Convolutional Neural Networks Applied to Flexible Pipes for Fatigue Calculations

The industry standard approach for the design of flexible pipes makes use of non-linear finite element analysis (FEA) in time domain to simulate the physical responses of the structure in different environmental conditions. Wave dynamics can be represented either by an irregular wave (IW) or an equivalent regular wave (RW) approach, which simplifies the analysis. Irregular wave modeling approximates better the structural responses, due to the stochastic nature of the environmental loading, having the drawback of being more computationally expensive. The computer processing time of IW-FEA often becomes intractable due to the large number of scenarios that need to be simulated, for different values of Hs (significant height), Tp (peak period) and different wave directions. Reducing the time needed to simulate each scenario would reduce significantly the total processing time. In order to achieve this, alternative hybrid methods have been proposed in the literature, combining FEA with machine learning models. This paper proposes the use of nonlinear autoregressive exogeneous convolutional neural networks (NARX-CNN) to predict tension and curvature responses along the length of a flexible riser. Experimental results show that the proposed model can generate more accurate responses than previous models. This work also extends the region analyzed by forecasting responses beyond the bending stiffener level, going down to the end-fitting and touch down zone locations. It is the first time that such regions, prone to fatigue issues, are evaluated with these types of algorithms for flexible pipes, as per authors’ knowledge.