A methodological approach for detecting multiple faults in wind turbine blades based on vibration signals and machine learning

Abstract Wind turbines generate clean and renewable energy for the international market. The most ‎‎important aspect of wind turbine maintenance is reducing failures, downtime, and operating and maintenance expenses. ‎This study aims to detect multiple faults exhibited by wind turbine blades; failures such as cracks (tip crack, mid-span crack, and crack ‎near the root) were observed in the blades at different locations. The research suggests a new approach, incorporating vibration signals and machine learning techniques to identify various failures in wind turbine blades. The technology of ranking features such as ReliefF algorithms, chi-squares, and information gains was adopted to discuss a method framework to diagnose several problems in wind turbine blades, such as cracks in different locations. The k-nearest neighbors (KNNs), support vector machines, and random forests are used to classify data based on measured vibration signals. The eight main time-domain features are calculated from the vibration signals. The proposed methodology was validated using four databases. The results showed good classification accuracy in four databases, with at least three non-conventional features in each database’s top nine features of the three classification techniques. The results also showed that when the ReliefF selection algorithm is applied with the KNN classification algorithm, it generates the highest classification accuracy under all failure conditions, and the value is 97%. Finally, the performance of the proposed classification model is compared with other machine learning classification models, and a promising result is obtained. ‎