Quantitative Characterization of Metallic Composite Particle Combustion using X-ray Phase Contrast Imaging and Machine Learning

Reactive metallic powders have a rich set of applications as energetic materials, including as additives in propellants, explosives, tools for bioagent defeat, and pyrotechnics. Microexplosions are often observed during the combustion of Al-Zr bimetallic powders in air, leading to a prolonged burn. Rapid bubble growth caused by the accumulation of nitrogen from air is hypothesized to be the underlying mechanism. To better understand this mechanism, x-ray phase contrast imaging was recently adopted to observe in situ bubble nucleation and growth during combustion of Al-Zr powders. We report here a high-throughput framework using machine learning to identify and track particles from x-ray phase contrast imaging videos. We use convolutional neural networks to identify particles, a Kalman filter to track trajectories, and additional tools to characterize burning particles. This framework enables us to automatically extract, organize, and analyze quantitative data from large amounts of video data. Using it, we are able to better determine the relationship between bubble growth and microexplosions. The framework can also be applied to combustion analysis of other energetic materials, and we believe it will be a valuable tool for quantitative characterization of combustion.