Ship-D

Ship-D is a dataset of parametric ship hulls to train machine learning models to design hull forms. The dataset contains 82,168 hull forms. 45 geometric parameters define each hull, allowing the large diversity of traditional mono-hull shapes and a larger design space. There are 5 subsets of hull designs in the Ship-D dataset: <br><br>

1) Constrained Randomized Set 1: 10,000 hulls, randomly sampled using the full ranges of all 45 parameters.<br>
2) Constrained Randomized Set 2: 10,000 hulls, randomly sampled, but do not contain bulbous bows or sterns.<br>
3) Constrained Randomized Set 3: 10,000 hulls, randomly sampled, but have strictly positive keel radii and zero deadrise angle. <br>
4) Diffusion Augmented Set 1: 41,752 hulls, sampled with a guided diffusion model to increase the displaced volume and reduce the hydrodynamic drag. (*Generated using ShipGen, citation below)<br>
5) Diffusion Augmented Set 2: 10,416 hulls, subsampled from Diffusion Augmented Set 1, but having randomized bulb parameters.<br><br>

For each hull design, the dataset contains the following:<br><br>

--------------------------------------------------------------------------------<br>
Geometric measurements at ten different drafts<br>
--------------------------------------------------------------------------------<br>
1) Displaced Volume<br>
2) Surface Area <br>
3) Waterplane Area <br>
4) Area Moments of Intertia in Roll Direction<br>
5) Area Moments of Intertia in Pitch Direction<br>
6) Longitudinal Center of Flotation (Center of Waterplane Area)<br>
7) Longitudinal Center of Buoyancy (Center of Displaced Volume)<br>
8) Vertical Center of Buoyancy (Center of Displaced Volume)<br>
9) Waterline length<br>
10) Height of draft measurement<br><br>

-------------------------------------------------------------------------------<br>
Other Geometric Measures<br>
-------------------------------------------------------------------------------<br><br>
1) Gaussian Curvature<br>
2) Largest Box that can be vertically lowered into each hull (MaxBox)<br>

-------------------------------------------------------------------------------<br>
Wave resistance calculations using the Michell Integral<br>
--------------------------------------------------------------------------------<br>
Thirty-two measurements across 4 drafts and 8 Froude numbers <br><br>

-------------------------------------------------------------------------------<br>
Visual Data<br>
--------------------------------------------------------------------------------<br>
1) Code to generate an STL mesh of each hull<br>
2) Blender Files that have code to generate Five Images of each hull (Front, Plan, Profile, Starboard Bow, and Port Stern)<br><br>

More details for the hull parameters, dataset, and papers can be found at
<a href=" https://decode.mit.edu/projects/ShipGen/"> https://decode.mit.edu/projects/ShipGen/</a> <br><br><br>

*Bagazinski, Noah J., and Faez Ahmed. 2023. "ShipGen: A Diffusion Model for Parametric Ship Hull Generation with Multiple Objectives and Constraints" Journal of Marine Science and Engineering 11, no. 12: 2215.
<a href="https://doi.org/10.3390/jmse11122215"> https://doi.org/10.3390/jmse11122215</a>

Ship-D是一款用于训练机器学习模型以开展船型设计的参数化船体数据集。该数据集共收录82168个船型，每个船型由45个几何参数定义，可覆盖传统单体船型的丰富形态，并拓展更大的设计空间。Ship-D数据集包含5组船型设计子集：

1) 约束随机集1：共10000个船体，基于全部45个参数的全范围随机采样生成。
2) 约束随机集2：共10000个船体，随机采样生成，但不含球鼻艏与球鼻尾。
3) 约束随机集3：共10000个船体，随机采样生成，但龙骨半径严格为正且斜升角（deadrise angle）为零。
4) 扩散增强集1：共41752个船体，通过引导式扩散模型采样生成，以提升排水体积并降低水动力阻力（*由ShipGen生成，引用信息见下文）。
5) 扩散增强集2：共10416个船体，从扩散增强集1中二次采样得到，但带有随机化的球鼻参数。

针对每个船型设计，数据集包含以下内容：

--------------------------------------------------------------------------------
十组不同吃水下的几何测量值
--------------------------------------------------------------------------------
1) 排水体积
2) 湿表面积
3) 水线面面积
4) 横摇方向面积惯性矩
5) 纵摇方向面积惯性矩
6) 漂心纵向位置（水线面面积中心）
7) 浮心纵向位置（排水体积中心）
8) 浮心垂向位置（排水体积中心）
9) 水线长度
10) 吃水测量对应的吃水深度

--------------------------------------------------------------------------------
其他几何指标
--------------------------------------------------------------------------------
1) 高斯曲率（Gaussian Curvature）
2) 可垂直容纳于每个船体的最大箱体体积（MaxBox）

--------------------------------------------------------------------------------
基于米切积分（Michell Integral）的波浪阻力计算
--------------------------------------------------------------------------------
共包含4个吃水与8个弗劳德数（Froude number）下的32项测量结果。

--------------------------------------------------------------------------------
视觉数据
--------------------------------------------------------------------------------
1) 用于生成每个船体STL网格的代码
2) 包含生成代码的Blender文件，可输出每个船体的5幅图像：正视图、顶视图、侧视图、右舷艏视图与左舷艉视图。

更多关于船体参数、数据集与相关论文的详细信息可访问：
<a href="https://decode.mit.edu/projects/ShipGen/">https://decode.mit.edu/projects/ShipGen/</a>

*Bagazinski, Noah J. 与 Faez Ahmed. 2023. “ShipGen：面向多目标与约束的参数化船体生成扩散模型”，《Journal of Marine Science and Engineering》（海洋科学与工程学报），第11卷第12期，第2215页。相关链接：<a href="https://doi.org/10.3390/jmse11122215">https://doi.org/10.3390/jmse11122215</a>

搜集汇总

背景与挑战

背景概述

Ship-D是一个用于机器学习船体设计优化的参数化船体数据集，包含82,168个船体形式，每个由45个几何参数定义，覆盖传统单船体形状并扩展设计空间。数据集分为5个子集，包括约束随机采样和扩散模型增强的船体，并提供几何测量、波浪阻力计算和视觉生成代码，适用于工程和机器学习研究。