A lazy learning approach for condition monitoring of wind turbine blade using vibration signals and histogram features. (February 2020)

• Record Type: Journal Article
• Authors: Joshuva, A.; Sugumaran, V.
• Is Part Of: Measurement. Volume 152(2020)
• Publication Date: 2020-02
• Subjects: Condition monitoring -- Wind turbine blade -- Histogram features -- Nearest-neighbour -- k-nearest neighbour -- Locally weighted learning -- K-star classifier; Weights and measures -- Periodicals; Measurement -- Periodicals; Measurement; Weights and measures; Periodicals; 530.8
• DOI: 10.1016/j.measurement.2019.107295
• Languages: English
• ISSNs: 0263-2241

Abstract

Highlights: Machine learning based condition monitoring is proposed for wind turbine blade. Crack, Erosion, Loose connection, Pitch angle twist and Bend faults are considered. Histogram features were extracted from the vibration signals. Feature classification was performed using different machine learning classifiers. Locally weighted learning shows the better result of 93.83%. Abstract: The main objective of the proposed research study is to discriminate different blade fault conditions which affect the wind turbine blades under operating condition through machine learning approach. A three bladed wind turbine was chosen and the faults like blade bend, blade cracks, blade erosion, hub-blade loose connection and pitch angle twist were considered in the study. This problem is formulated as a machine learning problem which consists of three phases, namely feature extraction, feature selection and feature classification. Histogram features were extracted from vibration signals and feature selection was carried out using J48 decision tree algorithm. Feature classification was performed using lazy classifiers like nearest neighbour, k -nearest neighbours, locally weighted learning and K-star classifier. The results of these classifiers were compared with respect to their correctly classified instances (accuracy percentage) and found that, locally weighted learning yielded a maximum accuracy of 93.83% with a computational time of 0.07 s.