Bearing Fault Detection using Empirical Mode Decomposition Technique

Ball bearings are one of the most important and used components in different industries. In time diagnosis of these elements can prevent human and financial losses. Early fault detection in bearings especially when vibration signals from these elements are nonlinear and nonstationary is difficult. Vibration analysis is one of the most effective methods in bearing fault detection. This research uses Empirical Mode Decomposition (EMD) method for processing vibration signals. EMD was developed by Huang in 1998 for analyzing nonlinear and nonstationary signals. Vibration signals acquired from bearing housing were decomposed in to Intrinsic Mode Functions (IMFs). By analyzing 8 IMFs from different bearing conditions namely: 1) Healthy bearing 2) Bearing with inner race fault 3) Bearing with outer race fault and 4) Bearing with ball fault, first IMF was chosen as the most sensitive IMF to different faults. Also bearing were tested under different loading conditions which were: 1) No load 2) 1 hp load 3) 2 hp load 4) 3 hp load. Six time feature extracted from first IMF in each load and bearing conditions. These features were used as input vector to an Adaptive Neuro Fuzzy Inference System (ANFIS) to detect each fault. The designed ANFIS network was able to detect different faults with 100% accuracy. Results from this research can be used as a novel method for bearing fault detection with excellent accuracy.