Model Updating Method for Jacket Platform Considering Different Component Degradation Based on Deep Learning

Abstract

Marine platforms are located in complex environments, and safety deteriorates throughout the day. It is necessary to analyze the jacket platform structure by the finite element method. Problems such as platform structure variation and fatigue corrosion lead to model deviation. In this paper, a finite element model correction method based on deep learning is proposed with a jacket platform as the engineering background. First, different platform design parameters are selected, and the corresponding fundamental frequencies are obtained by finite element simulation. Second, the input features are extended as necessary to increase the damage-sensitive information, with the nonlinear differences between the two reduced by an improved ResNet50 network. Finally, the correction values of the finite element model are obtained by combining the measured data with the inherent structural frequencies obtained by covariance-driven stochastic subspace identification (Cov-SSI). The results show that the error after correction is less than 4%, which can reflect the real marine platform state well.