Improved STFT Analysis Using a Kernel Method and Order Spectra

초록

Short-time Fourier transforms (STFT) are useful for analyzing signals with harmonics that vary with time. If the variation of the harmonics with time is smooth, such as in kinematic vibrations in vehicles, then it is possible to improve the STFT using correlation analysis and order spectra. In this paper, it is shown how correlation analysis can be performed using kernels and then it is shown how order spectra become simple to compute after this analysis. The results are illustrated by an analysis of axle and car body vibrations in the prototype high-speed train, HSR-350x, using acceleration data measured by researchers at the Korea Railroad Research Institute (KRRI).