Multilevel Dimensionality Reduction Combined with IBFO-Optimized VMD-SVM for Bearing Fault Classification

To address the challenges of excessive fault features and low identification accuracy in motor bearings, this paper proposes a bearing fault identification method integrating multilevel dimensionality reduction with Improved Bitterling Fish Optimization (IBFO)-optimized Variational Mode Decomposition (VMD) and Support Vector Machine (SVM). Specifically, Tent chaotic mapping is introduced to optimize BFO's initial population and mitigate local optima entrapment, while convergence parameters are refined to accelerate convergence and enhance robustness, and Cauchy variation is incorporated to improve local search capability. For VMD requiring tuning of decomposition layer count K and penalty coefficient α, IBFO optimizes these parameters to enhance decomposition performance, with features from optimal components forming eigenvectors subjected to multilevel dimensionality reduction. Furthermore, IBFO optimizes penalty factor C and kernel parameter g of SVM to boost recognition accuracy. Experimental results demonstrate 7.29% accuracy improvement through algorithmic parameter optimization and an additional 9.369% gain via multilevel dimensionality reduction, with simulations confirming its efficacy in enhancing bearing fault identification accuracy.