Multivariate Data Analysis in Conceptual Vessel Design - A Study of Offshore Construction Vessels

Abstract

This thesis investigates how the use of a design methodology based on multivariate data analysis can create a reliable and robust source of knowledge to reduce complexity, and support and complement the decision-making process of conceptual vessel design. Multivariate data analysis application in naval architecture is a fresh topic, and little literature exist in the area. A study of offshore construction vessels is performed, where all necessary steps from raw data sampling to knowledge harvesting is described and performed. Multiple regression analysis is performed on real vessel data from an offshore construction vessel database with lightweight, deadweight and cost as dependent variables. The causality of the regression equations is analysed and further used to bridge the gap between the vessel main dimensions and shape factors, and its functionality and performance yield. The approach is illustrated through a case study, where a vessel design solution designed and built after a traditional design spiral approach is investigated. Utilizing a previously conducted study of that specific design process, a new vessel design solution is proposed using the most cost-efficient solutions deduced from the multivariate analyses. With the existing design outcome as a benchmark, a comparative study of the existing and proposed vessel solution is performed. The multivariate approach is discussed as a part of a holistic vessel design methodology, and the findings from the comparative analysis is utilized as a foundation for evaluation and reflection of the design method utilized. The case and regression results indicates that multivariate data analysis can be a useful tool for the naval architect in the conceptual phase of the design process as it facilitates theoretically founded form to function mapping. The swiftness of estimation also indicates that the methodology can help elucidate the requirements to the vessel in this phase. There are however concerns regarding both data quality and quantity. The author concludes that multivariate data analysis can be an appreciated methodology in the naval architect s toolbox, and that further studies should be conducted to support the development of the methodology. The approach is also advised implemented in naval architecture education.