Optimal Level and Base Mother Wavelet Selection for Wavelet Analysis in Wind Turbine Health Monitoring

Chahinez Beldjaatit; Toufik Sebbagh; Samah Bouanik; Hocine Guentri

Chahinez Beldjaatit Department of Mechanical Engineering, Faculty of Science and Technology, 20 Août 1955 University, Skikda, Algeria
Toufik Sebbagh Department of Mechanical Engineering, Faculty of Science and Technology, 20 Août 1955 University, Skikda, Algeria
Samah Bouanik Department of Mechanical Engineering, Faculty of Science and Technology, 20 Août 1955 University, Skikda, Algeria
Hocine Guentri Department of Mechanical and Electromechanical Engineering, Faculty of Science and Technology, Abdelhafid Boussouf University center of Mila, Algeria

Keywords: Wind turbine, Bearing fault analysis, Discrete wavelet transform, Shannon entropy, Mother wavelet

Abstract

Wind turbines are gaining global attention as a renewable energy solution. However, issues with unexpected component faults in their powertrains are common. Vibration monitoring is commonly used to detect early signs of bearing and gear failures, aiming to improve productivity. Research shows that bearings are more prone to failures compared to gearbox gears. The wavelet transforms, particularly the discrete wavelet form, has recently become popular for analyzing non-stationary signals. Nevertheless, its effectiveness depends on choosing the right decomposition level and mother wavelet. The primary challenges of discrete wavelet transform (DWT) involve selecting the suitable mother wavelet for bearing signal analysis, as different choices lead to diverse results, and determining the optimal decomposition level to extract valuable features for anomaly identification. This study focuses on vibration signals from inner race faults and proposes an optimization approach for selecting the best decomposition level and mother wavelet based on criteria like Shannon entropy, energy-to-Shannon-entropy ratio, and reconstruction quality (root-mean-square error RMSE). The validity of the method is confirmed through correlation coefficient calculation and signal-to-noise ratio (SNR) analysis on the denoised vibration signal. These results prove the validity and robustness of the proposed method for the bearing fault signal analysis.