Page 313 - Practical Design Ships and Floating Structures

288

that does not directly involve a userdriven iterative design process. Conceptually, this situation can
be illustrated by a case in which a designer locates two designs both deviating somewhat from the
desired specification. The question arises as to how one can find out or “interpolate” from the design
data a new design (an emerging design) that meets the requirements of the given specification. The
reuse approach proposes an objective-driven search (eg. multiple objective optimisation) using
relevant ship design knowledge to identify a range of efficient solutions for further consideration.
These designs constitute the Pareto optimal set, where designs can only be improved with respect to
one objective at the expense of others. The mechanics of the “interpolation” process is automated so
that a designer is largely shielded from the background computations that are involved, partly as a
result of reusing higher level knowledge (e.g. rule induction based relationships between variables)
obtained from previous design efforts using the same database.
The reuse of design data can also be extended to satellite applications for which the data was not
originally prepared. For example, a designer may only have a relatively small set of data concerning
some aspects of ship performance. The designer may want to populate a relatively large database,
which does not currently contain data elements concerning these aspects of performance, with
approximated performance data elements based on the small set of design data that is available. In
such a situation an approximate response surface to a set of given performance data. Artificial Neural
Networks (ANN) and rule induction are two methods that can be employed within this reuse scenario
for extending the reuse of design data, through this response surfm approach, to satellite applications.

4 AN EXAMPLE APPLICATION
A small general cargo/container ship preliminary design application based on a decomposition and
reuse approach is presented as an illustrative example in this section. The ship design problem is
decomposed into two subproblems. Design reuse approach is then used to deal with one of the sub-
problems using the “interpolation” approach. The reuse of the design data is also extended to a
satellite application. The satellite application is illustrated via the extension of existing design data to
include seakeeping characteristics.
Suppose a set of design objectives is given: minimise transportation cost (FI), maximise annual cargo
carried (F2). minimise lightship weight (Fj), minimise the probability of machinery space flooding (Fd)
and minimise the probability of losing auxiliary power (F5). The first three objectives are driven by
economic considerations whereas the last two objectives are related to survivability of the vessel.
Mathematically speaking, a ship design problem can be represented by a set of objectives and
constraint equations over a finite number of design variables. For the illustrative example, the five
objectives mentioned above can be expressed in terms of five formal objective functions. In addition
to these five objective functions, eleven constraint equations are also to be taken into account to ensure
feasibility.

4.1 Design Decomposition

In this example the given problem is decomposed into two sub-problems using the hypergraph
partitioning approach mentioned in the previous section.
The objective functions can be represented in a tabular form. A partial table of the design problem is
shown in Table 1. In Table 1, the nodes (functions) are listed in the rows and the hyperedges (design
variables) are listed in the columns. For example, from Table 1, it can be seen that hyperedge (design
variable) xf connects functions FI, Fz, F3, ..., H9. In a complex problem (involving hundreds if not
thousands of nodes and hyperedges), it is not immediately clear as to how the problem can be