Page 118 - IJOCTA-15-4

Zhang et al. / IJOCTA, Vol.15, No.4, pp.649-669 (2025)

Figure 10. The first six intrinsic mode function (IMF) components (A) and their corresponding frequency
spectra (B) obtained from the empirical mode decomposition of the simulated signal

Figure 11. The first six intrinsic mode function (IMF) components (A) and their corresponding frequency
spectra (B) obtained from the ensemble empirical mode decomposition of the simulated signal

Comparing the spectra in Figures 7, 10, and As shown in Figure 13, the envelope spectra
11, it can be observed that the decomposition re- of the sensitive IMF1 component obtained from
sults of EMD and EEMD are not ideal. The IMF EMD and EEMD are both chaotic. Although
components obtained from both methods exhibit spectral lines exist at the fault frequency, there
varying degrees of mode mixing, with the IMF1 are many interference components, and the spec-
component being the most significant. The fre- trum does not exhibit regular variations. As a re-
quency spectrum of IMF1 shows a highly chaotic sult, the fault frequency could not be accurately
pattern, with spectral lines scattered across all identified, and the required valid information was
frequency ranges. This influences the accuracy of not obtained, thus complicating the fault diagno-
the features extracted from the signal. sis process. The primary reason for this issue is
that the EMD method suffers from mode mix-
However, it is necessary to select the IMF
components that effectively represent the signal ing due to its inherent algorithmic limitations.
characteristics in EMD and EEMD. The compre- While the EEMD method can address this is-
hensive evaluation factor for each IMF component sue to some extent, the white noise added to the
was calculated (Figure 12). signal cannot be completely neutralized, signifi-
cantly affecting the accuracy of the decomposition
The largest difference between EMD and method.
EEMD occurs between IMF1 and IMF2, as seen
in Figure 12. Therefore, IMF1 was chosen as the Based on the above analysis, it can be con-
sensitive IMF component, and its envelope was cluded that GJO–VMD can adaptively determine
demodulated. Figure 13 presents the resulting en- the most suitable key decomposition parameters
velope spectrum. according to the characteristics of the signal.
660