Automated wind turbine gearbox bearing diagnosis algorithm based on vibration data analysis and signal pre-whitening

In this paper an automated wind turbine gearbox bearing diagnosis algorithm is presented. The algorithm is based on most recent research results for separating discrete (gear) from random (bearing) frequency components using Cepstral Editing Procedure (CEP) based signal Pre-Whitening (PW). The proposed automated procedure builds up on the semi-automated procedure described by Sawalhi et al. in 2007. The procedure is updated with regards to the most recent achievements made concerning signal separation and extended with a frequency content identifier and rule based diagnosis to fully automate the diagnosis process. Furthermore, this paper gives a selection of important statements made in literature throughout the last decade to summarize the algorithms used and focus on the most important issues. Each of the involved processing steps is discussed with respect to its effectiveness based on real data.
The proposed procedure is applied to wind turbine data coming from seventeen wind turbines of the 2 MW class, for which vibration data are available containing both healthy and damaged states. Three application examples are given, where the automated procedure successfully diagnosed High Speed Shaft (HSS) bearing damages in the data sets.