Prediction of the superiority of the hydrodynamic performance of hull forms using deep learning

When designing a ship's hull form, a designer creates various candidate hull forms and performs a computational fluid dynamics (CFD) analysis to evaluate the performance of each candidate. Designers consider quantitative indicators, such as the total resistance and wake coefficient, and qualitative indicators, such as the wave height and pressure distributions, when evaluating the performance of a hull form. During the design process, quantitative and qualitative indicators are often used to determine the superiority of two hull forms. However, in the case of quantitative indicators, the difference between the two hull forms is often minimal; thus, superiority cannot be readily determined. Furthermore, because qualitative indicators are in the form of images, it is challenging to determine the superiority in many cases, even for experienced designers. To solve this problem, we propose a convolutional neural network-based model for predicting the superiority of hull form performance from a qualitative indicator of the image form derived from CFD analysis. The proposed prediction model received various types of hull form performance images. From these results, the hull form performance characteristics were well fused for prediction with high accuracy. CFD analysis images and quantitative indicators for 1600 hull forms were used to determine the superiority of the prediction model. The learned model was verified using 240 hulls. The result confirmed that the proposed model accurately predicted superiority with an accuracy of approximately 94%.