An investigation on parametric rolling prediction using neural networks

Parametric rolling is a recently identified phenomenon that affects vessels and is characterized by large roll angles that lead to severe rolling of the vessel and even capsizing. Parametric rolling is caused by a periodic variation of the self-righting characteristics of the ship. Due to its short inception time it is necessary to have an on-board warning system that monitors the current state and provides an advance warning of possible onset of parametric rolling. Vessels mainly encounter parametric rolling in monochromatic head seas. In this paper an application of artificial neural network technology to head seas parametric rolling prediction will be discussed. Neural network is an algorithm that imitates the mechanism of neurons in the brain. It can learn a function given by input-output pairs and return approximate outputs for inputs that were not given. Such algorithms are already used in naval architecture for approximation, control and classification. Here neural networks will be used in a recursive manner with discrete time-series to predict three to five future natural rolling periods to allow time to react and to counteract the phenomenon. The model is then improved in a remarkable way to include pitch and frequency data. The authors developed a systematic methodology and validation method which include the use of multiple initial conditions to avoid biased data. Experimental data were obtained in monochromatic head seas with a hull of a modern container vessel and a nonlinear numerical model using six degrees of freedom with terms defined up to third order derivatives Rodríguez (2010). This numerical model was shown to provide a good prediction of parametric rolling.