基于深度学习的无砟轨道砂浆层脱空病害识别

Identification of Voids in Mortar Layer of Ballastless Track Based on Deep Learning

为实现无砟轨道砂浆层脱空病害的快速准确检测,文章提出了一种基于冲击回波法、小波变换和深度学习的脱空病害识别方法:冲击回波法收集无砟轨道板的激振信号,利用小波变换将一维回波信号转化为二维频域特征图像,训练深度学习模型从图像中提取病害特征,从而实现病害的准确识别。为验证该方法的准确性,设计了预设脱空病害的全尺寸无砟轨道试验模型并通过人工激振获得冲击回波信号,经小波变换处理得到数据集,以此来训练与测试所设计的深度学习分类模型的病害识别准确性。结果表明:训练得到的深度学习分类模型的脱空病害识别准确率可达95.15%,满足实际工程应用的要求。

In order to realize rapid and accurate detection of voids in mortar layer of ballastless track,a void detection method was proposed on the basis of the impact echo method,wavelet transform and deep learning in this paper:the impact echo method was used to collect the excitation signal of ballastless track slab,the wavelet transform to transform the one-dimensional echo signal into characteristic images in two-dimensional frequency domain and training depth learning model to extract the defect features from the image,so as to realize the accurate recognition of the defect.In order to verify the accuracy of this method,a full-size ballastless track test model with preset void defect was designed and the impact echo signal was obtained through artificial excitation.Then,the data set was processed by the wavelet transform to train and test the defect detection accuracy of the designed deep learning classification model.The results indicated that the void detection accuracy of the trained deep learning classification model could reach 95.15%,which could meet the needs of practical engineering application.