Deep graph convolutional neural network for one-dimensional hepatic vascular haemodynamic prediction DOI Creative Commons

Weiqng Zhang,

Shuaifeng Shi,

Quan Qi

и другие.

bioRxiv (Cold Spring Harbor Laboratory), Год журнала: 2024, Номер unknown

Опубликована: Авг. 16, 2024

Abstract Hepatic vascular hemodynamics is an important reference indicator in the diagnosis and treatment of hepatic diseases. However, Method based on Computational Fluid Dynamics(CFD) are difficult to promote clinical applications due their computational complexity. To this end, study proposed a deep graph neural network model simulate one-dimensional hemodynamic results vessels. By connecting residuals between edges nodes, framework effectively enhances prediction accuracy efficiently avoids over-smoothing phenomena. The structure constructed from centerline boundary conditions vasculature can serve as input, yielding velocity pressure information corresponding centerline. Experimental indicate that our method achieves higher dataset with significant individual variations be extended involving other blood Following training, errors both fields maintained below 1.5%. trained easily deployed low-performance devices and, compared CFD-based methods, output along vessel at speed three orders magnitude faster. Author summary When using learning methods for analysis, simple point cloud data cannot express real geometric vessels, it necessary have additional extraction capability. In paper, we use predict parameters. vessels topology branch which improve strong generalisation ability. show highest flow simulation dataset, experimental human aorta also applied organs.

Язык: Английский

On the Preprocessing of Physics-informed Neural Networks: How to Better Utilize Data in Fluid Mechanics DOI
Shengfeng Xu, Yuanjun Dai, Chang Yan

и другие.

Journal of Computational Physics, Год журнала: 2025, Номер unknown, С. 113837 - 113837

Опубликована: Фев. 1, 2025

Язык: Английский

Процитировано

3
FE Reduced-Order Model-Informed Neural Operator for Structural Dynamic Response Prediction DOI
Laihao Yang, Xinzhe Luo, Zhibo Yang

и другие.

Neural Networks, Год журнала: 2025, Номер unknown, С. 107437 - 107437

Опубликована: Март 1, 2025

Язык: Английский

Процитировано

2
Spatiotemporal parallel physics-informed neural networks: A framework to solve inverse problems in fluid mechanics DOI Open Access
Shengfeng Xu, Chang Yan, Guangtao Zhang

и другие.

Physics of Fluids, Год журнала: 2023, Номер 35(6)

Опубликована: Июнь 1, 2023

Physics-informed neural networks (PINNs) are widely used to solve forward and inverse problems in fluid mechanics. However, the current PINNs framework faces notable challenges when presented with that involve large spatiotemporal domains or high Reynolds numbers, leading hyper-parameter tuning difficulties excessively long training times. To overcome these issues enhance PINNs' efficacy solving problems, this paper proposes a parallel physics-informed (STPINNs) can be deployed simultaneously multi-central processing units. The STPINNs is specially designed for of mechanics by utilizing an overlapping domain decomposition strategy incorporating Reynolds-averaged Navier–Stokes equations, eddy viscosity output layer networks. performance proposed evaluated on three turbulent cases: wake flow two-dimensional cylinder, homogeneous isotropic decaying turbulence, average three-dimensional cylinder. All cases successfully reconstructed sparse observations. quantitative results along strong weak scaling analyses demonstrate accurately efficiently flows comparatively numbers.

Язык: Английский

Процитировано

15
Flow-field reconstruction in rotating detonation combustor based on physics-informed neural network DOI Open Access

Xutun Wang,

Haocheng Wen,

Tong Hu

и другие.

Physics of Fluids, Год журнала: 2023, Номер 35(7)

Опубликована: Июль 1, 2023

The flow-field reconstruction of a rotating detonation combustor (RDC) is essential to understand the stability mechanism and performance engines. This study embeds reduced-order model an RDC into neural network (NN) construct physics-informed (PINN) achieve full-dimensional high-resolution flow field based on partially observed data. Additionally, unobserved physical fields are extrapolated through NN-embedded model. influence residual point sampling strategy observation spatial-temporal resolution results studied. As surrogate RDC, PINN fills gap that traditional computational fluid dynamics methods have difficulty solving, such as inverse problems, has engineering value for RDCs.

