Mechanical Systems and Signal Processing 208 (2024) 110983

Contents lists available at ScienceDirect

Mechanical Systems and Signal Processing

journal homepage: www.elsevier.com/locate/ymssp

Invited Review Paper

A review of distributed acoustic sensing applications for railroad

condition monitoring
Md Arifur Rahman a, Hossein Taheri a, *, Fadwa Dababneh b,
Sasan Sattarpanah Karganroudi c, Seyyedabbas Arhamnamazi d
a
The Laboratory for Advanced Non-Destructive Testing, In-situ Monitoring and Evaluation (LANDTIE), Department of Manufacturing Engineering,
Georgia Southern University, Statesboro, GA 30458, USA
b
Department of Industrial Engineering, German Jordanian University (GJU), Amman 11180, Jordan
c
Department of Mechanical Engineering, Université du Québec à Trois-Rivières, Drummondville, QC, Canada
d
Department of Mechanical, Chemical and Materials Engineering, University of Cagliari, Cagliari, Italy

A R T I C L E I N F O A B S T R A C T

Communicated by John E. Mottershead Accurate condition monitoring has been a major challenge among railroad management au­
thorities as they work to minimize collisions that lead to fatalities or damage to railroads infra­
Keywords: structure. Hence, research and technological developments in railroad maintenance and
Distributed Acoustic Sensing (DAS) inspection are vital. Several research studies, on the inspection and defect detection techniques
Structural Health Monitoring (SHM)
for railroads have been conducted by scholars. Despite the significant advancements made in this
Condition Monitoring (CM)
area, extensive studies are still required to enhance the accuracy of prognostic methods for
Rail tracking
Optical Time Domain Reflectometry (OTDR) railroad structural health monitoring (SHM) and condition monitoring (CM). Distributed acoustic
Data-driven intelligence method sensing (DAS) has been recognized as a promising measurement method because of its quick
sensing capabilities over long distances and for massive structures. DAS systems are classified
according to the optical sensing quality and sensing range. As DAS produces large noisy datasets,
in the case of railroad applications, algorithms for precise real-time and reliable analysis are
essential. Meanwhile, data-driven and machine learning (ML) methods for defect detection have
emerged as valuable approaches for SHM. Engineers and stakeholders can use ML algorithms to
examine the large volumes of data produced by SHM systems to identify patterns and behaviors
that might not appear in manual inspections. To support more precise and accurate maintenance
and inspection for railroad systems, methodologies used to detect, identify, and characterize
abnormal conditions in railroads using DAS and signal processing approaches for DAS large size
and noisy signals, must be reviewed. Accordingly, in this literature survey, the applications of
DAS methods for railroad CM are investigated. Among the variety of DAS methods, optical time
domain reflectometry (OTDR) is reviewed in more details, since it is the most common approach
in distributed fiber optic sensing. In addition, different OTDR-based DAS research for train
tracking and railroad SHM are reviewed, and a comprehensive summary of different railroad
defects is provided for further investigation. In all, this review paper provides a comprehensive
background on distributed fiber optic sensing, the importance and challenges in railroad
continuous CM, and the state-of-the-art application and future roadmap for the application of DAS
in the railroad industry.

* Corresponding author.
E-mail addresses: [email protected] (M.A. Rahman), [email protected] (H. Taheri), [email protected]
(F. Dababneh), [email protected] (S.S. Karganroudi), [email protected] (S. Arhamnamazi).

https://doi.org/10.1016/j.ymssp.2023.110983
Received 2 August 2023; Received in revised form 20 October 2023; Accepted 25 November 2023
Available online 13 December 2023
0888-3270/© 2023 Elsevier Ltd. All rights reserved.
M.A. Rahman et al. Mechanical Systems and Signal Processing 208 (2024) 110983

1. Introduction

The railroad system is efficient in terms of load capacity, fuel consumption, reliability, and safety compared to other transportation
systems for passengers and goods transportation. It is also superior to other transportation systems because of its design, structural
strength, and economic and environmental advantages. More specifically, high-speed railroads have been developing rapidly and
providing a safe and reliable infrastructure. Railroad systems reduce the generation of environmentally hazardous elements which can
improve air quality. For example, railroads are often 2–5 times more energy-efficient than cargo and aircraft, more suitable for pas­
senger transportation compared to cars and planes, and about three times more effective than lorries in freight transportation [1]. With
the implementation of planned railroads, total freight costs were reduced by 17 %, and CO2 emissions could also be lowered by around
20 % [2]. It has also been found that railroad transportation could contribute to lowering CO2 emissions per unit of GDP by 40 % per
year [3]. However, with the development of railroad systems, the number of railroad accidents has been growing as well.
According to safety statistics collected by the US Federal Railroad Administration (FRA), from 1992 to 2002, 2700 train crashes and
$302 million in costs (including $48 million in 2001) were incurred in the US, due to rail track failures [4,5]. To fulfill economical
demands, environmental advantages, and safety of the railroad system, the structural integrity of the railroad needs to be assessed and
monitored in real-time. Moreover, real-time monitoring systems need to be equipped with efficient and cost-effective sensors for the
detection and prediction of failures. Accordingly, technologists are focusing on safer operation of railroads using techniques such as
Distributed Acoustic Sensing (DAS) [6–9], Machine Learning (ML), and Artificial Intelligence (AI).
In recent years, Structural Health Monitoring (SHM) [10–16] systems using DAS are being widely applied for the damage iden­
tification, fault detection, and remaining useful lifespan estimation. Damage is described as variations to the systems’ physical
characteristics, such as changes to their interconnectivity and boundary conditions, which have a negative effect on how well they
function in operating conditions. Condition monitoring (CM) is possible through a wide range of non-destructive evaluation tech­
niques, DAS, and applying AI and ML. Meanwhile, in-situ CM of railroad systems can be described as real-time monitoring of train
tracks, level crossing, operational speed, safety, and security of boundaries. Continuous monitoring or in-situ monitoring ensures the
proper performance of infrastructures, identifies any defects at an early stage, and enables tracking of operational characteristics of the
railroad system. Infrastructures are commonly subject to various stages of physical, chemical, and dynamic loading during rail
operation. Even climate change may significantly deteriorate structural characteristics. In-situ damage detection uses sensing net­
works, rigorous data analysis from the damage, and an automated inspection process to minimize the number of unwanted mainte­
nance tasks and enhance financial gains [14].
In general, track recording cars, wayside equipment, or visual inspections are employed to evaluate the condition of the railroad.
However, the DAS sensing technique has been taking its place in the CM of railroad systems. The adoption of DAS using fiber optics
technology has been rapidly arising from the laboratory setting to real-world applications in recent years. DAS enables the mea­
surement of the desired parameters as a function of length along the fiber. It offers full, quick, and accurate coverage of every section
under testing, in contrast to any conventional Nondestructive Testing (NDT) [13,17–19] technique where the coverage or scanning
area of the sensors is relatively constrained. DAS is the most reliable fiber optic sensing technique in recent decades for CM of railroads,
as it can concurrently monitor and extract data from vibrations traveling over long distances at a high-frequency rate [20].
Despite the superb potential of DAS and its inspiring future applications, several challenges still impede the broad adoption and
application of DAS systems for CM and SHM. Among the major challenges for DAS systems are the need for an extensive fiber optic
sensing network, capabilities for detection, recording, and analysis of large size data from the DAS system, and the security of the fiber
optic sensing network. Although research tackles different aspects of these challenges, and a few review papers summarized the overall
effort by the research and industrial communities to address these limitations, the body of the available literature specifically for
railroad application is not well-organized and inclusive. Hence, a comprehensive survey on the fundamentals of distributed fiber optic
sensing, working principles and signal characteristics, potential application and limitations, and appropriate methods of signal pro­
cessing and analyses is necessary for railroad, CM, and SHM communities. Accordingly, this paper summarizes the state-of-the-art
research and application of DAS for railroad CM and outlines the needs for future work and research needs.
This review paper focuses on DAS techniques for CM of railroad infrastructures, and in particular on rail tracking and SHM of the
railroad. This paper aims to be a comprehensive informative article for researchers, engineers, and technicians who are particularly
working on SHM and CM of railroads. Through this article, the basic knowledge of the working principles of DAS and its capabilities for
CM of railroad systems will be discussed. This review paper distinguishes itself through providing a comprehensive coverage, practical
insights, guidance for researchers and practitioners, and identification of research gaps which makes it a valuable resource and a worth
reading article. The paper is organized as follows:
In section 2, the methodology of Systematic Literature Review (SLR) has been discussed with databases, timeline, and total number
of publications reviewed. In section 3, fiber optic sensors, principles of operation, optical sensing categories, and the concept of DAS
have been presented. As this research focuses on Rayleigh scattering (RS), in section 4, Rayleigh scattering principles and types of DAS
(OTDR, C-OTDR, and Ø-OTDR) are discussed including characteristics such as sensing range, strain sensitivity, and spatial resolution of
various DAS interrogators. In section 5, a brief review of various ML techniques for data-driven predictive maintenance has been
discussed including the questionnaires for diagnosis and prognosis activities of SHM. Finally, in section 5 the various applications of
RS-based DAS for rail systems have been reviewed. Section 6 is divided into three parts, real-time tracking of trains, SHM of rail-tracks,
and railroad safety.

2
M.A. Rahman et al. Mechanical Systems and Signal Processing 208 (2024) 110983

2. Methodology

2.1. Review strategy

This paper presents a systematic literature review (SLR) of Railroad anomaly detection and CM by using DAS-Fiber Optic Sensing,
Machine Learning, and Deep Learning methods. The focus of this review work is mainly Rayleigh backscattering-based DAS methods.
However, the review strategy was followed by the guidance published by Kitchenham et al. in 2007 titled “Guidelines for Performing
Systematic Literature Reviews in Software Engineering” [21]. The SLR process consists of three phases: Planning, Implementation, and
Reporting. At the outset of this systematic literature review, the research questions were carefully formulated. Subsequently, a
comprehensive search protocol was developed. This protocol outlined the resources including specific databases selected for the
search, identified key publications to be considered, and defined the set of keywords to be used in the search process. The purpose of
this protocol was to ensure a systematic and exhaustive retrieval of relevant literature for our study. In this study, various resources
including Google Scholar, ScienceDirect, Transport Research International Documentation (TRID), Railway and Transportation In­
dustry Journals, Specialized Railway and Sensor Conferences, Institutional Repositories, ProQuest, Scopus, Web of Science, PubMed,
and ACM Digital Library were utilized. These resources were chosen to enhance the scope of the evaluations and to ensure the coverage
of the intended Railroad DAS-Fiber optic-based anomaly detection and condition monitoring field. In implementing, appropriate
research works were determined among the enlisted resources according to relevant implementation criteria. The resources that
passed the implementation criteria were collected, and relevant information to the research questions was extracted. In the reporting
phase, a SLR article is prepared after the research questions were investigated by utilizing the collected information.

