【摘要】 潜艇耐压液舱结构复杂 ,往往需进行有限元分析以获得结构在极限载荷作用下的应力。此时 ,由于失效函数缺乏显式表达 ,故应用一般的可靠性计算方法将遇到困难。作者采用人工神经网络代替传统的多项式函数拟合失效面 ,并结合方向抽样技术 ,提出了基于有限元的结构系统可靠性计算的新算法。由于该方法引入了结构有限元分析 ,且无需进行失效模式间相关性的近似分析 ,因而计算精度好 ,适用范围广泛。潜艇耐压液舱结构系统可靠性分析的算例表明 ,该法有效地解决了复杂工程结构的系统可靠度计算问题。更多还原

【Abstract】 In this paper, an improved Response Surface Method is proposed to calculate structural system reliability. The Finite Element Analysis (FEA) is adopted by using MARC to attain stress results of the structures, and then the artificial neural network technique is used to replace the traditional polynomial functions to simulate failure surface. Finally they are combined with directional sampling techniques to get the failure probability. This newly developed algorithm, called RSM ANN (Response Surface Method based on Artificial Neural Network), does not need to perform approximate analysis of relativity among failure modes. Furthermore, it can be widely used in different situations and also has a good accuracy. In the final part of this paper, RSM ANN is applied to a typical submarine structure of pressure hull that has four primary failure modes to verify its feasibility in calculating structural system reliability. Good results are obtained and show that this method is particularly suitable to complicated structures. Also it achieves a balance between efficiency and accuracy.更多还原