An information-based K-singular-value decomposition method for rolling element bearing diagnosis.

Area of Science:

• Mechanical Engineering
• Signal Processing
• Condition Monitoring

Background:

• Sparse coding, particularly K-singular value decomposition (K-SVD), is widely used for fault feature extraction in rolling element bearings (REBs).
• Original K-SVD has limitations due to its dictionary update mechanism, which can reduce its effectiveness in identifying periodic impulses amidst complex interferences.

Purpose of the Study:

• To address the limitations of traditional K-SVD for fault diagnosis in REBs.
• To propose an enhanced information-based K-SVD (IK-SVD) method for more accurate fault signature extraction from degraded vibration signals.

Main Methods:

• Developed an information-based atom selection strategy (IASS) for dictionary updating in K-SVD.
• Introduced a harmonic ratio (HR) index as the evaluation criterion for selecting fault-related atoms.
• Implemented a residual estimation method for determining the threshold in the sparse coding stage.

Main Results:

• The proposed IK-SVD method effectively extracts fault signatures from complex vibration signals.
• Validation on simulated and practical locomotive bearing test rig data demonstrated the superiority of IK-SVD.
• The method shows good performance in fault detection of REBs, outperforming traditional approaches.

Conclusions:

• The IK-SVD method, incorporating an information-based atom selection strategy and harmonic ratio, significantly improves fault feature extraction for REBs.
• This approach offers a more robust solution for diagnosing bearing faults under challenging operational conditions.