3. Research questions

This SLR focused on railroad anomaly detection and condition monitoring by using ML/DL techniques, and Rayleigh Backscattering
based DAS-Fiber optic setup. The studies carefully searched multiple aspects, and the following five research questions (RQs) were
identified:
RQ.1: What is the fundamental of DAS and Fiber Optic Sensing Concept?
RQ.2: What are the variants of DAS, its basic parameters, capacity, and its application?
RQ. 3: What is the Fiber Optic Sensing Scattering Mechanism and Advantages of Rayleigh Scattering over Brillouin and Raman
Scattering?
RQ. 4: What are the data-driven predictive Maintenance Strategies for Anomaly Detection and Condition Monitoring?
RQ: 5: What is the proposed research works of DAS and data-driven approach for Railway infrastructures, its scopes, and
possibilities?
RQ.6: What are the challenges associated with the application of data-driven approaches for railroad infrastructure anomaly, and
condition monitoring?

3.1. Search approach

Conducting SLR in the field of DAS with a focus on anomaly detection, CM, SHM for Railway applications and by using ML, DL or
hybrid ML/DL data-driven method is a multifaceted endeavor that requires a structured approach. Current review begins by acquiring
a comprehensive understanding of the principles of fiber optic sensing, concept of DAS technologies, Elements, and operational
mechanisms. To achieve this, a thorough analysis of the literature from reliable sources, including Science Direct, Google Scholar,
Elsevier, ResearchGate, ASME, ASNT, Optica, and Google web search is necessary. However, to search following keywords have been
identified: Distributed Acoustic Sensing (DAS), Development from initial phase, Fundamental of DAS, Elements, Variants of DAS
technology, and Characteristics of DAS, Fiber Optic Sensing Fundamentals, Types of Fiber Optic Sensing and advantages of Rayleigh
Scattering over Raman and Brillouin Scattering, Data-Driven Predictive Maintenance using ML/DL, Anomaly detection using ML/DL
and Hybrid ML/DL, Common types of railroad defects, Railroad Anomalies, Current methods of railroad anomaly detection, Anomaly
detection of Railroad, Train Tracking using DAS-Fiber Optic Sensing, Condition Monitoring of Railroad, Structural health Monitoring
of Railroad using DAS-Fiber Optic Sensing, Monitoring Railroad safety Using DAS-Fiber Optic Sensing.

3.2. Exclusion and inclusion approaches

Authors collected the most up-to-dated research works except the research works having development, principle of DAS, Optical
Scattering, Fiber Optic Sensing. The inclusion/Selection and exclusion strategy for articles selection goes into a rigorous process. The
inclusion and exclusion of collected paper process is to define a boundary for SLR, which depends on the application and future
research works of the current authors. The focus of this review is on DAS-Fiber optic sensing (Rayleigh Scattering), ML/DL Date-driven
approach for CM, SHM of Railroad. However, a total of 145 articles were selected out of 450 based on the defined boundary, their
relevance, quality, and alignment with the current paper research goals. It was observed that although the research on DAS, Optical
sensing Fundamentals, Anomaly detection using ML/DL in other fields are very common in all the databases, the research work in
railroad anomaly detection using DAS-fiber optic sensing, and ML/DL technologies is too limited which is currently in development

3
M.A. Rahman et al. Mechanical Systems and Signal Processing 208 (2024) 110983

Fig. 1. Flow chart of Article Selection Process for the SLR.

phase. In addition, Google Scholar has the maximum number of publications with all the selected keywords of the current research
work, as the publications are summarized and linked to other databases mentioned above. Fig. 1 shows the article collection process
from various databases of this review work on Railroad CM, SHM using DAS and ML/DL.

3.3. Results

Starting at section 3, the review on selected articles from the SLR papers has been shown, and in section 7, a comprehensive
summary and the findings from the review work are presented.

4. Principles of fiber optic sensing and concept of DAS

Numerous sensors have been designed in recent years for in-situ SHM and CM of railroad systems, which can be used as vehicle-
mounted or track-mounted sensors. However, the system requires highly configured data-processing buses, that can collect from
multiple sensors, vehicle-based or track-based, along with advanced image processing tools [22]. The sensors for SHM are categorized
into piezoelectric sensors, microelectromechanical systems, and fiber optic sensors [14]. Most of the sensors are either expensive or
ineffective for the railroad’s harsh environment, high speed, and heavy loads. Moreover, owing to a higher range of distances and a
complicated network of rail structures, maintaining proper maintenance planning has been getting convoluted with financial in­
vestment, manpower, and scheduling of operations. Such a long distance and huge volume of infrastructure require a large amount of
equipment, sensors, and finance. Some other sensors are designed, using the Global Positioning System (GPS) for train tracking
[23,24]. However, GPS systems are not capable of transferring accurate signals for tracking rails. In certain cases, such as deep tunnels,

4
M.A. Rahman et al. Mechanical Systems and Signal Processing 208 (2024) 110983

Fig. 2. Single Point, Multi-point, and Distributed Sensors.

hillsides, and other restricted areas, the wireless or GPS will not properly work [25,26]. To determine the various states-condition and
health of structures, DAS can be a suitable solution for the health and CM of train infrastructure with lower cost and efficient real-time
updates [27–29].
Curtis D. Kissinger first secured a patent for optical fiber sensing in 1967, which is now a widely used technology [6]. The original
optical sensing device, referred to as photonic sensors, is made up of a branched fiber filament, with one-half of the bunch illuminating
a target while the second half receiving reflections from the targeted surface. The development and application of Optical Time
Domain Reflectometry (OTDR) [30,31] emerged in the late 70 s. The OTDR was developed to precisely detect faults and the magnitude
of defects in an infrastructure at specific fiber optic locations. Its functions include modulating and transmitting a string of swiftly
backscattered and mirrored light pulses inside the fiber line. Optical fiber optic sensors can be divided into two categories [32],
distributed and point-based sensing technologies. Fig. 2 shows a schematics comparison of the point sensor versus the distributed
sensor.
Optical fiber point sensors are traditionally passive point-based transducers that are designed to measure and transmit parameters
(such as temperature, strain, and sound) to a remote sensor control unit using interconnecting fiber optic cables. The sensor control
unit will typically transform the optical signal into a digital or analog value that can be processed by the control system. Fig. 3 shows
the operational principle of optical fiber classifiers, and illustrates how seism-acoustic pressure results in acoustic reflections.
DAS techniques have advanced considerably with substantial improvement in performance measures such as polarization,
coherence fading, and spatial resolution. Coherent optical time domain reflectometry (C-OTDR) and phase-sensitive coherent optical
time domain reflectometry (Ø-OTDR) are two categories of DAS systems that are based on OTDR. C-OTDR includes narrow sensors for
linewidth lasers, which have a consistent frequency and a substantially longer persistence duration, which is the key distinction be­
tween C-OTDR and traditional OTDR. The basic difference between OTDR and Ø-OTDR is the laser applied. In Ø-OTDR, the laser’s line
width is typically not more than 100 kHz, leading cohesion in length to be significantly longer compared to the pulse width [28].
By using the coherent Rayleigh backscattering of the probe laser in Fiber Under Test (FUT), the DAS device can determine the
physical field in the surrounding area. The external environment influences the characteristics of fiber optic cable. The environmental
influences include the physical effects, such as elastic-optical effect, thermos-optical effect, and thermal expansion. These effects
modulate the probe laser’s optical characteristics (amplitude, phase, frequency, etc.). After being identified and demodulated, these
features can be used to derive the physical field and enable sensing at every point along the fiber cable [34].
For the DAS setup, two essential system components are required, the processing of the interrogator system at the top, and the
arrangement of the fiber structure [35]. An interrogator unit, a processing unit, and a control unit are the three parts that make up the
surface section of the DAS system. A schematic shown in Fig. 4. The control unit, as suggested by its name, controls the interrogator and
processing device, and is like an industrial PC in both size and power usage. The processing unit is the “intelligence”, and the “muscle”
of DAS devices is the interrogation module. When the interrogator module shoots a laser pulse into a signal-mode fiber, the light that is
backscattered to the interrogator unit travels down the fiber’s length. The interrogator device processes the backscattered light to
separate the pressure wave (acoustic wave) signals from each position along the cable. The processing unit receives this data, after
which the acoustic impulses are analyzed and sent to a monitor.

4.1. DAS principle with OTDR technique

The initial DAS system was developed by utilizing the OTDR [36–38] technique. Short light pulses are sent through a fiber by an
OTDR. These light pulses are then scattered by fiber interruptions, for example defects, junctions, and clips. The backscattered signals
are then picked up and examined by the OTDR device. Events are characterized by measuring the signal strength over specific time

5
M.A. Rahman et al. Mechanical Systems and Signal Processing 208 (2024) 110983

Fig. 3. Operational principle of optical fiber classifiers [33].

Fig. 4. Component of the DA system.

intervals. The OTDR uses the following formula to determine distance:

c t
Distance = × (1)
n 2

where t is the time between the moment a pulse is launched and when it is received, n is the index of refraction of the fiber optic cable,
( )
and c is the speed of light (2.998 × 108 ms− 1) in a vacuum. The value of the refractive index (n) is around 1.5 and the factor 12 comes
from the fact that probed pulse travel and backscattering light travel are in opposite directions [39]. Fig. 5 shows the working principle
of OTDR.

6
M.A. Rahman et al. Mechanical Systems and Signal Processing 208 (2024) 110983

Fig. 5. OTDR and DAS working principle.

