Pre-Processing Method to Improve Cross-Domain Fault Diagnosis for Bearing

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Domain shift, caused by distribution differences between datasets, impedes accurate model performance in cross-domain scenarios.
• Bearing fault diagnosis is susceptible to domain shifts due to variations in operating conditions and system specifications.
• Existing cross-domain diagnosis methods struggle with different bearing systems due to signal characteristic variations.

Purpose of the Study:

• To develop a pre-processing method for domain adaptation in bearing fault diagnosis.
• To improve the accuracy of cross-domain fault diagnosis without requiring prior knowledge of fault frequencies.
• To validate the proposed method using vibration signals from distinct bearing systems.

Main Methods:

• Signals from two different bearing systems (Case Western Reserve University and Paderborn University datasets) were utilized.
• A novel pre-processing technique transformed vibration signals into a common pattern space.
• One-dimensional Convolutional Neural Network (1D-CNN) models were employed for diagnosis verification.

Main Results:

• The proposed pre-processing method significantly improved domain adaptation capabilities.
• Cross-domain fault diagnosis achieved high accuracy even with differing bearing system specifications.
• Comparison showed superior performance of the pre-processed dataset over the raw dataset.

Conclusions:

• The developed pre-processing method effectively addresses domain shift challenges in bearing diagnosis.
• This approach enables accurate cross-domain fault diagnosis for diverse rotating machinery.
• The method offers a robust solution for intelligent diagnosis systems facing varying operational and system conditions.