Fault diagnosis of wind energy conversion systems with sensor faults

Zhang, J.H.; Xiong, J.; Ren, Ming; Jiang, Minlin; Hou, Guang-Quan

doi:10.1049/cp.2013.1866

Cited by 1 publication

(1 citation statement)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Melero et al [13] proposed a method based on the Kalman filter to detect fault events and wrong wind data to improve the installation of a new wind farm. Zhang et al [14] have applied a filter‐based fault detection and isolation strategy for wind energy conversion systems with sensor faults, variable wind speed and measurement noises. Ferguson et al [15] presented a method of anomaly detection using temperature, rotational speed, and torque from wind turbine gearbox.…”

Section: Introductionmentioning

confidence: 99%

Wind turbines anomaly detection based on power curves and ensemble learning

Moreno

Coelho

Ayala

et al. 2020

IET Renewable Power Generation

View full text Add to dashboard Cite

Wind farms are increasingly important nowadays since in some countries can surpass conventional power sources. However, in countries where the exploration of this renewable source started recently, the lack of knowledge related to maintenance routines and efficient operation has led to fast performance degradation. In this context, wind turbine condition monitoring can detect anomalies in its performance as an unexpected failure, avoiding financial loss. In this study, machine learning approaches are applied as an online tool to detect abnormal wind turbine operation modes, evaluating the wind turbine operation in all regions of the power curve. The methodology has been validated with an original and real dataset collected from a large‐scale onshore wind turbine in Northeast Brazil. The results exhibit an expressive reduction of energy loss and indicate the ability of the proposed approach to assessing the abnormal modes even when a small number of recorded data are available. The standard classifiers reached on average 98.64% accuracy in the holdout data set. Additionally, an ensemble of classifiers is proposed which helped to improve in 12% the accuracy of the best classifier alone, increasing the confidence of alarms raised by the predictive maintenance tool.

show abstract

Section: Introductionmentioning

confidence: 99%