Reconstruction of the solid–liquid two-phase flow field in the pipeline based on limited pipeline wall information

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

Published: Feb. 1, 2025

Pipeline hydraulic transportation is the primary method for transporting deep-sea mineral resources and fossil fuels. blockage often causes excessive pressure in pipeline, leading to pipeline breakage or even cargo leakage, which severely impacts safety can easily trigger secondary disasters. Therefore, clarifying global flow field within pipelines, such as particle distribution, crucial monitoring controlling systems. This study uses a limited number of measurable wall sensor values inputs deep learning models reconstruction, with solid–liquid two-phase three-dimensional output. Three model frameworks from existing studies are summarized, their reconstruction effects compared. Based on this, new framework proposed. It expands low-dimensional same size using pseudo-decoder then processes them through an autoencoder. The results indicate that achieves further accuracy improvements compared previous three frameworks, R2 mean squared error reaching 0.933 5.13 ×10−4, respectively. Additionally, skip connection configuration model, dataset size, rate, well arrangement sensors accuracy, investigated. Finally, transferability demonstrated by reconstructing fluid velocity fields flow.

Language: Английский

---

Multi-scale boundary-enhanced diffusion network for high-resolution sea surface height reconstruction

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

Published: Feb. 1, 2025

Accurately reconstructing information-rich high-resolution sea surface height (SSH) from low-resolution SSH data has long been a critical challenge in oceanography. Despite significant advances generative methods, most existing approaches fail to adequately capture the multi-scale nature of oceanic phenomena and struggle resolve high-frequency features such as small-scale vortices boundary currents, particularly at high sampling factors. To address these challenges, we propose boundary-enhanced diffusion network (MBD-Net) for super-resolution. The key innovation our method lies design contextual squeeze excitation pyramid pooling module, which efficiently captures local global information across multiple scales, enabling model accurate reconstruction fine-scale structures while preserving large-scale patterns. Additionally, enhanced channel attention block, improves model's sensitivity details (particularly around complex vortex boundaries) strengthen its robustness by mitigating noise well. Experimental evaluations show that MBD-Net outperforms achieving average structural similarity indexes 0.983 4× 0.962 8× super-resolution ocean regions. These results demonstrate effectiveness versatility MBD-Net, establishing it promising tool high-fidelity environment.

Language: Английский

---

A physics-informed deep learning model to reconstruct turbulent wake from random sparse data

Physics of Fluids, Journal Year: 2024, Volume and Issue: 36(6)

Published: June 1, 2024

This study develops a flexible deep learning framework aimed at reconstructing the global turbulent wakes from randomly distributed sparse data. The is based on Generative Adversarial Networks where generator utilizes U-Net architecture and constraint module integrated into training process. It designed to overcome challenges posed by chaotic behavior of fields, randomness in sensor layouts, numbers. efficacy model validated across three high-fidelity datasets, including laminar wake behind circular cylinder, square cylinder. proposed demonstrates ability accurately reconstruct flow patterns both wakes, even utilizing merely 0.043% data target field. exhibits significant generalization capability, which means that has nearly independence distributions sensors robust adaptation inputs with unseen Ablation studies elucidate distinct complementary roles each within model. Additionally, bottleneck tensor analyzed through visualization, comparisons lift coefficient, quantitative analyses dimensionality reduction. These visualizations confirm extract distinctive phase information reliably data, thereby guiding reconstruction patterns. findings highlight potential for applications fluid dynamics collected variable manner.

Language: Английский

---

Aerodynamic characteristics of a surface mounted prism of small aspect ratio immersed in atmospheric boundary layer

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

Published: Jan. 1, 2025

In this study, we conducted three-dimensional large-eddy simulations to investigate the effect of atmospheric boundary layer on wake dynamics surface mounted square prisms small aspect ratio AR at Re = 20 000 and 18 600 for case 1 0.5, respectively. Here, is prism height H its width D. The currently employed in study results a further decrease average drag force weakening periodicity compared under turbulent with lower thickness. Statistical analyses were then performed, terms Reynolds stresses, mean field, visualization show variations flow dynamics. Immersed layer, causes shear recirculation bubble contract, altering morphology. inner vortex pair associated upwash stronger horseshoe surrounded ring vortices founded 0.5. Proper orthogonal decomposition (POD) method was analysis key features two within selected slices. far region, fluctuations are concentrated vertical direction. reduced concentration energy first few POD modes suggests diminished wake.

Language: Английский

---

Twin-rotor wind turbine wake characteristics under various configurations and states

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

Published: Feb. 1, 2025

The twin-rotor wind turbine, as a new concept, can effectively improve the power coefficient, accelerate wake recovery, and reduce cost of floating platforms mooring compared to single-rotor turbines. This paper investigates tandem parallel arrangements, setting up four different calculation models: vortex lattice (VL) filament conversion, extended blade root length, altered initial azimuth angle. Changes under tip speed ratios, steady unsteady, motion states were studied. study found that for research, distance needs be set more than 2D 5D (D is rotor diameter). In unsteady state turbines, coefficient difference between two rotors significant, with increased turbulence intensity, spacing ratio should appropriately increased, an optimal choice existing. For staggered arrangement has little impact on but significantly affects wake. Increasing length circular airfoil section smaller without changing larger rotor. Applying sinusoidal six degrees freedom turbine increase except yawing. research great significance twin-rotor, reducing platform costs, optimizing farm layout.

