Hull form optimization of fully parameterized small ships using characteristic curves and deep neural networks

Abstract

Designing a hull form typically involves beginning with a reference hull form based on ship owner requirements, editing the hull form to satisfy the requirements, and determining the most efficient hull form. Numerical analyses using Computational Fluid Dynamics (CFD) were employed to assess the performance of the hull form. However, these analyses require extensive computational resources, making it challenging to perform thorough analyses within the design timeframe. To address this issue, this paper proposes an approach that involves defining a range of hull forms with characteristic curves, predicting their performance using Deep Neural Networks (DNNs), and subsequently determining the optimal hull form based on these predictions. Initially, the hull form of a small ship was defined using four characteristic curves and parameterized using 29 variables. Fairness optimization was performed using these characteristic curves to define the hull form surface. By varying 29 parameters, 896 different hull forms were generated, with CFD analysis conducted for each variant. These data were then used to build a DNN model capable of predicting the performance based on hull form parameters. The accuracy of the DNN model was evaluated, resulting in a Mean Absolute Error (MAE) of 2.835%. Subsequently, the DNN model is combined with a genetic algorithm to identify the optimal set of parameters for the hull form, resulting in an optimal hull form. This optimization process revealed that the optimal hull form reduced the total hydrodynamic resistance by approximately 7% compared to the initial reference design. Consequently, this study demonstrates the effectiveness of the proposed method for deriving the optimal hull form for small ships.

Keywords

• Hull form design
• Characteristic curves
• Hydrodynamic performance
• Deep learning
• Deep Neural Network (DNN)