Scholarlink

<p>Optimizing engineering design problems, such as aerodynamic design, is computationally demanding due to the high cost and time required for numerical simulations. This study introduces an innovative approach to speed up this process by combining machine learning and optimization algorithms. A reduced-data neural network model is developed to predict key performance coefficients, specifically targeting the pitching moment and normal force coefficients. By strategically applying dataset reduction techniques and noise injection, we achieve over 58% reduction in the number of numerical simulations needed for training. This optimized neural network is then integrated into a design optimization framework. The particle swarm optimization (PSO) algorithm explores the design space and efficiently finds optimal solutions. The combined use of the reduced-data neural network and PSO significantly lowers computational cost and time while maintaining high accuracy. The model demonstrates a prediction error under 1% compared to high-fidelity numerical simulations. This work emphasizes the effectiveness of merging machine learning and optimization methods to improve computational efficiency in complex engineering design tasks, enabling faster and more effective exploration of design options.</p> Graphical abstract <p></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

Reduced-data neural network framework with PSO-based optimization for efficient aerodynamic design

  • 期刊论文

  • 发表日期：

    2026年2月11日

查看全文
  • Mohammad Hassan Shojaeefard,
  • Masoud Nobakhti

摘要

Optimizing engineering design problems, such as aerodynamic design, is computationally demanding due to the high cost and time required for numerical simulations. This study introduces an innovative approach to speed up this process by combining machine learning and optimization algorithms. A reduced-data neural network model is developed to predict key performance coefficients, specifically targeting the pitching moment and normal force coefficients. By strategically applying dataset reduction techniques and noise injection, we achieve over 58% reduction in the number of numerical simulations needed for training. This optimized neural network is then integrated into a design optimization framework. The particle swarm optimization (PSO) algorithm explores the design space and efficiently finds optimal solutions. The combined use of the reduced-data neural network and PSO significantly lowers computational cost and time while maintaining high accuracy. The model demonstrates a prediction error under 1% compared to high-fidelity numerical simulations. This work emphasizes the effectiveness of merging machine learning and optimization methods to improve computational efficiency in complex engineering design tasks, enabling faster and more effective exploration of design options.

Graphical abstract