Toward large scale assessment of railway site conditions using brodaband resonance frequencies from DAS DATA

The modern French railway network is equipped with optical fibers dedicated to telecommunication purposes, among which some remain unused. These so-called dark fibers can be exploited using Distributed Acoustic Sensing technology (DAS) to provide an effective tool for rapid assessment and long-term monitoring of site conditions along railway tracks. We present a methodology applied to a 19 km long DAS array operating under normal railway traffic conditions, highlighting the capability to perform continuous spatial analysis at kilometer scale with measurements every 4.9 meters. Despite the limited coupling associated with the on-conduit installation, corresponding to the standard operational conditions without any modification to the existing infrastructure, time windows selected before and after train passages allow the extraction of the resonance frequencies at each DAS channel, overcoming the low signal-to-noise ratio of the installation setup. Variations of resonance frequencies along the railway reflect changes in near surface soil conditions, related either to shear wave velocity or to variation in impedance contrast depth, with rapid spatial variation observed in karstic areas over only a few tens of meters. The novelty of this work lies in the use of resonance frequencies as a stable and repeatable site parameter derived from DAS data on a large scale. While this information does not quantify site amplification, it provides direct information on the frequency ranges that may be preferentially amplified. This makes them well suited for long term monitoring and for tracking temporal or spatial changes in site conditions under linear infrastructures, and for supporting future strategies to manage infrastructure evolution and time dependent variability. Seasonal variations of these resonance frequencies have been observed, further supporting this method as an effective tool for continuous site monitoring.