniques. They can be used to derive empirical relations between the ship main
characteristics and derived performances, such as resistance & propulsion [Holtrop
1984, Holtrop and Mennen 1982], or steel weight [Watson and Gilfillan 1976,
Harvald and Jensen 1992]. They are also used to generate syntactic knowledge, for
instance determine design lanes for parameters and/or parameter combinations [Smith
1992].
In the actual design process models the syntactic and interpretive knowledge is
usually given a priori, as in 1 to 4, and not obtained as part of the process itself.
However, in recent years there has been an increased focus on the integration of
learning and knowledge acquisition into the design process itself [Aamodt 1991,
Duffy 1993], and it is generally believed that the ability to continually develop and
adapt the knowledge base on which decisions are made will become a key
competitive factor in the future.
18
Many degrees of freedom exists in the problem statement owing to the fact that
the design problem specification tend to be under-specified
Feedback from the world is limited or delayed during problem solving much of
the feedback will have to wait for the artefact to be produced or put into regular
use, when a significant of the resources has already been expended
19
The input to the design process substantially consist of goals and intentions. The
output is a description of a design object
The design object must function independently of the designer the designer will
not participate in its use, and thus not be able to explain and perform its function.
The specification and delivery of the design object are temporally separated. This
enables the designer to model the performance of the design object before it is put
into production, thus reducing the risk of being wrong
Costs are associated with each and every action no effort towards reducing the
risk of being wrong, by for instance improving the knowledge base or extending
the exploration of the design space, comes for free
Answers are neither right, nor wrong, only better or worse a parallel to Simons
notion of design as a satisficing rather than an optimising activity [Simon 1982].
The problems tend to be large and complex
On the basis of the results from the protocol study, Goel & Pirolli proposed a
structure for the design problem space of the generic design problem, together with an
explanatory connection between this structure and the proposed characteristics of
the design task environment. Eight significant invariants were claimed:
20
Since there are no right or wrong answers, only better or worse, it is not
possible to derive a set of rational criteria for when to terminate the design
process. This is in contrast to more well-formulated problem-spaces where
stopping rules may be given by mathematically defined convergence criteria.
A limited commitment mode control strategy with nested evaluation cycles
Ideally, the designer would postpone all evaluation of the design object until it is
completely specified, thus avoiding sub-optimisation. However, given the cost,
time and complexity involved, this is not a feasible strategy. This leads to a
situation where the designer need to negotiate a constant tension between keeping
the problem open and keeping the problem tractable5. In the study by Goel and
Pirolli the designers chose a compromise evaluation strategy, where a three-step
process were devised for each decision: First evaluate the component/sub-problem
on its own and decide whether to accept or reject it, then evaluate it in the context
of the design so far, and third, re-evaluate it later in a more complete context. Goel
and Pirolli denoted this strategy as a limited commitment strategy mode.
The making and propagation of commitments
21
22
Design Task
Environment
Size & complexit y of
problem
Independent
funct ioning of art ifact
Design Problem
Space
Solut ion decomposit ion
int o leaky modules
Making and
propagat ion of
commit ment s
No right or wrong
answers - only bet t er
or worse
Many degrees of
freedom in design
problem st at ement
Figure 33: The design task environment and design problem space for a
general design problem, adopted from [Goel and Pirolli 1989]
23
Goel & Pirollis work is interesting because they show how the characteristics of the
problem space in design can be understood as a consequence of the external factors,
as given by the task environment. This is important to keep in mind in the
development of computer-based design tools, where a de facto problem space is
created by the functionality and interface of the system. Often, this problem space
seems to be as much a consequence of what is possible within the relevant technology
areas, as what is actually needed from the designers point of view. Thus, we may find
a mismatch between the designers intuitively created problem space (as given by for
instance the invariants of Goel & Pirolli), and the problem space he/she is forced to
work within when using computer-based design tools. Examples of such a mismatch
are:
Problem space created by human
designer
Opportunistic exploitation of
whatever methods are appropriate
and available (see Section 3.3.7)
It is obvious that if the mismatch between the two problem spaces is significant, it
will impede the efficiency and utility of the design tool. Thus, in the course of
devising methods for supporting the design problem-solving process, which is the
goal of this thesis, a thorough understanding of the relevant problem space is
necessary. This will be the main subject of Chapter 3, where I will look into the task
environment and corresponding problem space for ship design, based on the above
presented scheme of Goel & Pirolli. From this, a concept exploration model for
6
This capacity has been estimated to be about seven chunks [Simon, 1982, p. 80]. A chunk is a
maximal structure that can be recognized as one single item. E.g. PGH consist of three memory
chunks, while LCB only one (by a naval architect)
24
preliminary ship design will be proposed in Chapter 5. The content of this model will
to a large degree inherit from existing, prescriptive design models. Thus, in the
following section I will describe some of the more important of these.
model () 2. A preliminary pattern serving as the plan from which an item not yet constructed
will be produced. 3. A tentative description of a system or theory that accounts for all of its known
properties. () [American Heritage Dictionary, 1991]
25
chosen mainly because each category to some extent reflect a particular view of
design.
Energy
Mat erial
Energy
Funct ion
Signals
Mat erial
Signals
Figure 34: In the VDI model, the basic function in all technical systems involves the
conversion of energy, material, and/or signals [Pahl and Beitz 1984, p. 23]
The VDI model offers a problem oriented design strategy, where the emphasis is
placed on a detailed problem analysis and a structured procedure to identify a
solution. The first step is to identify the main function of the design object from the
problem description. The main function is then broken down into a hierarchy of subfunction. All functions are seen as a conversion of energy, material, and/or signal
(information), as illustrated in Figure 34. The transformation from a hierarchy of
function to a hierarchy of solution elements is by means of design catalogues, relating
elementary functions with alternative physical effect solutions. These solutions are
then synthesised into a complete design, and further improved in the embodiment
design phase.
