Deep learning-based analysis of aluminum particle trajectory and spatial size-velocity correlation in solid propellant combustion at 6 MPa

Abstract

Characterizing particle size, position, and velocity, particularly under elevated pressure, is crucial for understanding the underlying combustion dynamics and optimizing propellant formulations to enhance rocket propulsion efficiency. This study presents a deep learning-based data analysis method designed for the detection, tracking, and diameter uncertainty assessment of blurred combustion aluminum particles captured by microscopic imaging. The methodology employs a fine-tuned You Only Look Once model, enhanced by Slicing Aided Hyper Inference, to accurately detect aluminum particles, especially smaller ones. Weighted Distance Intersection over Union is proposed for robust tracking in scenarios with high particle concentrations. Innovatively, a Gaussian kernel-based blur estimation technique is introduced to quantify the uncertainty in particle size measurements. Qualitative and quantitative evaluation experiments have demonstrated the effectiveness of this method. Furthermore, time-resolved individual particle dynamics and spatial particle size and velocity profiles were thoroughly explored using frames captured from a single propellant strand burning test at 6 MPa. Two coupled physical mechanisms underlying particle size dynamics were discovered, while the multimodal behavior of combustion aluminum particles was further analyzed in depth. This research successfully expands the applicability of microscopic imaging, presenting an approach to studying combustion mechanisms under high-pressure conditions.