Fault diagnosis of gearbox using empirical mode decomposition and multi-fractal detrended cross-correlation analysis

A gearbox vibration signal is non-stationary and non-linear and has multiple components and multi-fractal properties, which make it difficult to effectively extract gearbox fault features. This paper proposes a method for gearbox fault feature extraction based on empirical mode decomposition (EMD) and multi-fractal detrended cross-correlation analysis (MFDCCA). First, EMD, a time-frequency analysis method, was employed to decompose the gearbox vibration signal into a number of intrinsic mode functions (IMFs). Second, the multi-fractal features hidden in the non-linear vibration signal were extracted by applying the MFDCCA to the selected major IMFs, thus highlighting the multi-fractality information which can be used to characterize different fault modes and severities of the gearbox. Third, for each IMF, three multi-fractal feature parameters sensitive to gearbox faults were selected from the multi-fractal features, further constructing the multi-fractal fault feature vector. Then, the principal component analysis (PCA) was introduced to reduce the dimensions of the extracted multi-fractal fault feature vectors and to enhance the accuracy of diagnosis. Finally, a radial basis function neural network was utilized to classify gearbox faults. Several commonly occurring faults were used to validate the proposed method in this study. Experimental results provide evidence that the extracted multi-fractal fault features can effectively distinguish different fault modes, even under slight variation in working conditions. Simultaneously, the results of comparison show that the performance of the proposed EMD-MFDCCA-PCA method outperforms that of EMD-MFDFA (multi-fractal detrended fluctuation analysis) combined with the traditional PCA.