Dynamic 3D surface reconstruction and instantaneous burning rate quantification in solid propellant via deep learning-driven stereoscopic imaging

Conventional measurement techniques typically approximate the burning surface as planar, neglecting its three-dimensional (3D) surface evolution during combustion. To address this limitation, this study proposes a high-speed stereoscopic imaging method for dynamic 3D burning-surface reconstruction and instantaneous burning rate quantification in solid propellants. The system acquires stereo image pairs at up to 2 kHz with a spatial resolution of 7.4 μm/pixel. A reconstruction pipeline integrating multiple deep learning models is then used to obtain transient 3D burning surfaces. From the reconstructed 3D meshes, temporal variations in surface area and volume are quantified, from which the instantaneous burning rate is determined. Experimental results illustrate three combustion stages, including flame propagation, steady combustion, and burnout. During the initial stage, the active burning region spreads across the surface before transitioning to inward regression, consistent with classical descriptions of flame spreading in solid propellants. These results demonstrate that the proposed method enables time-resolved 3D visualization of burning surface evolution and quantitative burning rate measurement in laboratory-scale solid propellant tests.