Fatigue Analysis of Flexible Riser Using Hybrid Machine Learning

Abstract This paper presents a novel hybrid machine learning methodology for evaluating the fatigue life of a flexible riser on long term environmental loading. Flexible risers are multi-layered structures made of several layers including pressure armour and tensile armour layers, out of which tensile armour layer is more prone to fatigue damage of flexible riser on long term environmental loading. The evaluation of stresses on the tensile armour layer is a complicated method due to friction acting between the internal layers. The friction coefficients between layers are calculated using analytical method proposed by Saevik [1] and used as a training data set for the Artificial Neural Network(ANN) model. Hybrid Machine Learning (Back Propagation Neural Network + Generic Algorithm) is used in the proposed fatigue damage estimation method. Global fatigue analysis of a flexible riser has been performed with environmental loading. The effective tension and bending moments are evaluated in the global analysis and used as inputs to the proposed Hybrid Machine Learning model to calculate the stresses in the tensile armour layer. The fatigue life of the riser based on the predicted stresses is computed using the rainflow counting method. A case study is described for a 12" diameter flexible riser with from turret moored FPSO with typical north-sea environmental loading is applied for demonstrating the methodology. The proposed hybrid method is a nonlinear parametric model where the parameters are optimised based on Genetic Algorithm based ANN model and the to predict stress on tensile armour from two primary input; bending moment and effective tension. The hybrid model is quite useful to evaluate the fatigue damage prediction on tensile armour layer. Comparing the fatigue life calculated from the proposed hybrid method and the analytical approach focuses on a nonlinear, friction coefficient dependent model [1], it is clearly visible that there is good agreement between both results. The proposed method is a Machine Learning based predictive model and can be used as an alternative tool for predicting fatigue damage in the flexible riser