Bearing Fault Diagnosis Using Multidomain Fusion-Based Vibration Imaging and Multitask Learning.

Area of Science:

• Mechanical Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Bearing fault diagnosis traditionally relies on domain expertise and limited feature extraction methods (time, frequency, or time-frequency domains).
• Vibration signals from bearing faults are complex, non-linear, and non-stationary, especially under variable operating conditions, posing challenges for existing techniques.

Purpose of the Study:

• To develop an autonomous diagnostic system for bearing fault identification that overcomes the limitations of traditional methods.
• To enable accurate fault detection under variable speed and load conditions.

Main Methods:

• A novel signal-to-image transformation technique, multi-domain fusion-based vibration imaging (MDFVI), is proposed to create composite images from raw time-domain signals, spectrum, and envelope spectrum.
• A convolutional neural network (CNN)-aided multitask learning (MTL) architecture is developed to process MDFVI images for fault identification.
• The system is trained and validated on two benchmark bearing datasets.

Main Results:

• The MDFVI technique effectively generates unique patterns for bearing faults, even under variable speeds and loads.
• The proposed MTL-based CNN architecture successfully identifies bearing faults concurrently under variable speed and health conditions.
• Experimental results demonstrate superior performance compared to state-of-the-art methods on both benchmark datasets.

Conclusions:

• The developed autonomous diagnostic system integrating MDFVI and MTL-CNN offers a robust solution for bearing fault diagnosis.
• The proposed approach significantly enhances diagnostic accuracy and reliability, particularly in challenging variable operating environments.
• This method represents a significant advancement in condition monitoring and predictive maintenance for rotating machinery.