An adaptive parallel non-probabilistic uncertainty analysis method for engineering structures using a kriging model

To enhance computational efficiency in structural analysis, an adaptive parallel non-probabilistic uncertainty analysis method is proposed. The approach employs the kriging surrogate model to replace computationally intensive finite element analysis, significantly reducing computational costs. By revising and controlling the expected improvement (EI) criterion, an innovative sample update mechanism is developed, enabling adaptive refinement of the kriging model, with particular emphasis on improving boundary region accuracy. Furthermore, a parallel computing framework for non-probabilistic uncertainty analysis is established based on pseudo-EI criteria, allowing simultaneous determination of multiple newly generated samples within each iteration cycle. Four numerical examples are presented to illustrate the accuracy and effectiveness of the proposed method, and the 150 kN biaxial tensile testing machine is used to demonstrate its applicability to engineering structural design. The results indicate that the method achieves superior computational efficiency while maintaining high accuracy in uncertainty analysis.