RUL forecasting for wind turbine predictive maintenance based on deep learning

In order to keep up with the rising demand, the wind industry is actively working to make it more viable and competitive, which means tackling some of the biggest challenges it faces [3,4]. A survey analysis [5] shows that approximately 45% of the overall budget might be set aside for operation and maintenance (O&M), as shown in Figure 1, posing as one of the biggest challenges faced by the wind industry. To counter this, preventive maintenance (PM) could be employed, which follows a periodically scheduled maintenance plan to reduce unplanned maintenance. However, this leads to unnecessary downtime, as often times the maintenance is not required [6-8]. This could be resolved if predictive maintenance (PdM) could be achieved. PdM predicts the optimal time for maintenance, ensuring it is performed precisely when needed and avoiding unnecessary machine stoppages [9]. One way to achieve this is by analyzing the remaining useful life (RUL) of the turbine and scheduling maintenance immediately prior to failure [10]. However, wind farms are often located in remote locations, usually spanning over many miles [11,12]; especially in the case of off-shore wind farms [13,14]; and timely arrival becomes an issue as a result.