Scholarlink

<p>This paper presents a novel fast prediction method for ship waves through the development of a neural network model called Ship Wave Residual Network (SWRN). The study employs a Wigley hull model for analysis, with datasets generated through Computational Fluid Dynamics (CFD) simulations. A comprehensive validation study addresses simulation uncertainties through the grid convergence index (GCI), Monte Carlo method, and Chebyshev inequality. The SWRN framework is applied to datasets encompassing various ship form parameters and Froude numbers. The accuracy of the neural network is evaluated by comparing ship-generated wave field and total resistance predictions from SWRN and CFD for hull forms and Froude numbers beyond the training datasets. Results demonstrate that SWRN training requires only 40% of the computational resources needed for a single CFD simulation. The trained model generates predictions for ship wave fields and total resistance under calm water conditions within seconds, maintaining acceptable accuracy levels. This prediction technology establishes a foundation for AI-accelerated solutions in ship-generated flow field analysis.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Fast Prediction Method in Deep Neural Networks for Three-Dimensional Ship Wave Field

  • 期刊论文

  • 发表日期：

    2026年4月1日

查看全文
  • Qiang Wang,
  • Cheng-sheng Zhan,
  • Bai-wei Feng,
  • Heng Li

摘要

This paper presents a novel fast prediction method for ship waves through the development of a neural network model called Ship Wave Residual Network (SWRN). The study employs a Wigley hull model for analysis, with datasets generated through Computational Fluid Dynamics (CFD) simulations. A comprehensive validation study addresses simulation uncertainties through the grid convergence index (GCI), Monte Carlo method, and Chebyshev inequality. The SWRN framework is applied to datasets encompassing various ship form parameters and Froude numbers. The accuracy of the neural network is evaluated by comparing ship-generated wave field and total resistance predictions from SWRN and CFD for hull forms and Froude numbers beyond the training datasets. Results demonstrate that SWRN training requires only 40% of the computational resources needed for a single CFD simulation. The trained model generates predictions for ship wave fields and total resistance under calm water conditions within seconds, maintaining acceptable accuracy levels. This prediction technology establishes a foundation for AI-accelerated solutions in ship-generated flow field analysis.