Bulletin of the American Physical Society
APS March Meeting 2022
Volume 67, Number 3
Monday–Friday, March 14–18, 2022; Chicago
Session A32: Differentiable and Machine Learning Infused Simulations in Fluid DynamicsFocus Recordings Available
|
Hide Abstracts |
Sponsoring Units: GDS Chair: Tiffany Summerscales, Andrews University Room: McCormick Place W-192B |
Monday, March 14, 2022 8:00AM - 8:36AM |
A32.00001: Emulating nonlinear dynamical systems from data using scientific machine learning Invited Speaker: Romit Maulik In this talk, I will present recent research that builds fast and accurate predictive models for various high-dimensional systems through a combination of data-driven and physics-based modeling. Moreover, through my examples, I will argue that such algorithm development must occur with consideration for data from multiple sources and fidelities, for deployment scenarios that leverage high-performance computing, and with appropriate emphasis on incorporating prior knowledge from physics-based modeling. I will give specific examples that highlight such themes for learning (both canonical and off-nominal) nonlinear dynamical systems from data. Some examples of the former include learning solutions to the advection-dominated viscous Burgers equations, and learning the chaotic nature of the Kuramoto-Sivashinsky equation. For the latter, I will discuss deployments of such learning algorithms for building reduced-order models for geophysical forecasting from numerical simulations as well as ship and satellite observation data. |
Monday, March 14, 2022 8:36AM - 9:12AM |
A32.00002: TBA Invited Speaker: Gabriel D Weymouth
|
Monday, March 14, 2022 9:12AM - 9:24AM |
A32.00003: Physics-guided surrogate models for fluid dynamics in complex geometries Varun Shankar, Robin Walters, Rui Wang, Rose Yu, Venkat Viswanathan Data-driven surrogate models based on deep learning have been shown to predict spatiotemporal dynamics orders of magnitude faster than traditional solvers. While much work on spatiotemporal surrogate models has dealt with canonical problems on gridded domains, many real-world problems involve complex geometry that may only be represented accurately through modern geometric deep learning approaches. Furthermore, pure data-driven surrogate models often fail to pertain to the underlying physical laws of the system. We propose a novel implementation of SO(3)-equivariant tensor convolutional networks to model moderate Reynolds number fluid systems in an arbitrary point-cloud domain. We use our model to predict turbulent quantities such as the Reynolds stresses and to forecast the velocity field with dynamics represented by a NeuralODE framework. Our model integrates physics principles by encoding symmetries in the architecture design. The efficacy of the encoded symmetries is validated with equivariance error and generalization to different geometries. This work aims to significantly expand the problem domain of deep learning surrogate models, contributing towards more efficient scientific modeling techniques in complex geometries. |
Monday, March 14, 2022 9:24AM - 9:36AM |
A32.00004: Deep learning based quasi-continuum theory for structural prediction of water and Lennard-Jones fluid in confined environments Haiyi Wu, Narayana R Aluru In this work, we propose a deep learning based quasi-continuum theory (DL-QT) to predict the concentration and potential profiles of a Lennard-Jones (LJ) fluid and water confined in a nano slit pore. In the first part, the deep learning model is built based on convolutional neural networks with the encoding-decoding process. The model is trained to relate the fluid properties to the fluid-fluid potential. We demonstrate that the well-trained model can accurately predict the fluid-fluid potential with a relative error < 5%. In the second part, the well-trained deep learning model is combined with the potential-based continuum theory to predict confined LJ fluid and water concentration profiles. We show that the DL-QT model has a robust predictive performance for LJ fluids confined in different channel sizes and under various thermodynamics states. In addition to predicting the properties of LJ fluid, we also demonstrate that the DL-QT model works well for confined water without introducing a coarse-grained potential model to the continuum theory. |
Monday, March 14, 2022 9:36AM - 9:48AM |
A32.00005: Learning hydrodynamic equations for active matter from particle simulations and experiments Rohit Supekar, Boya Song, Alasdair Hastewell, Gary Choi, Alexander Mietke, Jorn Dunkel Recent advances in particle-based simulation methods and high-resolution imaging techniques have enabled the precise characterization of collective dynamics in various biological and engineered active fluids. In parallel, data-driven algorithms for learning interpretable continuum models have shown promising potential for the recovery of underlying PDEs from continuum simulations. By contrast, learning macroscopic hydrodynamic equations and closure relations from microscopic particle simulations remains a major challenge. Here, we present a framework that leverages sparse regression learning algorithms to discover PDE models from coarse-grained microscopic data, while incorporating the relevant physical symmetries. We illustrate the practical potential through an application to a polar active particle model with alignment interactions mimicking those of swimming sperm cells. We further verify the framework with applications to recent micro-roller experiments and to tracked trajectories of the collective motion of animals. Our scheme succeeds in learning hydrodynamic equations that reproduce the characteristic dynamics observed in these systems, demonstrating how one can learn continuum theories directly from large-scale microscopic simulations and observations of complex systems. |
Monday, March 14, 2022 9:48AM - 10:00AM |
A32.00006: Modeling active fluids via physically constrained machine learning Matthew Golden, Roman O Grigoriev, Alberto Fernandez-Nieves, Jyothishraj Nambisan We investigate an experimental fluid flow driven by microtubules confined to an oil/water interface. Deriving a mathematical model of this active fluid from first principles is difficult, as not all the relevant physical processes are well understood. Instead, we use sparse physics-informed discovery of empirical relations (SPIDER) to learn the governing equations directly from experimental data. General physical constraints such as locality, causality, and symmetry are used to construct libraries of candidate relations between the flow field and the director field describing the orientation of microtubules. Sparse regression is then used to identify a parsimonious two-dimensional model of this system. Three PDEs are identified from data: an incompressibility condition and momentum balance describing the fluid flow and a separate equation for the director field. The latter two governing equations are distinct from those appearing in the literature. In particular, neither the advection terms nor the time derivative of the flow velocity appears in the momentum equation, consistent with the low Reynold's number of the flow. We also find that elastic effects cannot be described by weakly nonlinear terms in the evolution equation for the director field. |
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. |
© 2024 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