A Physics-Informed Data-Driven Approach for Boundary Layer Flows

What is it about?

Our research focuses on using a type of artificial intelligence called physics-informed neural networks (PINNs) to predict how fluid flows behave near surfaces, like the flow of air over an airplane wing. Traditional simulations of such flows rely on complex computations that can be time-consuming and expensive, especially for simulations that require high accuracy. Standard neural networks, while effective for many tasks, struggle to accurately model fluid flows unless trained on massive datasets, which are often hard to obtain. Instead, we employ PINNs, which incorporate the physics of fluid flows directly into the learning process by using equations that describe flow behaviors. By doing this, PINNs require less data and train faster than conventional methods. In our study, we tested PINNs on a classic boundary layer flow problem, using a simpler computational model to generate initial data. We then used the PINN to predict key flow properties, such as velocity profiles and boundary layer thickness. Our results show that PINNs provide more accurate predictions of these properties than standard neural networks, making them a promising tool for modeling wall-bounded flows efficiently. This approach could benefit various fields where fluid dynamics is critical, from engineering to environmental science.

Featured Image

Photo by Uriel SC on Unsplash

Photo by Uriel SC on Unsplash