Bearing Fault Diagnosis Method Based on Deep Convolutional Neural Network and Random Forest Ensemble Learning.

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Data-driven bearing fault diagnosis is crucial for machinery health monitoring.
• Existing methods struggle with feature representation and domain shift issues in real-world applications.
• Deep learning and ensemble learning offer potential solutions for improved accuracy and generalization.

Purpose of the Study:

• To propose a novel bearing fault diagnosis method combining deep convolutional neural networks (CNN) and random forest (RF) ensemble learning.
• To address the limitations of existing methods in feature extraction and domain adaptation.
• To enhance the accuracy and robustness of bearing fault diagnosis.

Main Methods:

• Vibration signals are transformed into 2D grayscale images using continuous wavelet transform (CWT).
• A LeNet-5 based CNN model is employed for automatic multi-level feature extraction from the images.
• An ensemble of multiple RF classifiers diagnoses bearing faults using the extracted multi-level features, combining local and global information.

Main Results:

• The proposed method demonstrates high accuracy in bearing fault diagnosis under complex operational conditions.
• Experimental validation on two distinct datasets (reliance electric motor and rolling mill) confirms the method's effectiveness.
• The approach outperforms traditional methods and standard deep learning techniques in diagnostic performance.

Conclusions:

• The integrated CNN and RF ensemble learning method provides a powerful tool for data-driven bearing fault diagnosis.
• Combining multi-level features, particularly low-level features with detailed characteristics, significantly improves diagnostic accuracy.
• The proposed method offers a robust and superior solution for real-world bearing fault detection challenges.