Fault Detection Method for Rolling Mechanical Equipment Based on EEMD and RF

Since rolling mechanical equipment is critical to manufacture systems, fault detection and diagnosis for them become more important and urgent with the improving need for safely and economical running. Aiming at streaming vibration signals, a fault detection and diagnosis scheme is proposed with an off-line training stage and on-line detecting stage. After de-noising by wavelet transformation, signals are decomposed into several IMF (Intrinsic Mode Functions) by EEMD (Ensemble Empirical Mode Decomposition) method. Fault features are expressed on definite IMFs rather than in a mixed mode. Based on these IMFs, the fault types of bearings can be represented by PSD (Power Spectral Density) to get good identification ability, and even to get the detailed physical meaning. The RF (Random Forest) method is used for classification of various fault positions and degrees with high efficiency and accuracy. Simulations with real data show the practicability of the fault detection scheme, and the earlier fault detection or fault forecast can be achieved by expanding classification training set for mixed fault data.