Dual-FBG bearing fault probe based on a CNN-LSTM-encoder network

Area of Science:

• Mechanical Engineering
• Optical Sensing
• Artificial Intelligence

Background:

• Bearing faults are critical in machinery, necessitating accurate and early detection.
• Traditional vibration sensing methods have limitations in frequency response and data accuracy.
• Developing compact, high-performance probes is essential for industrial monitoring.

Purpose of the Study:

• To propose a centimeter-sized bearing fault probe utilizing dual-fiber Bragg gratings.
• To enhance vibration measurement capabilities with a wider frequency range and improved accuracy.
• To develop a robust classification method for bearing faults under variable working conditions.

Main Methods:

• A dual-fiber Bragg grating vibration sensing probe was designed and implemented.
• Swept source optical coherence tomography and synchrosqueezed wavelet transform were employed for vibration measurement.
• A hybrid deep learning model combining Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and Transformer encoder was developed for fault classification.

Main Results:

• The proposed probe demonstrated multi-carrier heterodyne vibration measurements.
• The system achieved a wider vibration frequency response range and collected more accurate data.
• The CNN-LSTM-Transformer model achieved a 99.65% accuracy rate in bearing fault classification across variable conditions.

Conclusions:

• The centimeter-sized dual-fiber Bragg grating probe offers a promising solution for advanced bearing fault detection.
• The integrated sensing and AI-based analysis method significantly improves the accuracy and reliability of fault classification.
• This technology has the potential to enhance predictive maintenance and reduce machinery downtime.