What makes the VDI model interesting is the focus on a systematic approach to
design, offering relatively detailed procedures for the design engineer to follow. In
[Miles, 1994] the VDI-model is suggested as an epitome of the general approach to
design in German-speaking countries, as opposed to a focus on design analysis and
numerical techniques in the English-speaking world.
26
Figure 35: Design catalogue linking elementary functions with alternative solution
principles, from Beitz [Pahl, 1984 , p. 145]
27
Figure 36: The ship design spiral capturing the sequential and iterative nature of
practical ship design
Though the advent of computers have changed the content of the ship design process,
the spiral model still serve as a conceptual model both for the practical work in many
ship design offices, and for many computer-based design tools. Typically, the design
engineer specifies the values of a set of design variables that describes the ship, and in
some cases makes a preliminary selection of a hull form. These values serve as input
for calculation or simulation algorithms that predicts the performance of the ship in
areas such as hydrostatics, resistance & propulsion, weights, seakeeping, and cost.
This performance is then compared with the design requirements, and the design is
either accepted, or the analysis has to be rerun with a revised set of design variables.
A more detailed description of the content of each step in a process following this
scheme can be found in [Hills and Buxton 1988, Hagen 1992, Evans 1959,
Schneekluth 1987]. Examples of computer-based ship design tools that to a large
extent follow this conceptual scheme are Shipshape9, Hull-Calc, as part of the HullTech10 program suite, and Autoship11 [Ames and Lynaugh 1988].
28
Figure 37: An extension of the spiral, adding time as an extra dimension, from
[Andrews 1981]
The sequential, iterative design process, as captured by the spiral model, can be
captured in a task structure given by DESIGN-EVALUATE-REDESIGN, shown in
Figure 38. Using the basic inference processes described earlier in this chapter, the
three tasks can briefly be described
as:
DESIGN:
DESIGN
EVALUATE
REDESIGN
Propose a potential
solution to the design
problem,
either
based on syntactic
generation, on a
previous solution, or
on general domain
knowledge
and
experience
D = 4(Ks, Ki, V, I)
29
instance, if the proposed solution have been found unsatisfactory, there is usually
limited feedback on how (e.g. which variables, what direction, what stepsize) the
design should be changed in order to improve the solution. This relies to a large
extent on the designers expertise and experience.
A drawback that has been pointed out is that this model typically lock the designer to
his/her first assumption [Levander 1991]. This first assumption is often an existing
ship, perhaps slightly modified to account for differences in the functional
requirements, or a starting point within the mainstream of existing designs that is
based on empirical formulas. Since each iteration tend to be relatively timeconsuming, there is a tendency that only a limited part of the potential design space
around the first assumption is explored. This observation is also supported by a
protocol study of mechanical engineers by Ullman and Dietrich, referred in [Finger,
1989], concluding that designers pursue a single design concept, and that they will
patch and repair their original idea rather than generate alternatives
One argument in favour of this "basis ship" approach, based on a (minor)
modification of an existing solution, is that the continuous development into
specialised ship types during the last century has resulted in ship types close to an
"optimum" design. Hence, a single point initial assumption based on an existing ship
will be a good starting point.
While this might be true for a limited time in certain segments of the market, the
statement has not necessarily general validity. Technical progress, e.g. new materials,
new production methods, more efficient propulsion systems, together with an ever
changing environment for shipping activities in terms of stricter regulations, new
markets, and varying market conditions will always call for new solutions. Hence, the
assumption that an optimal solution will be found by minor adjustment of an
existing one can prove costly. Rather, it is desirable that the design problem is kept
open in the initial stages, by taking a larger set of possible designs into consideration.
30
minimize f ( x ),
subject to h( x ) = 0
g( x ) 0
x n
Here, f is an objective function to be minimised, corresponding to the design goals. f
may express the preference structure of a single design goal, or it may be a compound
function representing multiple design goals. h and g are functional constraints, and
is the set constraint defined over an n-dimensional design space n. In this form, the
design problem is a well-formulated problem, subject to solution by for instance
mathematical programming and operational research techniques.
Expressed in terms of the elementary inference processes discussed in the previous
section, optimisation can be modelled as a single deductive step directly from the
design problem formulation, as given by the transformation 2 (Equation 2.2). This
can be illustrated by the following example:
Ki1:
Ki2:
I1:
I2:
D:
DW(ShipA) = dw V(ShipA) = v
minimise TotalCost(ShipA)
AnnualCargoCapacity(ShipA) > minAnnualCapacity
Here, design knowledge (Ki) is given in the form of numerical relations between
design variables and the performance variables. Also the intended interpretations
design goals and constraints are expressed as numerical relations. In addition,
control knowledge concerning the transformation 2 itself is needed. Given that this
can be transformed into a well-behaved problem formulation, satisfying conditions
such as continuity, differentiability, well-boundedness, and unimodality, the
31
32
For real-world, complex design problems, the distance to this idealised mathematical
representation is often substantial. A vivid description of this is given in [Goldberg
1989]:
Theorists interested in optimisation have been too willing to accept
the legacy of the great eighteenth and nineteenth-century mathematicians who painted a clean world of quadratic objective functions,
ideal constraints, and ever present derivatives. The real world of
search is fraught with discontinuities and vast multimodal, noisy
search spaces.
Is this a valid description for the search space in preliminary ship design too? On the
basis of studies on the use of mathematical optimisation in preliminary ship design
(e.g. [Folkers 1973, Nowacki, et al. 1970, p. 367, Lyon 1982]), the answer seems to
be no. In these studies the design problem is modelled by the ship main
characteristics, a set of empirical relations, and an objective function usually
minimising either the steel weight or required freight rate, giving in most cases a
relatively well-behaved, convex search space.
