Distributed Fiber Optic Sensors (DFOSs) for Structural Health Monitoring (SHM) of Railway Infrastructure: A Critical Review

Abstract

Distributed fiber optic sensing (DFOS) has emerged as a transformative technology for structural health monitoring (SHM) of railway infrastructure, offering continuous, high-resolution measurements along extended optical fiber lengths, capabilities that conventional point sensors such as strain gauges and accelerometers cannot match. This review critically examines DFOS technology and its railway SHM applications, covering system components, interrogator units, optical fiber cables, and data acquisition systems, alongside the three principal scattering mechanisms: Rayleigh, Brillouin, and Raman, each offering distinct trade-offs in spatial resolution, sensing range, and measurand sensitivity. Field applications across track and sleeper monitoring, bridge health evaluation, tunnel lining assessment, and embankment stability are reviewed and critically compared. The integration of artificial intelligence (AI) and machine learning (ML) with DFOS data streams is discussed, demonstrating detection accuracy exceeding 97% in recent studies. Its main application rail embankment monitoring is discussed. Key challenges are identified, including high interrogator costs, large data volumes, installation complexity in retrofit scenarios, and environmental noise under operational train speeds. Future research priorities include lower-cost interrogation hardware, automated signal processing pipelines, digital twin integration, and standardized performance frameworks to accelerate large-scale adoption across railway networks worldwide.

1. Introduction

SHM has evolved significantly over the past century in rail infrastructures [1]. It has been transitioning from rudimentary manual inspections to sophisticated automated systems supported by AI that enhance the safety, longevity, and cost-efficiency of critical infrastructure [2]. The demand for advanced SHM systems has intensified as conventional inspection approaches have proven insufficient for ensuring the long-term durability of key structural elements such as rail, ballast, and sleepers [3,4,5]; particularly in critical zones such as bridges, tunnels, embankments and switch and crossings. The mid-20th century saw a shift with the use of electronic sensors (like strain gauges and accelerometers) which improved precision but still only monitored specific points, missing broader and distributed or subtler damage [6,7]. Recent advances in real-time SHM systems, powered by smarter sensors and data analytics such as MEMS, wireless sensor networks, and distributed fiber optic sensing, have transformed the field [8]. These systems offer lower electricity consumption, longer life span, and more durability and some offer continuous monitoring, enabling early damage detection, lowering maintenance costs, and improving safety—especially critical for rail tracks [9,10,11,12,13,14].

1.1. Railway Monitoring Challenges and Modern SHM Approaches

Track superstructure components (rails, fasteners, sleepers/ties, ballast) and substructure must be monitored, as well as associated civil works (bridges, tunnels, platforms, station buildings). Common defects include accumulated rail strains, track buckling or settlements, ballast degradation, misalignments and cracks. Any failure, whether in a sleeper, turnout, or a bridge span, can be catastrophic [1,15,16], interrupting operations or endangering passengers. Therefore, maintaining safety extends service life [17]. Traditional railway monitoring has relied on periodic inspections and localized detectors [17,18,19]. For decades, railways used manual surveys and equipment such as track geometry cars, wheel/axle counters or track circuits to identify problems. At the component level, point sensors, such as accelerometers, strain gauges, inclinometers, and acoustic/magnetic detectors, have been installed in select locations. However, those discrete measurements are labor-intensive and cannot easily cover thousands of kilometers continuously. Modern SHM is transitioning to more integrated solutions: for example, dense fiber optic sensing cables and wireless sensor networks can monitor long track sections in real time, while autonomous inspection vehicles (instrumented rail cars or drones) can collect high-resolution data across the network. By exploring current technologies, relevant case studies, and emerging trends, this paper underscores DFOS as a pivotal tool for ensuring the safety, resilience, and longevity of railway and pipeline infrastructure, where uninterrupted monitoring is essential to prevent failures and optimize maintenance.

1.2. Fiber Optic Sensing Applications in SHM

Optical fiber sensors detect changes in light properties—such as intensity, wavelength, phase, scattering, or polarization—caused by external physical parameters like strain, temperature, or pressure [3,7,17,20,21,22,23,24]. These changes serve as indirect indicators of structural behavior and integrity.

Table 1. Types of optical fiber sensors with respect to their principle of operation.

Sensor Type	Measurement Principle	Example	Data Type	Spatial Coverage	Measurement Capability	Typical Applications	Key Limitation	Ref.
Intensity-Based	Detects changes in transmitted or reflected light intensity caused.	Microbending sensors.	Discrete or semi-continuous (often analog).	Localized (single or multiple points).	Strain, displacement, curvature, pressure, crack detection.	Bridges, concrete structures, simple strain monitoring, leak detection.	Low accuracy, susceptible to signal loss and environmental interference.	[22,25,26,45,46]
Wavelength-Based	Uses periodic variations in the fiber’s refractive index to reflect specific wavelengths, sensitive to external parameter changes.	Fiber Bragg grating (FBG).	Discrete (typically digital).	Localized (single or multiple points).	Strain, temperature, pressure at discrete points.	Bridges, buildings, aircraft structures (FBG widely used).	Limited to point sensing, requires multiple sensors for large areas.	[20,21,26,27,28,29,47,48,49]
Phase-Modulated	Detects phase differences in light split and recombined across two fiber paths.	Examples: Mach–Zehnder interferometer, Sagnac interferometer, Fabry–Pérot interferometer (FPI).	Continuous (analog or digital) for Mach–Zehnder/Sagnac; discrete (typically digital) for FPI.	Localized or short-range segments (Mach–Zehnder, Sagnac); localized (single point) for FPI.	Vibration, strain, acoustic signals (Mach–Zehnder); high-precision strain, pressure, displacement (FPI).	Pipelines, perimeter security, seismic monitoring (Mach–Zehnder); tunnels, dams, composite materials (FPI).	Sensitive to environmental noise, requires stabilization (Mach–Zehnder); complex setup, costly for large-scale monitoring (FPI).	[26,29,30,31,32,33,34,35,36,37,38,50]
Scattering-Based	Uses scattering effects (e.g., Rayleigh, Brillouin, Raman) to measure parameters continuously along the entire fiber length.	Distributed fiber optic sensors (DFOSs).	Continuous (typically digital).	Extensive (kilometers).	Strain, temperature, deformation over long distances.	Railways, pipelines, bridges, geotechnical structures.	Higher initial cost requires specialized analysis equipment.	[9,17,39,40,41,42,43,44,51,52,53,54]
Polarization-Based	Detects changes in light polarization due to stress or magnetic fields.	Polarimetric sensors.	Discrete or continuous (analog or digital).	Localized or short-range segments.	Strain, stress, magnetic field detection.	Power lines, composite materials, niche SHM.	Limited sensitivity, complex calibration.	[55,56,57,58,59,60]

categorizes fiber optic sensors by their fundamental operating principles and expands on each type by outlining their data output, measurement capabilities, applications, and limitations. Intensity-based sensors are simple and cost-effective, suited for basic deformation or crack detection, but offer lower accuracy and are prone to signal noise [22,25,26]. FBGs are widely used due to their multiplexing capability and reliability for point-strain and temperature measurements, although they are limited to discrete locations [20,21,26,27,28,29]. Interferometric sensors like Fabry–Pérot offer exceptional precision in controlled environments but require complex setups [29,30,31,32,33,34,35,36,37,38]. DFOS, in contrast, stands out for offering continuous, high-resolution data along the entire fiber length, making it ideal for large-scale SHM applications [9,39,40,41,42,43,44]. This classification framework helps engineers and researchers select the appropriate sensing strategy based on project scale, monitoring needs, and environmental conditions.

1.3. From Conventional Sensors to Distributed Fiber Optic Sensing (DFOS)

To meet the demands of modern infrastructure, a range of advanced sensor technologies have emerged within SHM, including piezoelectric sensors, wireless sensor networks, and fiber-optic-based systems [3,10,61]. Piezoelectric sensors excel at detecting dynamic changes, such as vibrations, while wireless networks allow for flexible, large-scale deployments [10]. Among these, fiber optic technologies have gained prominence due to their ability to provide distributed measurements over long distances. DFOSs, in particular, have become a focal point of innovation, offering continuous, high-resolution data collection along the entire length of an optical fiber, unlike traditional point sensors [9,40]. Their growing use is driven by the need to monitor aging infrastructure, adapt to extreme weather, and improve resilience, issues especially relevant in places like Australia. This paper investigates the transformative role of embedded DFOS in SHM, with a particular focus on railway infrastructures which require continuous oversight. Railway networks worldwide are vast and diverse, ranging from China’s rapidly expanding high-speed system to Europe’s century-old railways. China’s rail network extends over 162,000 km (including about 47,000 km of high-speed lines), while the EU’s network is on the order of 2 × 10⁵ km with roughly 8500 km of dedicated high-speed lines [62,63]. Modern trains and heavier freight loads push these structures harder, and natural hazards (earthquakes, floods, landslides) plus aging infrastructure can accelerate damage. To prevent service disruptions and accidents, continuous SHM is now viewed as essential: monitoring feedback can increase reliability and safety and enables early damage detection that cuts maintenance costs and prolongs asset life [17,18].

1.4. Review Methodology

The initial literature pool comprised 200 sources. Following relevance screening, 13 sources outside the review scope were excluded, leaving 187 relevant publications. These sources were classified into three groups: 142 general FOS/DFOS studies covering sensing fundamentals, system components, and non-rail applications; 16 general railway-monitoring studies in which DFOS was not the principal monitoring method; and 29 core studies directly investigating DFOS applications in railway infrastructure. The selected evidence was subsequently synthesized according to sensing principles, system components, railway applications, field performance, AI and machine learning integration, limitations, research gaps, and future priorities (Figure 1).

2. DFOS Fundamentals

As infrastructure ages and climate-related stressors intensify, the demand for intelligent, resilient monitoring solutions continues to grow. DFOS addresses this demand by delivering robust, real-time insights that support proactive maintenance, extend asset lifespan, and enhance safety. With advancements in data analytics and integration, DFOS is poised to become a foundational element of smart infrastructure systems—bridging the gap between engineering performance and digital intelligence by providing continuous, high-resolution data over kilometer-scale infrastructure, such as railway sleepers, pipelines, and tunnels [9,39,40,41,42,43,44,46,64]. A typical DFOS system, as illustrated in Figure 2a, comprises three key components: the interrogator unit, optical fiber cable, and data acquisition system [64]. Unlike traditional point sensors, DFOS enables spatially distributed measurements by analyzing light scattering phenomena (Rayleigh, Raman, or Brillouin) along the entire fiber (Figure 2b). This makes DFOS uniquely capable of capturing subtle or distributed strain, temperature, and vibration variations, enabling early detection of structural issues such as crack initiation, delamination, or foundation settlement.

