A Network Science Approach to Understanding and Generating Ship Arrangements in Early-Stage Design.

In recent years, automated approaches for creating ship general arrangements in early-stage design have been developed. These approaches seek to avoid “black box” implementations by keeping the designer involved in the layout generation and selection process, but they do not avoid it entirely. Existing methods first generate layouts, next evaluate each layout’s quality, and subsequently filter out poor designs in an iterative process. In addition, desires to move toward full distributed system layouts in early-stage design have only led to more highly-refined CAD-style implementations requiring extensive modeling and computation time. This dissertation asserts that there is a need to shift away from the current trajectory toward higher-fidelity three-dimensional layout models and re-vector toward a perspective that focuses on understanding and inherently respects the fundamental underlying relationships among elements within those models. The research offered in this thesis uses network science to envision the layout problem from a new perspective. In this view, design relationships are information inputs into layout-related analyses rather than only post-processors for evaluating layouts. This is consistent with existing design processes in which human designers attempt to keep relevant relationships in the back of their mind at all times to inform decisions. Network nodes represent ship compartments and edges correspond to design constraints forming a relationship network. First, network concepts of centrality and hierarchy are used to highlight and rank the embedded drivers of an early-stage arrangement prior to developing spatial layouts by directly analyzing the relationship network in a methodical and holistic manner. The obscured design intent of a notional WWII naval vessel is exposed using the hierarchical approach. Second, a network partitioning method is used to cluster shipboard elements into communities of mutually-compatible elements to minimize the degradation of other items located in the same region of the ship. These communities can form the basis of functional zone definitions. Varying the number of partitions reveals a multi-scale depiction of the relationship network. Third, the communities are assigned to structural zones based on cumulative zone preference values. Finally, two new visualization techniques help designers establish connections between the network of inter-element relationships and spatial ship arrangements.PhDNaval Architecture & Marine EngineeringUniversity of Michigan, Horace H. Rackham School of Graduate Studieshttp://deepblue.lib.umich.edu/bitstream/2027.42/96126/1/justinwg_1.pd