Machine learning based bearing fault detection in rotating machineries using statistical features of Empirical Mode Decomposition

Fault detection using artificial intelligence methods has been widely used in recent years. Rotating machinery condition monitoring is very important for immediate diagnosis and prevention of major damages. A bearing is a very vulnerable component in rotating equipment. In this paper, we aim to detect bearing defects including inner race, outer race, and ball defects using a novel time-frequency domain feature extraction method. Firstly, the Intrinsic Mode Functions (IMFs) of the acceleration signals were extracted using the Empirical Mode Decomposition (EMD) method. Then we propose the application of statistical features such as Mean Absolute Value (MAV), Simple Sign Integral (SSI), Waveform Length (WL), Wilison Amplitude (WAMP), Zero Crossing (ZC), Slope Sign Change (SSC), Root Mean Square (RMS), Mean, Variance (VAR), Standard Deviation (STD), Skewness (SKW), Kurtosis (KURT), and Energy to the IMFs for the sake of feature extraction. After feature extraction, the feature vectors are used as the input of AI techniques including KNN, NB, LR, SVM, MLP, and CNN for fault classification. We applied the proposed feature extraction approach to the Case Western Reserve University (CWRU) bearing fault dataset. The results show that using EMD features in machine learning techniques increases the accuracy of detection up to 93.14% for different rotational speed test data in the KNN model.

Machine learning based bearing fault detection in rotating machineries using statistical features of Empirical Mode Decomposition Keywords:

Bearing fault detection , Statistical features , Empirical mode decomposition , machine learning methods , Time-frequency approach