Explore missing flow dynamics by physics-informed deep learning: the parameterised governing systems

Abstract

Gaining and understanding the flow dynamics have much importance in a wide range of disciplines, e.g. astrophysics, geophysics, biology, mechanical engineering and biomedical engineering. As a reliable way in practice, especially for turbulent flows, regional flow information such as velocity and its statistics, can be measured experimentally. Due to the poor fidelity or other technical limitations, some information may not be resolved in a region of interest. On the other hand, detailed flow features are described by the governing equations, e.g. the Navier-Stokes equations for viscous fluid, and can be resolved numerically, which is heavily dependent on the capability of either computing resources or modelling. Alternatively, we address this problem by employing the physics-informed deep learning, and treat the governing equations as a parameterised constraint to recover the missing flow dynamics. We demonstrate that with limited data, no matter from experiment or others, the flow dynamics in the region where the required data is missing or not measured, can be reconstructed with the parameterised governing equations. Meanwhile, a richer dataset, with spatial distribution of the control parameter (e.g. eddy viscosity of turbulence modellings), can be obtained. The method provided in this paper may shed light on data-driven scale-adaptive turbulent structure recovering and understanding of complex fluid physics, and can be extended to other parameterised governing systems beyond fluid mechanics.

arXiv commentUpdated

References in corpus (12)

• Machine Learning for Fluid Mechanics
• Turbulence Modeling in the Age of Data
• NSFnets (Navier-Stokes Flow nets): Physics-informed neural networks for the incompressible Navier-Stokes equations
• Physics Informed Deep Learning (Part I): Data-driven Solutions of Nonlinear Partial Differential Equations
• A composite neural network that learns from multi-fidelity data: Application to function approximation and inverse PDE problems
• PPINN: Parareal Physics-Informed Neural Network for time-dependent PDEs
• A Physics Informed Machine Learning Approach for Reconstructing Reynolds Stress Modeling Discrepancies Based on DNS Data
• Deep Learning of Vortex Induced Vibrations
• Physics Informed Deep Learning (Part II): Data-driven Discovery of Nonlinear Partial Differential Equations
• Quantifying and Reducing Model-Form Uncertainties in Reynolds-Averaged Navier-Stokes Simulations: A Data-Driven, Physics-Based Bayesian Approach
• Understanding and mitigating gradient pathologies in physics-informed neural networks
• Turbulence closure modeling with data-driven techniques: physical compatibility and consistency considerations

Cited by in corpus (5)

• Physics-informed PointNet: A deep learning solver for steady-state incompressible flows and thermal fields on multiple sets of irregular geometries
• Long-term predictions of turbulence by implicit U-Net enhanced Fourier neural operator
• Attention-Enhanced Neural Network Models for Turbulence Simulation
• Mean flow data assimilation based on physics-informed neural networks
• A Framework of Data Assimilation for Wind Flow Fields by Physics-informed Neural Networks