Rotating machinery fault diagnosis based on a novel lightweight convolutional neural network

Area of Science:

• Engineering
• Computer Science
• Artificial Intelligence

Background:

• Industry 4.0 and Industrial Internet of Things (IIoT) demand efficient models for fault diagnosis.
• Existing models often have high parameter counts and storage requirements, limiting their deployment.
• Accurate and automatic fault diagnosis of rotating machinery is crucial for industrial maintenance.

Purpose of the Study:

• To propose a lightweight convolutional neural network (LCNN) for intelligent fault diagnosis of rotating machinery.
• To reduce model parameters and storage space while maintaining high diagnostic accuracy.
• To address the 'black box' nature of deep learning models in fault diagnosis.

Main Methods:

• Construction of lightweight convolution blocks using spatial separable convolutions.
• Development of an LCNN model utilizing these lightweight blocks for fault diagnosis.
• Visualization of the network using Tensorboard and t-distribution stochastic neighbor embedding.

Main Results:

• The LCNN model achieved a diagnosis accuracy exceeding 99.9% with 6 lightweight convolutional blocks.
• The proposed model demonstrated significantly decreased parameters, enhancing robustness.
• Effective visualization of the LCNN network was achieved.

Conclusions:

• The LCNN model provides accurate, automatic, and robust fault diagnosis for rotating machinery.
• The reduced model size makes it highly suitable for deployment in Industrial Internet of Things (IIoT) environments.
• The method effectively balances accuracy, efficiency, and interpretability in intelligent fault diagnosis.