基于有限元-神经网络-Monte-Carlo的结构可靠度计算方法

Structure Reliability Calculation Method Based on Finite Element-Artificial Neural Network-Monte-Carlo

针对以往结构可靠度计算方法的不足,根据结构可靠度理论,引入神经网络理论和Monte-Carlo理论,结合结构有限元分析方法,提出基于有限元-神经网络-Monte-Carlo的结构可靠度计算方法;首先,建立结构有限元法分析模型,进行结构分析得到相应的结构响应量;然后,用得到的作用效应和结构响应量作为神经网络的训练样本和检验样本对神经网络进行训练,得到高度非线性映射关系的结构作用效应-结构响应模型;利用神经网络随机产生足够多的结构响应值;最后,用Monte-Carlo法计算结构的可靠度。该方法充分发挥了各方法的优点,相互弥补了不足,大大提高了计算效率,为结构可靠度计算提供了新的思路。

According to reliability theory, based on Finite Element, Artificial Neural Network and Monte-Carlo （F-A-M）, a new structure calculation method is developed. First,finite element analysis model is built up,inputting values of the structural parameters and loading at random, and then response of the structure is recorded.The number of the record is limited, but it is enough to train the Artificial Neural Network （ANN）. After ANN being trained with the training samples, a highly nonlinear mapping relationship between the efficacy and response of the structure is gained. Making use of the generalization capability of ANN,sufficient structural response value is developed at random. Then the reliability of the structure is calculated by the Monte-Carlo method. This approach which not only makes full use of the merit of ANN and Monte-Carlo, also makes the calculation more effective, provides a fresh method for the reliability of structure.