STRUCTURAL RELIABILITY ANALYSIS METHOD BASED ON ADAPTIVE ENSEMBLE LEARNING-SURROGATE MODEL

Structural reliability analysis requires calculation of the failure probability of a structure or system. However, the calculation of structural response is computation costly or difficult to carry out for the system with a relative low failure probability. The surrogate model can be used to represent the original performance function, which can ensure the accuracy and reduce the total number of runs of the original model significantly when combined with the adaptive experimental design. A structural reliability analysis method is proposed based on the adaptive ensemble learning-surrogate model, which integrates the versatile Kriging model and the recently developed PC-Kriging model. Based on the fact that these two surrogate models can provide prediction variance characteristics, a new ensemble learning function is proposed to identify the areas with high prediction errors and the failure boundaries. With experimental design stratgy, the new learning algorithm is used to add new samples in areas with large prediction errors and in areas close to the limit state, and iteratively update the ensemble learning-surrogate model. The proposed method is verified against three examples, which show that the method has low computational cost with high accuracy, compared with the single surrogate model-based structural reliability analysis methods (AK-MCS+U and AK-MCS+EFF methods).