Summary: | 碩士 === 國防大學理工學院 === 造船及海洋工程碩士班 === 99 === In this work, metamodeling techniques have been used to study a high-speed trimaran model of a scale of 1:100 that was experimented in a circulating water channel. The parametric analysis was investigated to obtain a robust design of minimizing the drag of the ship model. There are three control factors, that is, transverse span, longitudinal position and draft, and one noise factor-ship speed, which were used to conduct the parameter design in this work. An integrated approach, in which orthogonal arrays are combined with artificial neural networks and a genetic algorithm, is proposed seeking for the optimum parameter conditions. This approach can improve the shortcoming while one uses the Taguchi’s method alone, that is, the real optimum parameter values could not be obtained. The primary predicted values and parametric conditions for a minimum drag should be acquired by the Taguchi's method first. Then, treatment of each row in the orthogonal array together with its relative response is used to establish a set of training data (input/target pair) to the artificial neural network. The network by training with the data can describe precisely the nonlinear relationship between inputs and targets. Based on the nonlinear model estimated by the neural networks, a genetic algorithm is adopted to seek for the real optimum parameter levels. The result shows that the effect of the transverse span is of significance to the drag of the trimaran model, and the interaction between the transverse span and longitudinal position would be considered to the sensitivity of the drag. Finally, we were going to carry out the experiments on the confirmation to verify the optimum conditions whether it has reliability.
|