Predictions of Particle Trajectory Response to Reynolds Number in Turbulent Channel Flows Using ANNs

The present work details the development, implementation, and performance of a hybrid machine‐learning algorithm for predicting particle trajectories and dynamics in wall‐bounded turbulent flows, trained on direct numerical simulation and Lagrangian particle tracking (DNS‐LPT) data at Re τ = 180 and 300. The model accurately reconstructs the mean and fluctuating particle velocities and concentration profiles, across a range of Reynolds numbers, and maintains similar accuracy at an unseen intermediate Reynolds number ( Re τ = 240). This generalization demonstrates that the model is capable of learning the underlying physics rather than memorizing training data, enabling reliable predictions at conditions for which no DNS‐LPT training trajectories exist. By reducing the volume of high‐fidelity training data required while retaining DNS‐like accuracy, this approach offers a computationally cheaper alternative to traditional particle‐laden flow simulations.