Engine Fault Diagnosis Based on Independent Variational Mode Decomposition and Improved Kernel Extreme Learning Machine

In order to extract effective fault features from the cylinder head vibration signal under strong noise background and classify faults, an engine fault diagnosis method of independent variational mode decomposition (IVMD) combined with improved kernel extreme learning machine is proposed. Firstly, the matching waveform is selected to carry out the end extension of the original signal according to the spectral cyclic coherence coefficient. And the signal after end extension is decomposed into a number of intrinsic mode functions (IMFs) by using variational mode decomposition (VMD). The end effect in VMD is suppressed effectively. Then the selected effective IMFs and original signal are constructed as the input observation signals of kernel independent component analysis (KICA). After KICA, the noise and effective signal are separated further, the mode mixing is eliminated and the independent effective fault feature bands are obtained. Autoregressive model parameters, multi-scale fuzzy entropy and normalized energy moment of each frequency band are extracted to construct a joint fault feature vector. Lastly the improved KELM model based on social emotional optimization algorithm (SEOA-KELM) is constructed to classify the fault features in order to realize the engine fault diagnosis. The simulation and experimental results show that the proposed method can effectively suppress the end effect in VMD, improve the signal decomposition precision, eliminate the noise, separate independent and effective fault feature frequency bands, enhance feature parameter identification and improve the speed and accuracy of engine fault diagnosis finally. The average accuracy rate of engine fault diagnosis is up to 99.85%. © 2019, Editorial Department of JVMD. All right reserved.