Research On High-Speed Rail Corrugation Detection Based On Deep Learning

With the continuous development of high-speed railway,the problem of rail corrugation detection has also been paid more and more attention.In order to processing the massive data of rail corrugation,it is necessary to study the effective feature extraction algorithm and the fast and accurate detection method.In this paper,the deep learning technology is used to study and solve the problem of rail corrugation.Based on the improved dynamic time warping algorithm to augmentation the measured data,the deep learning method is used to detect rail corrugation,and good results are achieved.The main work is as follows:1.A method of rail corrugation detection and data automatic label based on synchronous extraction short time Fourier transform is proposed.The time-frequency characteristics of the vibration acceleration data of the high-speed railway axle box are analyzed.The instantaneous frequency of the signal is extracted by the synchronous extraction of short-time Fourier trans-form to complete the work of wave wear detection and label marking.The experimental results show that the time-frequency domain energy obtained by synchronous extraction of short-time Fourier transform is more concentrated,and the instantaneous frequency of the extracted signal is more accurate with better noise immunity.2.A rail corrugation detection method based on data enhancement and support vector ma-chine is proposed.The data enhancement method based on dynamic time warping is used to enhance the unbalanced measured data.At the same time,an improved algorithm of bacterial foraging optimization is proposed to optimize the kernel function parameters and penalty coef-ficients of SVM.The experimental results show that the algorithm proposed in this paper is effective and reasonable,and the advantages of the algorithm in rail corrugation identification are verified by the measured data.3.A rail corrugation detection method based on depth learning is proposed.The data en-hancement algorithm based on dynamic time warping is improved by density clustering and sample weighting.The spatial pyramid pooling is introduced into the deep convolution residual network structure to detect rail corrugation.The experimental results show that the algorithm proposed is effective in rail corrugation detection,and achieves high classification accuracy in several data sets of UCI database.