Fault diagnosis of wind turbine units using EMD, VDM, CEEMDAN decomposition methods with the help of feature analysis obtained by 1D CNN, SVM, and KNN machine learning algorithms.

Abstract

Renewable sources, with a focus on wind energy, play a crucial role in global energy sustainability. However, to maximize their utilization, the development of fault diagnosis tools for wind turbine units (WTUs) is necessary. Therefore, this research project aims to compare time series decomposition methods for generating features for classification algorithms applied to WTU fault diagnosis. The technique of decomposing signals before presenting them to classification algorithms has recently been used for fault diagnosis. However, the comparison between different current decomposition techniques is still limited in the literature, which motivates this work. Thus, the properties of WTU fault data will be analyzed using the Empirical Mode Decomposition (EMD), Variational Mode Decomposition (VDM), and Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) methods. Next, training will be conducted for the Convolutional Neural Network (1D CNN), Support Vector Machine (SVM), and K-Nearest Neighbors (KNN) classification models using the extracted features from the signal components. Finally, the test set will be classified to assess whether the decomposition tools can provide relevant features for fault diagnosis, and if so, to determine which of these decomposition methods will yield the best results in performance metrics using different classification algorithms.