Bulletin of the American Physical Society
77th Annual Meeting of the Division of Fluid Dynamics
Sunday–Tuesday, November 24–26, 2024; Salt Lake City, Utah
Session R12: Low-Order Modeling and Machine Learning in Fluid Dynamics: Other Applications I |
Hide Abstracts |
Chair: Patricio Clark Di Leoni, Universidad de San Andres Room: 155 B |
Monday, November 25, 2024 1:50PM - 2:03PM |
R12.00001: Realtime data-driven sensing of oscillatory crossflow using a fixed-wing drone Xiaozhou Fan, Fengze Xie, Julian Humml, Jacob Schuster, Yisong Yue, Morteza Gharib Flying into a highly turbulent atmosphere for fixed-wing drone remains challenging, as they are susceptible to atmospheric disturbances, such as twisters, due to small size, and that their onboard control surfaces and flight sensors are limited. To enable accurate and efficient reduced-order flow modeling, where efficiency is measured both in terms of the computational cost, and the amount of training data required, we built a fixed-wing drone equipped with two customized multi-hole probes that measure flight speed, angle of attack and sideslip, along with embedded wing pressure sensors. The model was mounted on a six-axis force transducer in a wind tunnel. We focused on introducing low frequency oscillatory yaw disturbance as a representative disturbance generated by a nearby twister system and trained the airplane to learn the signatures of the flow structure with onboard sensing and perform inference in real time. We hypothesize that the airplane would be able to infer the distance and orientation of the twister, thus, optimizing trajectory based on current prediction. i.e. whether to fly straight into it or plan a different route to avoid it. |
Monday, November 25, 2024 2:03PM - 2:16PM |
R12.00002: Data-driven prediction of unsteady loading on a 2D deforming airfoil Hamid Reza Karbasian, Wim M. van Rees In this study, we use a data-driven model enhanced with a design gate to predict the unsteady pressure and shear distributions along a 2D deforming airfoil. The deformation of the airfoil is governed by several design parameters, leading to an additional design space on top of the manifolds of the dynamical system. To generate training data for our data-driven approach, we extract unsteady traction distributions around deforming airfoils from numerical flow simulations across the design space. Subsequently, this dataset is used to build our data-driven model. The model relies on a pLSTM network architecture, which is a new variant of traditional LSTMs that embeds a design gate in the pLSTM cell. This new architecture helps overcome the well-known stability problems of LSTM, allows switching design conditions during the model operation, and increases the learning capacity of the neural network model for complex design spaces. We demonstrate that this data-driven model can predict the evolution of the surface loading for different deformation histories. Therefore, this problem can be combined with a structural model to perform two-way fluid-structure interaction problems in the future. Finally, at the end of the talk, we briefly consider how to extend other architectures like GRUs and transformers with design gates to improve their ability to dynamically predict flows across design space parameters. |
Monday, November 25, 2024 2:16PM - 2:29PM |
R12.00003: A Physics-Infused, Machine Learning Framework to Study Wind-Driven Runback Water Flows Pertinent to Aircraft Icing Phenomena Jincheng Wang, Charlelie Laurent, Suhas S Jain, Hui Hu Aircraft icing is one of the most dangerous weather hazards to threaten flight safety in cold weather. The transient behavior of wind-driven runback water film/rivulet (WDRWF/R) flows over aircraft wings would affect the dynamic glaze ice accretion process significantly. In the present study, a novel, flow-physics-infused, machine learning (ML) framework is developed for more accurate prediction of the transient characteristics of WDRWF/R flows pertinent to aircraft icing phenomena. A comprehensive experimental campaign is conducted in a wind tunnel by using a novel Digital Image Projection (DIP) technique to achieve spatiotemporal measurements of the film thickness fields of WDWF/R flows over a flat plate under different test conditions. The massive experiment data is used to train and test a specialized Physics-Guided Fourier Neural Operator (PGFNO) to learn the intricate characteristics of WDRWF/R flows. Physical knowledge is infused through a composite loss function, ensuring accurate pointwise flow reconstruction and mass conservation. The trained model was then used to predict the spatiotemporal evolution of WDRWF/R flows over a wide range of flow conditions. It was demonstrated that the physics-infused ML model can accurately predict the transient characteristics of WDRWF/R flows, such as film thickness height and its spectrum, for unseen wind speeds, water flow rates, and initial conditions. The model can aid in the accurate glaze ice accretion prediction when combined with freezing models. |
Monday, November 25, 2024 2:29PM - 2:42PM |
R12.00004: Utilizing Physics-Informed Neural Networks (PINNs) to Estimate Non-Uniform Surface Properties of Active Droplets Parvin Bayati, Stewart Mallory We have developed a novel PINN approach to estimate an active droplet’s non-uniform surface properties, such as interfacial tension and velocity. Unraveling these properties is essential for understanding droplet dynamics in various environments, but experimental determination poses significant challenges. While inverse methods are possible, solving inverse flow problems is often costly and requires complex formulations. |
