Toward neural network-based optical wave reconstruction for the supersonic turbulent cavity flow

Density fluctuations in the shear layer locally alter the effective index of refraction of the atmosphere, causing bore-sight errors that are characterized by an apparent shift in the target location. To address the lack of viable correction methods for supersonic and hypersonic aero-optical distortions, we perform a large-eddy simulation using the JENRE Multiphysics Framework to approximate the boundary-layer and shear-layer flow over a cavity operating at a free-stream Mach number of 2.3 and an altitude of 16 km. The optical path difference (OPD) is calculated from the high-frequency density sampling over a 0.0254m×0.0254m aperture located at the center of the cavity. Spectral proper orthogonal decomposition of the OPD reveals dominant flow structures contributing to wavefront aberrations. Using the simulated OPD data, we train an artificial neural network to process the Shack-Hartmann wavefront sensor outputs and reconstruct the original wavefront. This data-driven approach demonstrates potential for faster and more accurate correction of imaging errors compared to traditional methods, particularly when tailored to specific operational conditions.