Abstract
This paper describes the development of data-driven models for the prediction of mooring line tensions by separating the low- and wave-frequency components of the tensions, such that the former is approximated as quasi-static tensions while the latter is predicted using ANN models.
A bilinear model is used to interpolate the low-frequency quasi-static tensions between known values in a look-up table while a feed forward neural network model that utilizes the fairlead motions as input is used to predict the tension dynamics at the fairlead. The ANN models are trained using the results from numerical simulations, such as those generated during the engineering design and construction stages of the floating structure.
The predicted line tensions are compared with those obtained using coupled numerical simulations, including test cases for different wave realizations that are not included in the training dataset. Models trained for single and multiple directions of the environment are also assessed for prediction accuracy.
Initial comparisons show that good predictions of line tensions can be obtained for certain environments using the proposed approach, thus demonstrating the potential for use in applications where real-time predictions are required to enhance the safety and / or reliability of mooring systems.
