Research on Gear Fault Detection Based on VMD Optimization

Reliable gear-fault diagnosis requires transparent signal-processing and evaluation procedures. This study reports an adaptive secretary bird optimization algorithm (ASBOA) coupled with variational mode decomposition (VMD) and a bidirectional gated recurrent unit (BiGRU). ASBOA replaces the fixed SBOA stage schedule with probability-based operator selection using population diversity, convergence, and iteration progress. In 30 archived benchmark runs, the retained p-values were below the nominal 0.05 threshold for Sphere, Rosenbrock, and Ackley, but not for Rastrigin (p = 0.0748). Means and standard deviations are reported separately for each condition, but the statistical test used and the raw per-run values were not retained. The archived VMD records include a decomposition example labeled K = 8 and α = 2500 and a separate parameter-trajectory plot approaching α = 350; the files do not establish that these outputs came from the same configuration. The available classification records report best and mean accuracies of 0.840 and 0.787 for ASBOA-VMD-BiGRU, but the numbers of experiments per condition were unequal, and the training and test matrices were drawn from different sample sets. The archive also lacks the component-wise strategy-score mappings, the complete VMD objective and search bounds, and the full acquisition and BiGRU training protocol. Accordingly, the retained results are presented as descriptive evidence from completed experiments. No claim of universal superiority, quantified noise robustness, EMD superiority, or fair cross-model classification is made.