Research on Fatigue Crack Propagation Prediction for Marine Structures Based on Automated Machine Learning

In the field of offshore engineering, the prediction of the crack propagation behavior of metals is crucial for assessing the residual strength of structures. In this study, fatigue experiments were conducted for large-scale T-pipe joints of Q235 steel using the automatic machine learning (AutoML) technique to predict crack propagation. T-pipe specimens without initial cracks were designed for the study, and fatigue experiments were conducted at a load ratio of 0.067. Data such as strain and crack size were monitored by strain gauges and Alternating Current Potential Drop (ACPD) to construct a dataset for AutoML. Using the AutoML technique, the crack propagation rate and size were predicted, and the root mean square error (RMSE) was calculated. The prediction accuracy of the AutoML ensemble learning approach and the machine learning foundation model were evaluated. It was found that when the strain decreases by more than 3% compared to the initial value, crack initiation may occur in the vicinity of the monitoring point, at which point targeted measurements are required. In addition, the AutoML model utilizes ensemble learning techniques to show higher accuracy than a single machine learning model in the identification of crack initiation points and the prediction of crack propagation behavior. In the crack size prediction in this paper, the ensemble learning approach achieves an accuracy improvement of 5.65% over the traditional machine learning model. This result significantly enhances the reliability of crack prediction and provides a new technical approach for the next step of fatigue crack monitoring of large-scale T-tube joint structures in corrosive environments.