Physical Review Fluids
- Highlights
- Recent
- Accepted
- Collections
- Authors
- Referees
- Search
- Press
- About
- Editorial Team
Data-driven low-dimensional model of a sedimenting flexible fiber
Andrew J. Fox and Michael D. Graham
Phys. Rev. Fluids 9, 084101 – Published 16 August 2024
- Article
- References
- No Citing Articles
- Supplemental Material
PDFHTMLExport Citation
- Abstract
- Authors
- Article Text
Abstract
The dynamics of flexible filaments entrained in flow, important for understanding many biological and industrial processes, are computationally expensive to model with full physics simulations. In this paper, we describe a data-driven technique to create high-fidelity low-dimensional models of flexible fiber dynamics using machine learning; the technique is applied to sedimentation in a quiescent, viscous Newtonian fluid, using results from detailed simulations as the dataset. The approach combines an autoencoder neural network architecture to learn a low-dimensional latent representation of the filament shape, with a neural ordinary differential equationthat learns the evolution of the particle in the latent state. The model was designed to model filaments of varying flexibility, characterized by an elastogravitational number , and was trained on a dataset containing the evolution of fibers beginning at set angles of inclination. For the range of considered here (100–10000), the filament shape dynamics can be represented with high accuracy with only four degrees of freedom, in contrast with the 93 present in the original bead-spring model used to generate the dynamic trajectories. We predict the evolution of fibers set at arbitrary angles and demonstrate that our data-driven model can accurately forecast the evolution of a fiber at both trained and untrained elastogravitational numbers.
3 More
- Received 16 May 2024
- Accepted 11 July 2024
DOI:https://doi.org/10.1103/PhysRevFluids.9.084101
©2024 American Physical Society
Physics Subject Headings (PhySH)
- Physical Systems
Fibers
- Techniques
Machine learningStokes equations
Fluid Dynamics
Authors & Affiliations
Andrew J. Fox and Michael D. Graham*
- Department of Chemical and Biological Engineering, University of Wisconsin-Madison, Madison, Wisconsin 53706, USA
- *Contact author: mdgraham@wisc.edu
Article Text (Subscription Required)
Click to Expand
Supplemental Material (Subscription Required)
Click to Expand
References (Subscription Required)
Click to Expand
Issue
Vol. 9, Iss. 8 — August 2024
Access Options
- Buy Article »
- Log in with individual APS Journal Account »
- Log in with a username/password provided by your institution »
- Get access through a U.S. public or high school library »
Article part of CHORUS
Accepted manuscript will be available starting16 August 2025.Images
Figure 1
A flexible filament of length settling under an external force in a quiescent Newtonian fluid. The filament begins at an arbitrary initial inclination relative to the axis and evolves until it reaches a terminal shape. The fiber is modeled as a series of beads of radius connected by springs, with the center bead, which has position , shown in red.
Figure 2
The evolution of the shape of a filament settling in a quiescent Newtonian fluid from an initial orientation to a common terminal shape at for all initial angles of orientation within the training dataset.
Figure 3
The trajectories of the center bead of a filament settling in a quiescent Newtonian fluid at for all initial angles of inclination within the training dataset. The initial positions are denoted by symbol , and the terminal positions are denoted by the symbol .
Figure 4
The evolution of the shape of a filament settling in a quiescent Newtonian fluid from a common initial angle of orientation of to a terminal shape for all within the training dataset.
Figure 5
Block diagram for data-driven model combining the autoencoder and temporal-evolution scheme. The temporal-evolution neural network, expanded in red, can be separated into two distinct neural networks forecasting the evolution of latent representation of the shape and the shape-dependent change in position ; in practice, these can be forecasted by a single neural network.
Figure 6
(a)Loss over the testing data for each autoencoder architecture as a function of latent dimension. Block diagrams for the autoencoder neural network architectures: (b)No , (c)Encoder , (d)Decoder , and (e) Double .
Figure 7
Evolution of the shape of a filament settling in quiescent Newtonian fluid given an initial orientation from the best and worst forecasts (red) and the true evolution (black) at a given . Here, is within the training dataset, and the initial angles of inclination are not.
Figure 8
Evolution of the shape of a filament settling in quiescent Newtonian fluid given an initial orientation from the best and worst forecasts (red) and the true evolution (black) at a given . Here, neither nor the initial angles of inclination are within the training dataset.
Figure 9
Trajectory of the center bead a filament settling in quiescent Newtonian fluid given an initial orientation from the best (solid lines) and worst (dashed lines) forecasts and the true evolution at a given . The initial positions are denoted by symbol , and the terminal positions are denoted by the symbol , with the true and predicted positions in black and red, respectively. In (a), is within the training dataset but, at initial angles of inclination, not within the training data; in (b), neither nor the initial angles of inclination are within the training data.
Figure 10
Ensemble-average error vs time for forecasts of the evolution of a filament settling in quiescent Newtonian fluid at each (gray) and averaged over all (red). In (a), is within the training dataset but, at initial angles of inclination, not within the training data; in (b), neither nor the initial angles of inclination are within the training data.