ニューラルネットワークを利用した中高速艇速力性能初期推定ツールの構築

概要

論文の詳細を見る
In preliminary designs of high speed craft, it is often that main engines, reduction gears, and propellers are specified referring to accumulated trial data of actual craft, which is usually converted to design charts. These stern components should satisfy the requirements of both propulsive performance and practicable stern arrangement. The design knowhow of compromise between these requirements builds up by collecting the trial data. This paper proposes a trial database system for high speed craft, which consists of a collection of trial data, mapping neural networks (what we call a memory model , and a descriptive neural network. The memory model, whose input is a design condition (length, displacement, target speed) , and outputs are the required horse power and the standard propeller specification, is generated by learning of the trial data. The descriptive neural network, which denotes the frequency of actual craft with the similar design condition, indicates the designer how conservative the given design condition is. The proposed database system is based on the trial data of 36 craft. The constructed neural networks set standard for the stern components, quickly estimate the propulsive performance, and reduce the number of iteration of the design spiral. When additional trial data is available, it is easy to modify the constructed networks, taking advantage of learning ability of neural network.
社団法人日本船舶海洋工学会の論文