A Fault Diagnosis Method for Rolling Bearings Based on Feature Fusion of Multifractal Detrended Fluctuation Analysis and Alpha Stable Distribution

Table 6

Comparison of this work with some previous researches using CWRU bearings data set.

Reference	Features	Classifier	Defects considered	Classified states	Accuracy (%)
[4]	Permutation entropy of intrinsic mode functions (IMFs) decomposed by ensemble empirical mode decomposition (EEMD)	SVM + parameter optimized by intercluster distance (ICD)	N, B (0.18, 0.36, 0.54, and 0.72 mm), IR (0.18, 0.36, 0.54, and 0.72 mm), and OR (0.18, 0.36, and 0.54 mm)	2, 3, 4, 11	97.91–100
[6]	Statistical features from the time and frequency domains, wavelet packet energy, and complex envelope magnitudes	SVM, the -nearest neighbor classifier (-NN), and multilayer perceptron (MLP)	N, B (0.18, 0.54 mm), IR (0.18, 0.54 mm), and OR (0.18, 0.54 mm)	7	98.13–99.97
[7]	Skewness, kurtosis, standard deviation, root mean square (RMS), crest factor, and five entropies	SVM and artificial neural network (ANN) with different attribute filters	N, B (0.18, 0.54 mm), IR (0.18, 0.54 mm), and OR (0.18, 0.54 mm)	7	100
[18]	Features derived from higher order statistics analysis (HOSA) with feature reduction using PCA	SVM + one against all (OAA)	N, B (0.18, 0.36, 0.54, and 0.72 mm), IR (0.18, 0.36, 0.54, and 0.72 mm), and OR (0.18, 0.36, and 0.54 mm)	4	96.98
[19]	Multiscale entropy (MSE), hierarchical entropy (HE)	SVM + parameter optimized by PSO	N, B (0.18, 0.72 mm), IR (0.18, 0.36, 0.54, and 0.72 mm), and OR (0.18, 0.36, and 0.54 mm)	10	MSE + SVM: 97.75; HE + SVM: 100
[20]	Statistical features from the time and frequency domains, the energy of empirical mode decomposition (EMD)	Statistical locally linear embedding (S-LLE) + -NN, classification and regression trees (CART), and RBF-SVM	N, B (0.54 mm), IR (0.54 mm), and OR (0.54 mm)	4	97.26
[21]	Features extracted from wavelet kurtogram and quefrency cepstrum	Swarm Rapid Centroid Estimation (SRCE) + Hidden Markov Model (HMM)	N, B (0.54 mm), IR (0.54 mm), and OR (0.54 mm located at three different positions)	Drive end: 5; Fan end: 5	Drive end: 95.08; Fan end: 100
Present work	Features derived from MFDFA and ASD with feature fusion using KPCA	LSSVM + parameter optimized by PSO	N, B (0.18, 0.54 mm), IR (0.18, 0.54 mm), and OR (0.18, 0.54 mm)	7	98.6

N denotes normal, B denotes ball faults, IR denotes inner-race faults, and OR denotes outer-race faults.