Acoustic blade damage diagnosis based on improved symplectic geometry mode decomposition

Abstract Wind turbine blades are vulnerable to damage due to coupled operational loads. Acoustic detection demonstrates strong potential for blade monitoring, but its performance is often hindered by complex noise and imbalanced monitoring samples. Accordingly, an acoustic damage detection method is proposed based on improved symplectic geometry mode decomposition (SGMD). In the proposed method, an adaptive waveform matching extension is designed to mitigate endpoint errors during signal decomposition, while a full-pattern distance constraint is developed to enhance decomposition accuracy and pattern fidelity. Moreover, a comprehensive scoring index combining multi-scale local kurtosis and autocorrelation coefficients is designed for the precise measurement of transient impulsiveness to select effective signal components and eliminate noise. Finally, a weighted oversampling support vector machine (SVM) classifier is applied to improve recognition under class imbalance. Experimental results demonstrate that the proposed method outperforms conventional SVM models based on either traditional SGMD or other denoising techniques, highlighting its great practical application potential in wind turbine blade fault diagnosis.