Rolling Bearing Fault Diagnosis Based on Integrated Features of Vibration Signals and Support Vector Machine

Addressing the challenge of traditional time-frequency domain feature extraction methods in effectively capturing key information, this paper employs the energy values obtained through wavelet packet decomposition (WPD), along with the correlation dimension, to form the integrated features. An innovative fault diagnosis approach for rolling bearings is introduced, leveraging integrated features and employing support vector machine (SVM) for classification. Firstly, the original signal is decomposed using WPD, and then the energy features of the resulting frequency bands are calculated. The chaotic features embedded in the vibration signals are characterized by the correlation dimension, computed using the Grassberger-Procaccia (GP) algorithm. Subsequently, the energy features and the correlation dimension are combined to form the integrated features. To address the challenges of the traditional SVM model, which can often succumb to local optima and overfitting during classification, a novel approach utilizing Bayesian optimization for SVM parameter tuning is introduced. The method first constructs integrated features through calculation and then inputs these feature vectors into an SVM optimized using the Bayesian algorithm to achieve fault diagnosis. The method was tested on the Case Western Reserve University (CWRU) dataset, achieving a prediction accuracy of 97.8%. The experimental results confirm the practicality of the method proposed in this paper, and at the same time open up a new path for rolling bearing fault diagnosis within the field of underwater equipment.