Scholarlink

<p>This study presents a deep learning-based reduced-order modeling (ROM) framework that enhances extrapolation performance through manifold representation and data augmentation. The proposed framework integrates convolutional neural networks (CNNs) with various model-order reduction (MOR) techniques to efficiently handle high-dimensional aerodynamic data. A novel generative model is introduced, leveraging proper orthogonal decomposition (POD) for projection-based manifold learning while incorporating physical design space information. To assess the effectiveness of the proposed approach, aerodynamic predictions of NACA 4-digit airfoils are conducted using computational fluid dynamics (CFD) data. The results demonstrate that the projection-based manifold learning method significantly improves prediction accuracy by augmenting the dataset.</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

Improved Extrapolation Performance via Manifold Representation and Data Augmentation for Data-Driven Reduced-Order Model

  • 期刊论文

  • 发表日期：

    2025年3月1日

查看全文
  • Seongwoo Cheon,
  • Hyejin Kim,
  • Seokhee Ryu,
  • Haeseong Cho,
  • Hakjin Lee

摘要

This study presents a deep learning-based reduced-order modeling (ROM) framework that enhances extrapolation performance through manifold representation and data augmentation. The proposed framework integrates convolutional neural networks (CNNs) with various model-order reduction (MOR) techniques to efficiently handle high-dimensional aerodynamic data. A novel generative model is introduced, leveraging proper orthogonal decomposition (POD) for projection-based manifold learning while incorporating physical design space information. To assess the effectiveness of the proposed approach, aerodynamic predictions of NACA 4-digit airfoils are conducted using computational fluid dynamics (CFD) data. The results demonstrate that the projection-based manifold learning method significantly improves prediction accuracy by augmenting the dataset.