Fig 1.
Principle of matrix pencil mean frequency extraction.
Table 1.
Motor and bearing setup.
Table 2.
Sensor setup and data collection.
Table 3.
Experimental setup and data acquisition.
Fig 2.
Reconstructed vibratory signal using the matrix pencil method (M = 50 poles, SL = 300).
Fig 3.
Reconstructed vibratory signal using the matrix pencil method (M = 50 poles, SL = 300) for the healthy bearing.
Fig 4.
Results of the MPMF vibratory analysis for a healthy bearing state.
Fig 5.
Results of the MPMF vibratory analysis for bearings with an inner race fault.
Fig 6.
Results of the MPMF vibratory analysis for bearings with an outer race fault.
Fig 7.
Results of the MPMF vibratory analysis for bearings with a cage fault.
Fig 8.
Results of the MPMF analysis for bearings with a ball fault.
Fig 9.
Examples of MPMF vibratory shapes corresponding tovarious bearing conditions, including developing and fully developed faults.
Fig 10.
Proposed method for introducing additive white Gaussian noise to each MPMF vector to produce new MPMF vectors.
Table 4.
Performance summary of MLP classifier across seven different datasets.
Fig 11.
Confusion matrix for MLP classifier based on MPMF (Database cases 1, 2, 3).
Fig 12.
Confusion matrix for MLP classifier based on MPMF (Database cases 4, 5, 6, 7).
Table 5.
Classification results of the MLP model across seven datasets under various noisy environment conditions.
Fig 13.
Confusion matrix for MLP classifier based on MPMF (Database cases 1, 2, 3, 7).
Fig 14.
Confusion matrix for MLP classifier based on MPMF (Database cases 4, 5, 6).
Table 6.
Classification accuracies (%) on the CWRU bearing dataset [57,58].
Table 7.
Comparison of signal-processing and machine-learning techniques for fault detection.
Table 8.
5-Fold cross-validation results under different noise levels.