A Machine-Learning-Based Digital Inline Holography (DIH) for Airborne Water Droplet and Ice Crystal Characterization

Icing presents a significant safety hazard to Advanced Air Mobility (AAM) vehicles and substantially degrades their aerodynamic performance. Accurate prediction of icing events necessitates precise characterization of atmospheric icing conditions, particularly the morphological and dimensional properties of airborne supercooled water droplets and ice crystals within icing clouds. Digital inline holography (DIH) offers exceptional capability for capturing both amplitude and phase information of light waves, enabling three-dimensional reconstruction of particle positions, morphologies, and size distributions through sophisticated postprocessing algorithms. However, conventional postprocessing methods exhibit expensive computational cost that precludes real-time DIH measurements essential for AAM operations. This study presents a novel machine learning-based DIH framework designed to efficiently process holograms while maintaining accurate characterization of particle geometry and dimensions within three-dimensional volumes. The proposed framework achieves a 60-fold acceleration in postprocessing speed compared to traditional methods. Model training and validation utilized comprehensive experimental hologram datasets encompassing National Institute of Standards and Technology (NIST)-certified reference particles, airborne supercooled water droplets, and ice crystals. This advancement represents a substantial step toward implementing real-time DIH measurement systems for enhanced AAM icing hazard assessment and mitigation.