A schlieren image (SI) serves as the important technique for density volume measurement and reconstruction in a supersonic wind tunnel. Yet, current SI measurement techniques tend to work under multiple cameras, which lead to high complexity on optical windows. To facilitate the measurement, the feasibility of density volume reconstruction (DVR) via a single-shot SI only is investigated in this paper, via a data-driven paradigm. However, since the single-shot SI exists the schlieren void zone and ill-posed inversion for DVR, a significant challenge appears, especially for asymmetric density fields. To address these issues, the estimated surface aerodynamic performance (SAP) and three-dimensional (3D) geometry shape information of aerocrafts are collectively embedded into two-dimensional (2D) SI as feature enhancement, since SAP implies the latent flow field information to some extent, forming an aerodynamic-aware schlieren image (AAS) with 2D-cross-3D modality. Moreover, a novel convolution manner with elastic stride and sampling shape is developed to adaptively extract shockwave textures and aerodynamic-shape features in AAS. Additionally, the network architecture employs a multiphysics fusion structure with parallel aerodynamic pressure/heat channels for robust DVR, while embedding density-optical physical constraints into the loss function to ensure physical consistency. Validation shows the proposed method approaches multi-camera-level accuracy for both symmetric and asymmetric fields.
Long et al. (Fri,) studied this question.