Detecting Damage on Wind Turbine Bearings Using Acoustic Emissions and Gaussian Process Latent Variable Models

This paper presents a study into the use of Gaussian Process Latent Variable Models (GP-LVM) and Probabilistic Principal Component Analysis (PPCA) for detection of defects on wind turbine bearings using Acoustic Emission (AE) data. The results presented have been taken from an experimental rig with a seeded defect, to attempt to replicate an AE burst generated from a developing crack. Some of the results for both models are presented and compared, and it is shown that the GP-LVM, which is a nonlinear extension of PPCA, outperforms it in distinguishing AE bursts generated from a defect over those generated by other mechanisms.