Douglas Read

Date of Award


Level of Access

Open-Access Dissertation

Degree Name

Doctor of Philosophy (PhD)


Interdisciplinary Program


Michael Peterson

Second Committee Member

Donald Grant

Third Committee Member

Richard Kimball


During the initial phases of ship design, the naval architect would like to have as much information as possible about the design space. This information not only helps determine a good set of initial characteristics, it allows for informed design changes when reacting to evolving requirements. One of the most difficult performance measures to evaluate is the ship wave drag. This estimate is important in an optimization, because wetted surface and wave drag must be balanced. Multi-parameter optimization algorithms exist, but need a very fast and inexpensive fitness evaluation for them to be effective. Even though linear theory does capture some of the physics of the problem, it has long been out of favor due to its tendency to grossly over-estimate the wave drag. The other options available are parametric drag estimates and state-of-the-art boundary element codes. Here we present an intermediate method that makes a parametric correction to the linear theory using an artificial neural network. The method starts with a training set consisting of a large number of panel code evaluations for a systematic hull series, and then uses two approaches to the parametric correction. The first method uses the ratio of linear theory to panel code data as targets for an artificial neural network with parametric inputs. In the second method, we re-derive the linear theory with a new boundary condition, leading to a waterline integral term with unknown coefficients. The linear theory error is then used in a constrained minimization problem to solve for the unknown coefficients, which again provides targets for a neural network. Coupled with a mathematical hull form that can approximate realistic hull shapes, the results show promise for an intermediate wave drag estimation method that is fast enough to be used as a fitness evaluation for a multi-parameter optimization routine such as a genetic algorithm.