Deep learning for fatigue damage prediction in offshore structures

This thesis investigates a predictive maintenance approach for estimating the fatigue damage in crucial components of offshore floating platforms. The components subject of the study are the floating cables or pipe connected to the top side and to the seabed. The direct monitoring of these slender structures with sensors in critical sections is impractical and not cost efficient. Therefore, for the prediction of the fatigue damage, a so-called virtual sensor is implemented to forecast the mechanical loads in critical points, due to the motions of the floating part. Such virtual sensor consists of a recurrent neural network trained on synthetic data from FEM simulations. The results suggest the potentials and limits of the approach in estimating the loads and transferring this information to the fatigue damage estimation.