Fiber Optic Train Monitoring with Distributed Acoustic Sensing: Conventional and Neural Network Data Analysis.

Area of Science:

• Engineering
• Geophysics

Background:

• Distributed acoustic sensing (DAS) offers a scalable solution for monitoring extensive railway networks.
• Large, noisy datasets from DAS necessitate optimized algorithms for accurate train parameter extraction.

Purpose of the Study:

• To compare various data analysis strategies for DAS railway monitoring.
• To evaluate parameter uncertainties associated with different analysis methods.

Main Methods:

• Utilizing a commercial DAS system to record data from a Deutsche Bahn AG ICE 4 train.
• Localizing train signals along temporal and spatial directions.
• Applying peak finding and neural network algorithms for data enhancement.

Main Results:

• Both temporal and spatial localization methods achieved a velocity standard deviation below 5 km/h at 160 km/h.
• Enhanced data revealed individual bogie clusters, enabling train car counting.
• Determining velocity from bogie signals yielded a lower standard deviation of 0.8 km/h.

Conclusions:

• Optimized DAS data analysis provides precise train tracking capabilities.
• The presented methods support the development of real-time train monitoring and integrity analysis systems.
• Accurate train car counting and velocity determination enhance railway safety and operational efficiency.