A Deep Learning Method for Bearing Cross-Domain Fault Diagnostics Based on the Standard Envelope Spectrum

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Deep learning models for equipment fault diagnostics offer reliability but face challenges in cross-domain prediction accuracy.
• Existing methods often fail when applied to data from different operating conditions or equipment variations.

Purpose of the Study:

• To propose a novel deep learning fault diagnosis method for rolling bearings that enhances cross-domain diagnostic capabilities.
• To improve the prediction accuracy of fault diagnosis models when applied to data from different rotational speeds and bearing types.

Main Methods:

• A standard envelope spectrum is constructed based on the morphology of rolling bearing failure signatures.
• This reconstructed envelope spectrum is used to eliminate domain-specific differences (e.g., speed, model).
• A convolutional neural network is employed for feature learning and fault classification using the processed data.

Main Results:

• The proposed method demonstrated high diagnostic accuracy on diverse datasets, including publicly available and self-experimented data.
• It effectively handled variations in rotational speeds and bearing types.
• Performance was superior compared to several popular feature extraction techniques.

Conclusions:

• The reconstructed envelope spectrum deep learning method is effective for cross-domain fault diagnostics of rolling bearings.
• This approach significantly improves diagnostic accuracy across different operating conditions and equipment variations.
• The method provides a robust solution for reliable equipment operation through enhanced fault prediction.