[1D3] Fault diagnosis method for rolling bearings based on improved SVMD and refined composite multi-scale entropy
C Zhi and W Yimin
Zhejiang Sci-Tech University, Chin
Rolling bearings are critical components in large rotating machinery such as wind turbines, gas turbines and elevators. Prolonged operation and varying loads can lead to multiple types of failure, especially compound faults (for example simultaneous inner and outer ring defects). The complex nature of compound fault signals, coupled with environmental noise interference, makes traditional diagnostic methods less effective. To address this issue, this paper proposes a rolling bearing compound fault diagnosis method based on vibration signal analysis.
First, a dynamic model of rolling bearing compound faults is established to analyse the dynamic characteristics of different fault modes and extract key fault features. Second, an improved variational mode decomposition (SVMD) combined with refined composite multi-scale entropy is proposed to enhance signal denoising and fault feature extraction. By optimising the fitness function of SVMD, envelope entropy is used as the decomposition criterion and the best intrinsic mode functions (IMFs) are selected using a weighted combination of the logarithmic squared envelope spectrum Gini index and kurtosis. This enhances the effective fault components in the signal. Furthermore, refined composite multi-scale dispersion entropy is calculated to extract features that effectively characterise compound fault conditions.
Then, to address the difficulty in identifying different types of compound fault, a least-squares support vector machine (LSSVM) is employed to construct the fault diagnosis model. Since LSSVM performance is sensitive to parameter selection, an improved whale optimisation algorithm (IWOA) is proposed to optimise the model, significantly improving fault recognition accuracy.
Finally, the effectiveness of the proposed method is validated through rolling bearing compound fault experiments. The results demonstrate that the method accurately identifies various compound fault types and exhibits strong noise resistance and stability.
First, a dynamic model of rolling bearing compound faults is established to analyse the dynamic characteristics of different fault modes and extract key fault features. Second, an improved variational mode decomposition (SVMD) combined with refined composite multi-scale entropy is proposed to enhance signal denoising and fault feature extraction. By optimising the fitness function of SVMD, envelope entropy is used as the decomposition criterion and the best intrinsic mode functions (IMFs) are selected using a weighted combination of the logarithmic squared envelope spectrum Gini index and kurtosis. This enhances the effective fault components in the signal. Furthermore, refined composite multi-scale dispersion entropy is calculated to extract features that effectively characterise compound fault conditions.
Then, to address the difficulty in identifying different types of compound fault, a least-squares support vector machine (LSSVM) is employed to construct the fault diagnosis model. Since LSSVM performance is sensitive to parameter selection, an improved whale optimisation algorithm (IWOA) is proposed to optimise the model, significantly improving fault recognition accuracy.
Finally, the effectiveness of the proposed method is validated through rolling bearing compound fault experiments. The results demonstrate that the method accurately identifies various compound fault types and exhibits strong noise resistance and stability.
