This paper proposes a method using empirical mode decomposition (EMD) and histogram modeling for
bearing defects of a low-speed rotational induction motor. The proposed method minimizes signal distortion
caused by noise using the EMD and conducts histogram envelope modeling by normalizing a signal. Then,
the method extracts and selects unique features of each fault using partial autocorrelation coefficients
(PARCOR) and distance evaluation technique (DET). Using the extracted features as inputs, support vector
A Feature Extraction Method Using Empirical Mode Decomposition and HistogramModeling for Robust Bearing Fault Diagnosis
regression (SVR) classifies inner, outer, and roller faults of a bearing. For the optimal classification
performance, we vary a variable of the kernel function of SVR ranging from 0.01 to 1.0 and the number
of features ranging from 2 to 250. Experimental results indicate that the proposed method having a
parameter of 0.55 and 45 features outperforms conventional fault classification methods, yielding an average
of 95% classification performance.