4.2. DAS principle with C-OTDR technique

The DAS system, commonly referred to as coherent optical time domain reflectometry(C-OTDR) [37,38,40,41], entails detecting
extremely low levels of the backscattered signal by successively sending extremely coherent pulses traveling down an optical system.
The method relies on detecting Pico-strain level signatures in the fiber brought on by vibrations caused by activities near the fiber-optic
cable. The disturbance changes the scattering spots in the fiber core that result from the sub-wavelength heterogeneities created during
the drawing of the fiber on a molecular level. As a result, Rayleigh’s backscattered laser signal undergoes a distinctive and compre­
hensible alteration. This signal is then evaluated at the interrogator unit (IU) to produce several distinct, concurrently sampled acoustic
signals from the fiber [9]. Each signal is associated with a particular channel along the fiber, whose spatial resolution is determined by
the width of the brief (1–20 m) laser pulses pumped into the fiber cable. C-OTDR has a longer gauge length and notably improved
spatial resolution, the reason for which it has been employed for long-range applications such as railroad health monitoring [42]. Fig. 6
shows the principle of the basic experimental setup with OTDR (blue) and C-OTDR (red) techniques.

4.3. DAS principle with Ø-OTDR technique

Another well-known method of DAS is phase-sensitive optical time domain reflectometry (Ø-OTDR), which is introduced in articles
[29,34,27,43]. The backscattered coherent Rayleigh light is fed through an unbalanced Mach-Zehnder fiber interferometer (MZI) and
into a detector, similar to a typical OTDR system, after being launched into the fiber in a brief pulse. The gauge length is determined by
Ldelay/2, where Ldelay is the delay fiber length of the MZI, and the phase change is directly proportional to the change in strain between
the two locations. The sensing principle of the Ø-OTDR technique is shown in Fig. 7.
A three-port MZI and three detectors are utilized in the differential and cross-multiply phase demodulation system to avoid signal
fading. It is assumed that there are two groups of scattering centers, A and B separated by L meters. Using an MZI with a path imbalance
of twice this length 2L, the phase difference between the backscattered light from the two parts is compared. While ignoring fiber loss,
the backscattered field from A and B at the photodetector is expressed by [43]:
EA = E0 exp[i(ωt + φA ) ] (2)

Fig. 6. Basic experimental setup with OTDR (Blue) and C-OTDR (Red) techniques. (For interpretation of the references to color in this figure legend,
the reader is referred to the web version of this article.)

7
M.A. Rahman et al. Mechanical Systems and Signal Processing 208 (2024) 110983

Fig. 7. Sensing principle of Ø-OTDR technique; L, and represent the phase, length, and length change of the rail track, respectfully [27].

EB = E0 exp[i(ωt + φB ) ] (3)

where φA and φB are the phase of backscattered from the light sources A and B, respectfully, the angular frequency and magnitude of
the electric field of the backscattered light are presented by ω, and E0 with respect to the time t. The combined intensity of the
backscattered signal can be expressed as:

Idet = (EA + EB )(EA + EB )* = 2E02 cos(φA + φB ) (4)

An MZI with a 3-output pot coupler was used to overcome the fading issue. The 3 × 3 coupler’s three arms output intensity can be
written as:
I1 = I0 [M + Ncosφ] (5)

I2 = I0 [M + Ncos(φ + 2π/3)] (6)

I3 = I0 [M + Ncos(φ − 2π/3)] (7)

where I0 is the intensity of the input signal, M and N are constants, and φ is the phase difference of the light reflecting through the two
arms of the MZI. The output voltage can be written as:

Table 1
Comparison of the basic parameters of DAS interrogators.
Parameter IDAS3 CRI-4400 (Halliburton) MS-DAS2000 (Ovlink) QuantX HDAS Helios DAS
(Silixa) (Optasense) (Aragon) (Fotech)

Sensing Range without repeaters(km) 50 50 20 50 70 50

Strain sensitivity(ε) 10-9 10-9 10-9 10-9 10-9 10-9
Spatial resolution(m) 1 1 2 2 10 1

8
M.A. Rahman et al. Mechanical Systems and Signal Processing 208 (2024) 110983

√̅̅̅
VPh = 3φ (8)

Furthermore, various testing methods have been proposed by researchers, such as intelligent Distributed Acoustic Sensing (iDAS)
technology [32], Chirped-Pulse Phase-Sensitive OTDR (CP-φOTDR) [44,44], Frequency-Division-Multiplexing Coherent OTDR (FDM-
OTDR) [45], and other contributions based on OTDR. Various DAS interrogators have been developed in the last 10 years. In 2022, Zhu
briefly represented the comparison of different DAS interrogators device as summarized in Table 1 [46]:

4.4. Characteristics of DAS

Two major parameters of the DAS system are spatial resolution (SR), and signal-to-noise (SNR). SR helps to differentiate the po­
sitions and SNR assists in sensing the Rayleigh scattering signals within the range, depending on the fiber cable attenuation coefficient.
Coherent sensing is used to increase the SNR of the OTDR system. The laser source has a narrow linewidth and a long coherent length in
the C-OTDR, which differs from a conventional OTDR system [47]. As presented in Table 2, the DAS characteristics are collected from
different publications. In 2017, Waagaard et al. reported that the DAS technology was utilized at 148 km for standard single-mode fiber
√̅̅̅̅̅̅
(SMF), and 171 km for low-loss SCUBA 125 fiber cable with a self-noise range of 4.5 to 50 mrad/ Hz, strain resolution 1.4 ~ 0.3 pε/
√̅̅̅̅̅̅
Hz, and spatial resolution of 10 to 34 m without using any inline amplification [48]. In 2021, Yan et al. proposed an ultra-long
distributed sensor. Yan presented the forward transmission, continuous detection, and frequency-modulated signal delay lines for
higher-frequency ultrasound testing [49]. Continuous or coherent detection was utilized to recover the vibration-induced phase
fluctuations of the base-band signals, as well as the intermediate frequency signals produced by the frequency-modulated optical delay
line. At the far end of these two fibers, an acoustic-optic modulator and a time delay fiber were used to create the frequency-shifted
signal delay line. Due to the asynchronous operation, near to 125 m of spatial resolution could be achieved around 615 km of sensing
range for acoustic activities with a frequency greater than 1 kHz because of the characteristics of forward transmission and coherent
detection, the suggested technique had averages of extremely wide response, a longer sensing range, and straight forward sensing
structure. In the same year (2021), Masoudi et al. presented a DAS technique with a sensing range greater than 150 km, having low-loss
improved -backscattering fiber added to the far end of a regular single-mode fiber to extend the sensing capacity [50]. The minimum
measurable strain of 40 nε across a spatial resolution of 5 m was displayed by the combined system. Most recently, a report has been
published on the DAS development, which resulted in a field-lab hybrid link over 1000 km employing chirped pulses with correlation
√̅̅̅̅̅̅
detection and 20x frequency response and achieving a sensibility of 100 pε/ Hz at 20 m spatial resolution [51]. A novel high-sensitive
technique was proposed which was based on FDM-OTDR, illustrated the tests, and, compared the designed FDM-OTDR with traditional
C-OTDR techniques, resulting in the improvement of sensitivity by 8 dB compared to the conventional C-OTDR, concluded that at a
spatial resolution of 1 km, the 3 dB resolution and Rayleigh dead zone of traditional C-OTDR and proposed FDM-OTDR were similar
[52].

Table 2
Characteristics of DAS.
DAS, sensing techniques Sensing Length Frequency Self-Noise Strain Spatial Year Ref.
Response Resolution Resolution

Ø-OTDR 171 km 4.5 ~ 50 1.4 ~ 0.3 pε/ 10–30 m 2017

√̅̅̅̅̅̅ √̅̅̅̅̅̅ [48]
(Low-loss SCUBA) mrad/ Hz Hz
148 km(Standard
SMF)
Forward Transmission, 615 km >1kHz – – m [49]
Coherent detection, (20 kHz at 615)
Frequency-shifted optical
delay lines
√̅̅̅̅̅̅
Low-loss enhanced km ~ 100 Hz – ε/ Hz 5m
[50]
backscattering optical (Minimum at
fiber to single-mode fiber 1HZ)
√̅̅̅̅̅̅
C-OTDR, km 20x Frequency- 191.75 THz to 100 pε/ Hz 20 m 2022
[51]
Frequency-diversity Diversity 194.25 THz
chirped-pulse,Real-time Noise loading
coherent DP-16QAM channel.
transponder,
FDM-OTDR (Frequency- Tested at 250 km Sensitivity – 3 dB resolution km [27,52]
division multiplexing) /Relative frequency
improved by 8 dB
Ø-OTDR 1 km sensing fiber, 500–5000 Hz – 40 nε(SNR of 2m 2013 [43]
and 10 km for rail (Frequency Range). 1).20εn
tracking speed 360 83 Hz to 90 Hz (minimum strain
km/h. (Frequency resolution
resolution) measured)

9
M.A. Rahman et al. Mechanical Systems and Signal Processing 208 (2024) 110983

5. Fiber optic sensing scattering mechanisms

A fiber optic cable is capable of continuously and instantly measuring owing to a technology called distributed sensing. Distributed
sensing systems have been utilized for measuring the temperature and strain in the form of distributed temperature and strain sensing
(DTSS) [53], distributed temperature sensor (DTS), and distributed strain sensor (DSS) [19,14,20,21,22–26,29,30,54]. A DAS system
enables the detection of acoustic frequency strain signals across long distances and in severe environments. Distributed sensing, in
contrast to conventional sensors, relies on optical fiber rather than manufactured sensors to measure data at predetermined points.
Without any additional transducers in the optical route, the fiber optic cable serves as the both sensing element and data carrier.
Examining several studies shows that distributed scattering is best suited for long-distance and condition measurement for continuous
monitoring [28,55–57]. The scattering mechanism in fiber optics can be divided into three categories including Rayleigh, Brillouin,
and Raman Scattering [58–65]. In this review, only the DAS system designed based on the Rayleigh scattering is demonstrated for
railroad CM and SHM.

5.1. Rayleigh scattering

Rayleigh scattering (RS), known as elastic scattering, is a linear scattering process, where the scattered power is proportional to the
incident light compared to the wavelength of the incident light. On the contrary, Raman and Brillouin scatterings are both inelastic
scatterings since they include specific frequency modifications. The optical time domain reflectometry (OTDR) [56,30,66], and Optical
frequency domain reflectometry (OFDR) [67–69] integration systems are two Rayleigh scattering-based methodologies that now rule
the field of DAS technology.

