Improved VMD with hybrid grey wolf optimization for bearing signal fault analysis

• MBGWO, which improves the standard Grey Wolf Optimizer through Latin Hypercube Sampling for diverse population initialization, a “vision” mechanism inspired by Beetle Antennae Search to enhance local search, and a nonlinear convergence factor for balanced global and local exploration. We validated the effectiveness of the improved method through benchmark testing functions. • An improved VMD framework incorporating frequency distribution-guided center frequency updates, using exponential weighting to stabilize mode frequency estimation against noise. • An adaptive suppression mechanism for low-energy modes, minimizing redundant components and computational overhead. • Utilization of envelope entropy as a fitness function to optimize VMD parameters (number of modes and penalty factor), ensuring high-quality signal decomposition. Applied to simulated and real bearing signals, MBGWO-FA-VMD outperforms traditional VMD and GWO-optimized VMD, effectively extracting fault features and improving early fault detection in rotating machinery. At the same time, the signal decomposition and running time were compared under different signal-to-noise ratios and different numbers of sampling points. Variational Mode Decomposition (VMD) is effective for analyzing mechanical vibration signals, but its performance is hindered by noise interference and sensitivity to parameters like the number of mode components and penalty factor. This study aims to enhance VMD for bearing fault analysis. We propose a Mixed Grey Wolf Optimizer (MBGWO) integrating Latin Hypercube Sampling, a Beetle Antennae Search-inspired “vision” mechanism, and a nonlinear convergence factor to optimize VMD parameters. The VMD framework is improved with frequency distribution-guided center frequency updates and energy-adaptive mode suppression. The MBGWO-optimized VMD, coupled with envelope spectrum analysis, effectively decomposes bearing vibration signals and extracts early fault features. Simulation and experimental results on synthetic and real bearing datasets demonstrate the proposed method’s superior performance and robustness compared to traditional VMD, offering significant potential for early fault detection in rotating machinery