A Domain Adaption ResNet Model to Detect Faults in Roller Bearings Using Vibro-Acoustic Data

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Intelligent fault diagnosis of roller bearings faces challenges with data distribution discrepancies and noisy sensor signals.
• Traditional methods struggle with variations between training and testing datasets and sensor placement limitations.

Purpose of the Study:

• To develop a robust fault diagnosis model for roller bearings that addresses cross-domain data issues and sensor noise.
• To improve the accuracy and transferability of fault diagnosis models using multi-source acoustic and vibration data.

Main Methods:

• A domain adaption residual neural network (DA-ResNet) model was constructed, incorporating maximum mean discrepancy (MMD) and residual connections.
• MMD was employed to minimize distribution discrepancies between source and target domains, enhancing feature transferability.
• Acoustic and vibration signals from three directions were simultaneously sampled for comprehensive data acquisition.

Main Results:

• The DA-ResNet model demonstrated improved feature transferability by minimizing domain distribution discrepancies.
• Experimental results confirmed the necessity of multi-source data for enhanced diagnostic performance.
• The transfer operation significantly improved recognition accuracy in roller bearing fault diagnosis.

Conclusions:

• The proposed DA-ResNet model effectively addresses cross-domain challenges in roller bearing fault diagnosis.
• Utilizing multi-source acoustic and vibration data alongside domain adaptation improves diagnostic robustness.
• This approach offers a promising solution for intelligent fault diagnosis in industrial environments.