Fault diagnosis of rolling bearings using hybrid-domain features and dung beetle optimization-based support vector machine

Abstract

To address the challenges of insufficient feature extraction and poor robustness in rolling bearing fault diagnosis, this paper proposes a novel method termed Hybrid-Domain Features and DBO-Optimized SVM (Hybrid Features-DBO-SVM). Vibration signals are first preprocessed using Variational Mode Decomposition (VMD), with its key parameters optimized by the Osprey–Cauchy-enhanced Sparrow Search Algorithm (OCSSA) for effective denoising. Subsequently, Refined Composite Multiscale Permutation Entropy (RCMPE) is extracted to quantify signal complexity. Statistical time-domain features (e.g., kurtosis and peak value) are then fused with RCMPE to construct a discriminative hybrid-domain feature vector characterizing diverse bearing conditions. To mitigate the parameter sensitivity of the Support Vector Machine (SVM), the Dung Beetle Optimization (DBO) algorithm is employed to adaptively optimize its hyperparameters. Validation on the Jiangnan University bearing dataset demonstrates a diagnostic accuracy of 97.2%, surpassing comparative models including K-Nearest Neighbor (KNN), Random Forest (RF), Decision Tree (DT), Long Short-Term Memory (LSTM), and alternative feature extraction methods in both computational efficiency and classification precision. Further validation on the Case Western Reserve University (CWRU) dataset confirms the method’s robustness, advancement, and capability for rapid, accurate fault identification, providing a novel and robust solution for bearing condition monitoring and fault diagnosis.