5.1.1. OTDR for condition and structural health monitoring

In DAS systems, the interrogator functions similarly to a RADAR, which means it emits wave pulses. A series of pulses into the fiber
interact with the return of the naturally occurring scattered and backscattered signal of events over time [53]. This allows the dispersed
sensor to take measurements over the whole length of the cable. Tiny strain (or vibration) events inside the fiber optic that are brought
on by localized acoustic energy define interactions within the fiber that lead to backscatter, or light reflections, as this pulse of light
travels down the fiber optic. When it reaches the interrogation unit, the backscattered light ascends the fiber once more and is sampled
at Rayleigh frequencies. This process, known as OTDR, enables the backscatter event to be precisely mapped to a fiber distance. C-
OTDR is the foundation of the bulk of distributed acoustic sensor systems now available.
The key component in evaluating an optical fiber’s transmission loss is RS, which results from fused silica within the fiber. The
refractive index varies along the fiber as a result. The optical fiber’s inherent loss coefficient from Rayleigh scattering is written as:
A
αR = (9)
λ4
( )
where A ≈ (0.7 0.9).dBkm− 1 .μm− 4 , and α Unitkm− 1 is the coefficient of attenuation, of a typical SMF which is roughly 0.2 dB/Km at
the common optical communication wavelength of 1550 nm. RS is the major loss at that wavelength when the scattering coefficient is
between 0.12 and 0.16 dB/km. The following equation illustrates how the input optical power will decrease exponentially as a function
of propagation distance z due to attenuation loss: [70]
P(z) = Po e− αz
(10)

where incident light is P0 , total length is L, and at location z, and the transmitted power is denoted by P(z). If the attenuation coef­
( )
ficient, α Unitkm− 1 will be represented as dB/km, considering αdB ≈ 4.34α , then the transmitted power can be expressed as:

P(z) = Po 10− (αdB z/10)

(11)

The RS that occurs because of the inhomogeneity of the refractive index during the fiber manufacturing process causes the light
power to continuously deteriorate as a light wave propagates along the fiber optic cable. Therefore, when a laser pulse with peak power
of P0 and a pulse duration of T is delivered into one end of a fiber, with fiber attenuation (α), and z = Vg t/2n, changes from the fiber
length 0 to L, the RS signal will decrease by dPR across the length of fiber dz, which can be writing as:
dPR
= Cb αR Po e− 2αz
(12)
dz
The backward Rayleigh scattering ratio Cb, for the optical fiber, is:
( )2 (√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ )2
NA 1 n21 − n22 1
Cb = × = × (13)
nl m n1 m

The refractive index (RI) of the core and cladding are n1 and n2, respectively, while the numerical aperture is NA. The RI profile of
the fiber affects factor m. The recapture coefficient of a Gaussian beam waist ω(λ, z) is represented by:

10
M.A. Rahman et al. Mechanical Systems and Signal Processing 208 (2024) 110983

( )2
3 λ
Cb (λ, Z) = (14)
2 2πn1 ω(λ, z)
Solving the equations, the Rayleigh signal power is:
1
PR (z) = αR Po Vg TCb (z)e− 2αz
(15)
2
The formula z = Vg t/2n can be used to convert the delay time in OTDR to the distance, and the velocity series of light traveling
within the fiber cable can be captured with.
Vg(z) = c/ng (z). Moreover, the fault locations or the exceptional temperature, or strain sections are represented by the spatial
resolution Zmin of OTDR-distributed sensing properties, which can be written down as Δzmin = cτ/2n, where,τ is the pulse width of
incident light, n is the fiber core’s RI, and c is the speed of light in the vacuum.

5.1.2. OFDR for condition and structural health monitoring

OFDR can also be divided into two categories, incoherent OFDR (I-OFDR) and coherent OFDR (C-OFDR) [69,69,71–73]. The I-
OFDR technique amplifies a continuous wave (CW) optical carrier (probe signal), and a constant amplitude RF signal whose frequency
is periodically modulated across a specific frequency range, either step-frequency method or sweep frequency method (continuous).
The detection of frequency responses is usually done by the Rayleigh Backscattered Signal (RBS) or the back-reflected signal,
modulated in a vector signal analyzer (VSA) as a function of modulated frequency. On the contrary, in the C-OFDR category, without
mode hopping, the adjustable laser source’s optical carrier frequency is swept linearly in time. The frequency-shifted probe signal is
then divided into two paths, one to the device under test (DUT) and the other serves as a reference signal. The coupler is coherently
interfered with by the reference signal and the signal emanating from the test arm’s reflection sites. This interference signal has beat
frequencies that are seen as peaks on the network screen following the Fourier transform (FT) of the time-sampled characteristics.
In 2010, a comparison between OFDR and a high-resolution OTDR was presented by Liehr et al., showing that the OFDR provides a
wider dynamic range at shorter probe periods [74]. By utilizing OFDR, the strained fiber sections in pre-fluorinated polymer optical
fiber (POF) may be identified with better spatial resolution, at significantly quicker interrogation periods relative to the devices. The
OFDR sensor is effective in a variety range of senses because of its outstanding spatial resolution and broad dynamic range [75].
Various studies conducted on techniques of OFDR show that the frequency domain method is not complex but has drawbacks related to
sensing range. The implementation of the short-tuning-range-based method is restricted to high-accuracy measurement since it en­
hances the regularity of the test light at the cost of spatial resolution. Otherwise, OFDR is a smart choice for a high-accuracy DAS
system, which is widely utilized in the construction, energy, automotive, and aerospace industries, because of its better sensing ca­
pabilities at short distances, higher spatial resolution, and shorter sensing periods. The OFDR sensing capabilities also go beyond
temperature, strain, and fiber imperfection. The beating signals originate from the OFDR analyzer by optical interference between two
signals develop from a comparable linearly chirped highly coherent light source. Signal one Es (t) is the light that is reflected or
backscattered from an optical route of a fiber that is being tested, while the other, Er (t) is the optical reference from the reference path.
For a laser source with a linear adjustable frequency speed γ, the optical field Er(t) expressed [68] as:
{[ ]}
Er (t) = Eo exp j 2πfo t + πγt2 + e(t) (16)

where,f0 represents optical frequency, et is the nonlinear phase noise of the laser source, and E0 is the amplitude of the reference signal.
Reflected or the backscattered signal Es (t) with reflectivity can be written as:
√̅̅̅̅̅̅̅̅̅ {[ ]}
Es (t) = R(τ)Eo exp j 2πfo (t − τ) + πγ(t − τ)2 + 2πe(t − τ) (17)

where R(τ) denotes the reflective attenuation of the optical fiber that is connected to the time delay. As a result of the interference of
the local modulated signal Es (t) and the received signal Er (t) from the test area, the beating signal I(t) will be generated which can be
expressed as:
√̅̅̅̅̅̅̅̅̅ {[ ]}
Es (t) = R(τ)Eo exp j 2πfo (t − τ) + πγ(t − τ)2 + 2πe(t − τ) (18)
( )
where fb is the beat frequency fb = γτ , and the phase noise in the beating signal is represented by the phrase e(t) − e(t − τ).

5.2. Advantages of Rayleigh scattering over Brillouin and Raman scattering

He et al. presented advantages of the Rayleigh backscattering over Raman backscattering and Brillouin backscattering in 2021.
Compared with the DTS based on Raman backscattering and the distributed strain/temperature sensor based on the Brillouin back­
scattering, DAS based on Rayleigh backscattering revealed the following advantages. Firstly, the coverage of DAS is massive for
detecting events far away along the fiber cable, as the acoustic signal can propagate at a long distance in the optical medium. Secondly,
the elastic spreading of the incident light is the Rayleigh backscattering, which is highly sensitive due to interference of consistent
probing light within a spatial resolution. In contrast, constant light probing is not possible by Brillouin backscattering. The magnitude
of strain resolution of RS-based DAS is higher than the maximum strain, which is nearly equal to multiple strain sensors. Finally,
compared to Raman backscattering and Brillouin backscattering, the Rayleigh backscattering is stronger, highly sensitive, and has a

11
M.A. Rahman et al. Mechanical Systems and Signal Processing 208 (2024) 110983

higher sampling rate, which enables the collection of adequate information about the events within the range [47].
However, it was found from this study that Rayleigh’s Backscattering exhibits superior long-range event detection capabilities, as
illustrated in earlier sections of this study. When it comes to sensing events from railroad systems, it is evident that Rayleigh’s
Backscattering is the appropriate method compared to other scattering methods such as Raman and Brillouin Backscattering. In
particular, the C-OTDR setup, with its enhanced spatial resolution, long-range events sensing capabilities, and various other param­
eters as depicted in Table 2, strongly demonstrates that Rayleigh’s Backscattering is preferable to Raman and Brillouin Backscattering
in this application, such as SHM, CM, and anomaly detection of railroad.

6. Data-driven predictive maintenance for CM and SHM

The basic requirements of data-driven intelligence techniques for large scale infrastructures are substantial capacity, long distance,
and high SNR which can be achieved using fiber optic sensing methods [76]. With the growing need for predictive maintenance,
massive volumes of data have been produced due to recent trends toward digitalization in a wide range of industries. Expectedly, this
has led to an increase in the use of advanced ML algorithms [77–80] on big data. More sensors are being added to modern in­
frastructures (railroads, aircraft, pipelines, and civil structures), making ML-based diagnostic and prognostic approaches more popular
than traditional methods for creating preventive maintenance strategies. Planning for the operation, maintenance, and logistics is
derived from predictive maintenance, which provides an integrated solution for assessing the present or real-time, and future health
diagnosis of the entire system. Finding sufficient run-to-fail in-service data is exceedingly difficult since railroad systems must be in an
extremely safe and reliable mode of transportation. For railroads, a data-driven predictive model for determining health grade and
estimating Remaining Useful Life (RUL), run-to-fail data is necessary. The simulation strategy performed through the digital twin of
integrated preventive maintenance in any structure supports the maturity and validation of the ML-based diagnostics and prognostics
framework. Condition-based diagnostics and prognostics methodologies are classified into three major categories data-based, model-
based, or hybrid [78]. An ML-based Diagnostics & Prognostics (DnP) Framework has been developed to create a data-driven solution.
Information from the big data environment is ingested by the DnP framework along with sensor and built-in test equipment (BITE) data
produced by the integrated systems health monitoring (ISHM) simulation framework under various fault scenarios. Condition-based
anomaly detection using ML can be divided into three categories: supervised, semi-supervised, and unsupervised, depending on the
training method [73,81–84]. The various categories of ML-based anomaly detection are presented in Fig. 8.
Prognostics and diagnostics were first used in the medical profession [85]. Certain researchers described the terms diagnoses and
prognostics as follows. Switching from the conventional failure-and-fix methods to a predict-and-prevent methodology. Around a
decade ago, forecasting and RUL prediction were the only uses of prognostics that were addressed. Structural diagnosis is also part of
prognosis and health management. Structural diagnosis can be defined as the aim to isolate defects and identify the root cause of
failure. It may be summed up as the process of finding and determining the relationship between cause and effect. The prognosis shall

Fig. 8. State-of-the-art anomaly detection using Machine Learning and Deep Learning.

