Author(s): Ricardo Rebel; Christian Golla; Ramandeep Jain; Jochen Frohlich
Linked Author(s):
Keywords: Clustering Machine Learning Non-spherical particles Sediment transport DNS
Abstract: The motion of saltating particles in sediment transport is investigated with trajectories of non-spherical particles extracted from resolved direct numerical simulations of sediment transport. Machine learning is used to identify representative types of jumps exhibiting similar patterns of motion. Two different clustering approaches are compared for this purpose, clustering based on statistical metrics and clustering based on dynamic time warping. They are used to simultaneously determine an appropriate number of clusters as well as the composition of the individual clusters. Average trajectories and average orientations during the jumps are calculated for each cluster, providing the representative particle motion for the discovered types. The resulting method to identify and characterize a suitable number of clusters, i. e. types of jumps, constitutes a key result of this work. It is observed that in the data investigated four substantially different types of particle jumps occur during saltation.
Year: 2025