Scholarlink

<p>This study proposes an intelligent optimization framework integrating Bayesian-optimized Back Propagation Neural Networks (BO-BPNN) with Genetic Algorithm (GA) and Adaptive Monte Carlo (AMC) method for systematic marine propeller design. The methodological innovations include introducing Bayesian Optimization (BO) for adaptive hyperparameter tuning of BPNN and proposing an improved AMC algorithm with hybrid sampling and decaying search radius mechanisms. A data-driven surrogate model is established on 2 552 open-water performance records of AU-type propellers. Compared with conventional BPNN, the BO-BPNN model achieves significant improvements: an 89% reduction in Mean Squared Error (MSE = 0.000 3), a 66% decrease in Mean Absolute Error (MAE = 0.009 7), and a coefficient of determination <i>R</i><sup>2</sup> = 0.995 6. Systematic comparative analysis reveals that while both GA and AMC successfully identify identical optimal design parameters (the number of blades <i>Z</i> = 3, pitch ratio <i>P/D</i> = 0.7, disc area ratio <i>A</i><sub>E</sub>/<i>A</i><sub>0</sub> = 0.36, and advance ratio <i>J</i> = 0.5, yielding maximum open-water efficiency (<i>η</i><sub>0</sub> = 0.687), GA demonstrates superior convergence reliability despite comparable total computation time. CFD validation using STAR-CCM+ confirms the framework’s reliability with only a 0.6% relative difference. The framework establishes a reusable paradigm that synergistically integrates adaptive surrogate modeling with physics-based validation, significantly reducing computational costs and design cycles while maintaining high accuracy.</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

Hydrodynamic Performance Optimization of AU-Type Propellers Based on BO-BPNN Surrogate Model and CFD Validation

  • 期刊论文

  • 发表日期：

    2026年5月25日

查看全文
  • Hui Zuo,
  • Yong Zhao,
  • Xiaorui Zhang,
  • Jingbei Jia

摘要

This study proposes an intelligent optimization framework integrating Bayesian-optimized Back Propagation Neural Networks (BO-BPNN) with Genetic Algorithm (GA) and Adaptive Monte Carlo (AMC) method for systematic marine propeller design. The methodological innovations include introducing Bayesian Optimization (BO) for adaptive hyperparameter tuning of BPNN and proposing an improved AMC algorithm with hybrid sampling and decaying search radius mechanisms. A data-driven surrogate model is established on 2 552 open-water performance records of AU-type propellers. Compared with conventional BPNN, the BO-BPNN model achieves significant improvements: an 89% reduction in Mean Squared Error (MSE = 0.000 3), a 66% decrease in Mean Absolute Error (MAE = 0.009 7), and a coefficient of determination R2 = 0.995 6. Systematic comparative analysis reveals that while both GA and AMC successfully identify identical optimal design parameters (the number of blades Z = 3, pitch ratio P/D = 0.7, disc area ratio AE/A0 = 0.36, and advance ratio J = 0.5, yielding maximum open-water efficiency (η0 = 0.687), GA demonstrates superior convergence reliability despite comparable total computation time. CFD validation using STAR-CCM+ confirms the framework’s reliability with only a 0.6% relative difference. The framework establishes a reusable paradigm that synergistically integrates adaptive surrogate modeling with physics-based validation, significantly reducing computational costs and design cycles while maintaining high accuracy.