Supersonic flow field reconstruction using CNNs

The accurate prediction of a projectile’s aerodynamic coefficients is crucial in high-precision external ballistic calculations. The aerodynamic forces and moments exerted on a projectile in flight influence key performance parameters such as range, accuracy, time of flight and stability. A large body of work has therefore been dedicated to understanding the flow dynamics around projectile bodies and obtaining the critical force and moment coefficients. This has been traditionally achieved in aeroballistic range experiments, wind tunnel set-ups and through the use of numerical models. Nevertheless, a widespread still exists between different techniques, revealing the fluid physics is not yet fully understood. A better understanding of the aerodynamics at play is accessible through a combination of the three techniques. However, reliable wind tunnel results will require matching a series of similarity parameters imposed by the firing conditions, which will inevitably relate to the physical scale of the models used. The size of small calibre projectiles may prove challenging for measurement in wind tunnel set-ups, however upscaling the models inappropriately will result in unrepresentative flow fields due to wall interactions and blockage effects. On the other hand, sting supports for wind tunnel models disturb a smaller portion of the flow with increasing projectile scale, particularly in terms of wake perturbation - a key contributor to aerodynamic coefficients. Clearly, scale effects have important consequences, however they have not been explicitly treated in the supersonic projectile literature. This study aims to explore the effects and limits of projectile scaling in supersonic wind tunnels, through a series of experimental techniques (Schlieren visualization, pressure measurements, force balance measurements...) and numerical modelling. Additionally, we aim to develop the Background-Oriented-Schlieren technique a step further through the use of machine learning models to reconstruct complete flow fields from optical data.