However, extending this to assume that these search space characteristics can be
found for an arbitrary formulation of the preliminary ship design problem needs
caution. First, the functional relationship f, relating the design vector x to the design
goals, is not well understood in mathematical terms for multi-criteria design
problems. Different methods exist to bring the different criteria together. Examples
are the use of weighted value/utility functions [Keeney, 1976], the use of a ranking of
the design goals in a pre-emptive value function [Smith 1992], or a hierarchical
evaluation process [Sen and Yang 1994]. These methods may cause a non-convex
compound objective function, even if the single goals themselves are convex. One
example is in the design of a frigate using constraint violation as a criteria, showing a
highly multi-modal response function [Smith 1992, p. 168]. It is also a problem when
the optimisation algorithm relies on externally linked procedures, as for instance in
the ship design system NAPA [Priha 1989], or Engineous [Ashley 1992]. In these
cases it might be difficult to determine whether the requirements for using a specific
optimisation algorithm is satisfied.
Another problem with using mathematical optimisation is the relatively high level of
expertise that is needed both to formulate a well-behaved problem, and to verify that
the identified solution is in fact a global optimum within the search space.
To some degree the difficulties of using mathematical optimisation algorithms can be
hidden from the practising designer by a proper user interface, letting the designer
only change a limited set of design variables and parameters within a range known to
result in a well-behaved problem. However, this has the draw-back of limiting the
33
12
Meta-design means the design of the design process, that is, the modeling of the design process
with respect to the characteristics of the problem. The possiblity of explicitly suporting meta-design
will be further discussed in Chapter 5 in the context of concept exploration models
34
Templat e
Select ion
DSP
Compromise
DSP
Variables
Const raint s
information
Find
Satisfy
Minimise
deviation function13
13
The deviation function is a function that expresses the relative distance form the design goals in a
multi-objective goal formulation
35
can a correct and consistent mathematical be ensured? Bras gives an answer to this by
presenting a Design Guidance System [Bras 1992]. Here, the transition from a realworld problem to solvable computer code is achieved by the following steps:
1. The first step is to extract a natural language formulation, called a story, from the
real world problem.
2. This story is analysed using a
parser utility that identifies
relevant problem entities.
These entities are grouped
into a problem statement by
the parser.
3. The problem statements are
grouped with respect to the
keywords of the appropriate
support problem (e.g. givenfind-satisfy-minimise for the
compromise decision)
4. From
the
keyword
description a process network
is developed which will give
or several possible paths
towards a solution
5. In the last step this process
network is converted into
computer code that can be
directly interpreted by the
numerical solver.
14
In the DSPT, the numerical solver is called DSIDES, based on an adaptive linear search
algorithm [Mistree, 1992]
36
15
Other similar classifications are prototype refinement, prototype adaption, and prototype creation
in [Coyne, 1990], and original development, further development, and adaptive design in
[Wogerbauer, 1943]
37
Preliminary ship design can be classified as either of these three, dependent on the
particular situation. Sometimes the design requirements can be satisfied by a minor
modification of an existing vessel, thus doing variant design. This is perhaps the most
common type of design. In other situations a more fundamental change is needed.
Examples are the development of hatchless containerships, double hull tankers, and
self-unloading bulk carriers. Even though important sub-systems of the baseline
solution are changed, the overall solution principle is kept intact. Thus, these
situations belong to adaptive design. Original design within the ship domain is the
least common type. It requires the designer to be able to look beyond the boundaries
of existing solutions, emphasising skills such as creativity and inventiveness. The
development of the hydrofoil and the air-cushion vehicle are examples from recent
history16.
In the case of original design, it is clear that we do not posses the necessary design
knowledge inside the system boundaries. Accordingly, we are not able to arrive at a
design solution using the abductive inference step described on page 15. In that case it
was interpretive knowledge that provided the necessary link between the problem
specification and the design solution, e.g.:
I:
Performance I is required
Ki:
D:
16
The existence of original design has been questioned [Coyne, 1990], arguing that this is simply
adaptive design of a sufficient degree to be regarded as a novel idea. E.g. hydrofoils can be said to be
the adaption of an air wing into water, while the air cushion vehicles are based on ordinary
propellers/fans turned horisontally
38
beforehand, and mutually dependent you dont have Ki without D, and vice versa.
Thus both need to be part of the conclusion, as in the following example:
I:
ExcellentSeakeepingAbilities(ShipA)
Ki:
D:
IsType(ShipA, SWATH)
39
According to [Miles and Moore 1994], the ideal knowledge-based system should be
able to
With todays technology it is impossible to achieve all these goals. Most current
implementation in practical use are able to satisfy only the first two of these features.
However, new technologies are emerging, focusing on the unresolved features of the
above list. Some of these technologies will be further discussed at the end of this
chapter.
40
Speed
Displacement
Beam
Block
coefficient
Lengt h
Power
Draught
Dept h
St eelweigt h
Machinery
weight
Light weight
Deadweight
41
Unit s
Influence frame
Charact erist ic
Relat ionships
Influences
Expression
Reliabilit y
Pre-condit ions
St rengt h
Next
Next
No particular, fixed model of the design object is assumed, as is the case in design
applications based on procedural algorithms. Thus, the design object may be
modelled so as to fit the parallel mental model of the designer, and new design
aspects (Characteristics) and new knowledge (Relations and Influences) may be
added to the network without destroying the existing structure. By this, the metadesign process (the planning of the design process) and the design process are
seamlessly integrated.
42
43
satisfies a specified set of desired properties, while the latter centres on the problemsolving process itself, i.e. design as search in a state space, design as a mapping from
a performance space to a design space, or design as an intuitive leap where the
design solution almost instantaneously comes into the mind of the designer. This is in
compliance with the elementary inference processes discussed earlier in this chapter.
In performing the design task a set of assumptions is made about the domain. First,
the availability of a set of primitive components is assumed. In the ship design domain
this could for instance be design parameters (Lpp, B, Cb, DW), parts (propeller,
machinery components, structural parts) or system descriptions. This corresponds to
the design vocabulary. Second, there exist a set of primitive relations between the
components. Examples are relations between the hull design parameters
(Displacement = Lpp B T Cb) and part assemblies. This corresponds to syntactic
knowledge.