12
M.A. Rahman et al. Mechanical Systems and Signal Processing 208 (2024) 110983

be regarded as the health assessment and prediction process, which consists of detecting upcoming failures and forecasting RUL. A
diagnostic system includes the data collection system, signal preprocessing, and feature extraction module. A knowledge-based fault is
usually acquired from the physical model, specialist, and historical data. By comparing the feature extraction findings with the
knowledge base, it is ultimately determined what kind of failure has occurred and why. However, prognosis involves several additional
processes. Performance evaluation, degradation algorithms, fault detection, and prediction are used in prognostics in addition to
condition recognition and expertise-based faults, as in diagnostics. Therefore, prognostics seek to answer the following concerns: How
is the structure functioning right now? (Evaluation of performance), When will the system break down? (RUL), What significant faults
will cause downtime? (Root cause analysis (RCA)), Why does the error happen? (Diagnosis/ Prognosis).
Various algorithms based on deep learning and machine learning have been proposed by researchers and Engineers for anomaly
and condition monitoring in various sectors such as Deep Anomaly Detection for Industrial IoT (Internet of Things)[86], Real-time
anomaly detection in WAAM (Wire-Arc Additive Manufacturing) using CNN[82], Unsupervised ensemble learning for video anom­
aly detection[82], Semi-supervised anomaly detection framework based on statistics and deep learning for machine health prediction
[73], and Railway Internet-based anomaly detection using Extended Neural Networks [87]. The integration of such cutting-edge
technologies has opened up new avenues for improving the efficiency and safety of various industries. The field of railroad anom­
aly detection and condition monitoring is one area where these advancements have had a significant impact. However, the integration
of DAS, Machine Learning, and Deep Learning, can be a powerful tool combination for addressing the challenges in railroad infra­
structure maintenance, and safety. DAS converts optical fiber cables into a network of thousands of monitoring sensors, which rev­
olutionizes the way railroad condition monitoring and anomaly detection, allowing for continuous and high-resolution assessment of
vital assets like tracks, bridges, and tunnels. Anomalies in these structures can be detected in real-time or near real-time by applying ML
and DL algorithms to the DAS dataset, offering a proactive approach to maintenance and safety. This combination with DAS utilizes the
power of innovative technologies including IoT, GPS, CNN, LSTM, KNN, NN, SVM, and any hybrid deep learning or machine learning
models, to propose comprehensive and effective solutions for the rail industry. The connectivity of IoT devices to railroad in­
frastructures may add a novel approach to data collection and analysis, as these devices, equipped with various sensors, provide a
wealth of information about equipment performance, environmental conditions, and potential hazards. ML and DL algorithms have
the capability to analyze this data to identify anomalies or patterns indicative of structural weaknesses or issues with equipment
functionality. Additionally, GPS plays an important role in train tracking, and it can pinpoint the location of conditions by geographical
coordinates by using ML, or DL models. However, there are some limitations to IoT, GPS-based train tracking, condition monitoring,
and anomaly detection because of indirect sensitivity, lower resolution, structural configurations, and environmental hazard issues.
Nevertheless, for data processing and algorithm development, researchers proposed numerous ML methods for anomaly detection.
Anomaly detection techniques include both supervised and unsupervised machine learning. The major supervised machine learning
for anomaly detection is K-Nearest Neighbor (K-NN), Bayesian Network (BN), Supervised Neural Network (NN), Decision Tree (DT),
and Support Vector Machin (SVM), while the unsupervised machine learning mostly used for anomaly detection are K-Means, Fuzzy C-
Means, Unsupervised Niche Clustering, Unsupervised Neural Networks, and One-Class Support Vector Machine. The researchers found
that the supervised ML methods, significantly outperform the unsupervised machine learning methods, and the best-performance ML
methods achieved by the non-linear methods, like as SVM, multi-layer perception, and rule-based techniques [88].
On the other hand, for time series data analysis deep learning-based anomaly detection methods gained attention among re­
searchers, as numerous deep learning methods for feature extraction, statistical analysis, and events predictive modeling have already
proven their effective performance. Among the most basic forms of deep learning neural networks is the Multilayer Perceptron (MLP),
in which all layers are fully connected. As a result of their simplicity, their classification accuracy is usually outperformed by other
deep learning models that are designed specifically to capture common characteristics that exist in sequential data [89]. The appli­
cation of deep learning is rising in a variety of sectors such as smart manufacturing, production equipment, infrastructure facilities,
Logistic automation systems, smart energy management, network and farmwork, structural health, and so on [90]. One of the deep-
learning approaches is RNN which can recognize pattern sequences and predict events. Some of the extended versions of RNN are
LSTM and GRU. They have multiple gates: keep, and forget gates, and the models are suitable for long-term dependencies of sequential
data [91,92]. Anomalies can be determined by detecting the differences between predicted and actual signal data. This type of
anomaly detection strategy applies in the following research work [81,93–98]. For short-term data analysis, another DL approach
CNNs has been utilized by researchers for anomaly and condition monitoring sequential data. CNNs can capture complex features in
sequences by stacking the convolutional layers, and pooling layers. AlexNet, GoogleNet, Visual Geometry Group (VGG), Highway
network, Residual Network (ResNet), DenseNet, Residual Transform Network (ResNext), Residual Attention Network (RAN), Transfer
Learning, and High-resolution network (HRNet) are some of the CNN variants reviewed by the previous researchers [99].
Furthermore, hybrid deep learning algorithms have become popular for sequential forecasting projects in recent years. This type of
combination leverages the advantages of both conventional ML algorithms and DL methodologies. These methods are particularly
effective for complex structural anomalies with a variety of features or characteristics. The CNN-LSTM model is used to solve
sequential patterns forecasting by data reshaping into 3D tensors, enabling the model to handle vast amounts of data with higher
accuracy. Where the model captures both spatial and temporal dependencies in the data, making it useful for anomaly forecasting
[100–102]. In 2023 Han et al proposed a hybrid model Transformer- TimesNet for detecting multivariant time series anomalies such as
point anomalies, contextual anomalies, and collective anomalies [103]. Transformers useTimesNet model to convert the 1D time series
data into a 2D variant by using the frequency components and extracting important features from the transformed data; while the
Transformer uses the gained features to solve global dependencies. Anomaly detection in cloud data centers is based on Improved Grey
Wolf Optimization and Improved Convolution Neural Network (ImGWO-ImCNN) proposed by Garg et al. n.d. [104], where GWO is
used for multi-object feature extraction and CNN is used for anomaly detection with output result of both high precision and accuracy.

13
M.A. Rahman et al. Mechanical Systems and Signal Processing 208 (2024) 110983

However, the fault diagnostic models based on the hybrid ML and DL are popular among researchers. The following are some hybrids
DL-ML examples: CNN-SVM [105,106], CNN-KNN [107], CNN-Random Forest [108] and so on.

7. Application of DAS for Railroad Systems

For many years, DAS has developed an established application market, dealing with similar asset monitoring tasks and playing a
crucial role in a variety of sectors. Fiber optic cables along the track are used by DAS to find any anomaly indicators. DAS allows for the
measurement of an objective parameter as a function of fiber length. DAS offers full, quick, and accurate coverage of every section
under inspection, compared to any conventional NDT technique where the coverage or scanning area of the sensors is relatively
constrained [109]. Due to the long-range distance, multiple physical parameter measurements, and spatial sensitive advantages of the
DAS technique, its application has been rising in various sectors. DAS has been widely used for railroad health monitoring and tracking
[27,29,27], border control and security applications [110]. Pipeline monitoring [111–113], cable monitoring (communication and
power system) [114], and cement casting systems [115] are among the other application of the DAS systems. In environmental fields,
underwater critical infrastructures [116], and various geophysical-seismic monitoring [117] have used the advantages of DAS. As the
subject of this article is the application of Rayleigh scattering-based DAS for railroad systems, the contributions to rail structural
health, train tracking systems, railroad defects, and railroad security systems have been presented here.
Records on train derailments from the FRA (FRA.F.6180.54), between the years 2001 and 2010 were investigated for each type of
track, based on the frequency of emergence by reason and the number of cars damaged [118]. The investigation was performed to
assess the consequences of incident roots, track type, and accident velocity. According to the data, train derailments are most
frequently caused by broken rails and welds. As a result, measurements to prevent accidents with such a high frequency and severity
need and do receive significant and major attention. According to the article FRA16 published by FRA and the US Department of
Transportation, the three categories below are particularly applicable to the research and development of North American fiber optic
technology [119]. One of them is “Train-tracking”, which is defined as the detection of trains, analysis of timetables, control of tunnel
traffic, single-line protection, and level-crossing protection. The second category is “Tracking the condition of assets”, which focuses on
rail breaks, flat tires, catenary flashovers, infrastructure, point machines, and generators. The third one is “security”, which is the

Fig. 9. (a) Waterfall diagram and (b) transformation example for train overview (c) and rail view [120].

14
M.A. Rahman et al. Mechanical Systems and Signal Processing 208 (2024) 110983

protection of people, pets, workers, landslides, rock falls, and copper theft. Furthermore, the FRA report illustrated the following
categories that have been identified by the Association of American Railroads Fiber Optic Acoustic Detection (AAR FOAD) task force in
order of priority [119] including rail-break detection, tracking of the train, infrastructural health, track reliability, security, and
environmental hazards. Broken/loose ties, plates and bars, ballast degradation, and rail defects can affect the track reliability. Pitting,
shelling, and cracking are some of the major types of rail defects. In addition, thermal misalignment or deteriorated and crushed welds
can happen as a rail defect. The use of time-series data in real-time streaming has also been the subject of anomaly detection research.
Due to the severe environment, the scenario in the railroad is significantly more difficult than in the application of other structures. As
a result, the structure must be strong and proven to withstand these varying environments.

