Data-driven analysis of thin particle settling dynamics in air using experimental data

Thin, nonspherical particles can exhibit complex falling behaviors even in quiescent air due to their high Reynolds numbers and large density ratios. In recent years, data-driven approaches have emerged as powerful tools for modeling such intricate dynamical systems. To investigate the settling dynamics of thin particles, a dual-perspective shadowgraph experiment was conducted to collect trajectory data, capturing full three-dimensional velocity and acceleration. A predictive model based on a Long Short-Term Memory (LSTM) neural network was developed and trained using the acquired trajectory data. The model predicts diverse settling modes, including fluttering, tumbling, and irregular motions. Despite the limited number of experimental samples, model performance and robustness were significantly enhanced through data augmentation techniques involving coordinate transformations and the addition of Gaussian noise. Quantitative comparisons between predicted and observed trajectories confirm the model's effectiveness in capturing the complex settling behavior of thin particles in air.