To solve this task, a number of different design
processes are used, dependent on the characteristics of
the specific sub-task, the design domain, and the
preferences of the designer. According to
Chandrasekaran, the reason for this difference is mainly
the availability of knowledge. Hence, a prerequisite for
prescribing design methods for a particular domain (for
instance by developing computer-based design tools) is
a thorough survey of the design knowledge available in
this domain. This is also in compliance with my
previous discussion about different design models: The
appropriateness of a particular model, such as optimisation, design-evaluate-redesign, innovative abduction, or knowledge-based systems, is dependent on
the character of the available knowledge.
PROPOSE
CRITIQUE
MODIFY
44
elements of the design but not with the design as a whole, he needs tools that will help
to retrieve and assemble the elements in a simulation. When the designer has not had
experience in the domain, he will need tools that help him infer the constraints on the
design.
the exploitation of knowledge embedded in past design cases. That includes both
the use of individual past designs to aid particular design tasks, and in abstracting
and generalising new design knowledge. Traditionally, the knowledge embedded
in past design cases has been made available either as a single comparison ship
from which a new design solution can be derived, or through the compilation of
empirical formulas relating key design variables and the main performances.
Representative examples are [Watson, 1976, Holtrop and Mennen 1978].
45
However, we believe there is a potential to better utilise this knowledge by the use
of more intelligent techniques. Relevant technologies are artificial neural
networks and case-based reasoning
In the following some of the potential techniques mentioned above will be briefly
described.
46
To the ship designer these features are intuitively attractive, since a somewhat similar
approach has a long tradition through the use of so-called reference ship. Here, a
previous design is used as a starting point, and is then adjusted to fulfil the
requirements of the present task. However, the CBR technique goes some important
steps further, by allowing for qualified, historic design specifications to be stored in a
case base for later reuse. A case is explicitly defined in terms of both functions and
intentions, and the indexing scheme used for retrieving relevant cases is highly
dynamic, as opposed to traditional databases.
The search procedure is based on a coding of the parameter set, and not the
parameters themselves.
Genetic algorithms search from and maintain a population of points not a single
point
Genetic algorithms uses only objective function information to direct the search,
and are thus not relying on derivatives or other auxiliary knowledge
Genetic algorithms use probabilistic, not deterministic transition rules
Genetic algorithms perform exploration and exploitation of the design space
concurrently. In exploration the algorithm tries to avoid convergence in order to
evaluate designs throughout the space in the search for new solutions (or solution
47
elements). Exploitation focus on a local region, and tries to breed individuals with
high fitness within this region.
The basic genetic algorithm consist of three fundamental processes:
48
17
49
Such an integrated digital life cycle model is an important perspective on a model for
preliminary ship design. Of particular importance are the results from on-going
research projects aimed at ship relevant implementations. Some examples are:
3.5 Conclusion
In this chapter, we have looked into the fundamental reasoning processes in design,
such as deduction, abduction, and induction, and related these to practical design
situations. This is required for developing a formalised model of the ship design
process, which again is a prerequisite for understanding different design paradigms,
such as optimisation, simulation, concept exploration, etc.. In the next chapter we
will apply some of these theories and methodologies to the preliminary ship design
problem.
18
4
The Task Environment and Problem Space of
Preliminary Ship Design
The purpose of this chapter is to identify characteristics of the preliminary ship design
problem that distinguish it from both general design and general problems solving
situations. As a scheme I will use the concepts design task environment and design
problem space that were presented in Chapter 2, denoting the characteristics of the
design problems external and internal structure, respectively.
To provide a reference situation for the discussion, imagine for a moment an ideal
world. Here, a preliminary ship design situation might be described as follows:
We have perfect knowledge about all relevant aspects of the external design
environment, both the current state and all future states.
We also have ideal knowledge about ourselves. This implies that we are able to
explicitly state a value function that maps our true preferences between any
(design) alternative.
Third, we have a perfect model of the ship to be designed, that is, we have a
model that expresses all features influencing the value function, and accurately
maps these features into the preference structure in the design performance space.
And fourth, we have available unlimited computer capacity.
Given the situation described above, we can find a true optimal design solution by
brute force simply by parsing all possible design points, calculating the performance
52
and the corresponding value of each, and at the end select the one with the highest
value.
An alternative to the last point is a situation where the model of the ship-to-bedesigned is completely described by a set of mathematical relations that satisfies
requirements such as continuity, differentiability, and convexity within the actual
design space. Here, the design problem can be captured in the terms of a classical,
mathematical optimisation problem, as it was presented in Chapter 2.
The two situations described above are illustrated in Figure 41, using the conceptual
scheme of a design task environment and a design problem space. Given that the
underlying assumptions are true, we are able to locate optimum designs21.
Unlimit ed comput er
capacit y
Perfect knowledge
about t he ext ernal
environment
Perfect model of t he
ship t o be designed
Perfect model of t he
design problem
53
problem and the design environment that should be decisive in the choice of a
problem-solving paradigm, and thus is the focus of this chapter.
This is not to say that all ship design problem situations in all details have a task
environment corresponding to the above characteristics. Rather, it represent a
prototypical case a case which most designers would recognise as typical and to be
common for most ship design situations. In the following, a more detailed discussion
of each characteristic will be given.
21
A parallel to this perfect knowledge design situation is the dream attributed to Leibniz of a
universal algebra by which all knowledge, including moral and metaphysical truths, can someday be
brought within a single deductive system [Genesereth, 1988]
54
55
system itself. Examples are power supply, propulsion, fresh water, and cargo support.
This is contrary to many other engineering systems where many of these functions are
provided by external agents.
This results in a tight coupling between ship sub-systems and substantial secondary
and higher order effects on the performance as a consequence of design changes. Such
effects are particularly evident upon changes in weights and volumes, since each
alteration here in most cases affects the ships displacement directly22. The presence
of secondary and higher order effects contributes significantly to the complexity of
the design process, since it inhibits an effective partitioning of the overall design
problem. This has lead to the expression leaky modules describing the partitioning
of the top-level design problem into mutually dependent sub-problems. This will be
further discussed later in this chapter.
