Development, Verification and Validation of a Machine-Learned Actuator Line Propeller Model

Bowman, J. L., Bhushan, S., & Aram, S. (2025). Development, Verification and Validation of a Machine-Learned Actuator Line Propeller Model. Proceedings of the ASME 2025 Fluids Engineering Division Summer Meeting. Philadelphia, PA, USA. DOI:10.1115/FEDSM2025-158665.

Simulating ship maneuvering via computational fluid dynamics (CFD) faces significant cost barriers due to the intricate hull–propeller–rudder interactions under unsteady flow conditions. Body force-based reduced-order approaches such as the actuator disk model offer efficiency gains but oversimplify the flow physics, while higher-fidelity actuator line models (ALMs) better capture wake behavior yet rely on fixed lift and drag tables that do not reflect unsteady effects. To overcome these limitations, a machine-learned actuator line model (ML-ALM) for marine propellers is developed, leveraging blade-resolved CFD data to train a deep neural network that predicts local blade forces instead of static lookup tables. The ML-ALM integrates into the ALM framework by distributing hydrodynamic forces along rotating blade lines using two-dimensional sectional data, thereby capturing transient tip vortex dynamics and swirl with far lower computational cost than full blade-resolved simulations. In open-water validation tests, the ML-ALM closely reproduces thrust, torque, power coefficient curves, and near-wake velocity profiles, showing good agreement with experimental measurements and blade-resolved CFD benchmarks up to three diameters downstream. This work represents one of the first naval applications of an ML-enhanced ALM. It demonstrates a promising path toward efficient yet accurate CFD propeller modeling for realistic ship maneuvering scenarios, ultimately enabling faster design analyses for safe and efficient ship operations.