Monday, November 25, 2024 2:42PM - 2:55PM |
R12.00005: Machine learning models for unresolved capillary effects in multiphase flows Shahab Mirjalili, Chris James Cundy, Charlelie Laurent, Stefano Ermon, Gianluca Iaccarino, Ali Mani
|
Monday, November 25, 2024 2:55PM - 3:08PM |
R12.00006: Convolutional feature-enhanced physics-informed neural networks for the spatio-temporal reconstruction of two-phase flows Maximilian Dreisbach, Elham Kiyani, Jochen Kriegseis, George Em Karniadakis, Alexander Stroh Two-phase flow phenomena play a key role in numerous technical processes, including hydrogen fuel cells, spray cooling techniques and combustion. Optical measurement techniques, such as shadowgraphy and particle image velocimetry, provide insight through the measurement of the gas-liquid interface and internal velocity fields, respectively. However, these experiments are constrained to planar measurements, whereas the dynamics of the flow are generally three-dimensional (3D). Deep learning techniques based on convolutional neural networks offer a pathway for volumetric reconstruction of the experiments by leveraging spatial structure in the images and context-rich feature extraction. Physics-informed neural networks (PINNs) emerge as a promising alternative, as they incorporate prior knowledge encoded in the networks by training on governing equations, allowing for accurate predictions even from limited data. We propose a novel approach for convolutional feature-enhanced PINNs for the spatio-temporal reconstruction of two-phase flows from shadowgraphy images. The capability of the novel method is demonstrated by the accurate reconstruction of the 3D gas-liquid interface, velocity and pressure fields for an impinging droplet based on planar experimental data. |
Monday, November 25, 2024 3:08PM - 3:21PM |
R12.00007: Towards spatio-temporal prediction of cavitating fluid flow with graph neural networks Rui Gao, Shayan Heydari, Rajeev Jaiman We present a deep learning-based surrogate model for spatio-temporal prediction of cavitating fluid flow. Specifically, we introduce a finite element-inspired rotation equivariant hypergraph neural network for inferring and predicting dynamical behaviors of cavitating flow. We generate ground-truth spatial-temporal data by simulating a full-order variational system based on homogeneous mixture-based cavitation theory. We consider the flow past a NACA0012 hydrofoil to examine the predictive ability of the proposed graph neural network for cavitation dynamics. Results demonstrate that the network achieves stabilized and accurate temporal predictions of the system states, successfully forecasting the evolution patterns of individual cavitation events. Additionally, comparisons of predicted fluid loading coefficients are in good agreements with the ground-truth values. We also discuss some challenges encountered in the long-term prediction of flow patterns across multiple cavitation events. |
Monday, November 25, 2024 3:21PM - 3:34PM |
R12.00008: Efficient Estimation of Temporal Exceeding Probability for Ship Responses in Broad-Band Wave Fields Shayesteh Hafezi, Xianliang Gong, Yulin Pan Calculating extreme ship statistics is critical for ship design. This study aims to determine the temporal exceeding probability, i.e., the percentage of time a ship's response exceeds a threshold, as it navigates through an irregular broad-band wave field. Direct simulations are often infeasible due to the rarity of extreme events, high dimensionality of wave fields, and the computational expense of numerical models. To mitigate these challenges, we parameterize the wave field and simulate only the most informative groups using natural initial conditions (starting the simulation from several cyclic waves ahead). We adopt a Bayesian experimental design approach, leveraging the uncertainty in a Gaussian process surrogate model to define an acquisition function for sequentially selecting the next best sample, thus reducing computational costs. The approach's effectiveness is demonstrated through different wave parametrization thresholds and ship responses' thresholds using a nonlinear roll equation. |
Monday, November 25, 2024 3:34PM - 3:47PM |
R12.00009: Nonlinear energy amplification of turbulent flows over progressive surface waves Ziyan Ren, Anqing Xuan, Lian Shen We proposed a novel reduced-order model that extends resolvent formulation from flat walls to waving walls with a curvilinear grid for problems such as wind-wave interaction. Large-eddy simulations with various wave ages are performed to obtain the two-dimensional time-averaged flows. Utilizing Floquet theory with a curvilinear boundary-fitted grid, the nonlinear effects induced by the wave are described accurately. We investigate the mechanisms of nonlinear energy amplification for different wave ages and the associated energy transfer among different scales. The results enhance our understanding of the wave effects on resolvent modes and the corresponding energy transfer mechanism (triadic interaction) from an input/output perspective. |
Follow Us |
Engage
Become an APS Member |
My APS
Renew Membership |
Information for |
About APSThe American Physical Society (APS) is a non-profit membership organization working to advance the knowledge of physics. |
© 2025 American Physical Society
| All rights reserved | Terms of Use
| Contact Us
Headquarters
1 Physics Ellipse, College Park, MD 20740-3844
(301) 209-3200
Editorial Office
100 Motor Pkwy, Suite 110, Hauppauge, NY 11788
(631) 591-4000
Office of Public Affairs
529 14th St NW, Suite 1050, Washington, D.C. 20045-2001
(202) 662-8700