Detection of weak fault using sparse empirical wavelet transform for cyclic fault

Area of Science:

• Mechanical Engineering
• Signal Processing
• Condition Monitoring

Background:

• Early fault detection in rolling element bearings is crucial for predicting remaining useful life.
• Effective fault diagnosis relies on accurate signal processing for fault signature extraction.
• Existing methods may struggle with enhancing weak fault signals amidst system noise.

Purpose of the Study:

• To propose a novel diagnostic model for improving the signal-to-noise ratio (SNR) in fault detection.
• To enhance the extraction of fault signatures from rolling element bearings and gearboxes.
• To enable early fault detection for predictive maintenance applications.

Main Methods:

• Utilizing a kurtogram to identify the fault frequency band and filter system noise.
• Applying sparse-based Empirical Wavelet Transform (EWT) for signal processing.
• Implementing l1-regularized sparse regression for frequency domain defect signal analysis.

Main Results:

• Demonstrated significant improvement in the signal-to-noise ratio (SNR) of fault signals.
• Successfully extracted fault signatures for cyclic faults.
• Validated applicability for detecting faults in bearings and gearboxes.

Conclusions:

• The proposed sparse-based EWT diagnostic model effectively enhances fault signal detection.
• The method offers a robust approach for early fault identification in rotating machinery.
• Improved SNR facilitates more reliable remaining useful life predictions.