Statistical Data Mining through Credal Decision Tree Classifiers for Fault Prediction on Wind Turbine Blades Using Vibration Signals

Abstract

In a wind turbine, blades are the most important component of wind capture in wind turbines as they easily become unreliable due to environmental conditions. This paper demonstrates the malfunction characterization of wind turbine blades by the use of vibration data via the credal decision tree (CDT). The defects on the blades are replicated to model the defects through machine learning. The extraction of functions (statistical functions) and the selection of the component (algorithm of decision tree J48) were employed to identify the best framework for defect classification. Using the credal decision tree, 82.67% of classification accuracy have been obtained with the Kappa statistic of 0.792 and mean absolute error of 0.0768.