Compressive sensing for high-speed rail condition monitoring using redundant dictionary and joint reconstruction
Compressive sensing for high-speed rail condition monitoring using redundant dictionary and joint reconstruction
复制标题
DOI：
--
复制
发表时间：
2018
期刊：
影响因子：
--
通讯作者：
Si-Xin Chen;Yi-Qing Ni
中科院分区：
文献类型：
--
作者：
Si-Xin Chen;Yi-Qing Ni
在高速铁路（HSR）状态监测中，由于奈奎斯特定理，缺陷检测的分辨率和记录的数据量之间的冲突通常是一个问题。作为一种新兴的技术，压缩感知（CS）创造了一个机会，亚奈奎斯特采样时，目标信号具有稀疏表示在一个已知的域。然而，许多研究表明，稀疏性的缺乏限制了CS的适用性。此外，当多个压缩的测量向量是可用的，传统的CS算法恢复目标信号一次一个独立的不利用它们的稀疏表示之间的相关性。本研究将压缩感知应用于高铁状态监测，并采用两种方法来提高恢复精度。具体而言，CS的过程中，使用的轴箱加速度数据采集的高速列车运行在中国的一段铁路。在恢复结果的调查后，进行了相同的实验，除了离散余弦变换（DCT）矩阵被替换为冗余字典。另一系列实验假设信号在DCT域具有联合稀疏性，并同时重建它们。研究结果表明，在充分稀疏的条件下，对高铁状态监测数据进行欠采样和重构，可以得到误差较小的数据。即使压缩测量值相同，两种方法都被证明是有效的，以提高恢复性能，其中联合重建具有更好的性能。
In high-speed rail (HSR) condition monitoring, the conflict between the resolution of defect detection and the amount of recorded data is usually an issue due to the Nyquist theorem. As an emerging technique, compressive sensing (CS) creates the opportunity of sub-Nyquist sampling when target signals have a sparse representation in a known domain. However, many studies have shown that the lack of sparsity limits the applicability of CS. In addition, when multiple compressed measurement vectors are available, conventional CS algorithms recover target signals one at a time independently without exploiting the correlation among their sparse representations. This study applies CS to HSR condition monitoring and employs two methods to improve the recovery accuracy. Specifically, the process of CS is simulated using the axle box acceleration data acquired from a high-speed train ran on one section of railway in China. After the investigation of recovery results, the same experiments are conducted, except that the discrete cosine transform (DCT) matrix is replaced by a redundant dictionary. Another series of experiments assume that the signals have a joint sparsity in the DCT domain and reconstruct them simultaneously. The results show that the HSR condition monitoring data can be obtained through sub-Nyquist sampling and reconstructed with small errors when they are sufficiently sparse. Even if the compressed measurements are the same, both methods are proved effective to improve the recovery performance, in which joint reconstruction has better performance than the other.