A more thorough treatment of the effects of couplings between sub-functions of
engineering systems on the quality of design is given by Nam Suh [Suh 1990]. As one
of two invariant principles of good designs he asserts in his Independence axiom
that
an optimal design always maintain the independence of the functional
requirements
This is difficult to achieve in ship design. For instance, the functional requirements
(FRs)
FR1:
FR2:
22
The multiplicatory effect of the change of independent weights on the displacement can be
captured in Normands Number, after J. A. Normand who published the method in 1885
23
To avoid secondary effects in a decoupled design requires the FRs to be satisfied in the proper
sequence. A third category is uncoupled designs where the FRs are completely independent. In
uncoupled designs the sequence of the decisions does not matter.
56
Generally, preliminary ship design problems are coupled, and this coupling is one of
the primary causes for the traditional iterative, sequential design process. As
mentioned in Chapter 2, this process is often depicted by the design spiral, aimed at
outbalancing the conflicting design requirements. In an uncoupled or decoupled
design, no iterations would be necessary. Thus, the consequence of the coupling for
the design process as such is that it makes the prediction of design performance the
form-function mapping difficult.
The physical phenomena of moving a solid body at the boundary between two fluids
Some of the most important performance characteristics of the ship relate to the
interaction between the system itself and its environment, that is, the physical
phenomena of moving a solid body in water. Due to the complicated nature of the
flow around the surface of a ship hull, satisfactory analytical methods relating the hull
form to e.g. resistance, powering requirements, and seakeeping behaviour are yet not
developed or, at best, they are immature. Thus, in the preliminary stages of ship
design we still have to rely basically on empirically based methods, and/or predictions
based on similar vessels. Though offering only limited prediction accuracy, these
methods have the advantage of being computationally inexpensive and simple to use.
For more accurate predictions of hydrodynamic performance towing tank test are
used, with a corresponding high cost and delayed feedback. Recent advances in
Computational Fluid Dynamics (CFD) are likely to change this situation. Such
programs are already used in the hydrodynamic design process [Cardena 1992],
leading to the term the numerical towing tank. Due to limited accuracy in predicting
absolute performance levels, CFD tools are especially useful in comparative analyses
of hull forms, determining the marginal influences of the hull parameters
[MacPherson, 1993]. It has also been reported a development of interfaces to CFD
tools from standard design packages such as. NAPA [Marzi and Ye 1994], that
automatically models a suitable mesh for CFD analysis. The possibility of integrating
such tools in the preliminary design process will be further improved by the
development of neutral product data exchange standards, such as STEP. The
drawback of such an integration is the overhead it induces in terms of increased
problem size and additional processing requirements. Thus, the integration of more
advanced design tools into the preliminary stage in the future is believed to increase
the present level of complexity in the form-to-function mapping.
The complexity and stochasticity of the external environment
The physical external environment consist of the boundary between two fluids; water
and air. This environment exhibit a large variation along several dimensions, e.g.
57
waves, wind, and currents. All these variations need to be taken into consideration
when the overall performance of the vessel is determined.
The external environment of the ship also comprises the social and economic factors.
As a consequence of an international and very competitive market for the services
offered by ships, these factors typically show more variation and uncertainty here than
for most other engineering structures. This adds to the difficulty of assessing
important performance characteristics. Areas of uncertainty are e.g. long-term cost
predictions, freight rates, fuel prices, cargo supply/demand, and future transport
capacity, making estimations of life-cycle costs, optimal speed, optimal cargo
capacity, difficult.
All these three characteristics the integrated structure, the underlying physical
phenomena, and the external environment contributes to the complex form-tofunction mapping found in preliminary ship design. There is a continuous
development in ship analysis methods, improving the accuracy in predicting the
performance of the ship. Still, this development is only capable of removing the
apparent complexity of the mapping, by transforming it into a representation that is
solvable by an appropriate algorithm. As said by [Simon 1973 p. 150]:
In general, the problems presented to problem-solvers by the world
are best regarded as ill-structured problems. They become wellstructured problems only in the process of being prepared for the
problem-solvers. It is not exaggerating much to say that there are no
well-structured problems, only ill-structured ones that are formalised
for problem-solvers
58
Environmental protection
Fuel economy
discharging
Reliability in service
in
transit
and
for
Obviously, it is very important to avoid errors in the design process that result in
deviations of a magnitude corresponding to the latter two of these categories, since it
may lead to severe financial liabilities. As will be discussed later in this chapter, the
59
high cost of error has consequences for the design problem space in terms of
extensive performance modelling and verification of the proposed design, and need to
be taken into consideration when devising a model for supporting decision-making in
preliminary ship design.
60
such precedence is far more scarce and thus a more detailed specification is needed.
The impact on the design problem space is that the domain tradition generates a
number of design constraints - constraints that are not necessarily based on the design
specification, but implicitly assumed by all parties involved. Another consequence for
the design problem space is that it makes experience and knowledge embedded in past
design cases particularly important.
4.1.6 Strict time and resource constraints on the preliminary ship design
process
To stay competitive, a design office need to be able to respond quickly to tender
invitations24. At the same time, the outcome of the preliminary design process incur
serious obligations for the shipyard, both by deterring the utilisation of the production
resources for a long period of time, and by the committment made to the customer
through the contract (see Section 4.1.3). This require the technical and economical
feasibility of the proposed solution to be well founded. The negotiation of this tradeoff is difficult. As said in [Skipskonsulent 1991]:
"There is a perceived need for new, custom-made competitive design,
but there is far too little time for either of the parties to develop it. The
shipowner has his options badly reduced, and is forced to try to find a
standard design that meets the bill as well as possible, and within the
time limit available."
Because of reduced production times, some activities have been moved earlier relative
to the design time line. Thus, some decisions that earlier were made on the basis of a
detailed design description and/or production plans now need to be made on the basis
of knowledge available after the preliminary design stages. This put further demand
on the quality of design knowledge generated during the early stages.
