A method for generating multiple hull forms at once using MLP (Multi-Layer Perceptron)

doi:10.1016/j.oceaneng.2025.120659

Return

A method for generating multiple hull forms at once using MLP (Multi-Layer Perceptron)

2025-04-01

PRE

Jin-Hyeok Kim

Myung-Il Roh *

In-Chang Yeo

DOI：10.1016/j.oceaneng.2025.120659

Original

Original request for help

Save

Abstract

En 中文

When designing the hull form of a commercial ship, it is common practice to make manual adjustments to the shape of an existing reference hull form to meet design requirements. In efforts to generate a variety of hull forms, many shape optimization studies have either defined the hull form using parameters or directly modified the surface of the hull form, allowing for more flexible deformations. Recently, there has also been a surge of research utilizing deep learning-based generative models to generate new hull forms based on existing hull forms. While these methods offer more flexibility in shape deformation than traditional approaches, they have the limitation of needing to account for design requirements such as displacement and block coefficient of a ship. In this study, to address these limitations, we proposed a hull form generation method that enabled the simultaneous generation of multiple hull forms using MLP (Multi-Layer Perceptron) from deep learning. The MLP was used to deform the hull form, and a triangular mesh structure of the hull form was employed for surfacebased evaluation. Evaluation metrics included block coefficient, which is related to design requirements, the position of the center of buoyancy to promote the generation of diverse hull forms, and fairness, which relates to surface quality. By applying the proposed method, we were able to generate a variety of hull surfaces that met design requirements from a single mesh input. To validate the method, we applied it to publicly available hull forms such as the KCS (KRISO Container Ship), KVLCC2 (KRISO Very Large Crude Oil Carrier), and JBC (Japan Bulk Carrier). The results verified that the method successfully generated diverse hull forms while satisfying design requirements across multiple ship types.

Keywords:

Hull form

Hull form generation

MLP (Multi-Layer Perceptron)

Mesh deformation

Surface modification