A Fuzzy Fusion Rotating Machinery Fault Diagnosis Framework Based on the Enhancement Deep Convolutional Neural...

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Deep learning algorithms for rotating machinery fault diagnosis often rely on single signal features, leading to information loss.
• Robust nonlinear regression properties of artificial intelligence algorithms are valuable for fault diagnosis.

Purpose of the Study:

• To develop an enhanced fault diagnosis model for rotating machinery by overcoming limitations of single-feature deep learning approaches.
• To improve information utilization from vibration signals through multi-feature extraction and fusion.

Main Methods:

• Utilized Fast Fourier Transform (FFT), Short-Time Fourier Transform (STFT), and direct data slicing to create four distinct datasets from raw vibration signals.
• Developed four enhanced Convolutional Neural Networks (CNN) models, each processing a different data representation.
• Implemented a fuzzy fusion strategy to integrate outputs from the four CNN models, analyzing classifier importance and interactions.

Main Results:

• The proposed model demonstrated effective feature extraction capabilities on both artificial and real-world bearing fault datasets.
• The fuzzy fusion strategy successfully integrated information from diverse signal processing methods.
• The method exhibited good anti-noise characteristics and interpretability.

Conclusions:

• The multi-feature fusion approach using enhanced CNNs offers a superior alternative to single-feature methods in rotating machinery fault diagnosis.
• The fuzzy fusion strategy provides a more nuanced integration of classifier outputs compared to conventional methods.
• The developed model shows promise for practical applications in industrial machinery health monitoring.