A discussion on using Empirical Mode Decomposition for incipient fault detection and diagnosis of the wind turbine gearbox

Vibration signals from the gearbox of a wind turbine are essentially non-stationary and nonlinear in both time and frequency. Empirical Mode Decomposition (EMD) is an ideal method for dealing with this type of signal. Yet the signal containing the fault information was contaminated by the noise, which contains two different types of white noise and impact noise. This makes it so the vibration signal cannot be processed with EMD directly, since it will produce the spurious IMF (Intrinsic Mode Function). The signal has to be pre-processed before implementing EMD. In fact, a wavelet filter is perfect for white noise de-noising and the morphological filter is suitable for impulse interference. In this paper, a confederative filter, which is combined with the wavelet and morphological filter, is designed for signal preprocessing, and a standard processing program is proposed too. An experimental case shows the accuracy and efficiency of the confederative filter and the process program.