Abstract
In this work, we explore the potential of Physics-Informed Neural Networks (PINNs) to process and combine data from supersonic wind tunnel experiments with computational methods. Specifically, we aim to infer velocity fields for a family of turbulent shockwave-boundary layer interactions (SBLI) at Ma = 2, using Background Oriented Schlieren (BOS) measurements. We train a neural network constrained by a set of Euler equations to predict velocity, pressure, and density fields from limited and noisy BOS data in an inviscid region of the flow field. We compare the results with PIV measurements to validate the predictions and analyze the effects of the BOS line-of-sight integration. The results suggest that, in this inviscid case, it is possible to reconstruct the velocity fields with an RMSE of < 5% even though the source data is corrupted by three-dimensional effects and measurement noise.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have