【摘要】 为提高结构损伤识别的精度,提出基于动、静态数据融合的位置指标和完全基于频率的位置指标,并采用学习速率自适应调整的新型BP神经网络学习算法,其特点是在网络迭代过程中根据网络学习误差来调整学习速率的取值,该方法有效地克服了传统BP网络学习过程中容易陷入局部极小和收敛速度慢、学习效率不高的缺点,进一步讨论了参数输入方式对网络识别效果的影响,分别采用两步诊断法和一步诊断法进行损伤识别.结果显示,两步诊断法对损伤位置和程度的识别正确率较高,而一步诊断法识别效果却不令人十分满意;减少位置指标和程度指标的输入个数对损伤识别结果有显著的影响.更多还原

【Abstract】 To increase the precision of structural damage identification,two kinds of new location indices based on the fusion of the static,dynamic data and natural frequency respectively were proposed.In view of the disadvantages of traditional BP neural networks,a self-adaptive algorithm which could adjust the value of learning rate according to output errors during the training process was introduced.The updated algorithm was proved to avoid oscillation phenomenon and improve learning rate effectively.The influence of input manner on identification effect was also discussed.Two-step-method and one-step-method using self-adaptive networks were adopted for damage detection respectively.It turned out that two-step-method made more accurate estimation for damage location and extent than one-step-method did.The number of location index and extent index influenced the identification effect greatly.更多还原