A Surrogate Model to Predict Stress Intensity Factor of Tubular Joint Based on Bayesian Optimization Gaussian Process Regression

Abstract

Abstract The tubular joints located in the splash zone of the jacket platform are prone to damage or even crack due to long-term loads such as wind, wave, and current. If the crack development is not monitored and tracked, serious consequences will be caused. Aiming at the problem of long calculation time and low efficiency of stress intensity factor (SIF) in fracture mechanics, a method based on Gaussian process regression is proposed to construct the SIF surrogate model of tubular joints. By conducting mechanical simulation analysis on the tubular joint of the jacket platform under extreme storm load conditions, the dangerous position of the tubular joint is determined, cracks are introduced, and crack propagation simulation is carried out to obtain training data for the surrogate model. The Gaussian process regression surrogate model is established based on the composite kernel function, and the Bayesian optimization is introduced to optimize the hyper-parameters of the kernel function to determine the optimal surrogate model and verify the accuracy. The results show that the maximum mean relative error (MRE) of the SIF obtained by the proposed method is 4.94%, and the average value of MRE is only 0.41%. At the same time, the calculation time is reduced from about 4 h to 2.9 s, providing a method reference for real-time prediction of crack growth of jacket platform under the background of digital twin.