Fault Classification in Wind Turbine Systems Using Multichannel Vibration Signals: A Comparative Study of FFT and DWT-Based Features

Abstract

Reliable fault classification is essential for improving the operation and maintenance of wind turbine systems, the use of machine learning and multichannel vibration has also been used in this field, like the one proposed in this study. The methodology used combines signal segmentation with feature extraction using the Discrete Wavelet Transform (DWT) and the Fast Fourier Transform (FFT), then both techniques are compared under the same experimental conditions. For the FFT based representation, spectral energy is grouped into a reduced number of frequency bands then the trade off between classification performance and computational cost is studied. An SVM classifier is trained and evaluated using Group K-Fold cross-validation to avoid information leakage. The experimental evaluation shows that FFT derived features achieve better classification accuracy, greater stability, and lower computational cost than the DWT representation. The best compromise between accuracy and efficiency in FFT is obtained with 10 frequency bands, and the confusion matrices and feature space analysis revealed improved class separability. Concluding that FFT based features are an effective and computationally efficient representation for fault classification in wind turbine systems.