DFOS offers significant advantages for SHM, including continuous measurements along the entire fiber length, high spatial resolution (mm–m scale), and the ability to monitor long distances (up to tens of kilometers). It enables precise detection of strain and temperature (e.g., ~1 microstrain and 0.1 °C) and supports multi-parameter sensing through different scattering techniques. DFOS systems are also highly durable, being resistant to corrosion, electromagnetic interference, and harsh environmental conditions. However, several limitations remain. These systems involve high initial costs due to specialized equipment and installation requirements, and they generate large datasets that require advanced processing. Installation can be challenging, particularly for retrofitting existing structures, and DFOS may have reduced sensitivity to highly localized or high-frequency events compared to point sensors such as FBGs. DFOS is widely applied in infrastructure monitoring, including railway systems (track alignment and strain), pipelines (leak and corrosion detection), bridges and tunnels (deformation and cracking), and geotechnical systems (slope stability and settlement), as well as in offshore and wind energy structures. In practice, DFOS systems require low maintenance and can operate in extreme environments. Their flexible installation, either embedded, surface-mounted, or within conduits, supports a wide range of applications while enabling condition-based maintenance and reduced lifecycle costs. Each scattering mechanism interacts differently with the fiber’s properties (

Table 2. Characteristics of DFOS scattering mechanisms.

Mechanism	Optical Measurands	Spatial Resolution	Range	Sensitivity	Physical Parameters	Main Application Area	Ref.
Rayleigh	Amplitude, phase	Centimeter–meter (OFDR)	≲10 km	≈±1 µε, repeatability 0.1 µε	Strain, vibration	DAS for vibration monitoring, security, seismic detection, and strain mapping	[51,70]
Brillouin	Frequency, amplitude	1–10 m (standard BOTDA)	≲100 km	±10 µε, ±0.1 °C (BOFDA variants)	Temperature, strain	Distributed strain sensing (DSS) related to ground movements and their effect on infrastructures like pipelines, cables, dams, train tracks, bridges, tunnels, and buildings	[66,68,73]
Raman	Amplitude (anti-Stokes, Stokes)	≈1 m	≲30 km	±0.1 °C accuracy	Temperature	Fire detection in rail and road tunnels and buildings, power cable monitoring, pipeline monitoring, oil and gas in-well monitoring, geothermal/environmental	[68,69]

), each tailored to specific SHM needs [24,40,65,66,67,68,69,70,71,72]. Rayleigh scattering is caused by elastic interactions with microscopic density fluctuations, and it is used for high-resolution strain and temperature sensing, often with optical frequency-domain reflectometry (OFDR) or phase-sensitive optical time-domain reflectometry (φ-OTDR). Brillouin scattering involves inelastic interaction with acoustic phonons, enabling long-range strain and temperature measurements, typically via Brillouin optical time-domain analysis (BOTDA) or Brillouin optical frequency-domain analysis (BOFDA). Raman scattering results from interactions with optical phonons, primarily used for distributed temperature sensing (DTS) through Raman optical time-domain reflectometry (ROTDR).

It is also useful to compare fully distributed fiber optic sensing (DFOS) with the widely adopted fiber Bragg grating (FBG) sensing technology, as both are used for railway structural health monitoring but are optimized for different monitoring objectives [74,75,76,77,78,79,80]. FBG sensors are quasi-distributed sensors that provide highly accurate strain and temperature measurements at discrete grating locations, with high sampling rates suitable for capturing dynamic responses such as wheel–rail interactions, bridge vibrations, and cyclic loading at known critical locations. However, their monitoring coverage is limited to the installed grating positions; expanding coverage requires additional sensors and interrogation channels, and depending on the interrogation architecture, practical multiplexing is generally limited to several hundred to approximately one thousand gratings per fiber owing to finite wavelength bandwidth, optical power budget, and crosstalk effects. For localized monitoring applications, FBG systems generally require less complex and often lower-cost instrumentation than fully distributed systems, although this advantage diminishes as the number of monitored sensing points and required sampling rates increase. In contrast, scattering-based DFOS techniques (Rayleigh, Brillouin, and Raman) use the entire optical fiber as a continuous sensing element, enabling uninterrupted measurements over distances ranging from several kilometers to tens of kilometers without blind spots between sensing locations. Although Brillouin- and Raman-based DFOS generally offer lower temporal resolution than FBG systems, Rayleigh-based φ-OTDR/DAS can achieve comparable or higher sampling rates; distributed systems generally provide lower measurement precision or slower acquisition than point-based FBG sensing for dynamic applications, although modern Rayleigh-based OFDR systems can achieve strain resolutions comparable to FBG over shorter sensing ranges but offer comprehensive spatial coverage, which is particularly advantageous for detecting previously unknown defects, localizing anomalies, and monitoring long railway corridors, tunnels, embankments, and bridges. Consequently, the two technologies should be regarded as complementary rather than competing: FBG sensing is preferred where high-fidelity measurements are required at pre-defined critical points, whereas fully distributed DFOS is better suited to network-scale condition assessment and early anomaly detection over extensive railway infrastructure. Accordingly, integrated FBG–DFOS deployments are increasingly adopted in railway structural health monitoring to exploit the strengths of both technologies while mitigating their individual limitations.

The choice between single-mechanism and hybrid Rayleigh–Brillouin schemes is governed by the required sensing range, spatial resolution, measurand, and acceptable latency [40,51,54,72,81,82]. Rayleigh-based φ-OTDR/OFDR offers centimeter-scale spatial resolution and high sampling rates (kHz range), making it well suited for dynamic vibration and train-detection tasks, but is typically limited to tens of kilometers and is more sensitive to vibration-induced noise. BOTDA/BOTDR extends the sensing range to ~100 km with meter-scale resolution and greater robustness to environmental disturbances but requires temperature–strain compensation and has slower acquisition times (seconds to minutes per scan in conventional configurations). Raman-based distributed temperature sensing (DTS) is the slowest, requiring extensive averaging, and is primarily used for slowly varying temperature monitoring. Hybrid Rayleigh–Brillouin(–Raman) architectures are increasingly favored in long-distance railway projects because they enable simultaneous dynamic event detection and static strain/temperature monitoring from a single fiber, although they inherit the slower acquisition speed of the Brillouin channel.

2.1. Interrogator Unit

The interrogator unit is the core component of DFOS systems, serving as both the source and analyzer of the optical signals that enable distributed monitoring. It combines a coherent pulsed laser source, typically operating with pulse widths of 1–100 ns at 1550 nm for single-mode fibers, and a high-sensitivity photodetector [65]. This setup allows the interrogator to launch laser pulses into the fiber and analyze the backscattered light, generating spatially and temporally resolved profiles of strain, temperature, or vibration. By interpreting scattering phenomena such as Rayleigh, Brillouin, or Raman scattering, the interrogator provides critical data for a wide range of applications, from SHM to environmental sensing. Modern interrogators leverage advanced techniques to achieve high precision and extended monitoring ranges. Optical time-domain reflectometry (OTDR), which relies on Rayleigh scattering, is commonly used for basic sensing tasks and can achieve spatial resolutions as fine as 10 cm over distances up to 100 km. For applications requiring enhanced strain sensitivity, Brillouin optical time-domain analysis (BOTDA) is employed, offering spatial resolutions of approximately 1 m and strain sensitivities of ±1 µε. These capabilities are particularly valuable in scenarios such as railway monitoring [17,41], where interrogators detect microstrain changes in sleepers under cyclic loading, or in long-range pipeline leak detection, where improved signal-to-noise ratios (SNRs) are essential.

Commercial interrogators span a wide performance and price range, such as Luna ODiSI-B 5.0 [83], Sensornet Sentinel DTS-XXR (45 km, ~5 m spatial resolution) [84] for pipelines, and Silixa’s XT-DTS (50 km, 1–2 m resolution) [85]. However, their six-figure price tags (and rental rates of ~US $100/day) require a careful cost–benefit analysis for large-scale deployment. This evaluation must weigh the interrogator’s advanced capabilities, such as extended range and fine spatial resolution, against budget constraints and the specific requirements of the project.

2.2. Optical Fiber Cable

The optical fiber serves simultaneously as the sensing element and the data-transmission medium in DFOS systems. The fiber’s resistance to electromagnetic interference and corrosion makes it ideal for kilometer-scale applications. In most deployments, a standard telecommunications single-mode fiber, like Corning^® SMF-28^® (8.2/125 µm core/cladding), is used for its low attenuation (~0.16–0.17 dB/km at 1550 nm) and seamless integration with existing telecom infrastructure [86]. To transmit multiple signals simultaneously and decouple strain from temperature effects in complex scenarios, such as railway ballast settlement or mixed-loading pipe segments, multi-core fibers (MCFs) (e.g., three-core or seven-core designs) have emerged [87,88].

For choosing a fiber optic cable some consideration must be taken into account such as installation environment, deployment method, required data-transmission capacity, potential network expansion, and ongoing maintenance needs, which result in differences in cable structure, materials, and performance specifications. Table 3 summarizes the properties of single-mode and multi-mode fiber optic cables, including core and cladding materials, coating options, applications in standard and harsh environments, and approximate price ranges, based on industry standards and recent data. Silica (SiO₂)-based glass fibers dominate DFOS owing to their ultra-low attenuation (~0.2 dB/km at 1550 nm) and ability to sense over tens of kilometers with centimeter-level resolution using Rayleigh, Brillouin, or Raman scattering. On the other hand, plastic optical fibers (POFs) [89,90,91,92,93], typically made of PMMA or perfluorinated polymers, mainly employ Rayleigh scattering and offer a flexible, cost-effective solution for short-range applications (<500 m), despite higher attenuation (100–200 dB/km at 650 nm). Advances in optical frequency-domain reflectometry (OFDR) enable POFs to achieve subcentimeter resolution over hundreds of meters, ideal for embedding in concrete or geotextiles for precise SHM. However, POFs require UV- and moisture-resistant coatings to perform in harsh conditions, while glass fibers provide superior durability in extreme temperatures (up to 600 °C) and corrosive environments. Acrylate is the primary buffer coating, protecting against moisture and mechanical stress in normal environments. Polyimide is used for high-temperature resistance (up to 300 °C) [94,95], while fluoropolymers (e.g., ETFE, PTFE, PFA, FEP) provide chemical and mechanical durability (such as Zeus PEEK Reinforced Optical Fiber series [96]). In harsh environments, such as subsea pipelines or desert rail networks, specialty fibers with high-temperature or corrosion-resistant coatings are required [97,98,99,100]. Common polymers include polyethylene (PE) [101], polyvinyl chloride (PVC) [101,102,103], and low-smoke zero-halogen (LSZH) [104]. Each material is available in multiple grades, optimized for properties such as UV resistance, flame retardance, flexibility, or low-temperature performance. Reinforcing layers confer tensile strength and crush resistance. Options include metallic strength members (steel or aluminum wires), non-metallic aramid or fiberglass yarns, and polymeric tapes.

