关键词: 行星齿轮箱, 变分模态分解（ＶＭＤ）, 多尺度奇异值分解（Ｍ－ＳＶＤ）, 二次分解, 故障特征提取

Abstract: To address the difficulty in effectively separating fault components from noise components during the extraction of weak pitting fault features of planetary gearboxes, which leads to insufficient suppression of noise burrs and incomplete preservation of fault feature information, a feature extraction method combining Variational Mode Decomposition (VMD) and improved Multiscale Singular Value Decomposition (M-SVD) is proposed. Firstly, an objective function based on minimum envelope entropy is constructed, and a VMD signal decomposition method incorporating Sparrow Search Algorithm (SSA) is proposed to achieve adaptive decomposition of the original signal. Secondly, the square envelope spectrum kurtosis is selected as an indicator to screen and reconstruct the VMD decomposition components, thereby removing most of the noise burrs from the original signal. Then, considering the sensitivity of the mean to the overall trend variation of vibration signals, a signal processing method based on improved M-SVD isdesigned by combining the mean singular value with sample entropy. Finally, envelope demodulation is performed on the signal processed by improved M-SVD to extract the weak fault features of the planetary gearbox. Using SNR and RMSE as quantitative metrics, the proposed method is compared with the standalone SSA-VMD, the improved M-SVD, and the original M-SVD methods in pitting fault feature extraction experiments. The proposed method effectively suppresses noise components while maximizing the retention of fault feature information.

Key words:  planetary gearbox, Variational Mode Decomposition (VMD), Multiscale Singular Value Decomposition (M-SVD), second decomposition, fault feature extraction.