Station-keeping is one of the important factors in the design of offshore platforms. Some offshore platforms, such as Spar, Semi-submersible and FPSO, use mooring lines as a mean for station-keeping. Tensions in the mooring lines are one of the key factors in station-keeping. The design of an offshore platform and its mooring lines is based on computed motions of the platform and associated mooring line tensions from numerical simulations using a software code on the basis of metocean criteria.
This paper presents an Artificial Neural Network (ANN) model for the prediction of mooring line tensions based on the motions of the platform. This ANN model is trained with time histories of vessel motions and corresponding mooring line tensions for a range of sea states from the results of numerical simulations. After the model is trained, it can reproduce with great fidelity and very fast the mooring line tensions. In addition, it can generate accurate mooring line tensions for sea states that were not included in the training, and this demonstrates that the model has captured the knowledge for the underlying physics between vessel motions and mooring line tensions. The paper presents an example of the training and the validation of the model for a semi-submersible offshore platform for a range of sea states.
The training of the ANN model employed a back-propagation learning algorithm. In this algorithm the computed output error is back-propagated through the neural network to modify the connection weights between neurons. The training started with a small number of hidden neurons, and the model grew adaptively by adding hidden neurons until either the target output convergence is achieved or a maximum number of additional hidden neurons is reached. The ANN model discovers nonlinear relationships between the input and output variables during training.
The paper presents comparison of time series of mooring line tensions for sea states that were and were not included in the training between those from the numerical simulations and those computed by the trained ANN model. Fatigue assessment is also used to quantitatively measure the accuracy of the ANN model prediction of the time series of mooring line tensions. The paper presents the results of fatigue assessment using various stages of the ANN models with different number of hidden neurons. This shows that the additional hidden neurons improve the prediction of the ANN model of the mooring line tensions for sea states that were and were not included in the training.
This approach of prediction of mooring line tensions based on vessel motions using ANN model paves the way to the development of an ANN-based monitoring system. Also, this ANN study demonstrates a great potential for the use of a more general and comprehensive ANN model to help monitor the dynamic behavior of floating systems and forecast problems before they occur by detecting deviations in historic patterns.