Physics-informed neural network meets Kelvin-Helmholtz instability

What is it about?

Kelvin-Helmholtz instability (KHI) is a fluid instability that arises from velocity shear within a single continuous fluid or from a velocity difference at the interface between two fluids. This work utilizes the physics-informed neural networks to solve the inverse problems of KHI flows, enabling field reconstruction and parameter inference from sparse, noisy data.

Featured Image

Photo by NASA Hubble Space Telescope on Unsplash

Photo by NASA Hubble Space Telescope on Unsplash

Why is it important?

(1) Background. Kelvin-Helmholtz instability (KHI) is ubiquitous in nature (cloud formations, surface ripples on lakes, and exhaled breath) and industrial facilities (engines, jet cooling systems). Thus, prediction of KHI is a critical issue across various fields, including astronomy, meteorology, oceanography, life science, environmental science, and engineering. (2) Motivation. The required extensive high-fidelity data underscores the need for field reconstruction and parameter inference from sparse, noisy data, which is one of the capabilities of the physics-informed neural networks (PINNs). (3) Challenge. KHI flows have the spatiotemporal and magnitude multiscale, which challenges the PINNs due to the spectral bias and magnitude preference of neural networks. (4) Solution. The multiscale embedding and small-velocity amplification strategies are adopted to address the spatiotemporal and magnitude multiscale issues, respectively. The results demonstrate the accuracy of PINNs in field reconstruction and parameter inference, and validate the effectiveness of the proposed strategies.

Perspectives