7.1. Real-time train tracking using DAS

Several algorithms have been developed for train tracking in real-time using DAS, ML, and signal processing. Real-time tracking is
feasible by identifying the vibration of a moving train’s positioning. All contemporary railroad safety concepts are established on
tracking a train’s location on the railroad with high accuracy. A waterfall plot, that shows the amplitude of the vibration for each
sensor site with time, is a popular method for visualizing DAS data (shown in Fig. 9a). The diagonal signal pattern in this instance
represents a moving train, and the slope is speed dependent. The train length coincides with the pattern width and as a result, it is often
constant. With the train’s speed, the height varies along the time axis. The train front gradually turns into a horizontal line rail-
centered orientation Fig. 8b, or a Fig. 8b, or a vertical line train-centered orientation Fig. 9 c [120].
The position of the train is only known in a designated “block” between two signals in modern rail operations [121]. There can only
be one train operating at a time in a block, that is enough to prevent train crashes. Using a DAS interrogation, Wiesmeyr et al. collected
long-term DAS data from two distinct test locations. In one test site, the SMF optic cable was buried in a cable trench. In the other
location, the SMF optic cable was directly clipped to the railroad. Computing the spectrum energy in ten frequency bins is one of the
processes in vibration recognition. It is based on normalizing ten energy values so that they added up to one. This reduces ten feature
values to two and utilizes a trained Support Vector Machine (SVM) to classify the feature vector of length two as background vibration.
Extraction of edges from the data, connecting the edges to existing objects and creating new ones, and using a Kalman filter to smooth
out all the objects’ trajectories are the three steps that make up the final stages of the train tracking method [122]. Fig. 10 illustrates
results from the article where the left plot shows a waterfall from the raw data, and the right plot shows vibration (black color)
detection on white background. The experiment performed by Wiesmeyr et al. used a DAS interrogator device with C-OTDR. Wiesmeyr

Fig. 10. Experimentation findings, Raw subsampled signal on the left, display of the result of vibration detection on the right [121 121].

15
M.A. Rahman et al. Mechanical Systems and Signal Processing 208 (2024) 110983

(caption on next page)

16
M.A. Rahman et al. Mechanical Systems and Signal Processing 208 (2024) 110983

Fig. 11. Train tracking and reliability analysis experimental results [63].

used the following measurement parameters: 100 ns for laser pulse length, 1550 nm for the laser wavelength, 2000 Hz for the pulse
repetition rate, 150 kHz for the signal sampling rate, 16 bits for the signal sampling resolution, and 202, 208 m/s for the speed of light
fiber.
Kowarik et al. developed routines for near real-time train tracking and reliability analysis, using DAS and ML model [123]. For the
railroad used in Kowarik’s investigation, the sensing conditions were identified along the fast track of 35 km where the tests were
performed. The findings demonstrated that, as shown in Fig. 11a, the position of the train at each instant of time, or xcenter (t), is a
standard parameter to examine the train position and timing with DAS signals f(x, t). The identification of the train position required
significant data processing even though the raw data is subject to errors and noise with the processing of spatial and temporal samples.
The speed at each time can be estimated statistically once the train’s position xcenter (t) is known (Fig. 11b). In Fig. 10, the filtering
process (Fig. 11a) reveals the train location as a function of time. The variation of train velocity over time can be seen in Fig. 10b. Peak-
finding algorithms need 300 s to filter, locate, and align bogie clusters in the dataset. The use of the Artificial Neural Network (ANN)
offers advantages in processing speed and the capacity to look at various datasets.
At Kazakhstan testing facility for railroads, the OXY DAS system was implemented as represented in Fig. 12. A moving shunting
train is depicted in the C-OTDR image projection. The third component’s red color denotes a high level of energy that was transmitted
in the 60–150 Hz frequency range. This C-OTDR image also has a wider low-frequency range and a fuzzier shape. The actual local­
ization train is represented with excellent precision by the red line, which is indicated by a white arrow. The authors strictly followed
this line, demonstrating the explicit energy pulse associated with the train’s appearance at the location closest to the proper channel.
The fiber had been buried 10 m away from the monitoring object at a depth of 50 to 100 cm [33].
Peng et al. carried out a field testing of 10.2 km track while utilizing the phase-OTDR DAS method with a spatial resolution of 20 m,
a semiconductor laser with a narrow linewidth (~3 kHz), and a 200 ns laser pulse width was used to monitor the position and speed of
trains over a long distance in real-time, which is represented in the Fig. 13. In this experiment, the monitoring of trains and speed
depended on the acoustic signal detection along a 12.1 km optical wire, buried in the earth close to the railroad tracks between 0.7 and
1.5 m deep [125].
Gao et al. proposed a method utilizing the DAS system and continuous vertical wheel-rail force reconstruction. The authors used
Finite Element Method (FEM) to simulate the parallel path’s rail strain distribution, distancing 20 ± and 40 ± mm from the neural axis
of the rail, under different load conditions [29]. A temporal force is employed on the railroad, which is moved at a velocity of 250 km/
h, and the strain responses have been recorded and represented in Fig. 14.
In 2022, Zhang G Proposed a method for monitoring the train traffic using beamforming technique on the phase-squeezed signals
generated from 6 days continuous DAS data recording sensed by a 2 km fiber optic cable into time-velocity domain, that can auto­
matically detect the position and velocity information of various rail [126].

7.2. Structural health monitoring of railroad using DAS

Damage can develop anywhere and at any time since the railroad system is extremely complicated and includes many connecting
points. It requires a robust failure analysis, and types of faults and needs a prognostic, real-time algorithm for proper controlling and
predictive maintenance.

Fig. 12. Real-Time Data Processing scheme of OXY System [124].

17
M.A. Rahman et al. Mechanical Systems and Signal Processing 208 (2024) 110983

Fig. 13. Optical fiber cable, the relative position of train and railroads [125].

Fig. 14. Simulated FEA output of DAS signal [29].

7.2.1. Railroad defects

Internal rail defects normally require specific levels of stress to progress to a visible or detectable defect size. Defect planes are
identified in different orientations. Fig. 15 (A) describes different defect planes in relation to the rail section [127]. In the vertical
plane, stresses are developed in the longitudinal direction and are normal to the rail length. In the horizontal plane, stress propagation
is horizontal along the rail. Finally, in the transverse plane, stress propagation is in a transverse direction along the cross-section of the
rail. Various categories of defects in the railroad have been identified such as geometrical and structural defects [128]. The most
common structural defects in railroads are summarized in Table 3 and Fig. 15. There are three stages to the failure of metallic
structures caused by cracks: (a) crack initiation; (b) crack growth, and (c) a subsequent rapid crack growth that results in a total
fracture, which is presented in Fig. 14(G), and (H).

7.2.2. Structural health monitoring of railroad track defects

DAS technology is a promising sensing system, where the only medium for measurement and data transmission is fiber optic
sensing. Several DAS methods have been proposed by scholars. Detecting the defects before the fully broken rail with DAS is one of the
major concerns for railroad infrastructure. The final stage of broken rails occurs by developing various cracks or defects, some of which
are represented in Fig. 15. These defects are initiated by the continuous rolling contact fatigue at various stages from passing trains.
The major issue in utilizing DAS is to identify the cracks, broken rails, their types, and stages of defects generating the signal.
Measurement of rail dynamic strain using RB and SMF optical cable in lab and field was collected by Wheeler et al. in 2018. At a
length of 2.83 m and mid-span load, a rail was tested with a simply supported beam (Fig. 16a, and b), by applying both static and

18
M.A. Rahman et al. Mechanical Systems and Signal Processing 208 (2024) 110983

Fig. 15. Various railroad defects and terminology (Modified from) [127,128].

Table 3
Common defects in railroads.
Cause of Defect Generation Type of Potential Defect Relevant Image in Fig. 15

Manufacturing process Longitudinal and transverse defects Fig. 15 (B and C)

Operational process Handballing (wheel-burn defect) Fig. 15 (D)
Rail wear and fatigue Spalling, shelling, squats Fig. 15 (E)
Assembly and joining Bolt-hole cracks Fig. 15 (F)

Fig. 16. (a) Lab Experiment, (b) Fiber cable positioning, and (c) Field test (modified) [129].

dynamic cyclic loads. At vertical distances of 20, 50, 120, and 155 mm from the bottom of the rail, four fiber optic cables were
installed. The result from the lab test revealed that fiber cable should be installed near the foot or head of the rail, because of their
larger distance from the rail’s neural axis. With the results from the lab test, the fiber cable had been installed in the field (Fig. 16c), and

19
M.A. Rahman et al. Mechanical Systems and Signal Processing 208 (2024) 110983

data had been collected from the hi-rail vehicle and passenger train. The field result concluded that the system was not able to record
the higher vibration events, but it can be used in low-vibration conditions or slow-moving vehicles [129].
Wagner et al. distinguished certain rail issues such as a gap in rails, rail breakage, and gravel on rails using the DAS system and
analyzing the data with data-driven algorithms. Wanger considered the de-noising technique for reducing additional data from the
signals. The proposed technique is separated into two parts, rail detection, and rail defect detection, which is represented in Fig. 17.
Supervised ML methods were used to analyze the known defects. For the unknown defects of rail, unsupervised ML algorithms have
been employed. The key finding of the experiment was that DAS has an excellent probability of detecting broken rails, but the rail gaps
were most clear at velocities up to 25 km/h of the train [130].
In 2018 Wiesmeyr et al. proposed an algorithm, that can be utilized for Train Tracking and Condition Monitoring of wheel Flats by
using a traditional Machine learning method support Vector Machine (SVM), and DAS-Fiber Optic Sensing technology [131]. The
researchers focused on wheel flats as it causes continuous repetitive impacts on the rail, which may result in creating major damage of
rail. A total of 5 h data recorded along different rail tracks, different sampling rate and with a common Spatial Resolution of DAS was
0.68 m. Wiesmeyr et al. research found that the proposed method can detect flat wheels, in addition to the DAS-based train tracking
which is effective within 60 m and requires careful calibration, excluding challenging areas such as shunting yards.
According to L. Xie’s proposal in 2023, real-time recognition and categorization of track issues such as rolling contact fatigue (RCF),
corrugation, and unstable sleepers along heavy-duty trains might feasible [132]. Using SVM for detection and classification, the author
developed the real-time monitoring system. The research work used a phase-sensitive optical time domain reflectometer (Ø-OTDR) to
collect data. RCF had a recognition rate of 90.6 %, corrugation had a recognition rate of 97.04 %, and unsupported sleepers had a
recognition rate of 94.6 %.
However, some other research related to rail vibration-based anomaly detection with other sensing systems (Not DAS-Fiber Optic
Sensing) proposed by various researchers are Noise Sensors based [133], Acceleration Sensors based model for high speed rail Viaduct
[134], A combined GRAS ½ Microphone, Vertical vibration acceleration sensor (352C04) for sensing rail transit vibrations [135],
Velocity Sensor and GPS based Degree-of-Freedom model for gathering support condition of rail tracks [136], Vibration based
approach using tri-axial accelerometers to measure Rail Stress [137].

