Open-source BOS tomography dataset of high-speed flow over a flight body

Abstract

We present an open-source background-oriented schlieren dataset with 70 views of high-speed flow over a flight body. Sample analyses are performed using a neural-implicit reconstruction technique (NIRT) with total variation regularization as well as data assimilation via the 3D compressible Euler equations. Limited-data reconstructions based on nine views resolve sharp shocks that are consistent with the geometry, reproduce validation deflections with high fidelity, and exhibit minimal artifacts. Data assimilation recovers unmeasured fields, marking the first demonstration of 3D state estimation directly from experimental schlieren measurements. The NIRT also enables efficient uncertainty quantification, providing insight into well-resolved flow features and guiding design-of-experiments efforts. Public access to the data and code repositories is detailed at the end of this correspondence.

Figures (6)

• Figure 1: Flight vehicle model with fastening and routing regions shown in zebra print and individual design pieces in unique colors (left). Body dimensions are given (bottom) alongside images of the installation from four perspectives (right).
• Figure 2: Images of the flight body (top left) and experimental setup (bottom left). Schematic of the imaging setup and facility (right). Cameras are positioned to achieve similar fields of view and circles of confusion at the object plane, with the latter having an estimated diameter of 0.95 mm.
• Figure 3: Deflections from the master camera obtained by optical flow (top row) and reprojected through the reconstructed 3D field (bottom row). Blue borders indicate validation frames excluded from reconstruction. Deflections are color-coded by the bi-directional wheel (right), with white indicating no deflection. Masks used in reconstruction are shown: grayscale regions mark omitted pixels; the dotted pattern marks the freestream mask, where an ambient density boundary condition was weakly enforced.
• Figure 4: Streamwise slices of the non-dimensional density gradient field with key flow features labeled. Magnitudes are colored by the sign of the streamwise density gradient to distinguish expansion (blue) from compression (red).
• Figure 5: Density field shown as a centerline slice and two isosurfaces. The higher level ($\rho/\rho_\infty = 1.1$, red surface) captures compression structures as the flow encounters the body; the lower level ($\rho/\rho_\infty = 0.5$, blue surface) depicts the expansion off the tail.