Spatiotemporal reconstruction of unsteady bridge flow field via hierarchical graph neural networks with causal attention

Physics of Fluids, Journal Year: 2025, Volume and Issue: 37(1)

Published: Jan. 1, 2025

When calculating the transient flow around a bridge structure, traditional computational fluid dynamics methods are extremely time-consuming, especially for multiparameter optimization analyses. Inspired by development of deep graph neural networks with mesh this paper describes spatiotemporal prediction framework rapid reconstruction and flows on large-scale unstructured grids. To ensure stability reliability during self-supervised training, causal self-attention mechanism is employed in temporal model. The trained tested dataset containing 40 000 snapshots fields Reynolds numbers ranging from 104 to 105. relative mean square error model predicting velocity pressure found be order 10−3 does not exceed 10%. This demonstrates that capable reconstructing high-dimensional field information low-dimensional data. Furthermore, proposed achieves speedup two orders magnitude compared respect inference. validate its ability infer aerodynamic characteristics, used predict surface pressure, coefficients, streamlines, vorticity. results demonstrate has reliable accuracy, representation, identifying multiscale characteristics.

Language: Английский

---

Fast and accurate prediction of flow in multi-row cascade based on combined neural networks

Physics of Fluids, Journal Year: 2025, Volume and Issue: 37(1)

Published: Jan. 1, 2025

This paper proposes a fast and accurate method for predicting multi-row cascade flow based on framework of combined neural networks. The primary idea this is to decompose the whole-annulus into different types sub-regions, prediction surrogate models networks are constructed these sub-regions in rectangular computational domain by applying coordinate transformation technique. each sub-region then combined, continuity at interfaces among used iteratively compute cascade. main advantages proposed include reduced dataset generation cost network training through spatial decomposition, as well ability achieve combining sub-regions. test case two-dimensional stator-rotor interaction indicates that time developed approximately 5% required numerical simulation, with over 99% nodes field exhibiting normalized absolute error less than 0.05. approach can be further extended three-dimensional multi-stage turbomachinery.

Language: Английский

---

A convolutional neural network-based model for reconstructing free surface flow fields

Physics of Fluids, Journal Year: 2025, Volume and Issue: 37(1)

Published: Jan. 1, 2025

This paper introduces hydrological computational fluid dynamics model (HydroCFD), a deep learning based on the convolutional neural network U-Net framework designed for reconstructing free surface flow fields. With well-posed boundary and initial conditions, rapidly generates result that approximates two-dimensional (2D) shallow water equations, significantly improving efficiency of obtaining fields compared to traditional methods. The features an input layer integrates depth terrain (hydrological element variables), incorporates new loss function, coefficient variation function (CVLoss), improve accuracy stability. HydroCFD is trained validated two different datasets, open channel flows with groin, abrupt expansion. Error analysis demonstrated achieves high precision in 2D Furthermore, comparison six functions reveals CVLoss contributes improved

Language: Английский

---

Flow field prediction and optimization of non-axisymmetric aero-engine nacelles based on deep learning

Aerospace Science and Technology, Journal Year: 2025, Volume and Issue: unknown, P. 109990 - 109990

Published: Jan. 1, 2025

Language: Английский

---

A boundary-assimilation Fourier neural operator for predicting initial fields of flow around structures

Physics of Fluids, Journal Year: 2025, Volume and Issue: 37(2)

Published: Feb. 1, 2025

Repeatedly solving flow around structures with varying parameters using computational fluid dynamics (CFD) is often essential for structural design. This study proposes a boundary-assimilation Fourier neural operator (BAFNO) method to address the challenges of manually setting initial conditions CFD. The focus BAFNO on generalization ability predict fields without relying observational data. addresses boundary constraint requirements existing physics-informed models in parametric geometries. Inspired by ghost node method, domain are assimilated into loss function instead adding penalty terms. Meanwhile, structure boundaries damping source term level set function. can flexibly handle geometries different shapes and quantities. Subsequently, series numerical experiments flow-around conducted confirm performance BAFNO. results indicate that has strong capability, + CFD obtain dynamic stable faster than direct

Language: Английский

---

Transition of shock reflection with downstream expansion fan interference

Journal of Fluid Mechanics, Journal Year: 2025, Volume and Issue: 1007

Published: March 12, 2025

In this paper, the reflection of shock waves with downstream expansion fan interference in two-dimensional, inviscid flow is investigated, including steady Mach (MR) and unsteady transition process from regular (RR) to MR. A threshold for configuration based on non-dimensional wedge length proposed. The analytical model MR RR $\rightarrow$ established classical wave relations, whose prediction agrees well results obtained through numerical simulation. It found that significantly influences patterns, especially height stem shape slip line. interaction accelerates formation sonic throat, stabilizing structure rapidly, generally small heights. exposure triple point eliminates inflection line, slope increases smoothly. further affects time evolution during multiple-interaction stage process. appears modifications come curvature incident brought by interference. During stage, moves upstream along curved shock, where angle changes according curvature, resulting variation velocity.

Language: Английский

---

Attention-Based Multi-Modal Learning for Aircraft Engine Fan Fault Diagnosis

Aerospace Science and Technology, Journal Year: 2025, Volume and Issue: unknown, P. 110194 - 110194

Published: April 1, 2025

Language: Английский

---

Meshless Simulation of Multi-Propeller/Wing Interactions in Typical Distributed Electric Propulsion Configurations

Aerospace Science and Technology, Journal Year: 2025, Volume and Issue: unknown, P. 110287 - 110287

Published: May 1, 2025

Language: Английский

---

Deep learning-based reduced order model for three-dimensional unsteady flow using mesh transformation and stitching

Computers & Fluids, Journal Year: 2024, Volume and Issue: unknown, P. 106441 - 106441

Published: Sept. 1, 2024

Language: Английский

---

An effective convergence accelerator of fluid simulations via generative diffusion probabilistic model

Aerospace Science and Technology, Journal Year: 2025, Volume and Issue: unknown, P. 109917 - 109917

Published: Jan. 1, 2025

Language: Английский

---