Vortex-induced vibration prediction of an inclined flexible cylinder based on machine learning methods

Vortex-induced vibration (VIV) is a main cause of fatigue damage to slender flexible structures in offshore engineering, e.g., marine risers, mooring lines, submarine cables and pipelines. In this paper, attempts have been made to predict the VIV response of an inclined flexible cylinder commonly encountered in practice using different machine learning methods. Three different machine learning algorithms (i.e., BP neural network, support vector machine and Gaussian process regression) are employed to establish the prediction models. The data in our previous experimental tests (Xu et al., 2018) are adopted to establish the data set of the present research. The numbers of data samples for constructing the displacement and frequency prediction models in each direction are 10,500 and 500, respectively. The inclination angle, incoming flow velocity, axial tension and the position along the cylinder span are used as the input variables and the output variables are the root mean square values of the in-line and cross-flow displacements and the dominant oscillation frequencies. Compared to the previous studies, the effect of the axial tension and the large inclination cases are considered to establish more comprehensive prediction models. The results show that all the three methods can predict the displacement and frequency responses in the cross-flow and in-line directions with acceptable accuracy. After a detailed comparative analysis of the three machine learning methods, it is found that the BP neural network models tend to result in a certain degree of overfitting and some response characteristics predicted by the GPR models are not consistent with the conclusions from our previous experimental studies. Within the parameter space of the present research, the support vector machine is considered to be the most suitable method for predicting the VIV of the inclined flexible cylinder. •VIV of an inclined flexible cylinder is predicted using machine learning methods.•Effect of the axial tension and the cases with large inclination angles are considered.•BP neural network, SVM and GPR are employed to establish the prediction models.•SVM is found to be the most suitable method for predicting the VIV in the present study.