An Artificial Neural Network Approach for Fatigue Analysis of Slender Marine Structures

Abstract Fatigue life calculation of slender marine structures, such as risers and mooring lines, usually requires a high computational cost. This cost comes from using finite element-based numerical models to predict the stress response of such structures under the action of many fatigue-inducing environmental loadings that they will face during their operational life. Lately, alternative methods to reduce the computational cost of these predictions have been proposed. One consists of a hybrid method that combines the FEM (Finite Element Method) with Artificial Neural Networks (ANN). Most of the available hybrid FEM-ANN models are based on shallow neural networks, and the models are trained individually for each fatigue load case. The main goal of this work is to investigate a fatigue analysis methodology based on hybrid FEM-ANN models where the ANN modeling is developed using modern deep learning techniques, such as Long Short-Term Memory (LSTM). Besides, instead of using a case-to-case approach, this work presents a generalized deep learning model, i.e., the trained hybrid FEM-ANN model can make predictions in the top of the riser for various environmental loadings without the need for new training. The model is tested for a flexible lazy-wave riser and a free-hanging flexible riser. Results of the hybrid FEM-ANN model are compared to those from the complete FEM-based numerical analyses to show the former model’s accuracy.