A Contextual Multipartite Network Approach to Comprehending the Structure of Naval Design.

Abstract

Acquisitions fail due to complex interactions between many domains. Typical research focuses on one domain or another, e.g. process, product or organization. Systems Engineering is responsible for the bigger picture, but the complexity of naval acquisition still presents major challenges. Potential problems must be identified before mitigating their effects, requiring a complete comprehension of acquisition structure, predictive methods and lead indicators. The structure and challenges of design are a microcosm of acquisition; through multiple levels of context and increasing scale, fundamental relationships affect the outcome of a design, and thus acquisition as a whole. This thesis broadens the application of network theory for naval design from the analysis of physical systems to the general structure of design. The primary contribution is a network structure to represent the multiple domains of design simultaneously and in context, supported by unique methods for analysis and verification.

Specifically, a contextual multipartite network approach to represent the structure of naval design is developed with the application and extension of network mathematics, providing meaningful predictive insight. An algorithm for finding path lengths is adapted to quantitatively capture node to node influence across multipartite design networks, showing equivalency with a first order Taylor series expansion. The algorithm, termed path influence, is applied to predict the behavior of a ship design optimization formulation and verified using a full factorial design of experiments. A new metric, Winston centrality, is presented to compare the results of the algorithm with standard network centrality metrics. The flow of information across a design network is modeled with a continuous analogy based on Fick's second law of diffusion. Discrete information flows are then approximated using a version of the path influence algorithm, verified using discrete event simulation.