Prediction of Vessel RAOs: Applications of Deep Learning to Assist in Design

Abstract

In an age of high-paced design, a need arises for engineers to quickly estimate the feasibility of their ideas without spending weeks developing a computer model. At the same time, the use of machine learning models, or neural networks, in the maritime industry has grown substantially over the past years. By further extending the use of these predictive models in the design phase, marine engineers and naval architects can expedite their work. This paper focuses on the creation of a neural network that can estimate the Response Amplitude Operators (RAOs) of a vessel given its characteristic properties such as length, beam, and draft. A dataset was collected through a parametric design analysis of box barges using ANSYS AQWA, and the RAO was simulated for all 6 degrees of freedom. A critically damped spring equation was generated for each degree. A Keras Neural Network Model was trained on the three parameters and the wave heading angle, with the hidden layers and neuron count being adjusted to optimize the loss and maximize R-squared. To validate the results, a series of box barges with dimensions that were not a part of the training dataset were simulated in ANSYS, while the virtual model with identical characteristics was simulated with the Neural Network. The resulting RAOs were compared to validify the accuracy of the Neural Network. With this predictive model, engineers can quickly determine a hullform’s RAOs, and compare the response with the common sea states along the intended route. Additionally, the model can assist in design iteration. As the hull shape gradually changes, the new RAOs can be estimated to ensure that the design is progressing in an appropriate direction.