Constructing the general arrangement of a ship tends to be a very complex naval architectural problem. There are multiple systems that all work integrally. Some of these systems need to be placed together, others need to be separated and some systems probably need to be both at t
...
Constructing the general arrangement of a ship tends to be a very complex naval architectural problem. There are multiple systems that all work integrally. Some of these systems need to be placed together, others need to be separated and some systems probably need to be both at the same time for different objectives. The preliminary ship design phase is all about finding the balance between these objectives and possibilities within the arrangement. Automated general arrangement generation methods, such as the TU Delft pack- ing approach and IECEM, have created an algorithm and methodology that are able to generate thousands of feasible ship designs corresponding to the desired requirements. These approaches combine a bin-packing and a genetic algorithm in order to create a diverse set of possible designs. Although optimisation approaches can efficiently generate and search for a set of designs, it can not automatically select a design, van Oers et al. (2008). Out of the cloud of generated designs, choosing and evaluating those designs that are of most interest seems to be a more difficult task, DeNucci et al. (2008). Human rationale deems necessary to decide what properties of an arrangement can be analysed as “good” or “bad”. Using computational force to assist with design selection based on qualitative properties needs a way of capturing this rationale, which is demon- strated by DeNucci. Subsequently, this captured knowledge needs to be applied in a methodology to actually analyse the data by this rationale.
Design selection of computer generated arrangements is currently based upon performance parameters fo- cusing on numerical characteristics of the designs. Spatial relations within the interior layout can now only be manually analysed, which is merely possible for a singular design. This results in a lack of understanding of spatial and physical relationships, or qualitative properties, within the designed arrangement. This thesis proposes a method that is able to quantitatively compare these qualitative properties by creating a measure of effectiveness of the interior layout, or arrangement, of the designs. This is done by creating a simplified representation of the 3D arrangements using network theory.
By converting system objects and connections of these systems into nodes and edges, a mathematical net- work description of the arrangement is possible. Using eigenvector and betweenness centrality measures, rank and weights are given to these system objects. Design rules based upon rationale captured by DeNucci are set up and used to analyse connections between these systems. A scoring algorithm is created that is able to combine the ranks of the systems with the captured rationale to give a singular performance parameter, or measure of effectiveness of the interior layout characteristics of the designs. Application of this method to a data set of small cruise ship designs shows the capabilities of this methodology and the ability to quantita- tively compare the qualitative properties of these designs.
A data set of over 20,000 cruise ship designs created by the TU Delft packing approach is analysed by the use of the method. This demonstrates that the proposed method can be used on an entire data set allowing design selection based on the quality of the provided arrangements. The proposed method can be used to filter the data set of those arrangements whose quality seems to be too poor to be taken into account in design selection. Having these arrangements in the data set is still necessary, as this allows the genetic algorithm to create enough diversity within the design space. The proposed method can find specific physical properties of the arrangements and should lead to improved design space exploration of layout features of the generated designs.
Capturing the quality of the design into a single measure of effectiveness brings problems as well. It allows the possibility of direct comparison of a large number of designs, but does not relate back to why designs score differently. Splitting design rules into different design objective scores, and possibly summing this later, could allow the designer to relate back to the design rationale and why a design scores better or worse. This would lead to more understanding of possibilities and problems within the proposed arrangements.
