Sub-Nyquist Data Acquisition in High-Speed Rail Condition Monitoring by Means of Compressive Sensing

Abstract

Monitoring data of axle box acceleration is critical because it can be used for detecting track defects and assessing the operation condition of a high-speed rail (HSR) system. However, there is a conflict between data volume and acquisition resolution: a high sampling rate leads to huge volumes of data while a low sampling rate cannot capture high-frequency vibration ingredients. Being an emerging theory, compressive sensing (CS) enables sub-Nyquist sampling when the target signal has a sparse representation in a certain domain. In this study, we take advantage of CS to deal with the above conflict. Using the axle box acceleration data acquired from a train running on a high-speed railway in China, we simulate the process of CS and investigate different ways of sparse representation. Likewise, we leverage the joint sparsity of successive signals to improve the reconstruction accuracy. It is found that the redundant dictionary outperforms the discrete concise transform (DCT) matrix and the discrete Fourier transform (DFT) matrix in sparse representation, leading to smaller reconstruction errors. More importantly, the feasibility of sub- Nyquist data acquisition in HSR condition monitoring is verified. In addition, joint reconstruction proves to be efficient for improving the reconstruction accuracy.