Distributed optical fibers can be deployed using various installation techniques depending on the application and structural material. They may be surface-mounted, such as bonded with epoxy adhesives along rail tracks or structural beams, embedded within protective coatings (e.g., along pipelines), or integrated during construction, for instance, by placing them within tunnel linings, concrete formwork, or geotextiles before curing [61,65,102,105]. These approaches aim to optimize strain transfer from the structure to the fiber, which is essential for accurate sensing performance. A key factor influencing sensing accuracy is the mechanical coupling between the fiber and the host structure. Imperfect adhesion, debonding, or voids in the bonding layer can significantly degrade strain-transfer efficiency, resulting in attenuation, phase lag, or reduced sensitivity. In extreme cases, measurement accuracy may drop below 10 µε, compromising the reliability of SHM. Therefore, appropriate adhesives, fiber sheathing, and installation protocols must be carefully selected and validated based on the material system, environmental exposure, and expected strain levels.

Table 3. Overview of Fiber Optic Cable Properties.

Fiber Type	Material and Structure	Core/Clad (µm)	Attenuation	Sensing Range	Spatial Resolution	Temp. Range (°C)	Scattering Techniques	Typical DFOS Applications	Cost (USD/m)	Certification Standards
Single-Mode Glass	Silica core and cladding; acrylate (std) or polyimide/fluoropolymer (harsh) coatings	8–10/125	~0.2 dB/km @ 1550 nm	Up to ~100 km	cm–m (BOTDA, BOTDR)	−60 to +200 °C (acrylate) −60 to +600 °C (polyimide)	Rayleigh, Brillouin, Raman	Long-haul pipeline/tunnel/bridge/rail monitoring	0.5–2 (std) 2–5+ (specialty)	ITU-T G.652, G.657 [106,107]
Multi-Mode Glass	Silica core and cladding; acrylate or polyimide coatings	50/125 (OM2) 62.5/125 (OM1)	3–5 dB/km @ 850 nm	~1–10 km	1–10 m (BOTDA, Raman)	−60 to +200 °C	Brillouin, Raman, Rayleigh	Medium-range DFOS: industrial pipelines, structural segments, tunnels	0.3–1 (std) 1–3+ (specialty)	ITU-T G.651 (MMF) [108]
Multi-Mode Plastic (POF)	PMMA or perfluorinated core; PE/PVC or UV-stable jacket	980–1000/1000	100–200 dB/km @ 650 nm	Up to ~0.5 km	mm–cm (OFDR, OTDR)	−20 to +85 °C (PMMA) −20 to +120 °C (PF-POF)	Rayleigh (OFDR), OTDR	Short-range embedding in concrete/geotextiles, flood/slopes, wearable sensing	0.1–0.5	IEC 60793-1-1, IEC 60793-2-40 [109,110]

Table 3. Overview of Fiber Optic Cable Properties.

Fiber Type	Material and Structure	Core/Clad (µm)	Attenuation	Sensing Range	Spatial Resolution	Temp. Range (°C)	Scattering Techniques	Typical DFOS Applications	Cost (USD/m)	Certification Standards
Single-Mode Glass	Silica core and cladding; acrylate (std) or polyimide/fluoropolymer (harsh) coatings	8–10/125	~0.2 dB/km @ 1550 nm	Up to ~100 km	cm–m (BOTDA, BOTDR)	−60 to +200 °C (acrylate) −60 to +600 °C (polyimide)	Rayleigh, Brillouin, Raman	Long-haul pipeline/tunnel/bridge/rail monitoring	0.5–2 (std) 2–5+ (specialty)	ITU-T G.652, G.657 [106,107]
Multi-Mode Glass	Silica core and cladding; acrylate or polyimide coatings	50/125 (OM2) 62.5/125 (OM1)	3–5 dB/km @ 850 nm	~1–10 km	1–10 m (BOTDA, Raman)	−60 to +200 °C	Brillouin, Raman, Rayleigh	Medium-range DFOS: industrial pipelines, structural segments, tunnels	0.3–1 (std) 1–3+ (specialty)	ITU-T G.651 (MMF) [108]
Multi-Mode Plastic (POF)	PMMA or perfluorinated core; PE/PVC or UV-stable jacket	980–1000/1000	100–200 dB/km @ 650 nm	Up to ~0.5 km	mm–cm (OFDR, OTDR)	−20 to +85 °C (PMMA) −20 to +120 °C (PF-POF)	Rayleigh (OFDR), OTDR	Short-range embedding in concrete/geotextiles, flood/slopes, wearable sensing	0.1–0.5	IEC 60793-1-1, IEC 60793-2-40 [109,110]

2.3. Data Acquisition System

The data acquisition system for DFOS must handle high-throughput digitization, low-latency processing, and intelligent analysis to transform raw backscatter signals into actionable insights. At the front end, high-speed analog-to-digital converters (ADCs) sampling at ≥100 MHz with resolutions of 10 bits or more provide dynamic ranges around 60–80 dB, enabling detection of weak backscattered signals over tens or hundreds of kilometers of fiber, and an example system is Teledyne SP Devices’ digitizers [111]. The interrogator hardware, whether Rayleigh-, Brillouin-, or Raman-based, interfaces directly with these ADCs; its performance (noise floor, sampling rate, linearity) dictates the maximum sensing length, spatial resolution, and minimum detectable strain or temperature change [112,113,114].

Once digitized, raw data streams can reach the order of 1 GB/s for kilometer-scale systems (e.g., a 50 km deployment sampled at 1 m intervals), necessitating robust storage architectures such as multi-terabyte RAID arrays or high-throughput network-attached storage to archive both raw and processed data without bottlenecks. To address these demands and provide timely alerts, real-time processing is implemented on hardware-accelerated platforms: FPGAs embedded in the interrogator chassis or GPU-equipped edge servers handle initial signal conditioning (digital filtering, decimation) and inference tasks. For example, FPGA-based implementations of convolutional neural networks (CNNs) have been demonstrated to perform on-chip inference for vibration event recognition with submillisecond latency, maintaining accuracy comparable to offline models while operating within the power and space constraints of embedded systems [115].

Edge processing further reduces data transmission burdens by extracting salient features locally [116]. Instead of streaming entire backscatter profiles continuously, the system computes metrics such as peak strain values, spectral or time–frequency coefficients, and statistical descriptors over sliding windows. Multi-dimensional feature-extraction schemes, deriving time-domain, frequency-domain, and time–frequency-domain features from denoised signals, have been applied to classify disturbance events with high accuracy in DFOS contexts [117]. By selecting a compact set of discriminative features via algorithms like principal component analysis or ReliefF, edge nodes can condense data volumes dramatically (orders-of-magnitude reduction) while preserving event-detection fidelity. These summaries or alerts are then forwarded to central servers or cloud platforms, alleviating network load and enabling scalable monitoring over long fiber runs.

ML and AI-based processing strengthen both denoising and event classification. Deep learning models, especially CNNs, have been used to suppress coherent noise patterns in raw backscatter traces, improving SNR prior to downstream analysis; data-driven denoising methods (e.g., via generative models) can adapt to varied noise profiles encountered in field deployments. For event recognition, CNN architectures trained on labeled DFOS datasets distinguish between different disturbance types (e.g., train passage vs. environmental noise) with high reliability; FPGA-accelerated implementations demonstrate the practicality of embedding such inference directly in interrogator hardware, ensuring low-latency alerts. Beyond CNNs, recurrent or Transformer-based models may be explored for temporal pattern analysis when longer sequences must be interpreted. Moreover, ML-driven analytics applied to extracted features over time enable predictive maintenance: by identifying trends (e.g., gradually increasing support variability in a rail track), the system can forecast degradation and schedule interventions proactively.

Commercial DFOS offerings vary in architecture data processing, cost positioning, and AI integration. Distributed acoustic sensing (DAS) predominates, with several systems offering real-time processing [118] and edge feature extraction. For example, Ramsay et al. (2021) [119] report a three-order-of-magnitude data reduction through edge processing, while Mendoza et al. (2022) [120] detail digital signal processing that supports up to 10,000 samples per second. In systems that integrate AI, ML is used for event detection and signal enhancement; Fernández-Ruiz et al. (2022) [121] describe a convolutional neural network that achieves an 8.3× compression ratio via self-supervised denoising. Some vendors supply integrated platforms combining high-performance ADC modules (often tens of thousands of USD per channel), FPGA/GPU processing units, and storage in PXI or similar chassis; others deploy distributed edge units that perform feature extraction on-site and stream only condensed summaries. A common commercial example is the National Instruments PXIe-5162 10-bit digitizer (100 MS/s to 500 MS/s), priced at approximately US $20,000 per channel [122]. While academic literature seldom discloses precise hardware costs, industry analyses note that leveraging existing fiber infrastructure and shifting from scheduled to condition-based maintenance often outweighs upfront investments in advanced interrogators and computing hardware.

Architectures must accommodate intermittent connectivity, cybersecurity requirements, and variable data volumes, often combining local buffering with periodic bulk uploads.

Several key research gaps and limitations exist in the field of distributed fiber optic sensing systems: data management in DFOS remains challenging due to sustained high data rates (e.g., DAS systems generating > 100 MB/s) that exceed typical network capacities and demand extensive post-processing to unwrap and interpret phase data [119,120,123]. Noise mitigation under realistic operational conditions continues to limit DFOS deployment at full operating speeds or in harsh environments. Laboratory and slow-speed field trials have demonstrated feasibility, but vibration-induced noise and environmental disturbances degrade signal quality when trains or other dynamic loads move at normal velocities [124,125].

While systematic speed-resolved noise/attenuation curves remain scarce in the published railway DFOS literature, available data give a sense of achievable performance at representative speeds. In φ-OTDR-based DAS trials on operational lines, train localization from a passing ICE 4 service achieved a velocity standard deviation of under 5 km/h at a train speed of 160 km/h, indicating that signal quality remains usable for tracking purposes even at high operational speeds [126]. Other studies confirm speeds up to 160–250 km/h have acceptable signal quality [82,127]. Signal-to-noise ratio (SNR) in φ-OTDR systems has been shown to degrade as vibration frequency increases (corresponding to higher train speeds), with measurable drops observed in the hundreds of Hz to kHz range [128,129]. However, dedicated studies that report continuous noise-intensity or attenuation curves as an explicit function of train speed interval (e.g., 20–350 km/h) are largely absent from the literature; this remains a clear gap for future controlled field trials.

