A Multitask-Aided Transfer Learning-Based Diagnostic Framework for Bearings under Inconsistent Working Conditions.

Area of Science:

• Mechanical Engineering
• Machine Learning
• Signal Processing

Background:

• Sudden bearing failure in rotating machinery causes significant economic losses and safety risks.
• Traditional fault diagnosis methods struggle with the non-stationary and non-linear nature of vibration signals.
• Existing deep learning approaches face challenges in crack severity reasoning and simultaneous multi-health characteristic identification.

Purpose of the Study:

• To propose a novel bearing fault diagnosis framework robust to inconsistent operating conditions.
• To simultaneously identify machine speed and bearing health conditions.
• To accurately diagnose bearing faults with varying crack severities using transfer learning.

Main Methods:

• Preprocessing vibration signals using higher-order spectral analysis to extract distinct fault patterns.
• Employing a multitask learning (MTL)-based convolutional neural network (CNN) to process bispectral inputs.
• Integrating transfer learning (TL) to enhance fault identification across multiple crack severities.

Main Results:

• The proposed framework effectively identifies bearing faults under variable motor speeds, loads, and noise.
• Simultaneous identification of machine speed and health conditions was achieved.
• The TL-based approach demonstrated superior performance in diagnosing faults with multiple crack severities compared to state-of-the-art methods.

Conclusions:

• The developed diagnostic framework offers a superior solution for bearing fault diagnosis under challenging, inconsistent operating conditions.
• The combination of higher-order spectral analysis, MTL, and TL provides a robust and accurate diagnostic tool.
• This approach addresses key limitations of existing methods, enhancing machine reliability and safety.