7.3. Railroad safety using DAS

Wang et al. contributed to the development of a safe railroad infrastructure using phase-sensitive OTDR technology. The feasibility
of the method was tested in a field test of 20 km in length, which is illustrated in Fig. 18. The results presented by Wang show that,
along with other detrimental construction activities adjacent experiment location, the railroad traffic (speed and length of train) can be
monitored with the proposed set-up [110]. Furthermore, DAS-based smart fence systems for railroads were developed by prior Dibazar
et al. [138]. The “smart fence” can sense human footsteps, discriminate between rattle, kick, and climb, and can detect approaching
vehicles and their types.

Fig. 17. (a) Methodology (Left) of a broken rail detection technique and (b) Experimental Setup (Right), modified from [130].

20
M.A. Rahman et al. Mechanical Systems and Signal Processing 208 (2024) 110983

Fig. 18. Railroad safety in the construction field [110].

8. Conclusion

This paper systematically reviews the research evolution and application of DAS. The applications of DAS for long-term assessment
of potential risks, and enhancing safety and reliability were discussed. DAS can be used for early-stage damage detection which results
in prompt maintenance and prevention actions to avoid catastrophic failures in the railroad systems. Several DAS integrator devices
have been developed for various applications such as tracking and detection of failures. This survey study showed that tracking a train
is widely investigated by scholars compared to SHM or fault detection. SHM, prognosis, and diagnosis of railroad structure using the

Table 4
Summary of DAS technology, categories, and railroad applications.

21
M.A. Rahman et al. Mechanical Systems and Signal Processing 208 (2024) 110983

DAS technologies are still in developing states. Consequently, further investigation and experiment are required in addition to data-
driven fault detection analysis such as deep learning algorithms. Furthermore, compared to traditional point sensors, there is still a
substantial gap in terms of sensibility. A comprehensive understanding, of the relationship between pulse width, SNR, fading, sensing
distance, massive dataset, and surrounding noise must be developed. However, despite these limitations, DAS approaches may play a
significant role in many railroad industries because of advancements in the wide range of detection distance, sensitivity, multi-
dimensional monitoring, as well as integration with deep learning algorithms. Table 4 summarizes the survey and discussions in
this article on DAS technology, categories, and railroad applications.
This article can assist engineers, researchers, practitioners, and policymakers who are interested in or plan to work in the Rayleigh
backscattering-based DAS systems and railroad infrastructure monitoring by providing essential information about fundamental
background, working principles, signal characteristics, and data processing considerations. Understanding the developed techniques
and capabilities of DAS systems, associated challenges, and potentials for future development discussed in this paper will assist the
practitioners on implementation of DAS technique for railroads safety and preventive maintenance.
DAS has been already applied for several CM aspects of railroads such as tracking and positioning. Also, DAS shows many potential
applications for complicated tracking and position detection, structural integrity assessment, and safety observation in the railroad
industry need to be further investigated. Signal fading and its influence on the accuracy and repeatability of event detection is one of
the major topics which requires further investigation and research. Signal fading occurs in distributed fiber optic sensing due to
coherent interference of Rayleigh backscattered light and can affect the sensitivity of DAS. In addition, despite the several method­
ologies presented by researchers for DAS data processing, there remains much work that needs to be done on how DAS data may be
interpreted. Data processing challenges in DAS reveal many open research questions for the research community to explore. To
conclude as a take-home message, Incorporating distributed acoustic sensing (DAS) technology in railroad condition monitoring holds
immense potential for enhancing the safety, efficiency, and reliability of railway operations. This comprehensive review underscores
the diverse applications of DAS, from track defect detection and stress analysis to real-time monitoring of rolling stock. By harnessing
the power of acoustic sensing, this technology offers actionable insights, early fault detection, and data-driven decision-making for the
railway industry. Our analysis points to the need for further research, standardization efforts, and industry adoption to fully unlock the
benefits of DAS in railroad condition monitoring. With this review, we aim to inspire collaboration among researchers, practitioners,
and policymakers to advance this transformative technology, ensuring the longevity and performance of railway networks worldwide.

Funding

This work has been supported by funding from MxV Rail (Award number: 21-0825-007538) and Impact Area Accelerator Award
Grant from Georgia Southern University’s Office of Research.

Declaration of competing interest

The authors declare the following financial interests/personal relationships which may be considered as potential competing in­
terests: Hossein Taheri reports financial support was provided by MxV Rail.

Data availability

No data was used for the research described in the article.

Acknowledgment

The authors acknowledge the support Silixa Ltd and Georgia Southern University’s office of IT.

References

[1] Sz. A. Köllő, A. Faur, G. Köllő, A. Puskás, Environmental impacts of railway transportation systems, Earth Sci. Human Constr., 1, 1–5, 2021, doi: 10.37394/
232024.2021.1.1.
[2] J. E. Holler Branco, D. B. Bartholomeu, P. N. Alves Junior, J. V. Caixeta Filho, Evaluation of the economic and environmental impacts from the addition of new
railways to the brazilian’s transportation network: An application of a network equilibrium model, Transp. Policy (Oxf), vol. 124, 2022, doi: 10.1016/j.
tranpol.2020.03.011.
[3] M. Song, G. Zhang, W. Zeng, J. Liu, K. Fang, Railway transportation and environmental efficiency in China, Transp. Res. D Transp. Environ. 48 (2016), https://
doi.org/10.1016/j.trd.2015.07.003.
[4] A.W. Evans, The economics of railway safety, Res. Transp. Econ. 43 (1) (2013) 137–147, https://doi.org/10.1016/j.retrec.2012.12.003.
[5] S. Park, D.J. Inman, C.B. Yun, An outlier analysis of MFC-based impedance sensing data for wireless structural health monitoring of railroad tracks, Eng. Struct.
30 (10) (2008) 2792–2799, https://doi.org/10.1016/j.engstruct.2008.02.019.
[6] PROXMITY PROBE Curtis D. Kissinger, Goversville, N.Y., assignor to Me chanical Technology Encorporated, Latham, N.Y., a cor poration of New York Fied
Sept. 9, 1963, Ser. No. 307,676 5 Clains. (C. 88-14)”.
[7] H. Gabai, A. Eyal, On the sensitivity of distributed acoustic sensing, Opt Lett 41 (24) (2016) 5648, https://doi.org/10.1364/ol.41.005648.
[8] P. Paitz, et al., Empirical investigations of the instrument response for distributed acoustic sensing (Das) across 17 octaves, Bull. Seismol. Soc. Am. 111 (1)
(2021) 1–10, https://doi.org/10.1785/0120200185.
[9] D. Hill, Distributed acoustic sensing (DAS): Theory and applications, in Proceedings of Frontiers in Optics 2015, FIO 2015, OSA - The Optical Society, 2015. doi:
10.1364/fio.2015.fth4e.1.

