Hybrid Noise Reduction for Wind Turbine Blade Fault Diagnosis

Abstract

Accurate diagnosis of wind turbine blade faults relies heavily on the analysis of fan audio signals, which are often contaminated by noise. Effective noise reduction is essential to extract useful diagnostic information and ensure reliable analysis. Traditional spectral analysis techniques, such as Fourier transform-based filtering, face limitations in handling complex and non-smooth signals. Wavelet transform methods provide multi-resolution analysis but lack adaptive decomposition capabilities. Empirical Mode Decomposition (EMD) offers a promising solution by adaptively decomposing signals into intrinsic mode functions (IMFs) based on their inherent characteristics. This paper presents an improved noise reduction method that combines EMD thresholding with Savitzky-Golay filtering. High-frequency IMFs are processed using thresholding to preserve critical signal details, while low-frequency IMFs undergo Savitzky-Golay filtering to ensure smoothness. Experimental results on wind turbine blade audio signals demonstrate that the proposed hybrid approach achieves superior noise reduction performance compared to standalone EMD thresholding or filtering techniques, enhancing the reliability of fault diagnosis.