The purpose of this study is to delineate a novel Amplitude Factor (AF)-driven paradigm for discerning the most diagnostically salient Intrinsic Mode Functions (IMFs) from tri-axial vibration signals. By integrating univariate and multivariate analytical frameworks, the approach fortifies the feature abstraction, fault discernment and classification within gearbox condition monitoring.
Vibration signals were collected from an industrial multistage gearbox at a constant speed (1,000 RPM) across four loading conditions (0–30 Nm). Both univariate and multivariate signal decomposition techniques CEEMD and multivariate empirical mode decomposition were applied. A novel AF-based IMF selection strategy was proposed and compared with traditional methods like correlation analysis, kurtosis and permutation entropy. Signal-to-Noise Ratio was used as the comparative metric. Fault-informative IMFs were selected to isolate multidimensional features, including Root Mean Square, peak and standard deviation in the time domain and spectral kurtosis and total spectral energy in the frequency domain, thereby enriching the diagnostic framework. A brief computational evaluation was conducted to assess performance in terms of execution time and memory usage.
The proposed AF-based IMF selection method consistently yielded higher Signal-to-Noise Ratio and improved feature quality. The methodology proved highly effective in both univariate and multivariate settings, with Multivariate Empirical Mode Decomposition-based multivariate analysis achieving classification accuracies up to 97% for Multivariate Empirical Mode Decomposition and CEEMD both at elevated load levels. The proposed methodology successfully identified prominent harmonics of the gear mesh frequency such as 2.5 × GMF, 4 × GMF and 5 × GMF which serve as strong indicators of underlying critical faults in the gearbox system.
This study introduces a novel AF for IMF selection and demonstrates the underexplored potential of multivariate decomposition in industrial condition monitoring, offering a more precise and resilient fault diagnosis framework.
The peer review history for this article is available at: https://publons.com/publon/10.1108/ILT-06-2025-0278/
