Wavelet packet base selection for gearbox defect severity classification

Effective and efficient representation of the time-domain sensor data is critical to reliable signal discrimination and defect severity classification. This paper presents a wavelet packet base-selection approach that explores the Local Discriminant Bases (LDB) method. An optimal set of time-frequency subspaces are selected from a library of redundant wavelet packet subspaces to produce discriminant features that are capable of discriminating different classes. Selection of the wavelet packet subspaces contributes to constructing the best orthogonal base that enhances the accuracy of classification. Simulation and analysis of vibration data from a gearbox demonstrate that the developed signal processing method is well-suited for gearbox defect severity classification.