摘要
为实现结构的损伤定位,克服BP神经网络受病态样本影响大且抗噪声能力弱等缺点,采用SOFM(Self-Organizing Feature Map)网络,以结构振动组合损伤指标为网络输入,对一桁架结构进行了损伤位置识别研究。识别结果显示SOFM网络能够区分相似样本,由网络的拓扑图可以直观地评价网络训练和识别结果。在不考虑噪声情况下,网络可以正确识别损伤位置,在噪声水平不大于30%情况下,除个别单元外,网络对其他单元损伤的正确识别率均高于95%,显示出很好的抗噪声能力。
In order to identify the structure damage location and overcome the disadvantages of BP network,such as being affected greatly by ill samples and having poor antinoise ability,SOFM(Self-Organizing Feature Map)network is adopted to identify the damage location in a truss structure.The identification results show that SOFM network can distinguish similar samples and give intuitionistic judgement of the identification results through the topology map.When the noise is removed,the damage location can be identified correctly.When the noise level is under 30%,all the elements except individual element can be identified correctly in a probability over 95%,which shows that the SOFM network has a powerful antinoise ability.
出处
《振动与冲击》
EI
CSCD
北大核心
2007年第2期160-163,170,共5页
Journal of Vibration and Shock
关键词
自组织特征映射
神经网络
损伤位置识别
组合损伤指标
self-organizing feature map,neural network,damage location identification,combined damage index
