结构系统可靠性优化设计的神经网络方法

Reliability-based structural optimization using neural network

针对具有非正态随机参数的可靠性(优化)设计,提出了随机摄动-Edgeworth级数方法,采用该方法将可靠性概率约束转化为等价的确定型约束,可以迅速准确地获得优化设计信息。针对具有多失效模式的结构系统可靠性优化设计,提出了随机模拟-神经网络方法(MCS-NN),将随机模拟方法与神经网络技术有机结合,为结构系统可靠性优化设计提供了一种新方法。

In this study two methods are examined. In the first one, the probabilistic perturbation method and the Edgeworth series are employed to give the theoretical formula for reliability-based optimization with non-normal random parameters. Therefore, the probabilistic constraints are transformed into deterministic constraints, and then the reliability-based optimal design parameters could be obtained accurately and quickly. In the second one, neural network (NN) and Monte Carlo simulation (MCS) are applied in reliability-based structural optimization with more failure modes. The neural network is used to simulate the relation expression between the design parameters and the structural reliability based on Monte Carlo simulation. Therefore, the structural probabilistic constraints are transformed into a single equivalent deterministic constraint approximately by NN, and then the process of reliability-based optimization can be implemented conveniently.