Язык: Английский

Процитировано

13
An unsupervised deep learning model for dense velocity field reconstruction in particle image velocimetry (PIV) measurements DOI Open Access

Wei Zhang,

Xue Dong, Zhiwei Sun

и другие.

Physics of Fluids, Год журнала: 2023, Номер 35(7)

Опубликована: Июль 1, 2023

Supervised deep learning methods reported recently have shown promising capability and efficiency in particle image velocimetry (PIV) processes compared to the traditional cross correlation optical flow methods. However, learning-based previous reports require synthesized images simulated flows for training prior applications, conflicting with experimental scenarios. To address this crucial limitation, unsupervised also been proposed velocity reconstruction, but they are generally limited rough reconstructions low accuracy due to, example, occlusion out-of-boundary motions. This paper proposes a new model named UnPWCNet-PIV (an network using Pyramid, Warping, Cost Volume). Such pyramidical specific enhancements on holds capabilities manage boundary The showed comparable robustness advanced supervised methods, which based images, together superior performance images. presents details of architecture assessments its both

Язык: Английский

Процитировано

11
A framework of data assimilation for wind flow fields by physics-informed neural networks DOI
Chang Yan, Shengfeng Xu, Zhenxu Sun

и другие.

Applied Energy, Год журнала: 2024, Номер 371, С. 123719 - 123719

Опубликована: Июнь 20, 2024

Язык: Английский

Процитировано

4
A new composite neural network with spatiotemporal features extraction capability for unsteady flow fields predictions DOI
Cheng Xu, Zhengxian Liu, Xiaojian Li

и другие.

Physics of Fluids, Год журнала: 2025, Номер 37(2)

Опубликована: Фев. 1, 2025

Artificial intelligence based on neural network technology has provided innovative methods for predicting unsteady flow fields. However, both purely data-driven and single physics-driven can only perform short-term predictions fields are unable to achieve medium- long-term predictions. A composite CNN-GRU-PINN (CGPINN) is proposed by combining convolutional (CNN), gated recurrent unit (GRU), physics-informed (PINN). CNN GRU used learn the spatial temporal characteristics of flows, respectively. PINN adopted constrain field prediction data according physical laws. The around a circular cylinder employed verify performances CGPINN. test results show that compared PINN, reconstruction accuracy CGPINN improved about 86.10% average, 96.18%. Compared pure approaches, an average 65.71%. Additionally, exhibits better robustness, demonstrating insensitivity variations in sample size noise levels, thereby ensuring stable reliable across diverse conditions. This study more accurate robust method

Язык: Английский

Процитировано

0
Synergizing machine learning with fluid–structure interaction research: An overview of trends and challenges DOI
Muk Chen Ong, Guang Yin

Опубликована: Март 1, 2025

Язык: Английский

Процитировано

0
Machine learning for nonlinear integro-differential equations with degenerate kernel scheme DOI
Hui Li,

Pengpeng Shi,

Xing Li

и другие.

Communications in Nonlinear Science and Numerical Simulation, Год журнала: 2024, Номер 138, С. 108242 - 108242

Опубликована: Июль 18, 2024

Язык: Английский

Процитировано

2
Physics-Informed Neural Networks for Prediction of a Flow-Induced Vibration Cylinder DOI
Guang Yin, Marek Jan Janocha, Muk Chen Ong

и другие.

Journal of Offshore Mechanics and Arctic Engineering, Год журнала: 2024, Номер 146(6)

Опубликована: Авг. 1, 2024

Abstract Flow-induced vibration (FIV) is a common phenomenon in ocean engineering for subsea structures with circular-shaped cross sections. A large amount of computational resources or experimental efforts are required to predict measure the complicated motions and flow field surrounding circular cylinder undergoing vortex-induced (VIV). Physics-informed neural networks (PINNs) powerful deep learning techniques solving governing partial differential equations (PDEs) dynamic systems as an alternative complex numerical methods. In present study, framework built employing PINNs Navier–Stokes flows past FIV using sparsely distributed spatiotemporal data inside domain. The training process involves minimizing supervised loss at these sparse points residuals PDEs. For PINN model, moving frame around used collect from two-dimensional direct simulation results low Reynolds number. structural displacements also implemented developed PINN. performance evaluated by comparing predicted contours velocities data. hydrodynamic forces prediction achieved PINN-obtained predictions combined force partitioning method.

Язык: Английский

Процитировано

1