Jin-Hyeok Kim, Myung-Il Roh, In-Chang Yeo, "A Method for the Automatic Generation of Hull Form Surfaces Based on MLP (Multi-Layer Perceptron) Considering Design Requirements", Proceedings of G-NAOE 2024, Southampton, UK, 2024.11.05-09
| Abstract | When generating complicated geometries like the hull form of a commercial ship, the hull form deformation approach often involves slightly modifying the geometry to satisfy design requirements based on the hull form of a previously built reference ship. In the industry, methods such as the 1-Cp method or the Lackenby method have been used to edit the hull form by moving the section plane of the reference hull form longitudinally. While these methods have the advantage of quickly deforming the hull form when satisfying design requirements, they have the limitation of having a very limited degree of freedom in hull form deformation. To overcome this limitation, many hull form optimization research have used methods such as Free Form Deformation (FFD) and Radial Basis Function (RBF), which define control points and directly modify the surface of the hull form. Although this approach allows for very flexible deformation of the hull form, it has the limitation that the designer must define the position and direction of movement of the control points, and further consideration of design requirements is required. This study proposed an interpolation method that used the Multi-Layer Perceptron (MLP) of deep learning instead of the RBF method to solve this problem. The most significant characteristic of the MLP is that it can interpolate given points in various ways and perform multiple interpolations at once. Initially, the MLP was used to define multiple candidate surfaces. Points on the surface were extracted using surface meshes to approximately calculate values such as surface volume and fairness. Some of the vertices in the mesh were fixed points, while the rest were movable points. The extent to which these movable points move was automatically determined by the MLP. The calculated volume of the surface was used to satisfy design requirements such as the block coefficient, and fairness was used to improve the quality of the generated surface. Additionally, to generate a variety of surfaces, the centers of buoyancy of the generated surfaces were calculated to maximize their distance. The proposed method can generate various surfaces that interpolate the given points, especially finding multiple surfaces that satisfy the design requirements. Moreover, the proposed method did not require the definition of the location and direction of movement of control points, allowing for automatic deformation of the surface. The method was validated using the hull form of the publicly opened container ship, KRISO Container Ship (KCS). By using the proposed method, multiple hull forms were generated, and it was confirmed that the generated hull forms satisfied the design requirements. |
|---|---|
| Publication Date | 2024-11-05 |