Efficient adaptive Kriging approximation with active constraint identification and dynamic region sampling for reliability-based design optimization

The nested double-loop framework for reliability-based design optimization (RBDO) leads to substantial computational demands, which limit its practical engineering applications. While surrogate models enhance the computational efficiency of RBDO by substituting costly performance functions, striking an optimal balance between their accuracy and efficiency remains a significant challenge. This paper presents an efficient Kriging model updating strategy that integrates active constraint identification and dynamic region sampling (ACI-DRS) to address RBDO problems. An active constraint identification criterion is proposed to determine the validity of the Kriging model under each probabilistic constraint during iterations. During the update of the active Kriging model, the sampling center is dynamically switched between the current design point and the corresponding most probable point (MPP), and an improved expected feasibility function (IEFF) is proposed to select new sample points for model updating. The radius of this sampling region is determined by both the baseline reliability index and the degree of nonlinearity of the current Kriging model. The computational performance of the ACI-DRS method is evaluated against existing approaches. Results indicate that ACI-DRS achieves a substantial reduction in computational cost without compromising the requisite accuracy.