基于神经网络方法的复杂超静定结构失效概率分析  

Failure Probability Analysis of the Complex Hyper-static Structure Based on Neural Network Method

作　　者：王爱红[1] 徐格宁[1] 杨萍[2] 高有山[1] 

机构地区：[1]太原科技大学机电学院,太原030024 [2]兰州理工大学机电学院

出　　处：《中国安全科学学报》2007年第6期151-156,共6页China Safety Science Journal

基　　金：山西省自然科学青年基金资助(20051030);山西省自然科学基金资助(20041074);山西省教育厅重点学科资助项目(20045027-20045028);国家"十一五"科技支撑计划课题(2006BAK02B04-0102)

摘　　要：将整体结构按拓扑关系划分为若干模块,根据力的传递原理对模块结构进行失效概率计算,获得各模块结构的失效概率信息;运用有限元模拟分析获得整体结构的失效概率信息。再将模块结构的失效概率作为输入,整体结构的失效概率作为输出,构造样本集。以BP神经网络进行失效概率分析,既可提高计算速度和精度,也可利用其泛化能力对相同拓扑结构的超静定结构进行失效概率计算。算例中对包含5个模块结构的整体结构单元进行基于神经网络的失效概率分析,以网络外推能力计算了包含7模块结构的整体结构单元的失效概率,获得较好的计算精度,从而验证了该方法的有效性。Integral structure is divided into some module structure in the light of topological connection, then, module structure failure probability is computed and obtained according to force transfer principles. Information of integral structure failure probability is simulated by FEM （ finite element modeling）. Module structure failure probability is considered as input data, information of integral structure failure probability is considered as output data, and thus, specimen is obtained on input data and output data. Neural network structure with interpolation and extrapolation is gained by training specimen. Not only can the Computing speed and analyzing veracity be greatly improved, but also the failure probability of hyper-static structure with same topological structure can be computed with neural network generalization. In an example, failure probability of integral structure with five module structures is illustrated by neural network, and failure probability of integral structure with seven module structures is analyzed by neural network extrapolation, the result obtained has a good precision. This conclusion shows that this method is feasible and valid.

关 键 词：超静定结构 模块结构 整体结构 神经网络 失效概率 

分 类 号：X966[环境科学与工程—安全科学] TU311.3[建筑科学—结构工程]