Framework for data-driven propeller and rudder modeling for ship maneuvering

Abstract

The maneuvering characteristics of a vessel must be understood for safe and efficient operation in the ocean environment. During a maneuver, the propeller and rudder can operate in off-design conditions. Computational Fluid Dynamics is a useful tool to capture flow effects like viscous separation, but it is too costly to perform a large number of maneuvering simulations with a discretized propeller. The propeller has much smaller time and length scales compared to the maneuver, which drives up the computational cost when the propeller is discretized. Body-force propeller models provide a less expensive means of modeling a maneuvering vessel with CFD, but many propeller models can be inaccurate when the propeller operates off-design. This work develops a framework for a data-driven propeller and rudder model which is trained with high-fidelity simulations of the propeller and deflected rudder operating in the wake of the hull. The dimensionless velocity and oblique flow angle are used as features for the propeller and rudder model. The uncertainty of the data-driven model is dependent upon the error inherent in the training and validation data as well as the uncertainty induced by using a surrogate model to determine the response surface. The underlying uncertainties are examined. The calculated propeller and rudder forces from the data-driven model are compared to the experimental forces. The turning circle characteristics of the KRISO Container ship at model scale are examined with the implemented data-driven propeller and rudder model. The accuracy of the method used to train the model is maintained and the computational cost of modeling a maneuvering vessel is dramatically reduced compared to using a discretized propeller.