Capturing functional relations in fluid-structure interaction via machine learning.

Area of Science:

• Computational fluid dynamics
• Machine learning applications
• Structural mechanics

Background:

• Fluid-structure interaction (FSI) is crucial in fields like aerodynamics and cell biology.
• Traditional FSI simulations using Navier-Stokes equations are computationally intensive.
• Existing methods face significant computational overhead, limiting their practical application.

Purpose of the Study:

• To develop a machine learning (ML)-based strategy to bypass computationally expensive FSI simulations.
• To create accurate and computationally tractable models for fluid forces acting on immersed structures.
• To predict the dynamic behavior of immersed beams without detailed Navier-Stokes solvers.

Main Methods:

• Introduced dissipation into a beam model to simulate fluid effects.
• Decoupled time-varying forces using linear algebra to generate ground truth data.
• Employed symbolic regression (ML) to derive continuous functional forms for forces and moments.

Main Results:

• Symbolic regression successfully generated computationally efficient force and moment functions.
• ML-estimated forces and moments accurately predicted immersed beam deflections.
• The ML approach demonstrated conformity with detailed fluid-structure interaction solutions.

Conclusions:

• Machine learning, specifically symbolic regression, offers a viable alternative to traditional FSI simulations.
• The developed ML model accurately predicts beam deflections, reducing computational burden.
• This strategy enhances the efficiency of analyzing fluid-structure interactions in engineering and science.