Open AccessWind Turbine Failure Prediction Using SCADA Data
Author(s) -
Luiz Andre Moyses Lima,
A. K. Blatt,
J. Fujise
Publication year - 2020
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1618/2/022017
Subject(s) - overfitting , turbine , wind power , scada , computer science , reliability engineering , fault (geology) , fault detection and isolation , data mining , engineering , machine learning , artificial intelligence , artificial neural network , mechanical engineering , electrical engineering , seismology , geology , actuator
This paper proposes a failure prediction system for wind turbines using the Normal Behavior Model (NBM) approach. By using available SCADA data, the NBMs are trained to make predictions that reflect what would be a turbine’s normal operating condition. They are able to identify when a given operating condition is abnormal, which points towards probable component degradation. Alerts are raised based on the daily-averaged prediction error to help the O&M team in identifying turbines that need maintenance. The NBMs are comprised of numerous linear models with different inputs and training sets, according to an ensemble approach that aims to avoid overfitting and to reduce the amount of false-positive predictions. Description and insights on various development steps are presented, such as data treatment, model selection, error calculation and alerts generations. Two test cases are shown using operational data from existing wind turbines, highlighting the system’s ability to generate alerts weeks before a severe fault occurs.