In [Langenberg, 1982], the time allowed for the submission of a tender is estimated to be about
three to five weeks
61
62
63
example: Given 10 design variables, each to be evaluated at ten different values, the
total number of combinations becomes 1010. If we assume that each design evaluation
takes 1 millisecond, the total computer time needed will be 107 seconds more than 3
months. This is often referred to as the curse of dimensionality. We see the parallel
problem in chess. Even if we here have a complete formalised model often termed a
strong theory domain [Aamodt 1993] the deduction of an optimal plan leading to
victory is not possible because of the dimensionality of the problem. To compete with
human masters, the chess program must apply intelligent search methods in
addition to brute force.
Task Environment
Problem Space
Mult i-dimensional ,
part ly non-monet ary
performance
l ti
Complex mapping
bet ween form and
funct ion
Shallow knowledge
st ruct ure (high
soft /hard info. rat io)
Concluding this, I claim that satisficing rather than optimising is an important invariant of the preliminary ship design problem space. It is a consequence of some of
the characteristics of the task environment discussed in the previous section the
inability to formalise the real-world problem, the complexity and dimensionality of
the problem, and shallow knowledge structure of the domain. This is illustrated in
Figure 43.
64
25
For instance, for the formulation of the optimisation problem as in Section 2.2.3, the necessary
conditions for optimality are the Karush-Kuhn-Tucker (KKT) conditions, given as: 1) h(x*) = 0, g(x*)
0, 2) f* + Th* + Tg* = 0T, 0, 0, Tg = 0, where x* corresponds to the optimal design
solution
65
achieved by decomposition help to make the problem tractable for both the computer
and the human designer.
The main difficulty with decomposing the problem is to maintain a sufficient degree
of independence between the individual sub-problems. This is especially so in the
ship design domain, because of the high degree of interaction between the different
ship sub-systems (see Section 4.1.1). This has lead to the description leaky
modules.
There are different ways of dealing with these leakages. Traditionally, the main
strategy has been what is commonly termed outbalancing, as it is captured in the
ship design spiral. In this approach a change in one of the modules initiates the
recalculation of all other modules until a sufficient degree of consistency is reestablished. To what extent this is a viable approach depends to some degree on the
design task under consideration. For instance, the secondary effects of adding an
independent weight in a VLCC design model are far less problematic than for a
semisubmersible design model.
A second strategy is to seek a partitioning of the problem that keep the interactions
between the modules as small as possible. This was the strategy of axiomatic design
discussed in Section 4.1.1. A third alternative is to try to couple the different subproblems by the explicit modelling of the interaction effects. An example of this is
given in [Smith 1992], where the determination of the main characteristics of the hull
and the propeller for a frigate are modelled as two separate decision problems, but
with the interaction effects taken care of by the solver.
My main point here is that the complexity of the design problem will in most practical
design situations require a decomposition strategy and the choice of this strategy is
important for the quality of the design result. Hence, when proposing a design model
and a design paradigm, the support for an efficient decomposition strategy need to be
taken into consideration.
66
But this is not an all-or-none choice; in real-world design situations there is a constant
negotiation between making commitments and keeping the problem open. A possible
strategy towards a favourable trade-off is to keep the solution on a high level of
abstraction as long as possible. An example of this is reported in [Levander 1991]
from the cruise ship design domain. Here, the importance of keeping the solution on a
functional level as long as possible is emphasised, thus maintaining the option to
select different conceptual solutions. On the other hand, if we leave the functional
domain, we easily get locked into exploring the design space along a set of predefined dimensions, even if we keep the problem open in the sense that we postpone
the assignment of numerical values to the design variables.
An alternative to consecutively determining the value of particular design variables is
to make commitments in terms of systematically constraining the search space. This is
intuitively done by experienced designers for instance when an appropriate range
for the design variables is selected at the outset of the design process. It is obviously a
reasonable trade-off against exploring the whole domain of real numbers. Still, by
every addition of a constraint to the design space the designer faces the risk of
discarding a potentially favourable solution.
For the real-world design of complex systems like ships, the necessity of making
commitments at some level of detail along the design process is an inescapable fact.
The development of more advanced design tool is not likely to change this. Rather,
the focus should rather be on how to provide support for the rational negotiation of
such commitments. Alternative strategies for this will be further discussed in
Chapter 8.
For instance, this extrapolation may be accomplished by using the the total differential method.
E.g. the change in displacement with respect to a change in an independent weight Wi may be
67
that produces a set of empirical relations that may be useful for a larger range of ship
sizes and types. Well-known examples are the regression analyses of model and trial
test results at the Netherlands Ship Model Basin, aimed at developing a functional
relationship between the ship main characteristics and ships resistance and
propulsion properties [Holtrop 1977], and the work by [Watson and Gilfillan 1976],
developing empirical formulas for a large range of ship characteristics.
The point here is that the extensive reliance on knowledge embedded in existing
design solution is an important characteristic of the design problem space. In Chapter
6 I will discuss further how this can be reflected in the structure of the design model.
Level "zero":a top-level description of a potential design based on the ship main
characteristics
Level "one":a more detailed description of the ship. Both level "zero" and "one"
are parameter based descriptions
Level "two": introducing detailed geometry of the ship, including compartmentation. The corresponding design analysis is based on the defined geometry
(and not on the design parameters)
Level "three":introducing the geometry of the appendages. Thus, still more exact
calculations are possible in order to determine the corresponding performances
expressed as = Wi/(1- -1Wii), assuming that the individual weights may be calculated by
formulas following the pattern Wi = ki
68
The use of abstraction is crucial to all design in complex domains, since we are not
able to model all aspects of reality in the computer. The reality need to be simplified,
by ignoring those aspects that are of less relevance to the current context.
