Data-Driven Propeller Modeling for Ship Maneuvering

Abstract

The analysis and understanding of the maneuvering characteristics of seafaring vessels is critical for their successful operation. Ships operate in the harsh ocean environment and must be able to either withstand the storms that they sail through or alter course to avoid particularly severe storms. As vessels become more automated and design practices shift toward reducing emissions it is important to gain an understanding of the expected maneuvering characteristics of vessels both in calm water and in waves. One way to determine the maneuvering characteristics of a vessel is by using Computational Fluid Dynamics (CFD) to account for effects like transient flow separation over appendages and green water over the deck of the ship. Both the propeller and the rudder are critical components of this analysis, but it is difficult to accurately and efficiently determine the forces generated by these surfaces. The differences in time and length scales between the propeller and the hull create a bottleneck for efficiently modeling a vessel with numerical methods since the limiting time step size is that of the propeller. There is complex flow interaction between the hull, propeller, and rudder which is important to capture to properly model the multi-dimensional forces on each surface.

One way to alleviate the cost of directly modeling the rotating propeller is to apply a propeller model. Existing propeller models often use simplifying assumptions and may predict incorrect forces especially in off-design conditions. A propeller model is useful to reduce computational cost, but it is desirable for the propeller model to be as accurate as a high-fidelity method.

The objective of this work is to develop a framework for a data-driven propeller and rudder model that can predict the forces with sufficient accuracy, such that the propeller model maintains the accuracy of a high-fidelity method but can be implemented in a maneuvering simulation at a significantly reduced cost. The data-driven model is trained with CFD simulations of the propeller operating in the behind condition with the rudder deflected to account for the flow interaction between the hull, propeller, and rudder. Different data-driven techniques are compared and evaluated. The rudder forces are also directly modeled by the data-driven model.

The data-driven propeller and rudder model is trained, evaluated, and implemented for two different model scale vessels. The first demonstration is on the KRISO Container ship which is a single screw model container ship with a semi-horned rudder. The second case study is on the twin-screw twin-rudder Office of Naval Research Tumblehome model scale surface combatant. The accuracy of the data-driven model is evaluated with different techniques and the accuracy is considered in terms of the underlying discretization and turbulence modeling uncertainty. Maneuvering simulations with the data-driven propeller and rudder model are performed for each vessel performing turning circle maneuvers in calm water and in waves.

This work demonstrates how a data-driven propeller and rudder model can be trained, validated, and implemented in a CFD maneuvering simulation. The computational cost is significantly reduced in both case studies performed and promising results are found.