Signal Feature Extraction And Diagnosis Of Rotor Vibration Faults Of An Aeroengine

For aeroengine fault diagnosis technology, the fault feature identification and extractiondirectly affect the accuracy and effectiveness of fault diagnosis. Currently the widely-usedmethod of aeroengine fault diagnosis is vibration signal analysis. Diagnosis is conductedbased on the fact that aeroengine vibration signal has different characteristics under differentengine conditions.Aiming at the problems that traditional signal analysis method is difficult to describenon-stationary rotor vibration signals and evaluate signal feature quantitatively, this researchputs forward a method of feature vector extraction based on entropy and empirical modedecomposition. Energy method is then used to select the component of the main faultinformation, and the accuracy of the selected main component is verified using the correlationcoefficient method. On the basis of the main component selected, singular spectrum entropy,power spectrum entropy, wavelet energy spectrum entropy, and wavelet packet spacecharacteristic spectral entropy are calculated in time domain, frequency domain andtime-frequency domain. The four types of information entropies are used to form rotor faultdiagnosis feature vector.Aeroengine fault diagnosis is performed by the method that combines SVM withevidence theory, and then multiple-classification SVM is used to diagnose the three typicalrotor faults of aeroengines. According to evidence theory, information fusion is conducted tothe signals from different measuring point and different operating conditions. The output is inthe form of probability. This method can greatly enhance the accuracy of aeroenginediagnosis，and the result can be used to observe the degree of failure more intuitively.