Artificial neural network based on wavelet transform and feature extraction for a wind turbine diagnosis system

This paper presents a Wavelet Transform (WT) based Artificial Neural Network (ANN) input data pre-processing scheme and outlines the results of localized rotor angular speed defect of a wind turbine system by employing this proposed methodology. The methodology consists of calculating Daubechies 4-order (DAUB-4) dilation WTs with the Multi-Resolution Analysis (MRA) of the data, and then extracting predominant wavelet coefficients distributed to certain levels of these WTs using the Parseval's theorem. The features extracted from the dominant wavelet coefficients are used as inputs to ANN classifier to evaluate its performance. The simulation results show that the backpropagation network trained with a reasonably small number of training sets is able to recognize and classify signals of residues efficiently and can achieve high accuracy rate under various test cases.