Reykjavik UniversityHassanian, R.Helgadóttir, I.Gharibi, F.Beck, A.Riedel, M.2025-11-172025-11-172025-02-01Hassanian, R, Helgadóttir, I, Gharibi, F, Beck, A & Riedel, M 2025, 'Data-driven deep learning models in particle-laden turbulent flow', Physics of Fluids, vol. 37, no. 2, 023348. https://doi.org/10.1063/5.02517651070-66312365237604bea84f9-75b4-4de7-a747-aba84f0f926085217894949https://hdl.handle.net/20.500.11815/6074Publisher Copyright: © 2025 Author(s).The dynamics of inertial particles in turbulent flow are complex, and in practice, gravity influences particle dynamics. However, the effects of gravity have not been appropriately investigated using numerical approaches. This study provides the first empirical evidence of a data-driven deep learning (DL) model to predict the velocity, displacement, and acceleration of inertial particles in a strained particle-laden turbulent flow. This study introduces a DL model to experimental data from Hassanian et al., who investigated distorted turbulent flow within a specific range of Taylor microscale Reynolds number, 100 < R e λ < 120 . The flow experienced a vertical mean strain rate of 8 s−1 under the influence of gravity. Lagrangian particle tracking technique was employed to capture each inertial particle's velocity field and displacement. The deep learning model relies on experimental particle-laden turbulent flow, demonstrating all effective parameters, including turbulence intensity, strain rate, turbulent energy dissipation rate, gravity, particle size, particle density, and small and large-scale effects. The forecasting model demonstrates significant capability and high accuracy in generating predictions closely aligned with the actual data. Model training and inference are run on the high-performance computing DEEP-DAM system at the Jülich Supercomputing Center. The proposed approach can potentially enhance the understanding of inertial particle dynamics and the parameters that affect them. Furthermore, data-driven models can offer new insights into particle motion and the underlying differential equations within physics-based deep learning frameworks.2960331eninfo:eu-repo/semantics/openAccessComputational MechanicsCondensed Matter PhysicsMechanics of MaterialsMechanical EngineeringFluid Flow and Transfer Processes/dk/atira/pure/researchoutput/researchoutputtypes/contributiontojournal/article10.1063/5.0251765