Deep Learning-Based Monitoring of Tension and Fatigue in a Hybrid Mooring System of a Floating Offshore Wind Turbine

Mooring integrity monitoring of floating offshore wind turbines (FOWTs) is constrained by the cost and limited reliability of subsea tension sensors. This study develops a deep learning framework that predicts the dynamic tension of a hybrid polyester–chain mooring system for a 15 MW semi-submersible FOWT from the platform six-degree-of-freedom motion alone. A database of fully coupled OrcaFlex simulations was generated for sea states off Hsinchu, Taiwan, and split at the simulation level with cluster stratification. Long Short-Term Memory (LSTM), Bidirectional LSTM (BiLSTM), and Convolutional Neural Network–LSTM (CNN-LSTM) architectures were compared, and the predicted tension was processed by rainflow counting and Miner’s rule to evaluate fatigue damage. On the test set, all three models exceeded an R2 of 0.999, with BiLSTM the most accurate at an RMSE of 11.15 kN, despite the nonlinear axial stiffness of the polyester segment. The predictions were insensitive to the random wave realization, and the accuracy was maintained on sea states and loading directions absent from training. Applied to the recorded conditions of Typhoon KRATHON, in which wind and wave are misaligned by up to 54°, the framework reproduced the one-day cumulative fatigue damage of the simulated tension to within about 4% for LSTM and BiLSTM. With correlated measurement noise on the motion input, the tension error approximately doubles. BiLSTM is recommended for offline monitoring and the unidirectional LSTM for real-time prediction.