Reliability-based Design Optimization By Adaptive Kriging Surrogate

As engineering structures have been becoming more and more complicated,many uncertain factors should be considered in the engineering constructing process in order to meeting the desired requirements.Reliability-based design optimization(RBDO)that can quantify these uncertainties fairly has been widely used.However,it still needs unsustaintable calculation cost,especially for those highdimensional and small failure probability problems.With regard to reducing this kind of cost,using the adaptive Kriging surrogate to substitute the real complicated simulation or tests,and carry out reliability analysis with Monte Carlo Method has been gained a lot of attentions in the RBDO fields.This work mainly focuses on the construction strategies of adaptive Kriging surrogate and the application in the RBDO field.The main contents are as follows:(1)Comparison of initial sampling methods.When build the surrogate without any prescribed information,the expected initial sampling should be scattered in the whole design space uniformly.In this paper,we judge the uniformty of six initial sampling methods by CL2 criterion,and evaluate the performance in single-Kriging surrogate reliability analysis.(2)Comparison of learning function and stopping criteria.During the constructing process,take full advantage of the prior points information,develop a new method that can construct the surrogate efficiently,correctly and robustly.In this paper,we combine three learning functions and two stopping criteria randomly,and choose four examples from linear,dimensional and applied situation.Use these examples to compare the performance in the aspects of correctness,efficiency and robustness,then form a new strategy adding the most important points.(3)Apply the new strategy in the RBDO process.In the process of RBDO,using double-loop surrogate to substitute the probabilistic constraint,utilitze the inner surrogate to substitute the performance function,and outer surrogate is used to substitute failure probability function.Then take advantage of sequential quadratic programming to perform single-loop RBDO.