Development of a physics-informed neural network to enhance wind tunnel data for aerospace design

Machine-Learning-Enabled Foil Design Assistant

“…These were addressed by employing a low-dimensional, robust, and accurate foil parametric model for the foil geometry description coupled with simple performance indicators and the lift and drag profiles over a range of angles of attack, which permit a user-friendly and intuitive definition of geometry or performance requirements by an engineer even at the early design stages. An obvious drawback in this setup is the lack of meta-information commonly used in advanced physics-informed neural networks (PINNs), which embed foil physics and are able to more-accurately predict foil performance and, consequently, enhance design optimization; see, for example, some recent relevant PINNs used in foil performance prediction in [31,32]. To address this last issue, we opted to include the integral characteristics (geometric moments) of foil profiles, which are commonly correlated with the performance criteria and can, therefore, provide a physics-informed flavor without requiring expensive calculations, as was recently demonstrated by Shah et al in [26].…”