克里金
概率逻辑
替代模型
数学优化
可靠性(半导体)
计算机科学
信任域
最优化问题
内环
数学
机器学习
人工智能
功率(物理)
生物
物理
量子力学
计算机安全
农学
半径
控制器(灌溉)
作者
Yuanzhuo Ma,Xiangxiang Jin,Xi-Long Wu,Chang Xu,Hongshuang Li,Zhenzhou Zhao
标识
DOI:10.1016/j.ress.2023.109386
摘要
With the development of the surrogate-assisted Reliability-Based Design Optimization (RBDO) methods in recent decade, efficiency has been continuously improved by various state-of-the-art learning schemes. However, ensuring sufficient accuracy and feasibility at the optimum is still challenging, especially for real scenarios with complex probabilistic constraints and high fidelity. In order to achieve a good balance between efficiency and accuracy and feasibility at the optimum, both a single-loop and a double-loop adaptive Kriging-based RBDO methods are proposed. By directly approximating the probabilistic constraints using Kriging models in the single-loop method, the original RBDO problem is converted into a trivial deterministic optimization, which can be solved by any type of optimization method. The real values of failure probabilities are estimated by Generalized Subset Simulation (GSS) at the initial construction or during each update of the Kriging models. To further raise efficiency in the double-loop adaptive Kriging-based method, GSS is substituted by an inner-loop Kriging-based failure probability estimation, which includes each local enrichment at the limit states corresponding to the active probabilistic constraints based on an initial global enrichment within the augmented reliability space. Once a certain Kriging model meets the requirement during the refining of all, only the others are kept updating until the accuracy of all are accepted. Four examples are used to demonstrate the performance of the proposed two methods.
科研通智能强力驱动
Strongly Powered by AbleSci AI