Main Bearing Fault Prognosis in Wind Turbines based on Gated Recurrent Unit Neural Networks

The transition from onshore to offshore wind farms isan imminent fact in the future. It supposes to face hard challengeslike difficulties to carry out offshore maintenance operations dueto increased downtime (because of several causes like continuouslybad environmental conditions) on wind farms. That is why, thereis a need to improve maintenance and monitoring practices likethose involved in condition-based area. This work proposes amethodology based on three key points: (i) a semi-supervised modelbuilt from a gated recurrent unit (GRU) neural network and byusing only healthy real SCADA data, (ii) propose a fault prognosisindicator (FPI) to trigger warnings or fault alarms as such, and (iii)detect the main bearing fault several months in advance on a faultywind turbine. The reported results show the excellent performanceof the GRU trained model to predict the main bearing temperatureas output by exploiting the capabilities of GRUs (recurrent-basedneural networks) to decide what information to forget or preservethrough time. In the FPI construction, the use of exponentiallyweighted moving average (EWMA) helps at the results to avoid thepresence of false alarms that is very useful in any detection strategy.Finally, the stated methodology lets to detect the main bearing faulton a WT two months in advance at least, which contributes toplan maintenance actions ahead of time. Furthermore, in this way,the lifespan of this large component may be extended and windturbine’s uptime may increase in a significant percentage.