Integration of EEMD and ICA for wind turbine gearbox diagnosis

Gearbox failure becomes a major concern for reliability of wind turbine because of complex repair procedures, long downtime and high replacement costs. Prior studies showed that the majority of gearbox failures were initiated from bearing failures. Because of the low signal-to-noise ratio (mixture of bearing defect signals and gear meshing signals) and transient nature of bearing signals, it poses significant difficulty for bearing defect diagnosis in wind turbine gearbox at the incipient stage. To address it, this paper presents an effective fault component separation method that integrates ensemble empirical mode decomposition (EEMD; an adaptive signal decomposition method in time-frequency domain) with independent component analysis (ICA; a blind source separation technique), without requiring a priori information on the rotating speeds or bandwidth. The method firstly decomposes one-channel vibration measurements into a series of intrinsic mode functions as pseudo-multi-channel signals, by means of EEMD. ICA is performed on the intrinsic mode functions to separate bearing defect-related signals from gear meshing signals. Envelope spectrum analysis is performed on the bearing defect-related signals to identify bearing structural defects. The effectiveness of the developed method in separating bearing defect-related signals from gear meshing signals for more effective fault diagnosis in bearings is evaluated and confirmed, numerically and experimentally. Copyright (c) 2013 John Wiley & Sons, Ltd.