Language: Английский

---

Aerodynamics prediction of vertical-axis wind turbines based on meta learning under regional interactions

Energy Conversion and Management, Journal Year: 2025, Volume and Issue: 332, P. 119727 - 119727

Published: March 13, 2025

Language: Английский

---

Intelligent Prediction of Manta Ray Flow Field Based on a Denoising Probabilistic Diffusion Model

Acta Physica Sinica, Journal Year: 2025, Volume and Issue: 74(10), P. 0 - 0

Published: Jan. 1, 2025

The manta ray is a large marine species that exhibits both highly efficient gliding and agile flapping capabilities. It can autonomously switch between various motion modes, such as gliding, flapping, group swimming, based on ocean currents seabed conditions. To address the computational resource time constraints of traditional numerical simulation methods in modeling ray's 3D large-deformation flow field, this study proposes novel generative artificial intelligence approach denoising probabilistic diffusion model (surf-DDPM). This method predicts surface field by inputting set parameter variables. Initially, we establish for ray’s mode using immersed boundary spherical function gas kinetic scheme (IB-SGKS), generating an unsteady dataset comprising 180 sets under frequency conditions 0.3-0.9 Hz amplitude 0.1-0.6 body lengths. Data augmentation then performed. Subsequently, Markov chain noise process neural network generation are constructed. A pretrained embeds parameters step labels into data, which fed U-Net training. Notably, Transformer incorporated architecture to enable handling long-sequence data. Finally, examine impact hyperparameters performance visualize predicted pressure velocity fields multi-flapping postures were not included training set, followed quantitative analysis prediction accuracy, uncertainty, efficiency. results demonstrate proposed achieves fast accurate predictions characterized extensive high-dimensional upsampling. minimum PSNR SSIM values 35.931 dB 0.9524, respectively, with all data falling within 95% interval. Compared CFD simulations, AI enhances efficiency single-condition simulations 99.97%.

Language: Английский

---

A novel model for the study of future maritime climate using artificial neural networks and Monte Carlo simulations under the context of climate change

Ocean Modelling, Journal Year: 2024, Volume and Issue: 190, P. 102384 - 102384

Published: May 17, 2024

This paper proposes a new model to study future coastal maritime climate under change context. combines statistical analysis, Monte Carlo simulations and Artificial Neural Networks (ANNs). Statistical analysis are used extrapolate wave context at regional level ANNs propagate these projected sea states obtained in deep water the coast. The use of allows for utilization large amounts data very low computational cost, enables generation projections level. combination two methodologies results accurate (MSE 0.02 m 1 s) computationally inexpensive hybrid that considering change. methodology has been validated applied Western Mediterranean long-term regime extreme events, obtaining increases events up 1.5 height 1.8 s period by 2050.

Language: Английский

---

Machine learning and numerical simulation research on specific energy consumption for gradated coarse particle two-phase flow in inclined pipes

Physics of Fluids, Journal Year: 2024, Volume and Issue: 36(9)

Published: Sept. 1, 2024

In deep-sea mining engineering, accurately predicting the energy required per unit length of pipeline to transport a mass solids (dimensionless specific consumption, DSEC) is crucial for ensuring conservation and efficiency in project. Based on our previous work, we utilized machine learning (ML) computational fluid dynamics (CFD)–discrete element method (DEM) study characteristics flow field variations gradated coarse particles inclined pipes (gradated refer solid mixed size quantity ratios). First, collect 1185 sets data from 13 experimental literature, after analyzing processing them, an ensemble model based four other ML models developed. Both pure substance (PS) (MP), prediction accuracy this relatively higher (PSs are spherical with uniform density, MPs different shapes, sizes, densities). Then, CFD-DEM process operating conditions include low velocity volume concentration (2 m/s 2.5%), high 7.5%), (4 2.5%). Under concentrations, as well DSEC hardly changes variation pipe inclination angle. high-concentration conditions, gradually becomes vertical, value increases.

Language: Английский

---

An attention-neural-processes approach to reconstructing unsteady water load for seaplanes

Physics of Fluids, Journal Year: 2024, Volume and Issue: 36(9)

Published: Sept. 1, 2024

The distribution of holographic unsteady water load offers important information for evaluating the hydrodynamic performance seaplanes. However, traditional tank test is limited by number sensors that can be deployed on bottom hull, thus only providing sparse data estimating and leading to inaccurate evaluation seaplane performance. To achieve accurate rapid reconstruction distribution, a machine learning load-reconstruction model based Attention Neural Processes proposed. This performs spatiotemporal modeling utilizing sensor data. It directly learns patterns across multiple time steps employs modules capture spatial load. Comparisons with alternative methods demonstrate model's superior ability simultaneously temporal dependencies In addition, robust generalization capability also validated reducing in training results indicate proposed exhibits high prediction efficiency, accuracy, which great significance comprehensively

Language: Английский

---