Rolling Bearing Fault Diagnosis Using Hybrid Neural Network with Principal Component Analysis

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Fault prognostics and health management (PHM) increasingly utilizes deep learning for rolling bearing fault diagnosis.
• Existing models often lack verified generality and robustness under complex, extreme variable loading conditions.

Purpose of the Study:

• To propose an end-to-end hybrid deep neural network model for intelligent rolling bearing fault diagnosis.
• To enhance model generality and robustness, particularly under challenging variable load scenarios.

Main Methods:

• Principal Component Analysis (PCA) for feature dimensionality reduction and data pre-processing.
• A hybrid deep learning architecture combining Convolutional Neural Network (CNN) for denoising/feature extraction, Bi-directional Long Short-Term Memory (BiLSTM) for time-series feature extraction, and an attention mechanism for optimal weight assignment.

Main Results:

• The model achieved 100% accuracy under constant load and nearly 90% under variable load.
• Demonstrated 72.8% accuracy under extreme variable load conditions (2.205 N·m/s to 0.735 N·m/s and vice versa).
• Performance is comparable to existing deep learning models, with significant improvements in robustness.

Conclusions:

• The proposed hybrid deep learning model exhibits reliable robustness and generality for rolling bearing fault diagnosis.
• The model effectively handles complex and extreme variable loading conditions, advancing PHM capabilities.