Algorithm for Time-Resolved Background-Oriented Schlieren Tomography Applied to High-Speed Flows

Quantitative flow-visualization techniques like background-oriented schlieren (BOS) are important tools for studying complex, time-varying flows, such as high-speed jets in propulsion systems. BOS captures 2D deflections related to path-integrated density gradients, these data can be combined across multiple views to reconstruct 3D density fields. This work evaluates the neural-implicit reconstruction technique (NIRT) for time-resolved BOS tomography, benchmarking its performance on two synthetic jet flows. Using a thermally driven jet snapshot, we validate NIRT's range of operability and compare its accuracy to state-of-the-art methods. Reconstruction stability is quantified through deep ensembles, which are used to assess the resolvable length scale. Our results showcase NIRT's robustness to sparse views and measurement noise. To explore temporal dynamics, we reconstruct a time-resolved BOS dataset of a high-speed jet, employing smoothness-based regularization to ensure physically plausible solutions. NIRT reliably resolves both large-scale structures and fine flow features with high fidelity. Future extensions include data assimilation using the governing equations, enabling the inference of additional fields like velocity and total energy. Our framework lays the groundwork for high-fidelity 4D flow field analysis from experimental BOS data.