Abstract: A data-driven approach for the preliminary design of ship hull forms has been proposed to address the issues of long design cycles and high manual effort in traditional ship hull form design methods. By focusing on the digital representation of hull lines, database construction, and data storage, classification and retrieval, a method for constructing a hull form database is proposed to enable the visualization of functionalities such as adding, deleting, viewing, modifying, and matching of hull lines. To fully utilize the existing data in the database, a feature extraction function for hull lines is developed. This function segments the ship hull surface and calculates the normal vector, Gaussian curvature and mean curvature of the hull surface, thereby facilitating dimensionality reduction of the hull surface features. A convolutional neural network is then employed with the reduced-dimensional features as inputs to predict the ship resistance in static water. Experimental results show that the database can effectively manage the data of the ship hull form, and the error of the total ship resistance coefficient predicted by the neural network is within 10%. This work enables the inheritance of high-quality data into the preliminary hull form design of new ship types.

Key words: data-driven, hull form design, feature dimensionality reduction, neural network

摘要： 针对传统船舶型线设计方法设计周期长、人工耗时多的问题,该文提出基于数据驱动技术的船舶型线初步设计方法,围绕船体型线的数字化表达、数据库构建,以及数据存储、分类与调用等,制定船体型线数据库的构建方法,实现型线可视化添加、删除、查看、修改和匹配功能。首先利用数据库中已有数据,建立船体型线特征提取函数,对船体曲面进行分割并计算曲面的法向量、高斯曲率和平均曲率,实现船体曲面特征降维;然后建立卷积神经网络模型,以降维后的特征参数作为输入,预测船舶的静水阻力性能。测试结果显示：该数据库能便捷地管理船舶型线数据,且神经网络对船舶的总阻力系数预测误差在10%以内,还可将优秀的数据继承到新船型的型线初步设计中。

关键词: 数据驱动, 型线设计, 特征降维, 神经网络