【摘要】 为实现对复杂结构损伤位置的识别,采用分步识别策略,将原结构划分为多个广义子结构,先进行广义子结构的识别,再进行损伤单元的识别.为克服广义子结构划分的主观性,提出了基于聚类分析的广义子结构划分方式.以悬臂桁架结构为数值算例,对比了3种划分方式的网络训练收敛情况以及网络对带噪声检验样本的识别结果.研究结果表明,按基于聚类分析的广义子结构划分方式构造的网络易于收敛,且对带噪声检验样本的正确识别率高于另外2种划分方式1%~5%.更多还原

【Abstract】 In order to improve the identification effect for damage location of complex structures,a multi-step identification method was adopted.With this method,the damaged sub-structure can be found after a structure is decomposed into several sub-structures,then the damaged element in the sub-structure can be recognised.To solve the problem that the conventional modes to decompose a structure will lead to incorrect results,a decomposition mode of sub-structures based on a clustering analysis was put forward.Three decomposition modes were adopted by taking a suspension truss structure as an example.By comparing the training convergence and identification results of their neural networks with noise samples,the result shows that the decomposition mode of sub-structures based on a clustering analysis can make it easier to train neural networks,and compared with the other two decomposition modes,its correct identification rate increases by 1% to 5%.更多还原