Data annotation and feature extraction in fault detection in a wind turbine hydraulic pitch system

Abstract

The performance of wind turbines can be improved by processing supervisory control and data acquisition (SCADA) data. SCADA data can be processed in a reasonable time to enhance decisions made about maintenance schedules. The pitch system is critical in improving wind turbine operation by analysing data of the most relevant SCADA features. This study gathers the most significant pitch faults, and by implementing the adaptive neuro fuzzy inference system (ANFIS) technique it demonstrates the fault detection potential of this technique. The proposed approach includes the detailed pre-processing of SCADA data, emphasising the labelling process, in which a modified power curve monitoring method is used. During the implementation of the ANFIS, different combinations of the selected parameters were tested for their effects on the performance of fault detection. This methodology was implemented at a windfarm, commissioned in 2004, in five 2.3 MW fixed-speed onshore wind turbines equipped with a traditional servo-valve controlled hydraulic pitch system. Overall, data on 10 years of the operation of each wind turbine were utilised, and a total of nine pitch events were considered. Individual measurement for each blade angle was available for detecting pitch faults. Results demonstrated above 86% achievement of F1-score for pitch fault detection.