Deep Learning applicato all'analisi strutturale per cavi elettrici sottomarini di turbine eoliche offshore flottanti

Structural dynamic and non-linear analysis is complicated and computationally expensive to produce. The computational cost of generating simulations is even greater when we consider the multiple environmental factors that occur when the structure is positioned in a harsh environment. Having at our disposal a not so expensive and reliable numerical model for generating time series of structural responses would save money and time. The work of this thesis focuses attention on performing condition monitoring, by predicting the time series of tensions in some located regions, for power cables connected to a Floating Offshore Wind Turbine (FOWT). In an attempt to accomplish this task, a hybrid methodology is applied, which consists in using short time series of displacements and tensions, generated with dynamic Finite Element Analysis (FEA), for training Deep Learning (DL) based models. The latter would then be used for prediction and analysis instead of the more computationally expensive FEA. The aim of this thesis is therefore to exploit the versatility and learning power that characterises these deep learning models to obtain results comparable to those of FEA, which is more resource intensive. Experimental results showed how it was possible for these models to capture significant temporal and physical correlations between points located at considerable different distances from each other with most of the time minimal loss of accuracy. The most significant result, which also concern the most relevant application of all, can be attributed to the experiments involving several sea states, in which case all models show great predictive behaviour in most cases, considering the wide variety of different input and output combinations used for the evaluations.