Signal quality issues, uneven sensitivity, fading points in phase demodulation, SNR degradation under operational perturbations, and sampling-frequency limits over long fibers further constrain reliable measurements [116,119,120,121,123,130,131,132]. Calibration, validation, and standardization of DFOS measurements over long durations and varied conditions remain underdeveloped. Factors such as fiber aging, temperature drift, and installation-induced variability can introduce biases or drift in strain or temperature readings over time [43,124]. Edge-based and AI-driven denoising and feature extraction partially mitigate these burdens, but generalized models that adapt across environments and robust coherent detection schemes are still needed. Cost and implementation gaps include opaque hardware and project cost breakdowns, high expenses for specialized components (e.g., interrogators, FBGs), and limited multi-point handling in many systems; these factors slow adoption beyond single-use cases like leak detection. Integration of different sensing modalities (DTS, DAS) lacks standardized frameworks, hindering richer, multi-parameter insights [114,116,118,131,133]. Future work should target scalable data-management architectures (advanced compression, edge computing), enhanced signal processing (adaptive filtering, ML-based denoising), transparent cost models and lower-cost hardware solutions, broader multi-application frameworks, and seamless fusion across sensing types to fully leverage DFOS for continuous, large-scale monitoring.

2.4. AI-Driven Data Interpretation and Damage Detection

Distributed fiber optic sensing technologies generate massive volumes of continuous, high-resolution data when deployed along infrastructure assets. Interpreting this data effectively requires the integration of AI and ML methods, which can identify meaningful patterns, classify events, and detect anomalies in real time [134,135]. Stork et al. [136] note that DAS, which is increasingly used for microseismic monitoring in industrial settings, generates large data volumes (~650 GB/day for a 2 km cable at 2000 Hz, 1 m sampling), requiring advanced processing for near-real-time analysis. For example, AI models are being applied in civil infrastructure such as tunnels, pipelines, and bridges to detect structural fatigue, corrosion, and mechanical stress well before visible damage appears [128,137,138].

These systems utilize techniques such as unsupervised learning to model baseline conditions and supervised learning to classify specific structural responses, helping decision-makers plan timely interventions. In the context of DFOS, AI-based techniques including anomaly detection, support vector machines, neural networks, and ensemble methods are commonly employed [54,117,120,123,128,137,138,139,140,141,142]. These models can detect sudden changes in strain, vibration, or temperature, which may indicate physical threats such as cracking, leakage, or unauthorized activity. Across various infrastructure types, whether urban tunnels, high-voltage power cables, or buried pipelines, AI augments the capabilities of DFOS by distinguishing between operational noise and critical anomalies. For instance, ML has been used to detect seismic activity in buried assets and to monitor load cycles in bridge structures to assess long-term fatigue risk [136,138].

In railway applications, this synergy has proven particularly transformative [128,143,144,145]. In Du et al.’s study [146], a 12 km DFOS installation continuously monitored a railway test track, and AI models were used to sift the data for unusual events: “thanks to the data collected using the fiber, ML can be used to detect and classify these intrusions and alert the relevant authorities”. As Yürekli et al. [147] note, this “data-driven analysis” approach to railway sensor data “significantly contributes to increasing railway safety and allows the development of proactive measures”. Common AI techniques in DFOS analysis include both anomaly detection and supervised classification. In anomaly (or novelty) detection, models are trained on normal fiber optic signal behavior so that deviations stand out. For example, Xie et al. [148] used an unsupervised convolutional autoencoder on DAS data from a high-speed rail site: by learning the “normal” vibration patterns, the system could flag abnormalities (e.g., a person climbing a barrier) with 91.5% accuracy, outperforming prior supervised methods. Rahman et al. [134], in one study, employed a deep convolutional neural network–long short-term memory–sliding window (CNN-LSTM-SW) hybrid in monitoring train position and normal/abnormal railroad conditions over a 4.16 km fiber cable DAS system and reported 97% detection accuracy for rail condition monitoring. In another study [149], the team explored railway tracks’ structural health and condition monitoring using DAS data extracted from a high tonnage loop (HTL)–fiber optic bed in MxV Rail facilities (Pueblo, CO) and applied a deep long short-term memory–sliding window (DLSTM-SW) model and recorded accuracy greater than 97% with processing times of only a few seconds. The detection targets of the study were defective location, non-defective location, and train position. In a different approach, Karapanagiotis et al. transfer learning applied to crack detection achieved a mean average precision of 0.968 with data processed in less than 0.05 s per 10,000 points [141].

In supervised approaches, labeled examples of known events (trains passing, rockfalls, etc.) train classifiers like support vector machines, random forests, or neural networks. Ensemble methods, combining several algorithms, often yield the best performance. For instance, a recent field trial on a mountainous Turkish railway used a “voting” ensemble of SVM, random forest, XGBoost and gradient boosting models [147]. This hybrid model automatically distinguished rockfalls, falling trees and other hazards in real time, achieving about 98% accuracy. Such techniques can also estimate event severity by extracting features (signal amplitude, frequency content, etc.) from the raw DFOS time series before classification. The AI-driven approach dramatically improves damage and hazard detection compared to manual monitoring. In practice, automated DFOS analysis detects faults or intrusions earlier and more reliably. The Karabük railway study found that the ensemble ML model “accurately detects environmental events” with 98% success, enabling real-time alerts of rockslides and tree falls. Early anomaly detection yields big safety and cost benefits: as one survey notes, spotting “unusual events” sooner “improves passenger safety”, avoids service interruptions and reduces maintenance costs. In fact, AI can extend warning lead times for emerging hazards [150]. For example, Zhao et al. showed that ML analysis of DFOS seismic data improved the accuracy of landslide detection and “extend[ed] early warning times” compared to traditional methods [151]. Overall, these case studies demonstrate that AI enables DFOS to function as an intelligent “nervous system” for railways, sifting through big data to provide highly accurate, real-time damage and intrusion alerts that would be missed or delayed with conventional monitoring.

The ability to generate real-time alerts enables faster responses, potentially averting service disruptions or catastrophic failures. Moreover, ML enables long-term infrastructure health assessment by identifying trends, wear patterns, and degradation over time. Case studies have shown that integrating AI into DFOS platforms can extend early warning times for landslides, track bed deformation, and other structural threats. As AI algorithms continue to evolve, and datasets become more diverse and comprehensive, DFOS systems are expected to play a leading role in the digital transformation of infrastructure monitoring across transportation, utilities, and energy sectors.

3. DFOS Market Projection and Key Players

The market for distributed fiber optic sensors has shown notable expansion in recent years. Valued at USD 1.83 billion in 2024, it increased to USD 2.01 billion by 2025. Forecasts indicate sustained growth over the coming years, with an anticipated compound annual growth rate (CAGR) of 9.27%, projecting the market to reach USD 3.12 billion by 2030 [152]. This growth is fueled by the demand for real-time monitoring across industries such as oil and gas, infrastructure, and security. While exact market share percentages are not specified due to limited data, key players dominate based on their prominence and specialized roles. Leading vendors include Halliburton with DFOS services [153], Luna Innovations (post-Silixa acquisition) [154], OptaSense (Luna) holding an ~18% DAS share [155], Yokogawa in DTS [155], AP Sensing [156], and OgMentum [157] as a key fiber supplier per market reports [152]. Exact share data are proprietary; repeated listing in syndicated analyses indicates these firms’ leadership positions.

Rayleigh-based DFOS (OFDR/DAS) dominates modern rail monitoring due to submeter spatial resolution and kHz-level sampling, ideal for real-time train tracking, ballast health assessment and fault detection. DFOSs are widely used for monitoring infrastructure like railways and pipelines, but finding the latest numbers on active projects can be challenging. The most recent comprehensive data is from 2017, when the Fiber Optic Sensing Association (FOSA) reported over 1300 installations globally [158], spanning more than 20,000 miles in more than 75 countries. Since then, updated global deployment figures have not been publicly available, though the total is likely to have grown considerably given increasing interest in infrastructure resilience. Key insights from that time include China leading with approximately 11.3% of installations, followed by Germany at 9.4%, the United States at 6.5%, and South Korea at 4.8%. The assets most frequently monitored included power cables (22.2%), tunnels (20%), pipelines (13.5%), and perimeters (8.4%), with a total length spanning more than 20,000 miles (33,300 km). This gap suggests a need for updated surveys to capture the current scale of DFOS deployment (Figure 3).

Regional differences in railway operation modes further shape both the promotion difficulty and cost-effectiveness of DFOS adoption. High-speed passenger-dominant networks (e.g., China, Japan, and parts of Europe) are more willing to accept higher upfront costs for high-resolution, continuous monitoring given greater safety requirements and centralized funding, whereas freight-heavy networks (e.g., North America and Australia) tend to favor lower-cost, targeted deployments at critical locations (bridges, tunnels) given lower operating speeds and different risk profiles. These regional patterns broadly track the wider DFOS market [158,159]: North America held over 30% of global DFOS revenue in 2025, driven substantially by oil and gas, telecom, and energy-infrastructure investment rather than rail-specific spending, while Asia Pacific is projected to record the fastest CAGR through 2033, driven by rapid industrial expansion, large-scale infrastructure investment, and accelerating smart-infrastructure and 5G rollout across China, India, and Southeast Asia; trends consistent with the region’s continued expansion of high-speed rail networks. Rail infrastructure monitoring was the largest application segment by revenue in the global DFOS market in 2025, driven by increasing demand for continuous assessment of track conditions, structural integrity, and environmental factors affecting rail operations, underscoring the sector’s importance even though most published regional figures report market-wide rather than rail-specific data. Existing trackside fiber infrastructure and legacy signaling systems further affect deployment costs: regions with mature dark fiber networks can achieve substantially lower marginal costs by repurposing existing cables. These factors help explain observed regional variations in DFOS adoption rates, though rail-specific regional breakdowns remain scarce in published market data.

4. Implications of DFOS for Resilient Railway Infrastructure

DFOS has been applied across rail infrastructure, including track and sleeper monitoring (e.g., continuous rail strain, support continuity assessment, and even train position) [160], bridge health evaluation (deflection measurement as shown in Figure 4, prestress loss detection or damage, dynamic response capture) [161], tunnel lining deformation monitoring under external loads, and embankment stability and stiffness tracking. Unlike conventional point-based methods (track circuits, wheel counters, periodic visual inspections), DFOS provides continuous, high-resolution profiles of strain, temperature, or vibration along structures, functioning as an “artificial nerve” and enabling early detection of localized anomalies [162]. Leveraging existing telecom fibers alongside tracks can reduce installation costs, as a single fiber replaces numerous discrete sensors. However, DFOS generates large data volumes, especially from high-frequency DAS, which necessitates advanced signal-processing algorithms and data-management frameworks to distinguish meaningful events from noise. Overall, DFOS offers superior coverage and real-time monitoring capabilities that yield more accurate, comprehensive assessments of railway asset condition and operational safety [17].

As this review focuses on railway infrastructure,

Table 4. Infographic of pros and cons of DFOS applications in railway monitoring.

