An Approach for Obtaining Dynamic Responses of Slender Marine Structures Assisted by Generalized Artificial Neural Networks

Fatigue life evaluation of slender marine structures, such as flexible lines, are usually performed through stochastic dynamic analyses in the time domain. Traditionally, these analyses are performed in order to obtain the dynamic responses’ time series of the risers (axial force and bending moments). These analyses are usually conducted through Finite Element Method (FEM) which presents high degree of confidence in its results, but also can generate high computational costs. So, many works seek alternative methods to obtain these dynamic responses and some of them have introduced the use of artificial neural networks (ANN) in this process to obtain computational gain. These works presented interesting results using a hybrid FEM-ANN models, but most of them treat each riser and environmental condition separately, that is, the models are trained individually for each response evaluation. The present work is an extension of the previous one [1] and its main goal is to develop a network that from just one single training it is able to generate accurate predictions of the axial tensions and bending moments time series of flexible risers connected to different platforms and also for different wave load conditions. The model is tested for a free-hanging flexible riser. The network’s results are compared to those from the full FEM-based numerical simulations to show the accuracy and robustness of the presented model.