A Comparative Study of VMD Methods Based on Different Optimization Algorithms in Non-Stationary Signal Decomposition of Bucket Runners

Bucket runners are subjected to periodic discrete jet impacts during high-speed operation, yielding stress signals characterized by strong impact, wide bandwidth, and high non-stationarity, which demand high-frequency separation accuracy. Since VMD decomposition quality depends on the mode number K and the penalty factor α, empirical parameter selection for such non-stationary signals often leads to under-decomposition or modal aliasing. This study employs four metaheuristic algorithms—PSO, WOA, GWO, and SSA—to adaptively optimize VMD parameters. Eight evaluation metrics are established: the optimal mode number, optimized penalty factor, false mode number, over-decomposition score, under-decomposition score, under-decomposition proportion, residual energy proportion, and computation time. The experimental results demonstrate that SSA-VMD yields mode numbers consistent with true signal components and achieves a full over-decomposition score, a 33.33% under-decomposition ratio, and a 3.78% residual energy ratio, exhibiting the best comprehensive performance. In the stress signal decomposition of a hydropower station bucket runner, SSA-VMD extracts eight IMF components. After removing one false mode, the center frequencies of seven effective modes correspond to rotation frequency harmonics and structural natural frequencies, with a residual energy as low as 0.01%. These results verify the method’s effectiveness in non-stationary signal feature extraction.