Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
126 tokens/sec
GPT-4o
47 tokens/sec
Gemini 2.5 Pro Pro
43 tokens/sec
o3 Pro
4 tokens/sec
GPT-4.1 Pro
47 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Deep Learning of Vortex Induced Vibrations (1808.08952v1)

Published 26 Aug 2018 in physics.flu-dyn, cs.CE, cs.LG, math.AP, and stat.ML

Abstract: Vortex induced vibrations of bluff bodies occur when the vortex shedding frequency is close to the natural frequency of the structure. Of interest is the prediction of the lift and drag forces on the structure given some limited and scattered information on the velocity field. This is an inverse problem that is not straightforward to solve using standard computational fluid dynamics (CFD) methods, especially since no information is provided for the pressure. An even greater challenge is to infer the lift and drag forces given some dye or smoke visualizations of the flow field. Here we employ deep neural networks that are extended to encode the incompressible Navier-Stokes equations coupled with the structure's dynamic motion equation. In the first case, given scattered data in space-time on the velocity field and the structure's motion, we use four coupled deep neural networks to infer very accurately the structural parameters, the entire time-dependent pressure field (with no prior training data), and reconstruct the velocity vector field and the structure's dynamic motion. In the second case, given scattered data in space-time on a concentration field only, we use five coupled deep neural networks to infer very accurately the vector velocity field and all other quantities of interest as before. This new paradigm of inference in fluid mechanics for coupled multi-physics problems enables velocity and pressure quantification from flow snapshots in small subdomains and can be exploited for flow control applications and also for system identification.

Citations (341)

Summary

  • The paper demonstrates that deep neural networks accurately predict lift, drag, and pressure fields in vortex induced vibrations using scattered data.
  • It replaces computationally expensive CFD methods with efficient, data-driven models for solving complex fluid-structure interactions.
  • The study’s approach generalizes to multiphysics problems, paving the way for advanced flow control and real-time structural assessments.

Deep Learning of Vortex Induced Vibrations

The paper "Deep Learning of Vortex Induced Vibrations" presents an innovative approach to solving inverse problems in fluid mechanics using deep neural networks (DNNs). Specifically, this research focuses on predicting the lift and drag forces on structures subjected to vortex-induced vibrations (VIV), a phenomenon that occurs when the vortex shedding frequency matches the natural frequency of a structure. Traditional computational fluid dynamics (CFD) methods, which are often computationally expensive and limited in versatility for high Reynolds numbers and complex geometries, are replaced with a data-driven deep learning methodology.

Research Overview

The authors propose two primary methods for predicting vital fluid dynamics parameters. The first involves using scattered space-time velocity data and the structure's motion to estimate structural parameters, and to reconstruct the time-dependent pressure fields, velocity vector fields, and the dynamics of the structure using four coupled deep neural networks. The second method uses only concentration field data—such as dye or smoke visualizations—enabling the inference of velocity fields and other dynamic quantities through five coupled deep neural networks.

Numerical Results and Claims

The paper demonstrates high accuracy in inferring pressure fields and structural parameters. It also indicates that the method delivers reliable predictions when provided with either velocity data or simply snapshot visualizations of concentrations, without prior data on pressure. Such an approach is not only computationally efficient but surpasses the capabilities of standard CFD approaches in certain complex scenarios. Further, the research showcases the potential of DNNs to generalize solutions in multi-physics problems across small subdomains.

Implications and Future Directions

The implications of this work are substantial for both theoretical and practical applications. On the theoretical side, the use of Navier-Stokes informed deep neural networks suggests a shift in how fluid mechanics problems could be tackled, leaning on the intrinsic capabilities of deep learning to assimilate scattered data efficiently. Practically, such methodologies could be transformative in areas such as flow control, structural safety assessments in engineering, and real-time monitoring systems in various engineering applications.

The potential advancements in AI and data-driven models could expand with applications beyond incompressible flow scenarios. Future research could explore chaotic and turbulent regimes, enhancing the predictability and efficiency of solutions obtained from machine learning frameworks. Additionally, this paper opens pathways for modifying the proposed algorithms to handle non-Newtonian flows and compressible scenarios, extending both the depth and breadth of physics-informed machine learning.

In summary, the paper effectively fuses deep learning with fluid mechanics to offer a scalable and robust framework for predicting vortex-induced vibration dynamics, providing a foundation for extending data-driven approaches into more complex physical simulations in the field of computational fluid dynamics.