Design Task
Environment
Design Problem
Space
69
Mult i-dimensional ,
part ly non-monet ary
performance
l ti
Complex mapping
bet ween form and
funct ion
Personalised
evaluat ion funct ion
and st opping rules
FLEXIBILITY
Sat isficing rat her
t han opt imising
Shallow knowledge
st ruct ure
St rong domain
t radit ion
"One-of-a-kind" and
"engineered-t o-order"
solut ions
EXPRESSIVENESS
Decomposit ion
int o leaky modules
PERCEIVABILITY
Ext ensive use of
empirical relat ions
and previous cases
EFFICIENCY
Mult iple abst ract ion
hierarchies
Delayed/limit ed
feedback from t he
world
Figure 44: The preliminary ship design problem space and task environment, with
corresponding critical factors by which to evaluate the design model
70
Design Task
Environment
provide t he
set t ing for
Design Model
Figure 45: The structure of the design model is determined by the design problem
space, which again is determined by the task environment
4.3.1 Flexibility
By flexibility is understood the freedom the designer has to meta-design a design
process that is both in conformance with the characteristics of the design problem
under consideration, and in accordance with his/her personal preferences and level of
expertise.
The focus on flexibility as a critical factor is founded on several of the invariants of
the design problem space. One is the existence of multiple abstractions and
abstraction hierarchies, that was discussed in Section 4.2.6. This requires a flexible
model of the ship that can evolve along the process, starting from a coarse description
in the beginning to a detailed description in the end. Part of the problem of
incorporating this kind of flexibility lies in keeping the model consistent when more
detail is added to the structure, or the focus is changed from a high level abstraction
(e.g. V/Lpp-ratio, Lpp/.33-ratio, form coefficients) to low level abstractions (e.g. hull
dimensions, detailed hull geometry).
Second, flexibility is needed to add new knowledge to the model as knowledge is
gained through the design process. Only in very few cases the state of knowledge at
the outset of the design process is sufficient to formulate a complete model containing
all aspects required to solve the design problem. Examples are for instance the
addition of new constraints to limit the search space, the reformulation of goals, the
revision of the parameters in empirical relations, or the addition of an ad-hoc rule to
guide the search process.
And third, flexibility is needed in focus the attention of the design problem to those
aspects of the design problem that are most relevant in order to determine the overall
performance of the vessel. This is strongly related to the presence of personalised
evaluation functions and stopping rules as an invariant of the design problem space.
The support for a flexible meta-modelling capability varies substantially among
existing design tools. The most common is a relatively static model that is implicitly
71
defined by the set of analysis procedures on which the application is based, with little
or no possibility for the designer to make changes. However, in Chapter 2 we also
saw an example of the opposite in the DESIGNER application, where the basic
structure of the design model allowed for a very high degree of freedom in the metamodelling of a particular design problem. This example also illustrate that the support
for flexibility need to be founded on the low-level representation of the relevant
domain concepts.
However, the support for flexibility is not achieved without a cost. Compared to a
highly structured model with a pre-defined, fixed set of design variables and a given
template of design analysis procedures, a model explicitly supporting meta-design
tend to be considerably more complex, and thus less efficient with respect to the
computational efficiency and size of the computer implementation.
4.3.2 Expressiveness
The expressiveness of a model refers to the ability to capture the relevant knowledge
content of a particular domain. In the context of ship design this comprises all
knowledge useful for generating, analysing, and evaluating potential design solutions,
ranging from first principles facts to experience-based and intuition-like rules-ofthumb. A parallel to expressiveness is the notion of epistemological27 adequacy in
[McCarthy and Hayes 1969], stating that a representation is called epistemologically
adequate for a person or machine if it can be used practically to express the facts that
one actually has about the (relevant) aspects of the world.
In practice, epistemologically adequacy need to be weighted against the added
complexity it invokes for the design model. Thus, types of knowledge only marginally
relevant for the domain may be discarded in the interest of simplicity. Knowledge
types may also be discarded because the corresponding inference procedure operating
on this knowledge is either non-existent or difficult to implement.
In most existing, commercially available ship design tools, the only type of
knowledge that is explicitly supported is numerical knowledge. This knowledge is
expressed basically in a procedural form, usually implemented as design analysis
programs operating upon a design object implicitly defined by a collection of
attribute-value pairs. Though this representation is preferable from an operational
point-of-view, and has in most respects been the only realistic alternative28, it is also
27
epistemology
1. The division of philosophy that investigates the nature and origin of
knowledge. () [American Heritage Dictionary, 1991]
28
It may be argued that for instance a rule-based representation has been a realistic alternative. The
primary argument against this stand is the fact that this representation has not found wide-spread use in
commercially available ship design applications
72
clear this representation lead to important parts of the semantic content of the domain
knowledge being lost. This will be further discussed in the next chapter, in relation to
the knowledge content of the preliminary ship design process.
4.3.3 Perceivability29
In [Rzevski 1980], perceivability is defined as the ability to hold all the important
aspects of a problem and all their interrelationships in mind at one and the same time.
He asserts that:
People tend to reject, ignore, or undermine the importance of systems
whose purpose, aims, function, or organisation they cannot perceive
If people are forced to interact with an imperceivable system they are
likely to commit an inordinately large number of mistakes
To what degree a particular design model is regarded as imperceivable are strongly
correlated to the complexity of the model. Correspondingly, measures to improve
perceivability are closely connected to complexity reduction techniques in
information systems modelling. Many of these techniques are associated with objectoriented programming, and comprise for instance methods such as the encapsulation
of information and procedures within objects that have a simple and well-defined
external interface, and the use of inheritance that enable attributes and operations to
be shared among information entities based on a hierarchical relationship.
As was the case with flexibility as a critical factor, there are limits to what measures
should be taken in order to increase the perceivability of a model. This stems from the
dual role of the model in the design process30, as illustrated in Figure 46. On the one
hand the model serve as a basis for communication and understanding of the realworld design problem, and on the other hand it serves as a definition of the problem to
be solved by the computer. For the first of these roles, a high degree of abstraction
and simplicity is preferable, while for the latter, a detailed and concrete model is
necessary in order to be computable.
