This thesis describes the design and implementation of an optimisation system for America's Cup Class (ACC) yachts. The system, named VESPA, uses a measure of merit that closely approximates the actual America's Cup race format; a round-robin match-racing tournament, held over many races between a population of candidate designs, using a stochastic wind model. VESPA was used by the Alinghi team to provide design recommendations for the 2007 America's Cup. The optimisation of racing yachts is a problem that has been considered resistant to full analysis due to its complexity. Consequently, attempts at yacht design optimisation to date have been restricted to simplified subsets of the problem. While Velocity Prediction Programs (VPP) have been widely used to provide details of sailing performance for one or more yachts, the statistical models on which these programs are based have not been sufficiently accurate to allow optimisation of hull shapes. Other efforts to automate yacht design optimisation have used an objective function that evaluates the performance of each boat using Computational Fluid Dynamic (CFD) analysis of the hull. This approach suffers from long execution times, which may result in the adoption of a restricted measure of merit, such as hull resistance at a small number of forward speeds, heel and yaw angles. In order to permit the use of the chosen measure of merit while retaining acceptable performance, a sparse sample of designs, derived from a parent hull using a novel parametric transformation method, had their hydrodynamic characteristics calculated by the SPLASH potential flow code. The output from SPLASH was subsequently used to train a set of neural-network based hydrodynamic metamodels for use by the VPP. The need to assess a population of designs for the tournamentbased measure-of-merit makes the problem well suited to stochastic, population based optimisation methods. As a result, a Genetic Algorithm (GA) was chosen to perform the optimisation, using a parsimonious Race Modelling Program (RMP) to simulate a tournament of races based on performance data provided for each boat by the VPP. Each component within the VESPA system was validated to ensure confidence in the optimisation results. Optimisation runs were performed over several months using multiple parent models to investigate the effect of changes to various design variables. Finally, a design optimised by VESPA was tank tested at 1/3 scale, confirming the improvements over its parent design predicted by SPLASH. VESPA proved itself capable of making genuine design improvements to an existing parent model while retaining reasonable execution times. VESPA also revealed several unexpected insights into the nature of the solution space for the design of ACC yachts, including multiple optima and the potential for intransitivity in the solution when interactions between boats at rounding marks are considered.
History
Publication status
Unpublished
Rights statement
Copyright 2010 the author. all rights reserved Thesis (PhD)--University of Tasmania, 2010. Includes bibliographical references