Regresar

Fault Prognosis for Wind Turbines' Main Bearing based on SCADA data

Abstract:

The rapid growth of large-scale wind turbines (WT) has changed the requirements in terms of operation and maintenance strategies. WTs are required to be safe and profitable systems. A great option is failure prognosis, aiming to reduce maintenance and operating costs, and forecast service life based on condition. In this work, the analysis of the data from the supervisory control and data acquisition (SCADA), already present in industrial sized WTs, and work orders (repair and maintenance data) allows a normality based methodology to be implemented using convolutional neural networks (CNNs). In this work, two years of SCADA data from a real under production wind farm, composed by 12 wind turbines is used. The main bearing failure occurred in one of this wind turbines, and thus it could be used to verify the performance of the proposed strategy in a real life situation. Finally, a proposed indicator helps us pinpoint in advance the occurrence of the fault. The obtained results provide a warning time (alarm) for the main bearing fault up to 5 months in advance.