No.	Challenges (Cons)	Description	Benefits (Pros)	Description
01	High Setup and Operational Costs	Setting up DFOS and operation can cost more than traditional systems.	Early Issue Detection	Spots rail issues early, keeping trains and passengers safe.
02	Complex Installation	Challenging retrofitting processes for existing structures which can disrupt operations.	Smarter Maintenance for Sustainability	Enables precise repairs through continuous monitoring, reducing material waste and environmental impact.
03	Complex Data Management	Requires robust systems for vast data volumes and advanced tools to process large datasets.	Low Maintenance	Tough against weather and interference, needing less upkeep.
04	Environmental and Connection Issues	Weather changes and poor site conditions can reduce data reliability.	Smart Data Precision	Uses AI and digital twins to predict track issues accurately.
05	System Reliability Issues	Fiber optic cables may break or harsh conditions, and long-term data can drift, reducing reliability.	Climate Resilience and Security	Warns of landslides and intruders for resilient railways.
06	Implementation Challenges	Evolving standards and specialized training make scaling DFOS across networks difficult.	Broad Network Coverage	Monitors entire tracks in real time, scalable across networks.

provides a visual summary of the key advantages and challenges of using DFOS for railway monitoring. Specifically, DFOS, as an emerging transformative technology, enhances railway monitoring by enabling early detection of rail cracks, deformations, and broken-rail events through high-resolution, continuous monitoring with DAS and Brillouin sensing. Additional sensors, such as cameras and temperature sensors, support environmental monitoring for risks like frost heaves or landslides. The image also depicts real-time intrusion detection via phase-sensitive OTDR and data integration with AI and digital twins for predictive maintenance, showcasing how these technologies enhance railway safety, reliability, and sustainability. It supports sustainability through condition-based maintenance by using strain and temperature profiling to target interventions, reducing material waste and emissions from unnecessary repairs. Additionally, DFOS data streams integrate with digital twins and ML models to accurately detect anomalies such as ballast washouts or cracked sleepers, enabling automated alerts and prioritized maintenance [13]. It should be noted that DFOS-based crack detection at the sleeper level does not require instrumenting every sleeper across a network, which would be impractical with current technology and cost constraints. In practice, this approach is applied selectively, targeting a representative subset of sleepers at known high-stress locations, such as transitions, switches and crossings, or within dedicated instrumented test and monitoring sections. The integration of DFOS data with digital twin frameworks and AI-based extrapolation models is being explored as a means of inferring the likely condition of non-instrumented sleepers from a limited number of instrumented reference points, offering a pathway toward broader network coverage without full-scale sensor deployment [163].

For climate resilience and security, DFOS detects temperature and strain changes to warn of frost heaves or landslides, while phase-sensitive OTDR acts as a virtual security fence to identify intrusions. Despite these advantages, challenges remain, as fragmented pilot projects hinder cross-network comparisons, though emerging standards like the IEC 61757 series aim to harmonize performance and installation specifications [17,39,74,164].

Over the past few years, numerous studies have validated that DFOS technologies, including Brillouin-based sensors (BOTDR/BOTDA) for distributed strain, Rayleigh-based OFDR for high-resolution local strain, DAS for vibrations, and FBG for quasi-distributed high-accuracy points, can greatly improve the monitoring of railway infrastructure. Environmental factors like temperature changes necessitate careful calibration or compensation (e.g., using separate “temperature fibers” or correction algorithms) to distinguish thermal strain from mechanical strain. Installation methods also influence performance; fibers can be embedded inside structures during construction for optimal strain transfer (as done in new concrete members) or retrofitted onto existing structures by bonding to surfaces (which is more flexible for deployment but requires ensuring good coupling and protecting the fiber).

4.1. Track and Sleeper Monitoring

Wheeler et al. (2018) [162] demonstrated that Rayleigh backscatter DFOS (optical frequency-domain reflectometry) can capture distributed dynamic strains along rails under train passages. Laboratory bonding of fibers (on rail foot and head) required minimal surface preparation and enabled curvature inference, revealing support variability over tens of meters in slow-moving tests. The study’s dual-fiber approach enabled calculation of rail neutral axis shifts and curvature, something not possible with a single strain gauge. However, passenger trains at normal speeds produced vibration-induced noise beyond the system’s reliable range, indicating sensitivity limits for high-frequency loading and the need for slow-speed or quasi-static scenarios for consistent data. Additionally, only a short rail segment (7–10 m) was instrumented, implying that current high-resolution Rayleigh-based systems may need modular deployment to cover long track lengths (due to sampling rate and fiber length trade-offs).

In a follow-up, Wheeler et al. (2019) [165] used Rayleigh DFOS on a 7.5 m rail section under a deliberately slowed freight train (~10 km/h) to compute per-sleeper loads by combining distributed strain profiles with digital image correlation (DIC). This approach identified individual sleeper support conditions (e.g., voids) that would be masked by aggregate measurements, underscoring the benefit of DFOS as a multi-sensor along the rail. This study demonstrated that DFOS can turn a rail into a multi-sensor that measures how each sleeper supports the rail in real time. Yet, practical deployment is hindered by sensitivity to vibration (requiring slowed trains), limited coverage per interrogator, and the necessity of complementary displacement measurements to interpret strain data. High train speeds caused signal “fading” and noise, which is a technical hurdle for Rayleigh-based DFOS in this context. The monitoring covered only a short length of track; scaling this up to kilometers would require multiple fibers or interrogators and better handling of fiber signal attenuation and noise. Additionally, the need to integrate DFOS data with DIC (camera-based) measurements for deflection indicates that DFOS alone might not yet measure absolute displacements without complementary data or assumptions (it measures strain, which must be integrated with displacement with some boundary conditions).

Bednarski et al. [45] deployed DFOS using both composite and monolithic sensors bonded directly to rails across more than 2000 m of track in central Poland (Figure 5a). The same sensors were connected to different optical interrogators (Rayleigh, Brillouin, and Raman), allowing simultaneous generation of strain, displacement, and vibration profiles from a single fiber installation. Measurements were conducted both on a long-term cyclical basis under changing environmental conditions and on a short-term dynamic basis during actual train passes. This study demonstrated that a single DFOS installation can function as a multi-output sensing platform, replacing several discrete sensor types at once. Unlike traditional point sensors that capture behavior at a single sleeper bay or rail section, the distributed profile revealed the full strain and displacement response along the entire instrumented length in real time, enabling direct detection of localized anomalies such as stress concentrations or support irregularities that would otherwise be missed. The comparison between composite strips and monolithic direct-bonded fibers revealed a key practical trade-off: direct bonding offers more immediate and faithful strain transfer, but is more vulnerable to damage, while composite strips provide greater durability and protection at the cost of a slight measurement lag (Figure 5b,c).

Milne et al. (2020) [166] used DAS to convert existing or installed fibers into arrays of virtual microphones, capturing strain waves as trains passed. By processing DAS signals, the study inferred rail deflection bowls and per-sleeper load distribution over hundreds of meters, validated against strain gauges and high-speed camera measurements. A key contribution was showing that a single DAS system can cover hundreds of meters of track, revealing variations in track stiffness and performance continuously, whereas traditional sensors provided only sparse data. DAS’s high spatial coverage and sample rates enable continuous monitoring of track stiffness variations and early detection of weak spots. However, massive data volumes, complex signal processing, calibration needs (especially for loosely coupled telecom fibers), and sensing-range limits (attenuation over tens of kilometers) pose practical challenges, meaning very large networks may need multiple DAS units and careful data stitching. Additional hurdles include vibration-induced noise at operational train speeds, fiber coupling reliability, and the need to integrate DFOS with complementary methods (e.g., DIC) for full interpretation of strain data. DFOS techniques on track segments reveal support variability, sleeper conditions, and dynamic behaviors not accessible via discrete sensors. Rayleigh-based methods excel under low-speed or controlled conditions; composite installations improve durability; DAS offers long-range, continuous coverage but demands advanced analytics and robust fiber coupling.

Beyond strain and deflection monitoring, DAS is also being deployed for right-of-way security and hazard detection. In June 2025, Sensonic GmbH partnered with the Chicago Transit Authority on a DAS pilot program designed to automatically detect intrusions and fallen objects along the rail right-of-way, using fiber optic cables for continuous real-time monitoring to strengthen hazard detection and improve transit system reliability. This application illustrates a complementary use case for track-side DFOS distinct from structural condition monitoring: rather than characterizing gradual deformation or damage accumulation, the system is optimized for rapid anomaly detection and operational safety response, an increasingly important capability for urban transit systems operating in constrained, publicly accessible corridors [159].

A recent example of quasi-distributed FBG monitoring applied to slab track is the work of Zhang et al. (2024) [167], who instrumented CRTS III ballastless track specimens with FBG sensors connected in series via wavelength-division multiplexing and subjected them to low-cyclic reversed loading to simulate seismic pseudo-static effects. By embedding the fiber gratings both within pre-slotted reinforcement and within PVC-encased pipes at multiple points across the base plate and self-compacting concrete layer, the authors were able to track strain evolution, residual strain, and cross-sectional strain distribution as damage progressed through three stages: cracking at the base-plate anchors, deformation of the shear reinforcement, and eventual crushing of the concrete layer. Their results showed that strain concentrations consistently developed at the concave corners of the track slab’s groove region, that transverse tensile strain exceeded longitudinal tensile strain throughout loading, and that fiber gratings embedded directly in reinforcement provided more sensitive and reliable readings than those embedded in PVC pipe, which also showed a lower sensor survival rate under cyclic load. This study illustrates how quasi-distributed FBG arrays can resolve highly localized strain gradients around structural discontinuities (e.g., grooves, anchors) that would be difficult to capture with sparse conventional strain-gauge instrumentation, reinforcing the case for FBG/DFOS deployment in slab-track health monitoring under dynamic and seismic loading conditions.

4.2. Bridge Monitoring

Kishida et al. (2025) [161] instrumented a 1 km high-speed railway bridge in Taiwan with multiple DFOS lines (Figure 6a), capturing both static strains and dynamic responses (via integrated DAS) during normal train operations and post-earthquake events (capturing the bridge’s deformation in 3D). The continuous static strain profiles enabled rapid post-event safety assessment, while dynamic measurements recorded vibrations from passing bullet trains (Figure 6b,c). This large-scale deployment highlights DFOS’s ability to replace many point sensors, offering thousands of sensing points along spans, passive operation, and immunity to electromagnetic interference. Challenges include significant installation effort for fiber routing and protection over long spans, high interrogator costs (though competitive versus many discrete sensors), and the need for automated data processing and integration into maintenance workflows. Fortunately, as shown, the benefits (especially in emergency scenarios and for aging bridges) can outweigh these challenges.

