Vibration sensor data denoising using a time-frequency manifold for machinery fault diagnosis

Area of Science:

• Mechanical Engineering
• Signal Processing
• Data Science

Background:

• Vibration sensor data is crucial for machinery fault diagnosis.
• Background noise often hinders the accurate identification of fault signatures.
• Existing denoising methods may struggle with complex non-stationary signals.

Purpose of the Study:

• To introduce a novel denoising method for machinery fault diagnosis using the time-frequency manifold (TFM) concept.
• To enhance the reliability of fault signature identification in noisy vibration sensor data.
• To demonstrate the effectiveness of the TFM-based method compared to traditional approaches.

Main Methods:

• Introduction of the time-frequency manifold (TFM) concept for signal analysis.
• Development of a denoising method by synthesizing the TFM using time-frequency synthesis and phase space reconstruction (PSR).
• Evaluation using a clustering-based statistical parameter and a new diagnostic approach, frequency probability time series (FPTS) spectral analysis.

Main Results:

• The proposed TFM denoising method achieves satisfactory noise suppression while preserving the inherent time-frequency structure of the signal.
• The method demonstrates superior performance in machinery fault diagnosis compared to two traditional denoising techniques.
• Analysis of vibration sensor data from defective bearings validates the effectiveness of the TFM approach.

Conclusions:

• The TFM-based denoising method offers a significant advancement for reliable machinery fault diagnosis.
• The approach effectively handles noise in vibration data, enabling clearer identification of fault signatures.
• The TFM concept provides a robust framework for analyzing non-stationary signals in mechanical systems.