Affiliations: School of Jet Propulsion, Beijing University of
Aeronautics and Astronautics, Beijing, China
Note: [] Corresponding author: Cheng-Wei Fei, School of Jet Propulsion,
Beijing University of Aeronautics and Astronautics, Beijing 100191, China.
Tel.: +86 152 0112 7536; E-mail: feicw544@163.com
Abstract: In order to correctly analyze aeroengine whole-body vibration
signals, Wavelet Correlation Feature Scale Entropy (WCFSE) and Fuzzy Support
Vector Machine (FSVM) (WCFSE-FSVM) method was proposed by fusing the advantages
of the WCFSE method and the FSVM method. The wavelet coefficients were known to
be located in high Signal-to-Noise Ratio (S/N or SNR) scales and were obtained
by the Wavelet Transform Correlation Filter Method (WTCFM). This method was
applied to address the whole-body vibration signals. The WCFSE method was
derived from the integration of the information entropy theory and WTCFM, and
was applied to extract the WCFSE values of the vibration signals. Among the
WCFSE values, the W_{FSE1} and W_{CFSE2}
values on the scale 1 and 2 from the high band of vibration signal were
believed to acceptably reflect the vibration feature and were selected to
construct the eigenvectors of vibration signals as fault samples to establish
the WCFSE-FSVM model. This model was applied to aeroengine whole-body vibration
fault diagnosis. Through the diagnoses of four vibration fault modes and the
comparison of the analysis results by four methods (SVM, FSVM, WESE-SVM,
WCFSE-FSVM), it is shown that the WCFSE-FSVM method is characterized by higher
learning ability, higher generalization ability and higher anti-noise ability
than other methods in aeroengine whole-vibration fault analysis. Meanwhile,
this present study provides a useful insight for the vibration fault diagnosis
of complex machinery besides an aeroengine.
