基于相关函数和卷积神经网络的结构损伤识别

Structural damage identification based on correlation function and CNN

为了改善基于振动信号的结构损伤识别效果,提出一种基于相关函数和卷积神经网络相结合的结构损伤识别方法。以一铁路钢梁桥结构为例,首先通过对结构的振动响应进行自相关运算来提高振动信号的信噪比,再使用自相关样本作为卷积神经网络(convolutional neural network,CNN)的输入可以显著提高其识别精度,且当振动信号中的噪声水平越高时,自相关样本作为CNN输入的识别精度的提升效果越明显,且自相关运算具有比快速傅里叶变换(fast Fourier transform,FFT)更强的抗噪性。使用互相关函数对结构上布置的多传感器的数据进行融合,再将融合后的信号作为CNN的输入,互相关在对2个传感器数据特征有效融合的前提下可以成倍地削减数据集的维度,减少网络运算的参数量,从而减少用时提高训练效率,且互相关样本作为网络输入同样具有较高的识别精度和较强的抗噪性。

In order to improve the structural damage identification effect based on vibration signal,a structural damage identification method based on the combination of correlation function and convolutional neural network is proposed.Taking a railway steel girder bridge structure as an example,firstly,the signal-to-noise ratio of the vibration signal is improved by performing autocorrelation calculation on the vibration response of the structure,then the autocorrelation sample is used as the input of convolutional neural network,which can significantly improve the recognition accuracy.When the noise level in the vibration signal is higher,the improvement effect of the recognition accuracy of the autocorrelation sample as the convolutional neural network input is more obvious,and the autocorrelation operation has stronger noise immunity than that of the fast Fourier transform.The cross-correlation function is used to fuse the data of the multi-sensors arranged on the structure,then the fused signal is used as the input of the convolutional neural network.Under the premise of effective fusion of the data characteristics of the two sensors,the cross-correlation can double the dimension of the data set and reduce the number of parameters of the network operation,thereby reducing the time and improving the training efficiency,and the cross-correlation sample as the network input also has high recognition accuracy and strong noise immunity.