DYWIDAG’s Smart Tendon system represents a further example of industrialized DFOS deployment, embedding an optical fiber within bonded post-tensioning strands to monitor strain, friction losses, and local defects using Brillouin optical frequency-domain analysis (BOFDA), with a measurement range of up to 25 km and spatial resolution of 20 cm [168]. While primarily applied to pre-stressed bridge and geotechnical anchor structures rather than rail track directly, it illustrates the broader commercial maturity of embedded DFOS relevant to railway bridge infrastructure. Ye et al. (2020) [169] embedded BOTDR cables and discrete FBG sensors in pre-stressed concrete bridge girders to track immediate and long-term pre-stress losses. In this study, BOTDR cables were embedded along the length of four 11.9 m pre-stressed concrete beams during fabrication (after pre-stressing strands were tensioned, before pouring concrete). Alongside the distributed fibers, the team also embedded FBG sensors at discrete points. The BOTDR cables measured total strain, while FBGs measured both strain and temperature (the FBGs were in pairs, one to capture temperature so that thermal strain could be subtracted). The results showed that the distributed strain measurements closely matched theoretical predictions from design codes (Eurocode 2 and other models) for time-dependent pre-stress losses, once temperature effects were accounted for. This validated that DFOS can accurately capture subtle strain changes (on the order of tens of microstrains over months) in massive concrete members. The combination of continuous (DFOS) and point (FBG) data provided confidence in distinguishing genuine structural strain from noise or thermal drift. Embedding fibers during construction ensures protection and longevity, but this approach is limited to new builds or major rehabilitations. Retrofitting a similar capability in an existing pre-stressed bridge would be difficult (external bonding of fiber might not capture the internal tendon losses directly). The system complexity was relatively high. It requires dual interrogator systems (DFOS and FBG), careful calibration, and assumptions about composite action between tendons and concrete. Despite these cons, the successful outcome suggests the technique is quite viable for critical bridges (where long-term monitoring justifies the cost).

Cocking et al. (2021) [170] used dense FBG networks (quasi-distributed), instead of using Brillouin or Rayleigh scattering, on a repaired skewed historic masonry arch, capturing dynamic strain distributions at ~50 locations per train passage at 1 kHz or more. This quasi-distributed array captured longitudinal strain distributions along the arch, revealing strain discontinuities over cracked sections, load paths, and direct measurements of crack opening displacements in both in-plane and out-of-plane directions under Class 185 passenger trains. Analysis of multiple passages demonstrated that heavier trains later in the day and higher speeds produced localized dynamic strain amplification, findings that suggest current bridge design codes may underestimate speed-related effects in arches with shallow ballast cover. The system’s microstrain sensitivity (tens of µε) enabled reliable quantification of subtle structural responses and early indicators of degradation (e.g., crack widening or stiffness loss) well before visual detection. However, deploying such an array entails extensive installation, protection, and calibration of individual sensors, leading to higher upfront costs, complex signal demodulation (including temperature compensation), and a proliferation of potential failure points. Although FBG multiplexing yields high-resolution, high-frequency insights unattainable by sparse gauges or static DFOS alone, gaps between sensors mean truly continuous coverage requires very dense placement. Finally, the sheer volume of data (each train generating a large dataset across dozens of sensors) demands data handling and expert interpretation. The authors managed this by statistical analysis and identifying patterns, but it requires specialized knowledge, and not every bridge owner has an SHM team on call to do this, which can be a barrier to practical adoption.

DFOS on bridges, from long spans to discrete embedded fibers, enables comprehensive health monitoring: continuous static strain mapping, dynamic vibration capture, and long-term behavior (e.g., pre-stress losses). Passive fiber sensors reduce wiring complexity and can operate under harsh environmental conditions. Yet, widespread adoption requires addressing installation logistics, interrogator affordability, data processing automation, and integration with digital twins or maintenance platforms. Embedding new structures and targeting critical bridges can maximize benefits while managing costs.

4.3. Tunnel and Embankment Monitoring

Gómez et al. (2020) [171] deployed Brillouin-based DFOS along the lining of the L9 metro tunnel in Barcelona to monitor strain continuously during nearby excavation. The fiber was glued in both longitudinal and hoop directions, and data collected over several months was processed with advanced filtering to remove noise. Measured strain trends aligned with theoretical models and conventional geotechnical instruments, enabling detection of subtle deformations that sparse sensors would miss. This continuous “nervous system” approach offers far greater coverage than manual surveys, works in constrained underground environments, and provides early warning of tunnel distress during adjacent construction. However, retrofitting fibers in existing tunnels demands careful, labor-intensive installation in limited-access windows and poses risk of damage; environmental influences (e.g., temperature fluctuations, train-induced vibrations) can introduce noise requiring expert data processing; and segmental linings may exhibit joint movements not fully captured by continuous strain alone, necessitating complementary discrete measurements. Despite these challenges, the study confirms DFOS as a viable, real-time SHM method for tunnels, particularly where nearby activities pose a risk.

Minardo et al. [172] used BOTDA-based DFOS glued along the sidewalls of a 200 m Italian national railway tunnel crossing an active landslide to monitor structural deformation over three years, successfully detecting and tracking strain peaks at tunnel joints consistent with independent inclinometer and satellite data and demonstrating the system’s potential as an early warning tool, though fiber breakage in the most deformed zones remained a key practical limitation (Figure 7).

Maggio et al. [173,174] conducted active seismic surveys using DAS along a Victorian-era railway embankment to assess seasonal stiffness changes due to moisture and freeze–thaw effects. By recording controlled seismic waves through the fiber array at different times of year and comparing travel-time profiles with geophone data, they demonstrated that DAS can detect subtle reductions in wave velocity correlating with embankment weakening. This approach offers dense spatial coverage and the ability to repeat surveys frequently, similar to an “ultrasound” for earthworks, potentially leveraging existing trackside fibers for real-time health monitoring. However, integrating active-source generation on live lines is challenging, and using trains as passive sources complicates interpretation. Data analysis requires expertise to disentangle moisture, temperature, and other influences, and long embankments may demand enhanced coupling or amplification for reliable sensing. Despite these hurdles, the study indicates that DAS-based geotechnical monitoring can significantly improve proactive embankment inspection and resilience against climate-induced deterioration.

Beyond deformation and seismic response, the long-term survivability of the sensing fiber itself in humid, corrosive underground conditions remains a key open question for tunnel and embankment applications. Prolonged exposure to high relative humidity (>95%), condensation cycles, hydrogen sulfide (H₂S), and biogenic sulfuric acid can trigger degradation mechanisms such as moisture-induced swelling or delamination of hygroscopic coatings (e.g., polyimide), hydrogen darkening, jacket corrosion, and biofouling. Mitigation strategies focus on robust coating and jacketing solutions, including direct femtosecond-laser inscription through factory-applied polyimide, hermetic carbon coatings, gel-filled tubes, and corrosion-resistant metallic packaging (e.g., titanium). For example, Ams et al. (2026) [175] achieved stable operation of laser-written polyimide-coated FBG sensors in titanium enclosures for over 5.5 years in high-H₂S wastewater headspaces with minimal drift or biofouling. These point-by-point femtosecond-laser-written FBG sensors, inscribed directly through commercial polyimide coating, were deployed in wastewater infrastructure with relative humidity exceeding 95% and hydrogen sulfide concentrations above 400 ppm (with short-term exposure above 30,000 ppm); conditions broadly comparable to humid, corrosive tunnels and embankment environments. While this study concerns FBG-based temperature/humidity sensing rather than distributed strain sensing, it demonstrates that fabrication method and packaging design are critical, underexplored levers for long-term DFOS durability in the kind of humid, corrosive underground conditions found in railway tunnels and embankments. Complementary sewer studies further support this: Alwis et al. (2017) [176] evaluated polyimide-coated FBG sensors in aggressive sewer environments, Bremer et al. (2014) [177] demonstrated sensor resilience under highly alkaline conditions (pH 13.4), and Rente et al. (2019) [178] reported extended in-sewer humidity monitoring performance. These results highlight the need for dedicated long-term durability trials in railway tunnel and embankment settings that assess combined mechanical, hygroscopic, and chemical degradation pathways.

DFOS provides continuous strain and seismic-response monitoring for underground and geotechnical assets, enhancing early detection of distress due to adjacent activities or climate effects. Benefits include remote data acquisition in inaccessible environments and high spatial coverage. Practical hurdles involve installation constraints in existing tunnels, advanced signal processing to filter environmental influences, ensuring fiber coupling and long-term sensor durability under humid and corrosive conditions, and coordinating active or passive seismic sources for embankment testing.

Table 5. Summary of some Examples of DFOS Rail Monitoring Studies with Field Trial Data.

