Application of adaptive surrogate models in time-variant fatigue reliability assessment of welded joints with surface cracks

The time-variant fatigue reliability of welded joints subjected to stochastic loading is assessed based on the PHI2 method, which takes the time-variant reliability problem as a two-component parallel system reliability problem. The complex fatigue process that takes place at the weld toe is modeled as a semi-elliptical surface crack growth process. Surrogate models, representing surface crack sizes at a time instant, are applied to solve the reliability problem involving time-consuming fatigue crack growth analyses. Polynomial regression models and Kriging interpolation models are both employed. Their corresponding adaptive procedures are also adopted to improve the reliability results. Some adjustments of the active refinement algorithm based on the Kriging model are proposed to make it more suitable for the present study. The time-variant fatigue reliability assessment of a T-plate welded joint is carried out. Without using the adaptive procedures, the Kriging models can better represent the crack sizes compared with the polynomial models. The outcrossing rate formulated by the PHI2 method is sensitive to the reliability index of the first component of the two-component parallel system. The effectiveness of the adaptive procedures, which improve the reliability results, is demonstrated.