Scholarlink

<p>Accelerated computation of geometrically parameterized problems has become a key task in computational science and engineering. The hyper projection-based reduced order models (HPROMs) serve as essential physical tool for speeding up these computations. However, HPROMs require users to have intrusive access to the source code to collect the intrusive snapshots of nonlinear operators and to embed the projection and approximation processes within code. It’s challenging for most users of commercial software. Accordingly, we propose an innovative nonintrusive reduced-order methods including finite-orthogonal-basis geometrical-information reduced-order model (FBG-ROM) and its hyper-reduced form, finite-orthogonal-basis geometrical-information hyper-reduced-order model (FBG-HROM). These methods directly discretize PDEs via reduced order bases <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\Phi\)</EquationSource> </InlineEquation> and its corresponding differential operator <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(\mathcal{L}(\Phi )\)</EquationSource> </InlineEquation> which implies the geometry deformation information, rather than via element-wise traversal used in PROMs and HPROMs. This discretizing approach gives FBG-HROM a nonintrusive characteristic in terms of model discretization during the online phase, allowing users of commercial software to discretize the PDEs without relying on numerical computation models that have a high barrier, and a nonintrusive characteristic in terms of data acquisition during the offline phase, allowing users to obtain all necessary snapshots through commercial software. We demonstrate the accelerated computations of the FBG-HROM on three classic cases of fluid dynamics and solid mechanics—curved channel flow, blood flow in a stenotic vessel, and a plate with a hole—highlighting the excellent generalization performance and acceleration effects of the proposed frameworks.</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 nonintrusive finite-orthogonal-basis reduced-order model for accelerated computation of geometrically parameterized problems

  • 期刊论文

  • 发表日期：

    2025年6月28日

查看全文
  • Hongjiang Wang,
  • Lijie Qiao,
  • Han Dong,
  • Chaohui Huang,
  • Weizhe Wang,
  • Yingzheng Liu

摘要

Accelerated computation of geometrically parameterized problems has become a key task in computational science and engineering. The hyper projection-based reduced order models (HPROMs) serve as essential physical tool for speeding up these computations. However, HPROMs require users to have intrusive access to the source code to collect the intrusive snapshots of nonlinear operators and to embed the projection and approximation processes within code. It’s challenging for most users of commercial software. Accordingly, we propose an innovative nonintrusive reduced-order methods including finite-orthogonal-basis geometrical-information reduced-order model (FBG-ROM) and its hyper-reduced form, finite-orthogonal-basis geometrical-information hyper-reduced-order model (FBG-HROM). These methods directly discretize PDEs via reduced order bases \(\Phi\) and its corresponding differential operator \(\mathcal{L}(\Phi )\) which implies the geometry deformation information, rather than via element-wise traversal used in PROMs and HPROMs. This discretizing approach gives FBG-HROM a nonintrusive characteristic in terms of model discretization during the online phase, allowing users of commercial software to discretize the PDEs without relying on numerical computation models that have a high barrier, and a nonintrusive characteristic in terms of data acquisition during the offline phase, allowing users to obtain all necessary snapshots through commercial software. We demonstrate the accelerated computations of the FBG-HROM on three classic cases of fluid dynamics and solid mechanics—curved channel flow, blood flow in a stenotic vessel, and a plate with a hole—highlighting the excellent generalization performance and acceleration effects of the proposed frameworks.