Study	Country	Application	Data Collected	Fiber Length	Study Duration	Major Findings	Major Limitations
Bocheńska et al. [179]	Germany	DFOS monitoring of traffic-damaged Itztal Bridge (tram + road overpass) for crack and pre-stressing wire fracture detection	Distributed strain at pier heads and superstructure soffit, monthly measurement sessions	3 sensors, ~80 m each (~70 m active)	~18 months (November 2023–May 2025)	Detected and localized pier-head cracking and stress concentrations; enabled year-over-year degradation tracking	Coverage limited to instrumented zones only; monthly (not continuous) sampling may miss transient events
Klug et al., 2016 [164]	Austria	Continuous track deformation monitoring (field experiments)	Distributed strain along rail under natural and simulated loads (thermal, lateral, vertical)	Several tens of km capability; field tests on meters	Days/weeks during tests	Detected small local deformations (millimeter-scale) due to thermal effects, landslide simulation, locomotive/wagon loads; complements point measurements	Rigid coupling (clamps/glue) required; accurate temperature compensation (parallel loose fiber); calibration effort; data interpretation complexity
Lienhart et al., 2019 [180]	Austria	Railway tunnel monitoring	Strain and temperature in tunnel lining	Multiple kilometers	Long-term (construction phase)	Comprehensive monitoring of tunnel linings, shafts, and structures, suitable for long-term maintenance	Installation challenges, data processing complexity
Sun et al., 2024 [181]	Canada	Curved track thermal buckling indicator monitoring	Axial strain, curvature, lateral deflection along curved rail	~20 m section	~1 month (summer peak)	DFOS captured thermal-induced axial strain and curvature; GPR models predicted buckling risk with uncertainty bounds; confirmed need for multiple temperature fibers	Single temperature fiber insufficient for full cross-section compensation; section length limited; requires integration with geotechnical/thermal models
Cheng et al., 2024 [182]	Italy	Masonry arch rail bridge dynamic monitoring	Distributed dynamic strain under train loads	One span (tens of meters)	Proof-of-concept sessions	Quantified full-field dynamic strain patterns of masonry arch under train loads; identified potential anomalies/crack locations; feasibility of Luna ODiSI system	Single-span proof-of-concept; short monitoring periods; environmental influences; requires careful installation on heritage structures
Minardo et al., 2013 [183]	Italy	Rail traffic monitoring	Strain from train passage, train identification, axle count, speed, load	60 m	Short-term field test	Real-time monitoring with BOTDR, enabled train identification, axle counting, speed detection, load calculation	Limited to short section, potential calibration needs
Bednarski et al. [41]	Poland	Monitoring rail deformation on railway embankment and concrete bridge	Axial strain, temperature, curvature, vertical displacement, vibrations (via DSS, DTS, DDS, DAS)	2200 m	One year (starting early 2023, with 12 monthly sessions)	High-resolution strain profiles detected temperature-induced compressive stresses and sleeper effects; enabled curvature and displacement calculations; largest DFOS rail system globally	Non-uniform rail restraint affected strain distribution; required complex thermal compensation for large temperature gradients (−15 °C to +45 °C); installation and data processing complexity; lack of automated measurement solutions
Mohamed et al., 2007 [184]	Singapore	Twin-tunnel interaction monitoring (Circle Line)	Distributed strain and temperature in shotcrete lining during adjacent tunneling	~40 m section	During construction phases (weeks per stage)	Provided hundreds of sensing points; captured strain/temperature variations due to neighboring TBM operations; improved understanding of tunnel interaction effects	Embedding fibers in shotcrete requires timing coordination; calibration/synchronization with other sensors; data noise from construction activities
Kishida et al., 2025 [161]	Taiwan	Elevated high-speed railway bridge monitoring	Static strains, dynamic DAS, post-earthquake profiles	~1 km span	~1 year	Enabled 3D deformation calculation, rapid safety confirmation after M6.4/M6.8 earthquakes, dynamic response to individual train cars	Installation complexity on live structure; cost and need for robust bonding/protection of fibers; large data volume to process
Hsu et al., 2021 [185]	Taiwan	In-service rail strain and temperature monitoring	Continuous Brillouin-based strain and temperature along rail; geometric irregularities detection	~100 m test section	Days to weeks	Measured temperature differences (~12.1 °C vs. ambient with ~1.5 h delay), detected rail irregularities (−0.3 to +0.4 mm), automatic anomaly alerts possible	Single temperature fiber cannot capture cross-sectional variations; needs multiple fibers/temperature compensation; data transfer and processing infrastructure required
Gue et. al., 2015 [186]	UK	Cast-iron tunnel long-term strain monitoring	Continuous distributed strain along tunnel lining	Entire cast-iron tunnel length (order of 100 m)	Long-term (months to years)	Continuous and distributed strain monitoring possible without interrupting traffic; early detection of lining deformation trends; supports maintenance planning	BOTDR-based systems require calibration; potential fiber aging; data management over long durations; ensuring fiber survival in harsh tunnel environment
Cocking et al., 2021 [170]	UK	Skewed masonry arch railway bridge dynamic SHM	Distributed dynamic strain under live loads	One span (tens of meters)	Weeks/months	Detailed 3D strain responses, crack movement measurements, sensitivity to speed/temperature, post-repair integrity	Installation complexity on heritage structures, FBG/DFOS integration, limited coverage, environmental variability

provides a summary for application of DFOS in rail monitoring.

4.4. Technology Readiness and Deployment Maturity

Several patterns emerge from this comparison. Bridge monitoring applications are generally the most mature, ranging from field-validated large-scale deployments (Kishida et al. [161]) to fully commercial products for new-build construction (Ye et al. [169]; DYWIDAG Smart Tendon [168]). This relative maturity likely reflects the more controlled installation environment of bridge structures, where sensors can be embedded during construction or retrofitted with more predictable access than track or embankment applications (

Table 6. Technology readiness and deployment maturity of DFOS applications in railway infrastructure.

Application	Representative Evidence	Maturity Assessment
Track/sleeper strain monitoring (Rayleigh-based)	Wheeler et al. [162,165]; short segments (7–10 m), slow/quasi-static speeds only	Early field trial (TRL 4–5), validated in controlled/slowed conditions; not yet robust at operational train speeds
Track monitoring, full-scale industrialized installation	Bednarski et al. [41]; 2000+ m operational deployment, mechanized/robotic bonding	Operational deployment (TRL 7–8), demonstrated at full operational scale on in-service track
Track deflection/support monitoring (DAS)	Milne et a. [166]; hundreds of meters, validated against strain gauges	Field-validated (TRL 5–6), good spatial coverage demonstrated, but data volume/processing remain barriers to scale-up
Right-of-way security/hazard detection (DAS)	Sensonic/Chicago Transit Authority pilot [163]	Pilot deployment (TRL 6–7), active transit-authority pilot, not yet widespread standard practice
Bridge monitoring, large-scale distributed strain	Kishida et al. [161]; 1 km high-speed rail bridge, static + dynamic	Field-validated, near-operational (TRL 6–7), large-scale deployment with demonstrated post-earthquake assessment value
Bridge embedded pre-stress monitoring	Ye et al. [169]; DYWIDAG Smart Tendon [168]	Commercially available (TRL 8–9) for new-build/major rehabilitation, reflects existing commercial maturity
Bridge dense FBG arrays	Cocking et al. [170]	Field trial (TRL 5–6), high-resolution research deployment; cost and data-handling barriers limit routine adoption
Tunnel lining monitoring	Gómez et al. [171] (Barcelona L9); Minardo [172] (3-year landslide monitoring)	Field-validated (TRL 6), multi-month to multi-year deployments confirming viability; retrofit installation remains labor-intensive
Embankment seismic/geotechnical monitoring	Maggio [173]	Early field trial (TRL 4–5), proof-of-concept seasonal monitoring; interpretation complexity remains a barrier

).

Track and embankment applications, by contrast, remain concentrated at the early field-trial stage (TRL 4–6), with the notable exception of Bednarski et al.’s [45] industrialized installation, which demonstrates that scale-up is achievable once robotic/mechanized bonding removes the labor-intensive installation bottleneck that otherwise limits track-based DFOS to short trial segments. This suggests that installation methodology, rather than sensing technology itself, is currently the primary barrier to track-level deployment at scale. Tunnel monitoring occupies an intermediate position, with several multi-year field deployments (Minardo et al. [172]; Lienhart et al. [180]) demonstrating long-term viability, though retrofit installation into existing tunnel linings remains labor-intensive compared to new-build applications.

Overall, no railway DFOS application reviewed here has yet reached full, standardized operational deployment across a rail network in the way that, for example, axle-counters or track circuits have. The constraints underlying this gap are installation practicality, data volume management, and the absence of standardized interpretation protocols.

5. Key Constraints for Railway DFOS

Distributed fiber optic sensors promise continuous, high-resolution monitoring of rails, sleepers, bridges, tunnels and embankments. Despite their advantages, the high cost of distributed fiber optic sensors (DFOSs) remains a major barrier to large-scale deployment. Unlike point sensors such as strain gauges or accelerometers, which can be selectively installed at critical locations, DFOS requires continuous lengths of optical fiber, complex interrogation units, and specialized installation procedures. This makes full-line coverage across thousands of kilometers of railway track prohibitively expensive, especially for aging rail networks or budget-constrained operators. Additionally, maintenance and data processing costs further increase the total investment over the system’s lifecycle. Beyond cost, DFOS implementation faces several real-world challenges.

5.1. Embedment and Installation Difficulties

Various installation methods have been studied, including gluing to surfaces, embedding in grooves, and integrating during casting [187,188,189]. While subsequent installation can provide good strain transfer, integrated sensors generally offer more reliable results [188]. Monolithic DFOS designs have shown superior performance in crack detection and width measurement compared to layered cables [53,188]. Proper selection of sensing fibers, installation techniques, and data processing methods are crucial for accurate measurements, particularly in reinforced concrete applications [190]. DFOS technology enables the assessment of various mechanical properties, including initial strain states, bond behavior, and deflections [190]. However, careful evaluation is necessary to avoid misinterpretations, especially in complex loading scenarios [41,53]. Embedding fibers in new concrete elements (sleepers, viaduct decks) risks damage from high-pressure compaction and vibrating tables, creating “dead zones” where sensing fails. Retrofitting onto existing rails or sleepers via surface grooves is even more challenging, which demands adhesives that transfer strain reliably under ballast tamping, yet conventional epoxies often crack or debond after repeated load cycles [17]. The selected sensing cable and anchor spacing allow the system to withstand high strain and effectively detect rail cracks several centimeters in length [164].

Groove-based installation, in which the bare fiber is sealed into a shallow surface groove (typically 1–2 mm wide) with a suitable polymer adhesive, has been shown to provide strain transfer performance comparable to cast-in fibers, even in retrofit applications [53,167,191,192]. Adhesive selection is critical [193,194]: stiffer epoxies improve strain-transfer fidelity but may amplify local strain concentrations at cracks, while softer adhesives reduce peak stresses at the cost of some sensitivity. Surface-bonded (non-grooved) installation is simpler but generally more prone to debonding under cyclic loading. Additional effective measures include thorough surface preparation, use of fatigue-resistant epoxy formulations, and controlled pre-tensioning of the fiber prior to bonding. Collectively, groove embedding with an appropriately stiff, fatigue-resistant adhesive and proper surface preparation represents current best practice for enhancing interfacial coupling in retrofit railway monitoring installations subject to cyclic train loads.

5.2. Harsh Environmental and Mechanical Conditions

Building on the durability challenges discussed in Section 4.3, railway tracks and sleepers (whether made of concrete, timber, or composite materials) and bridges are subjected to extreme mechanical loads from passing trains, dynamic vibrations, and environmental factors such as temperature fluctuations (−20 °C to +40 °C), moisture ingress, and freeze–thaw cycles. Optical fibers, while highly sensitive and capable of detecting strain and temperature changes, are inherently fragile. The constant cyclic loading and high-impact forces in railway systems can lead to fiber breakage or degradation, leading to optical loss and signal drift, compromising the sensor’s reliability over time. Ensuring optical fibers survive these conditions without losing functionality is a significant engineering challenge. Temperature changes also shift Brillouin and Raman spectra, requiring careful compensation algorithms to isolate strain from temperature [17,74,94,98,156,195].

