Fault Diagnosis for Rotating Machinery Using Multiscale Permutation Entropy and Convolutional Neural Networks

Area of Science:

• Mechanical Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Existing rotating machine fault diagnosis methods often struggle with single-scale signal analysis.
• Effective fault detection requires analyzing complex vibration data from multiple sensors.

Purpose of the Study:

• To develop an advanced fault diagnosis method for rotating machinery.
• To overcome limitations of single-scale analysis by integrating multi-scale entropy and multi-channel deep learning.

Main Methods:

• Multi-scale permutation entropy (MPE) was used to analyze vibration signals at various scales, generating permutation entropy (PE) feature vectors.
• A multi-channel fusion convolutional neural network (MCFCNN) was designed, with each channel processing MPE features from individual sensors.
• Unsupervised learning within MCFCNN extracted and fused features from multiple channels before final fault identification via multi-layer perceptron.

Main Results:

• The proposed MPE and MCFCNN method demonstrated high accuracy in diagnosing faults in planetary gearboxes and rolling bearings.
• Comparative analysis showed superior performance over single-channel CNN and existing CNN-based fusion techniques.
• The method achieved significant improvements in diagnostic accuracy, stability, and speed.

Conclusions:

• The integration of MPE and MCFCNN offers a robust and efficient solution for rotating machine fault diagnosis.
• This approach effectively leverages multi-scale information and spatial correlations from sensor data.
• The developed method provides a promising advancement for predictive maintenance in rotating machinery.