Empirical mode decomposition based features for diagnosis and prognostics of systems

Original publisher: Adelphi, MD : Army Research Laboratory, [2008] OCLC Number: (OCoLC)318692436 Subject: Hilbert-Huang transform. Excerpt: ... Next, we examine the Fourier transforms of the IMFs. The Fourier transforms of the IMFs were obtained for each block separately and then the amplitudes were averaged over the twelve blocks. Further, the amplitudes were averaged over 10 frequency samples so that the resultant samples are averaged over 24 Hz. The results are shown in figures 11-20 for the six data sets, two faulted and four good. Only the first 10 modes are shown since the last two modes do not show much variation. Note that the frequencies of the IMFs and the modulation rates decrease as the IMF number increases. There is a difference in frequency where the amplitudes peak and in the frequency bands over which the amplitudes are relatively high for the good and the faulted data sets. We use these differences to generate features for discriminating between the good and faulted data sets. Figure 11. Amplitude of the Fourier transform of the first IMF of six data sets. 9