Reported thermal-compensation performance varies substantially with the sensing technique, interrogation method, calibration procedure, and the magnitude and rate of temperature change. Under controlled laboratory conditions, Brillouin-based distributed sensing systems have demonstrated temperature resolutions on the order of 1 °C over long sensing distances, while several studies have reported compensation errors increasing as the imposed temperature variation becomes larger [40,65,181,196,197]. However, these performance metrics are generally obtained under carefully controlled experimental conditions and should not be interpreted as representative of field performance. Although numerous thermal compensation approaches have been proposed, relatively few have been validated under long-term field conditions encompassing the large seasonal temperature variations, rapid diurnal thermal gradients, and heterogeneous thermal fields encountered in railway infrastructure. Recent work on field-deployable thermal compensation has further emphasized that no universally applicable compensation procedure currently exists, because the thermal response depends not only on the sensing principle but also on fiber installation, bonding conditions, host-structure properties, and environmental boundary conditions [198]. Consequently, practical implementation requires application-specific calibration and field validation rather than a universal “plug-and-play” approach. These observations suggest that compensation errors reported under laboratory conditions are likely to represent a lower bound on the uncertainties encountered during long-term railway operation, where solar radiation, moisture, rapid diurnal temperature gradients, imperfect thermal coupling between the optical fiber and the host structure, and combined thermo-mechanical loading introduce additional sources of uncertainty that are seldom fully represented in laboratory validation.

Distributed fiber optic sensors face challenges in harsh environments, necessitating specialized fiber protection strategies. Optical fibers for sensing in extreme conditions require optimized waveguide design and thermally stable coatings to maintain mechanical strength at high temperatures [98,195,199]. To resist abrasion, alkaline attack (pH > 12 in cement pore water) and chemical ingress (chlorides, oils), fibers are sheathed in polymer tubes or metal capillaries. A two-layer protective construction can effectively reduce environmental noise sensitivity in fiber optic components, acting as an acoustic shock absorber or filter [200]. However, thick jackets increase the minimum bend radius and introduce strain-transfer errors, degrading measurement accuracy by up to 10% unless calibrated.

5.3. Construction and Maintenance Hazards

During track laying and sleeper handling, cranes, forklifts and tamping machines can sever or crush exposed fibers, particularly in dynamic construction environments. In tunnels, ballast movement and periodic under-track interventions further risk fiber displacement unless trenched or encased, solutions that forgo direct structural sensing [75]. A 2019 case study by the Cambridge Centre for Smart Infrastructure and Construction at Hooley Cutting near south London describes engineers abseiling from the crest to secure dual-row fiber cables on soil nails at 20 m intervals over a 100 m span, a process that heightened exposure to slope instability and increased the risk of falls or equipment failures that could damage the fibers. The study emphasizes that, while robust protective measures, such as durable housings and secure anchoring, are essential, the physical stresses imposed by construction equipment and the unstable terrain remain significant hazards in such installations [201]. The need for tailored protective measures, such as robust cable ducts and careful installation practices, is evident, especially given the dynamic and hazardous nature of railway environments.

5.4. Data Volume, Noise and Interpretation

A single DFOS channel can produce gigabytes of backscatter data per day. Rail-induced vibrations, thermal drift and ballast settlement generate background noise that must be filtered to detect meaningful events (e.g., cracks, settlements). Advanced signal-processing or machine learning models improve anomaly detection (>90% accuracy) but require extensive labeled datasets and on-site calibration. Computational strategies can effectively mitigate these issues. Computational distributed fiber optic sensing using ghost imaging techniques can significantly reduce sampling rates, offering simplification and cost reduction [202]. SNR analysis shows that this approach requires twice the averaging time compared to conventional methods but reduces sampling rate requirements [203]. Data compression techniques, particularly fast wavelet transform (FWT), demonstrate promising potential for real-time compression of distributed optical fiber sensor data, addressing storage and transmission challenges [204]. Edge processing strategies can convert large data streams into diagnostics or processed products, enabling efficient data streaming and real-time diagnostics. This approach allows for the separation of diverse signals and targeted data upload to remote servers, facilitating the digitalization of engineering and geoscience assets [205].

A related but distinct challenge in dense railway installations is the reliability of monitoring when multiple DFOS channels are deployed in close proximity, such as parallel or crossed fiber runs along sleepers, slabs, or embankments. Rather than optical leakage between separate fibers, the dominant risk is data-interpretation crosstalk [51,206,207]: within a single Rayleigh- or Brillouin-based fiber, the finite spatial (gauge-length) resolution can blur strain contributions from closely spaced features into a single measurement cell, while quasi-distributed FBG arrays multiplexed at high grating density are subject to well-documented spectral-shadowing and multiple-reflection crosstalk that degrades multiplexing capacity. In addition, when several fibers are embedded close together in the same structural element, overlapping strain fields in the host material itself can make it difficult to attribute a measured response unambiguously to a specific fiber or location. Mitigation strategies include adequate physical and spectral separation of sensing channels, careful gauge-length selection relative to the smallest feature of interest, optimized WDM/multiplexing schemes for FBG arrays, and dedicated signal-processing algorithms to disentangle overlapping strain responses. Future railway DFOS deployments in dense track sections should explicitly characterize and quantify these interpretation-level interference effects to ensure long-term monitoring reliability.

6. Future Directions for DFOS Research in Rail Monitoring

A promising alternative to backscatter-based DFOS is forward-transmission laser interferometry, recently demonstrated by Wang et al. (2025) [208] for high-speed railway health inspection. Rather than installing dedicated sensing fiber, this approach repurposes existing telecom fiber cables already running within railway cable ducts, turning them into distributed vibration sensors. Using daily passing trains as the excitation source, the system monitored a 12 km rail section continuously over 14 months, employing an average power spectral density indicator to assess infrastructure health and successfully detecting subtle creep deformations in two railway sections, as well as distinguishing pre- and post-maintenance track conditions. Because the technique operates on forward-transmitted light rather than weak backscattered signals, it offers higher sampling capability, larger dynamic range, and improved compatibility with high-speed train operation, addressing key limitations of backscatter-based DFOS such as the need for slow-speed testing and limited long-distance sensing performance. Its reliance on existing telecom infrastructure also removes the need for dedicated sensor installation, suggesting a scalable pathway for network-wide railway health monitoring that merits further investigation alongside conventional DFOS approaches.

The future of DFOS in rail monitoring centers on durable hardware, cost-effective interrogators, automated data processing, advanced analytics, expanded applications, and standardized frameworks. Research into hybrid silica–polymer fiber coatings aims to enhance durability in harsh installations (e.g., concrete embedment) without degrading sensitivity [74,189]. Lower-cost Rayleigh interrogators and repurposing telecom fibers for DAS could cut capital expenses by 30–50% but require field validation versus Brillouin systems, including thermal compensation algorithms proven in lab-to-field trials handling −15 °C to +45 °C variations [41].

Automated pipelines for noise filtering and temperature correction are essential to manage terabyte-scale backscatter data. ML models (CNNs/RNNs) demonstrate >95% anomaly detection accuracy in analogous settings [120,123,126,128,140], yet open-source libraries for Rayleigh, Brillouin, and Raman signal processing remain scarce. Integrating DFOS feeds into digital twins via standardized schemas and real-time cloud platforms enhances live visualization and predictive maintenance [39,163,209]. Broader deployments include bridge joint monitoring over >100 m spans and tunnel freeze–thaw detection using OFDR-based high-resolution sensing, as well as catenary wire and signaling asset health [64,210].

Edge-AI over future 6G networks and secure logging (e.g., blockchain) can enable low-latency event detection. Scaling requires cross-sector consortia, fiber suppliers, interrogator vendors, and software developers to co-design sensors with embedded analytics and share open datasets for benchmarking. Harmonizing DFOS standards with railway norms (e.g., via IEC TC 235 alignment) and leveraging state-of-the-art insights from tunnel monitoring reviews will ensure interoperability and certification, accelerating adoption across networks [211,212]. Finally, the installation and challenges of DFOSs in railway systems, particularly during construction and maintenance, are critical areas of study, given their increasing adoption for SHM.

7. Conclusions

This review has critically examined DFOS technology, its fundamental principles, field applications, and emerging research directions for SHM of railway infrastructure. The findings demonstrate that DFOS represents a significant advancement over conventional monitoring approaches, and the key conclusions are summarized below:

• DFOS converts ordinary optical fiber into a spatially continuous measurement array, delivering high-resolution profiles of strain, temperature, and vibration over kilometer-scale assets, a capability that conventional point sensors such as strain gauges, accelerometers, and track circuits fundamentally cannot replicate. Unlike discrete sensors that capture conditions only at specific locations, DFOS functions as an artificial nervous system for railway infrastructure, enabling early detection of anomalies such as rail cracks, ballast degradation, sleeper voids, and foundation settlement across entire track sections simultaneously.
• The three principal scattering mechanisms, Rayleigh, Brillouin, and Raman, offer distinct and complementary strengths across different monitoring needs. Rayleigh-based DAS excels at centimeter-scale spatial resolution and high-frequency vibration capture for real-time train detection and intrusion monitoring; Brillouin-based systems provide long-range strain and temperature sensing with accuracies approaching ±1 µε and ±0.1 °C for embankment stability and bridge pre-stress loss assessment; and Raman-based sensing delivers subdegree temperature accuracy suited to tunnel fire detection and environmental monitoring. Field trials reviewed here confirm that these techniques have moved well beyond laboratory demonstration, with deployments validated across short instrumented rail sections, kilometer-scale elevated bridges, historic masonry arch bridges, and Victorian-era embankments.
• The integration of AI and ML with DFOS data streams has proven transformative for railway monitoring, with deep CNN, LSTM architectures, ensemble classifiers, and self-supervised anomaly detection methods all achieving detection accuracies consistently above 97% in recent field studies. FPGA-based edge inference now enables submillisecond latency classification directly within interrogator hardware, allowing DFOS to function not merely as a passive recorder but as an active, intelligent monitoring platform capable of real-time hazard alerts, predictive degradation forecasting, and automated maintenance prioritization.
• Key challenges remain that currently limit deployment to critical sections rather than full network coverage. High interrogator costs, typically six-figure capital outlays, combined with large DAS data volumes exceeding 100 MB/s on operational lines, installation complexity in retrofit scenarios, adhesive debonding under cyclic loading, temperature compensation requirements, and the absence of harmonized performance standards collectively represent significant barriers. The emerging IEC 61757 series addresses standardization only partially, and vibration-induced signal fading at operational train speeds continues to constrain the reliability of Rayleigh-based sensing under real-world conditions.
• Future research should prioritize four interconnected directions: development of lower-cost interrogation hardware and repurposing of existing telecommunications fibers to reduce capital barriers by an estimated 30–50%; construction of automated signal-processing pipelines combining adaptive noise filtering, temperature compensation, and physics-informed ML to handle terabyte-scale data at operational speeds; integration of DFOS feeds with digital twin frameworks and standardized cloud architectures for network-scale predictive maintenance; and harmonization of DFOS performance metrics with international railway engineering norms to accelerate regulatory acceptance and large-scale commercial adoption. With the global DFOS market projected to reach USD 3.12 billion by 2030, the technology is well positioned to become a foundational element of the digital railway infrastructure ecosystem, supporting safer, more resilient, and more sustainably maintained rail networks worldwide.