Fault feature selection method of rolling bearings based on multiple metric weighting

To better screen the fault features of the original high-dimensional vibration signals and improve the accuracy of rolling bearings fault diagnosis, a fault feature selection method based on weighted optimization of feature evaluation metrics was proposed. The smoothness priors approach was adaptively used to decompose the non-stationary vibration signals, and the various time-domain, frequency-domain and time-frequency domain features were extracted to construct an initial fault feature set. Then, four feature performance evaluation indexes of monotonicity, discrimination, identification and robustness were integrated, and a weighted linear combination based on the sine-cosine optimization algorithm was used to comprehensively evaluate the fault features performance, followed by the screening of sensitive fault features. The proposed method was applied to rolling bearings experimental data, and the support vector classifier was used as the diagnostic machine to verify the effectiveness of the proposed fault feature selection method. © 2024 CIMS. All rights reserved.