A novel framework for predicting active flow control by combining deep reinforcement learning and masked deep neural network

Physics of Fluids, Journal Year: 2024, Volume and Issue: 36(3)

Published: March 1, 2024

Active flow control (AFC) through deep reinforcement learning (DRL) is computationally demanding. To address this, a masked neural network (MDNN), aiming to replace the computational fluid dynamics (CFD) environment, developed predict unsteady fields under influence of arbitrary object motion. Then, novel DRL-MDNN framework that combines MDNN-based environment with DRL algorithm proposed. validate reliability framework, blind test in pulsating baffle system designed. Vibration damping considered be objective, and traditional DRL-CFD constructed for comparison. After training, spatiotemporal evolution 200 time steps motion predicted by MDNN. The details field are compared CFD results, relative error within 5% achieved, which satisfies accuracy serving as an interactive algorithms. frameworks then applied find optimal strategy. results indicate both achieve similar performance, reducing vibration 90%. Considering resources expended establishing database, resource consumption reduced 95%, response during each episode decreased 98.84% framework.

Language: Английский

---

A novel method for predicting fluid–structure interaction with large deformation based on masked deep neural network

Physics of Fluids, Journal Year: 2024, Volume and Issue: 36(2)

Published: Feb. 1, 2024

Traditional fluid–structure interaction (FSI) simulation is computationally demanding, especially for bi-directional FSI problems. To address this, a masked deep neural network (MDNN) developed to quickly and accurately predict the unsteady flow field. By integrating MDNN with structural dynamic solver, an system proposed perform of flexible vertical plate oscillation in fluid large deformation. The results show that both field prediction structure response are consistent traditional system. Furthermore, method highly effective mitigating error accumulation during temporal predictions, making it applicable various deformation Notably, model reduces computational time millisecond scale each step regarding part, resulting increase nearly two orders magnitude speed, which greatly enhances speed

Language: Английский

---

Fast fluid–structure interaction simulation method based on deep learning flow field modeling

Physics of Fluids, Journal Year: 2024, Volume and Issue: 36(4)

Published: April 1, 2024

The rapid acquisition of high-fidelity flow field information is great significance for engineering applications such as multi-field coupling. Current research in modeling predominantly focuses on low Reynolds numbers and periodic flows, exhibiting weak generalization capabilities notable issues with temporal inferring error accumulation. Therefore, we establish a reduced order model (ROM) based Convolutional Auto-Encoder (CAE) Long Short-Term Memory (LSTM) neural network propose an unsteady method the airfoil high number strong nonlinear characteristics. attention mechanism physical constraints are integrated into architecture to improve accuracy. A broadband excitation training strategy proposed overcome accumulation problem long-term inferring. With only small amount latent codes, relative reconstructed by CAE can be less than 5‰. By LSTM signals, stable dynamic evolution achieved time dimension. CAE-LSTM accurately predict forced response complex limit cycle behavior wide range amplitude frequency under subsonic/transonic conditions. errors predicted variables integral force 1%. fluid–structure interaction framework built coupling ROM motion equations structure. predicts series pitch displacement moment coefficient at different frequencies, which good agreement computational fluid dynamics, simulation savings exceed one magnitude.

Language: Английский

---

Transfer machine learning framework for efficient full-field temperature response reconstruction of thermal protection structures with limited measurement data

International Journal of Heat and Mass Transfer, Journal Year: 2025, Volume and Issue: 242, P. 126785 - 126785

Published: Feb. 12, 2025

Language: Английский

---

Intermittent flow influences plant root growth: A phytofluidics approach

Physics of Fluids, Journal Year: 2024, Volume and Issue: 36(4)

Published: April 1, 2024

The challenges of food security are exacerbated by the world's expanding population and diminishing agricultural land. In response, hydroponic cultivation offers a potentially more sustainable approach to growing nutrient-dense crops compared traditional methods. Motivated this understanding, we conducted series experiments explore behavior Brassica juncea (Pusa Jaikisan) plant roots under various flow configurations within controlled environment. considered were no-flow/flow (NF/F), continuous flow, flow/no-flow (F/NF), stagnation. Additionally, anatomical sectioning study how different affect cellular structure root cross section. We also performed numerical simulations investigate internal stress generated conditions. observed that an increased number cortical cells developed in response higher case which protected inner vascular bundle from excessive biological stress. Comparing designs, found resulted longer length F/NF NF/F configurations. per unit average power was highest for 2 h case, followed NF/F, 3 F/NF, cases. This suggests periodic conditions (F/NF NF/F) with lower power, necessary requirement economical use, led lengths. Furthermore, nitrogen uptake configuration flow. Consequently, infer cultivation, altering type could be cost-effective less nutrient solution wastage, promoting better growth scenario.

