Fault detection and isolation of wind turbine gearbox via noise-assisted multivariate empirical mode decomposition algorithm

The wind turbine power transmission system exploits a planetary gearbox due to its large power transmission. In comparison with the common rotating systems, the wind turbine (WT) gearbox is assumed a complex system. Therefore, condition monitoring and fault detection isolation (FDI) of such systems are not straightforward and conventional signal processing methods (e.g. Fast Fourier transform) are not applicable or do not have an acceptable output accuracy. This paper proposes a new FDI approach for wind turbines based on vibration signals’ signatures derived from the multivariate empirical mode decomposition (MEMD) algorithm. Vibration signals are measured from a 750 kW planetary wind turbine gearbox on a dynamometer test rig provided by National Renewable Energy Laboratory (NREL).  In WT applications, to gather enough data with high accuracy and to avoid losing local information, multiple sensors must be utilized to collect data from different locations of the gearbox yielding a multi-sensory dataset. In standard EMD, joint information of multi-sensory data will be lost. Additionally, the intrinsic mode function (IMF) groups may not have the same characteristic features. To capture cross information of the dataset and to remove the effect of noise on the output results, a noise-assisted MEMD (NA-MEMD) algorithm is employed. Vibration signal features are also extracted by using discrete wavelet transform (DWT). Three major faults of the WT gearbox are detected using NA-MEMD and a comparison between NA-MEMD and DWT methods confirms the capability of the NA-MEMD method.