Fault detection and analysis for wheelset bearings via improved explicit shift-invariant dictionary learning

Area of Science:

• Mechanical Engineering
• Signal Processing
• Condition Monitoring

Background:

• Wheelset bearings are critical for high-speed train operation.
• Extracting fault impulse signals from noisy vibration data is challenging for bearing diagnostics.

Purpose of the Study:

• To propose an improved explicit shift-invariant dictionary learning (IE-SIDL) method for accurate extraction of wheelset bearing fault impulse signals.
• To enhance the efficiency and accuracy of fault detection in wheelset bearings.

Main Methods:

• Developed an IE-SIDL method utilizing circulant matrices for shift-invariant dictionary construction.
• Implemented a three-flips method for fast dictionary construction and frequency-domain reconstruction for dictionary updates.
• Employed an indicator-guided subspace pursuit (SP) method based on sparsity of envelope spectrum (SES) for improved sparse coding.

Main Results:

• The IE-SIDL method demonstrated excellent capacity in extracting fault impulse signals from simulated and experimental wheelset bearing vibration data.
• The processed signals exhibited improved time- and frequency-domain characteristics, facilitating fault detection.
• The method effectively overcomes noise and irrelevant components present in vibration signals.

Conclusions:

• The proposed IE-SIDL method is effective for extracting wheelset bearing fault impulse signals.
• The enhanced signal processing aids in reliable fault detection and behavior analysis of high-speed train wheelset bearings.
• This approach offers a robust solution for condition monitoring in critical railway infrastructure.