Hybrid Rubbing Fault Identification Using a Deep Learning-Based Observation Technique
IEEE Transactions on Neural Networks and Learning Systems
|October 8, 2020
Summary
This study introduces a hybrid deep learning and control theory method for early detection of turbine rub-impact faults. The approach enhances vibration signal analysis for accurate fault diagnosis in industrial settings.
Area of Science:
- Turbomachinery diagnostics
- Nonlinear system analysis
- Artificial intelligence in engineering
Background:
- Rub-impact faults in turbines are complex, nonlinear issues.
- Early detection is challenging due to computational costs of traditional methods.
- Existing techniques struggle with nonlinear signal characteristics.
Purpose of the Study:
- To develop a computationally efficient hybrid approach for diagnosing turbine rub-impact faults.
- To improve the accuracy of fault detection in early stages.
- To create a scalable framework suitable for industrial applications.
Main Methods:
- System modeling using autoregressive with eXogenous input Laguerre (ARX-Laguerre) technique.
- Employing an ARX-Laguerre proportional-integral observer (PIO) to enhance vibration signal estimation.
- Applying a deep neural network to PIO output for fault diagnosis.
Main Results:
- The hybrid approach effectively diagnoses rubbing faults of varying intensities.
- Improved fault differentiation capabilities for complex nonlinear signals.
- High fault classification accuracy achieved, demonstrating robustness.
Conclusions:
- The proposed hybrid framework offers a practical solution for industrial turbine monitoring.
- Combines control theory and deep learning for superior fault diagnosis.
- Suitable for real-world applications requiring accurate and efficient fault detection.
