Looseness Identification of Track Fasteners Based on Ultra-Weak FBG Sensing Technology and Convolutional Autoencoder...

Area of Science:

• Civil Engineering
• Mechanical Engineering
• Signal Processing

Background:

• Subway fastener looseness poses a significant risk of train derailment.
• Current methods for identifying loose fasteners are often random and insufficient.
• Vibration responses from track beds offer potential for detecting fastener issues.

Purpose of the Study:

• To develop and validate a novel method for identifying loose subway fasteners.
• To leverage convolutional autoencoder (CAE) networks for analyzing vibration signals.
• To enhance the safety and reliability of subway infrastructure through early detection.

Main Methods:

• A field experiment was conducted on an actual subway line to collect vibration data from loosened fasteners.
• Vibration signals were converted into 2D images using a pseudo-Hilbert scan.
• A two-stage convolutional autoencoder (CAE) network was employed for feature extraction and recognition.

Main Results:

• The proposed pseudo-Hilbert scan combined with CAE achieved superior performance compared to raster and Hilbert scans.
• Accuracy, precision, recall, and F1-score showed significant improvements (at least 23.8%, 9.5%, 20.0%, and 21.1% higher, respectively).
• t-distributed stochastic neighbor embedding (t-SNE) visualization confirmed the method's strong ability to distinguish loose fastener features.

Conclusions:

• The CAE-based method with pseudo-Hilbert scan is effective for identifying loose subway fasteners.
• This approach offers a reliable and accurate solution for real-time monitoring of subway track integrity.
• The findings contribute to improving subway safety by enabling proactive maintenance.