基于神经网络重要抽样法的结构可靠度评估

Important Sampling for Structural Reliability Based on Artificial Neural Network

文章提出基于神经网络重要抽样法的结构可靠度评估方法,该方法通过训练神经网络,建立随机变量与结构响应之间的非线性映射关系,模拟结构极限状态方程,采用优化算法计算设计点.在重要抽样法中利用训练好的神经网络代替有限元求解器进行数值计算.结果表明,所训练的神经网络能够有效地模拟真实的结构极限状态方程,结构失效概率计算结果具有很高的精度,调用有限元分析的次数显著减少,有效地提高了计算效率.

Calculations based on finite element analysis are time-consuming, thus decreasing the efficiency. An important sampling approach for structural reliability evaluation based on artificial neural network （ANN） is proposed in the paper, in which ANN is trained and can map the structural responses and random variables, and limit state functions of structures is simulated with an optimization algorithm applied to calculate the design point. The results show that the trained ANN can simulate the limit state functions of structures, and the proposed approach in reliability evaluation is accurate. Besides, procedure calls using finite element solver to calculate the value of the limit state functions of structures are significantly reduced, which enhances the efficiency in structural reliability evaluation.