TY - JOUR
T1 - Model-based fault detection for generator cooling system in wind turbines using SCADA data
AU - Borchersen, Anders Bech
AU - Kinnaert, Michel
PY - 2016/4
Y1 - 2016/4
N2 - In this work, an early fault detection system for the generator cooling of wind turbines is presented and tested. It relies on a hybrid model of the cooling system. The parameters of the generator model are estimated by an extended Kalman filter. The estimated parameters are then processed by an appropriate statistical change detection algorithm in order to detect faults in the cooling system. To validate the method, it has been tested on 3 years of historical data from 43 turbines. During the testing period, 16 faults occurred; 15 of these were detected by the developed method, and one false alarm was issued. This is an improvement compared with the current system that gives 15 detections and more than 10 false alarms. In some cases, the method detects the fault a long time before the turbine reports an alarm. A further advantage of the method is that it is based on supervisory control and data acquisition data that are available for the operator of all modern turbines. Thereby, the method can be implemented without the need to modify or install additional components in the turbines. Copyright © 2015 John Wiley & Sons, Ltd.
AB - In this work, an early fault detection system for the generator cooling of wind turbines is presented and tested. It relies on a hybrid model of the cooling system. The parameters of the generator model are estimated by an extended Kalman filter. The estimated parameters are then processed by an appropriate statistical change detection algorithm in order to detect faults in the cooling system. To validate the method, it has been tested on 3 years of historical data from 43 turbines. During the testing period, 16 faults occurred; 15 of these were detected by the developed method, and one false alarm was issued. This is an improvement compared with the current system that gives 15 detections and more than 10 false alarms. In some cases, the method detects the fault a long time before the turbine reports an alarm. A further advantage of the method is that it is based on supervisory control and data acquisition data that are available for the operator of all modern turbines. Thereby, the method can be implemented without the need to modify or install additional components in the turbines. Copyright © 2015 John Wiley & Sons, Ltd.
KW - SCADA data, extended Kalman filter, model-based fault detection, wind turbine, validation on real data
U2 - 10.1002/we.1852
DO - 10.1002/we.1852
M3 - Journal article
SN - 1095-4244
VL - 19
SP - 593
EP - 606
JO - Wind Energy
JF - Wind Energy
IS - 4
ER -