29
4.3.4 Efficiency
73
Human designer
Design
model
Basis for implement at ion
and processing
Comput er
31
This may for instance be a design analysis procedure where the designer has access only to the
compiled version, and not the source code nor a proper documentation or explanation a rather
common problem in many existing ship design tools
74
primarily directed towards enhancing the human efficiency in the design process.
These factors will usually have a negative effect if we focus on the computational
efficiency of the system alone. For instance, when enabling a high degree of
modelling flexibility, we will at the same time run the risk of sacrificing the efficiency
that can be gained from a simpler knowledge representational formalism and
inference methods. As an example, consider the representation of a simple constraint,
e.g.
LBRATIO: LPP/B < 5.9
(Constraint)
tag='LBRATIO'
leftP
rightP
(NumElem)
value = '5.9'
(OperElem)
value = '/'
leftP
(PropElem)
value = 'LPP'
rightP
(PropElem)
value = 'B'
32
These considerations can be captured in Coggins Law of Software Engineering, which states
that Pragmatics must take precedence over elegance, for Nature cannot be impressed
75
Our knowledge about the external conditions of the ship system is limited. The
performance of the ship is a result of its interaction with the surrounding
environment, both physically, economically, and socially. Thus, in order to
evaluate the design solution correctly we need to model all relevant aspects of this
environment. This is difficult both because of the inherent uncertainty involved in
predicting the future, and because it would imply a very large and complex design
model.
Our knowledge about our own preferences and trade-offs is limited. The
consequence of this is that we are usually not able to explicitly formulate goals
that covers all relevant performance aspects completely, which is a necessary
requirement for identifying optimal solutions. The result is that real-world ship
design is a process of finding satisficing solutions, where the evaluation of
design alternatives relies to a large degree on human intuition and experience.
Our knowledge about the design object is limited. That is, given a description of a
ship, we are not able to accurately predict all aspects of the corresponding
performance at the design stage, even under idealised conditions.
And even if we were able to derive a perfect model covering all the three
preceding aspects, it would be intractable from a computational point-of-view.
The situation would be like for chess programs: even though we are able to devise
a complete model of the domain, we still are unable to develop a program that is
capable of supporting optimal decisions, as a result of finite computational
capacity.
76
These deviations from an idealised world need to be taken into consideration when
proposing a model for ship design.
5 References
Ames, R. M. and Lynaugh, K. M. (1988) "A Review of Hullform Design Systems for
the Marine Industry", Marine and Offshore Computer Applications, September 1988
Andrews, D. (1981) "Creative Ship Design", The Naval Architect, November 1981
Andrews, D. (1985) An Integrated Approach to Ship Synthesis, The Naval
Architect, April 1986
Antonsson, E. K. (1987) "Developing and Testing Hypothesis in Engineering Design
Research", J. of Mechanisms, Transmissions and Automation in Design, Volume 09,
January 1987
Ashley, S. (1992) "Engineous Explores the Design Space", Mechanical Engineering,
February 1992
Balachandran, M. (1993) "Knowledge-Based Optimum Design", Computational
Mechanics Publication, London
Booch, G. (1994) "Object-Oriented Analysis and Design with Applications",
Benjamin/Cummings, Redwood City, California
Bras, B. A. (1992) "Foundations for Designing Decision-Based Design Processes",
Ph.D. Dissertation, University of Houston, Houston, Texas
Braun, R. D., Kroo, I. and Gage, P. (1993) "Post-Optimality Analysis in Aerospace
Vehicle Design", AIAA Aircraft Design, Systems, and Operations Meeting, August
11-13, 1993, Monterey, California
Cardena, E. M. (1992) "Use of CFD Tools in the Hull Lines Design Process", Design
of Marine and Offshore Structures (CADMO92)
78
References
79
80
References
Gage, P. and Kroo, I. (1993) "A Role for Genetic Algorithms in a Preliminary Design
Environment", AIAA Aircraft Design, Systems, and Operations Meeting, August 1113, 1993, Monterey, California
Gamma, E., Helm, R., Johnson, R, Vlissides, J. (1994) Design Patterns - Elements of
Reusable Object-Oriented Software, Addison-Wesley Publ., Reading, Massachusetts
Genesereth, M. R. and Nilsson, N. J. (1988) "Logical Foundations of Artificial
Intelligence", Morgan Kaufmann Publishers Inc.
Georgescu, C., Verbaas, F. and Boonstra, H. (1990) "Concept Exploration Models for
Merchant Ships", CFD and CAD in Ship Design, Wageningen, The Netherlands,
Elsevier Science Publishers B. V.
Goel, V. and Pirolli, P. (1989) "Motivating the Notion of Generic Design within
Information-Processing Theory: The Design Problem Space", Design Studies, Volume
Spring 1989
Goel, V. and Pirolli, P. (1989) "Motivating the Notion of the Generic", AI Magazine,
Spring 1989
Goldberg, D. (1989) "Zen and the Art of Genetic Algorithms", 3rd International
Conference on Genetic Algorithms, George Mason University, Morgan Kaufman
Publishers
Goldberg, D. E. (1989) "Genetic Algorithms in Search, Optimization, and Machine
Learning", Addison-Wesley, Reading, Massachusetts
Grigoropoulos, G. J. and Loukakis, T. A. (1992) "A New Method for Developing Hull
Forms with Superior Seakeeping Qualities", Design of Marine and Offshore
Structures (CADMO92)
Hagen, A. (1992) Prosedyrer for Prosjektering av Skip (Procedures for the Design of
Ships), Dept. of Marine Systems Design, Norwegian Institute of
Technology/MARINTEK,Report no. 260292
Hagen, A. (1993) "The Framework of a Design Process Language", Ph.D.
Dissertation, Department of Marine System Design, Norwegian Institute of
Technology, Norway
Hales, C. (1987) "Analysis of the Engineering Design Process in an Industrial
Context", PhD Dissertation, Gants Hill Publication
81
82
References
83
84
References
85
86
References