基于STFT和DCNN的伤损识别方法

Damage identification method based on DCNN and STFT

为解决脉冲涡流检测(PECT)信号特征提取耗时且需要专家经验等问题,提出了一种基于短时傅里叶变换(STFT)和两层深度卷积神经网络(T-DCNN)的智能伤损识别方法。该法采用短时傅里叶变换将一维的PECT信号转换为二维时频信号;构建由两层卷积层、BN层、池化层和全连接层构成的DCNN,并将二维时频信号作为输入,进行端对端伤损识别。结果表明,该法与其他经典网络构架的伤损识别方法(VGG11、VGG16)相比,具有精度高、耗时短等优点,更符合工程领域应用需求。

In order to solve the problems of timp-consuming feature extraction of pulse turbine detection signal and requiring expert’s experience,an intelligent damage identification method based on pulse eddy current technology and two layers deep convolutional neural network(T-DCNN)is proposed.Firstly,the pulsed eddy current(PEC)testing technology is used to obtain one-dimensional time signal,and then the short-time Fourier transform(STFT)method is used to convert it into two-dimensional time-frequency signal,which is grayed as the input of T-DCNN.Secondly,it is compared with VGG11,VGG16 and the methods of ensemble empirical mode decomposition(EEMD)and Synchrosqueezed Wavelet Transforms(SSWT),the recognition performance of the self-built network is proved.Finally,the superiority of STFT method is proved.Experiments show that the proposed method has the advantages of short time,good accuracy and it fulfills the requirements of engineering field.