Wind Turbine Multi-Fault Detection based on SCADA Data via an AutoEncoder

Abstract

Nowadays, wind turbine fault detection strategies aresettled as a meaningful pipeline to achieve required levels of effi-ciency, availability, and reliability, considering there is an increasinginstallation of this kind of machinery, both in onshore and offshoreconfiguration. In this work, it has been applied a strategy that makesuse of SCADA data with an increased sampling rate. The employedwind turbine in this study is based on an advanced benchmark,established by the National Renewable Energy Laboratory (NREL)of USA. Different types of faults on several actuators and sensed bycertain installed sensors have been studied. The proposed strategy isbased on a normality model by means of an autoencoder. As of this,faulty data are used for testing from which prediction errors werecomputed to detect if those raise a fault alert according to a definedmetric which establishes a threshold on which a wind turbine workssecurely. The obtained results determine that the proposed strategyis successful since the model detects the considered three types offaults. Finally, even when prediction errors are small, the model isable to detect the faults without problems.