Language: Английский

---

Identification of the form of self-excited aerodynamic force of bridge deck based on machine learning

Physics of Fluids, Journal Year: 2024, Volume and Issue: 36(1)

Published: Jan. 1, 2024

This paper introduces an intelligent identification method for self-excited aerodynamic equations. The is based on advanced sparse recognition technology and equipped with a new sampling strategy designed weak nonlinear dynamic systems limit cycle characteristics. Considering the complexity of experiment condition difficult priori selection hyperparameters, information criteria ensemble learning proposed to derive global optimal model. first validated by simulated data obtained from some well-known equations then applied flutter wind tunnel experiments. Finally, reasons different results under sizes candidate function space are discussed perspective matrix linear correlation numerical calculation.

Language: Английский

---

Prediction of turbulent channel flow using Fourier neural operator-based machine-learning strategy

Physical Review Fluids, Journal Year: 2024, Volume and Issue: 9(8)

Published: Aug. 12, 2024

The implicit U-Net enhanced Fourier neural operator (IUFNO) combines the loop structure of FNO (IFNO) with U-Net, leading to long-term predictive ability in large-eddy simulations (LES) turbulent channel flow. It is found that IUFNO outperforms traditional dynamic Smagorinsky model (DSM) and wall-adapted local eddy-viscosity (WALE) at coarse LES grids. predictions both mean fluctuating quantities by are closer filtered direct numerical simulation (fDNS) benchmark compared models, while computational cost much lower.

Language: Английский

---

Fourier neural operator based fluid–structure interaction for predicting the vesicle dynamics

Physica D Nonlinear Phenomena, Journal Year: 2024, Volume and Issue: 463, P. 134145 - 134145

Published: April 4, 2024

Language: Английский

---

Data assimilation method and application of shear stress transport turbulence model for complex separation of internal shock boundary layer flow

Physics of Fluids, Journal Year: 2024, Volume and Issue: 36(5)

Published: May 1, 2024

Traditional turbulence models suffer from low accuracy and weak applicability when predicting complex separated flows, such as those that occur in shock boundary layers. To overcome this problem, the present paper considers a cavity-ramp structure calibrates model parameters using deep neural network (DNN) surrogate genetic algorithm (GA). The non-intrusive polynomial chaos expansion method is used to quantify uncertainty of shear stress transport (SST) determine effects these on wall pressure, allowing suitable feature identification be selected for DNN model. compared with traditional method, results highlight advantages construct Finally, GA optimize calibrate SST based experimental data. Experimental show highly accurate, predicted achieving coefficient determination above 0.998. has higher precision, stronger extraction ability, faster prediction times than method. calibrated produces pressures are close data, verifying feasibility proposed It expected approach will improve calculation

Language: Английский

---

Augmentation of piezoelectric thin-film flapping velocimetry turbulence strength detection via machine learning

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

Published: Jan. 1, 2025

Qualitatively evaluating the fundamental mechanical characteristics of square-fractal-grid (SFG)-generated turbulent flow using piezoelectric thin-film flapping velocimetry (PTFV) is rather time-consuming. More importantly, its sensitivity in detecting high-frequency, fine-scale fluctuations constrained by high-speed camera specifications. To reduce dependency on imaging future PTFV implementations, regression models are trained with supervised machine learning to determine correlation between piezoelectric-generated voltage V and corresponding local equivalent velocity fluctuation. Using tip deflection δ data as predictors responses, respectively, Trilayered Neural Network (TNN) emerges best-performing model compared linear regression, trees, support vector machines, Gaussian process ensembles trees. TNN from (i) lower quarter, (ii) bottom left corner, (iii) central opening SFG-grid provide accurate predictions insert-induced centerline streamwise cross-sectional lateral turbulence intensity root mean square-δ, average errors not exceeding 5%. The output predicted response, which considers small-scale across entire surface, better expresses integral length scale (38% smaller) forcing (270% greater), particularly at corner SFG where eddies significant. Furthermore, effectively captures occasional extensive excitation forces large-scale eddies, resulting a more balanced force distribution. In short, this study paves path for comprehensive expedited dynamics characterization detection via PTFV, potential deployment high Reynolds number flows generated various grid configurations.

Language: Английский

---