A NBM based on P-N relationship for DFIG wind turbine fault detection

Abstract

Supervisory control and data acquisition (SCADA) data has been widely applied to identify abnormal conditions in wind turbine generators (WTG). One approach was to apply Artificial Intelligence (AI) to SCADA data, comparing the predicted power output of a WTG and its actual output and using the prediction error as an indicator to detect faults. However, complicated training processes limit its application. This paper presents a normal behavior model (NBM), based on power output-generator speed (P-N) curve, to analyze SCADA data from modern pitch regulated WTGs for detecting anomalies. Through analysis of the operational characteristics of the pitch regulated WTG, it is found that inaccuracies in wind speed measurement, the inertia of the rotor, yaw and pitch misalignments, and air density fluctuation may affect the performance of the power curve monitoring algorithms. This paper shows that under normal conditions the P-N curve based NBM performs better when fitting the SCADA data to WTG output under normal conditions than the power curve. Results demonstrate that it can give alarm to forthcoming faults earlier than the existing condition monitoring system (CMS).