22
M.A. Rahman et al. Mechanical Systems and Signal Processing 208 (2024) 110983

[10] V. Giurgiutiu, Structural health monitoring: with piezoelectric wafer active sensors. Elsevier, 2007. Accessed: Jun. 04, 2023. [Online]. Available: https://www.
google.com/books/edition/Structural_Health_Monitoring/AG5h8Hu-MdUC?hl=en&gbpv=1.
[11] N. M. M Maia, Shock and vibration digest A review on vibration-based structural health monitoring with special emphasis on composite materials a review on
vibration-based structural health monitoring with special emphasis on composite materials.
[12] D. Balageas, C.-P. Fritzen, A. Güemes, Structural Health Monitoring, vol. 90, John Wiley & Sons, 2010.
[13] F. Magalhães, A. Cunha, E. Caetano, Vibration based structural health monitoring of an arch bridge: from automated OMA to damage detection, Mech. Syst.
Signal Process 28 (2012) 212–228, https://doi.org/10.1016/j.ymssp.2011.06.011.
[14] D. A. Tibaduiza Burgos, R. C. Gomez Vargas, C. Pedraza, D. Agis, F. Pozo, Damage identification in structural health monitoring: A brief review from its
implementation to the use of data-driven applications, Sensors (Switzerland), vol. 20, no. 3. MDPI AG, Feb. 01, 2020. doi: 10.3390/s20030733.
[15] F. Pozo, D. A. Tibaduiza, Y. Vidal, Sensors for structural health monitoring and condition monitoring, Sensors, 21 (5) MDPI AG, pp. 1–6, Mar. 01, 2021. doi:
10.3390/s21051558.
[16] J.O. Willberry, M. Papaelias, Structural health monitoring using fibre optic acoustic emission sensors, Sensors (switzerland) 20 (21) (2020) 1–31, https://doi.
org/10.3390/s20216369.
[17] C. A. Jones, C. Jones, Application of Phased Array Eddy Current (PAEC) Nondestructive Technology for Stress Corrosion Cracking (SCC), 2022. [Online].
Available: https://digitalcommons.georgiasouthern.edu/research_symposiumhttps://digitalcommons.georgiasouthern.edu/research_symposium/2022/2022/
25.
[18] D.M. McCann, M.C. Forde, Review of NDT methods in the assessment of concrete and masonry structures, NDT and E Int. 34 (2) (2001) 71–84, https://doi.org/
10.1016/S0963-8695(00)00032-3.
[19] M. Willcox Jiang Li, M. Willcox, G. Downes, The Old Cider Mill Kings Thorn Herefordshire HR2 8AW Telephone Fax Email Web Site A Brief Description of NDT
Techniques A Paper By. [Online]. Available: www.InsightNDT.com.
[20] Z. He, Q. Liu, Optical fiber distributed acoustic sensors: a review, J. Lightwave Technol. 39 (12) (2021) 3671–3686, https://doi.org/10.1364/JLT.39.003671.
[21] Guidelines for performing Systematic Literature Reviews in Software Engineering, 2007.
[22] R.W. Ngigi, C. Pislaru, A. Ball, F. Gu, Modern techniques for condition monitoring of railway vehicle dynamics, J. Phys.: Conf. Series Institute Phys. Publish.
(2012), https://doi.org/10.1088/1742-6596/364/1/012016.
[23] J. Otegui, A. Bahillo, I. Lopetegi, L.E. Diez, A survey of train positioning solutions, IEEE Sens J 17 (20) (2017) 6788–6797, https://doi.org/10.1109/
JSEN.2017.2747137.
[24] R.I. Rajkumar, P.E. Sankaranarayanan, G. Sundari, GPS and ethernet based real time train tracking system, in: Proceedings of the 2013 International Conference
on Advanced Electronic Systems, ICAES, 2013, pp. 282–286, https://doi.org/10.1109/ICAES.2013.6659409.
[25] L. Wang, et al., Three-year review of the 2018–2020 SHL challenge on transportation and locomotion mode recognition from mobile sensors, Front Comput Sci
3 (2021), https://doi.org/10.3389/fcomp.2021.713719.
[26] Z. Sun, K. Tang, X. Wang, M. Wu, Y. Guo, High-speed train tunnel navigation method based on integrated MIMU/ODO/MC navigation, Appl. Sci. (Switzerland),
vol. 11, no. 8, 2021, doi: 10.3390/app11083680.
[27] D. Milne, A. Masoudi, E. Ferro, G. Watson, L. le Pen, An analysis of railway track behaviour based on distributed optical fibre acoustic sensing, Mech Syst
Signal Process 142 (2020), https://doi.org/10.1016/j.ymssp.2020.106769.
[28] Y. Shang et al., Research progress in distributed acoustic sensing techniques, Sensors, vol. 22, no. 16. MDPI, Aug. 01, 2022. doi: 10.3390/s22166060.
[29] L. Gao, C. Zhou, H. Xiao, Z. Chen, Continuous vertical Wheel-rail force Reconstruction method based on the distributed Acoustic sensing technology,
Measurement (lond) 197 (2022), https://doi.org/10.1016/j.measurement.2022.111297.
[30] K.I. Aoyama, K. Nakagawa, T. Itoh, Optical time domain reflectometry in a single-mode fiber, IEEE J Quantum Electron 17 (6) (1981) 862–868, https://doi.
org/10.1109/JQE.1981.1071237.
[31] M.K. Barnoski, M.D. Rourke, S.M. Jensen, R.T. Melville, Optical time domain reflectometer, Appl. Opt. 16 (9) (1977) 2375–2379, https://doi.org/10.1364/
AO.16.002375.
[32] C. Conway, Bandweaver, Beng, An introduction to fibre optic Intelligent Distributed Acoustic Sensing (iDAS) technology for power industry applications, 2015.
[33] A.V. Timofeev, Monitoring the railways by means of C-OTDR technology, World Acad. Sc., Eng. Technol. 9 (2015) 717–720.
[34] Z. Wang, B. Lu, Q. Ye, H. Cai, Recent progress in distributed fiber acoustic sensing with Φ-otdr, Sensors (Switzerland), vol. 20, no. 22. MDPI AG, pp. 1–26, Nov.
02, 2020. doi: 10.3390/s20226594.
[35] History and analysis of distributed acoustic sensing (das) for oilfield applications, 2013. Accessed: Jun. 04, 2023. [Online]. Available: https://oaktrust.library.
tamu.edu/handle/1969.1/150995.
[36] A.M. Rizzo, et al., Known and unknown event detection in OTDR traces by deep learning networks, Neural Comput. Appl. 34 (22) (2022) 19655–19673,
https://doi.org/10.1007/s00521-022-07634-2.
[37] Measurement method by otdr and terminal station apparatus: US7215415, May 8, 2007, Accessed: Jun. 04, 2023. [Online]. Available: https://patents.google.
com/patent/US7420666B2/en.
[38] Systems and methods for OTDR tracing and mapping, US20040015309A1, Accessed: Jun. 04, 2023. [Online]. Available: https://patents.google.com/patent/
US20040015309A1/en?oq=US20040015309A1.
[39] Introduction to Distributed Temperature Sensing ,DTS-Primer-2022, Accessed: Jun. 04, 2023. [Online]. Available: https://silixa.com/wp-content/uploads/
DTS-Primer-2022.pdf.
[40] J.P. King, D.F. Smith, K. Richards, P. Timson, R.E. Epworth, S. Wright, Development of a coherent OTDR instrument, J. Lightwave Technol. 5 (4) (1987)
616–624, https://doi.org/10.1109/JLT.1987.1075523.
[41] O. Gautheron, J. B. Leroy, P. Marmier, COTDR Performance Optimization for Amplified Transmission Systems, 1997.
[42] L. Chamoin, S. Farahbakhsh, M. Poncelet, L. Chamoin, An educational review on distributed optic fiber sensing based on Rayleigh backscattering for damage
tracking and structural health monitoring,” 2022. [Online]. Available: https://hal.archives-ouvertes.fr/hal-03701630v2.
[43] A. Masoudi, M. Belal, T. P. Newson, A distributed optical fibre dynamic strain sensor based on phase-OTDR, Meas Sci Technol, vol. 24, no. 8, 2013, doi:
10.1088/0957-0233/24/8/085204.
[44] M. R. Fernández-Ruiz, L. Costa, H. F. Martins, Distributed acoustic sensing using chirped-pulse phase-sensitive OTDR technology, Sensors (Switzerland), vol. 19,
no. 20. MDPI AG, Oct. 02, 2019. doi: 10.3390/s19204368.
[45] NTT Technical Review: Frequency-division-multiplexing Coherent OTDR for Realizing Effective Construction and Maintenance of Submarine Optical Cable
Systems. Accessed: Jun. 04, 2023. [Online]. Available: https://www.ntt-review.jp/archive/ntttechnical.php?contents=ntr201410ra1_s.html.
[46] H. H. Zhu, W. Liu, T. Wang, J. W. Su, B. Shi, Distributed acoustic sensing for monitoring linear infrastructures: current status and trends, Sensors, 22 (19).
MDPI, Oct. 01, 2022. doi: 10.3390/s22197550.
[47] Z. He, Q. Liu, Optical fiber distributed acoustic sensors: a review, Journal of Lightwave Technology, vol. 39, no. 12. Institute of Electrical and Electronics
Engineers Inc., pp. 3671–3686, Jun. 15, 2021. doi: 10.1109/JLT.2021.3059771.
[48] O.H. Waagaard, et al., Real-time low noise distributed acoustic sensing in 171 km low loss fiber, OSA Contin 4 (2) (2021) 688, https://doi.org/10.1364/
osac.408761.
[49] Y. Yan, F.N. Khan, B. Zhou, A.P.T. Lau, C. Lu, C. Guo, Forward transmission based ultra-long distributed vibration sensing with wide frequency response
[Online]. Available: J. Lightwave Technol. 39 (7) (2021) 2241–2249 https://opg.optica.org/jlt/abstract.cfm?URI=jlt-39-7-2241.
[50] A. Masoudi, M. Beresna, G. Brambilla, 152 km-range single-ended distributed acoustic sensor based on inline optical amplification and a micromachined
enhanced-backscattering fiber, Opt. Lett. 46 (3) (Feb. 2021) 552–555, https://doi.org/10.1364/OL.413206.
[51] E. Ip et al., DAS over 1,007-km Hybrid Link with 10-Tb/s DP-16QAM Co-propagation using Frequency-Diverse Chirped Pulses; DAS over 1,007-km Hybrid Link
with 10-Tb/s DP-16QAM Co-propagation using Frequency-Diverse Chirped Pulses, 2022.

23
M.A. Rahman et al. Mechanical Systems and Signal Processing 208 (2024) 110983

[52] H. Iida, Y. Koshikiya, F. Ito, K. Tanaka, High-sensitivity coherent optical time domain reflectometry employing frequency-division multiplexing, J. Lightwave
Technol. 30 (8) (2012) 1121–1126, https://doi.org/10.1109/JLT.2011.2170960.
[53] DISTRIBUTED ACOUSTIC SENSING. Accessed: Jun. 04, 2023. [Online]. Available: https://www.bandweaver.com/fiber_optic_sensing_technology/distributed-
acoustic-sensing/.
[54] M. C. Fenta, D. K. Potter, and J. Szanyi, Fibre optic methods of prospecting: a comprehensive and modern branch of geophysics, Surveys in Geophysics, vol. 42,
no. 3. Springer Science and Business Media B.V., pp. 551–584, May 01, 2021. doi: 10.1007/s10712-021-09634-8.
[55] X. Zhang, Y. Lu, F. Wang, H. Liang, Y. Zhang, Development of fully-distributed fiber sensors based on Brillouin scattering, Photonic Sensors 1 (2011) 54–61,
https://doi.org/10.1007/s13320-010-0019-7.
[56] S.I. Furukawa, K. Tanaka, Y. Koyamada, M. Sumida, Enhanced coherent OTDR for long span optical transmission lines containing optical fiber amplifiers, IEEE
Photon. Technol. Lett. 7 (5) (1995) 540–542, https://doi.org/10.1109/68.384537.
[57] D. Krohn, Fiber Optic Sensors: Fundamentals and Applications, 2015. Accessed: Jun. 04, 2023. [Online]. Available: https://www.photonics.com/images/web/
Webinars/80/Fiber_Optic_Sensors_Fundamentals_and_Applications.PDF.
[58] A. K. Arora, M. Rajalakshmi, T. R. Ravindran, V. Sivasubramanian, Raman spectroscopy of optical phonon confinement in nanostructured materials, J. Raman
Spectroscopy, 38 (6). John Wiley and Sons Ltd, pp. 604–617, 2007. doi: 10.1002/jrs.1684.
[59] L. Maksimov, A. Anan’ev, V. Bogdanov, T. Markova, V. Rusan, O. Yanush, Inhomogeneous structure of inorganic glasses studied by Rayleigh, Mandel’shtam-
Brillouin, Raman scattering spectroscopy, and acoustic methods, in: IOP Conference Series: Materials Science and Engineering, 2011. doi: 10.1088/1757-899X/
25/1/012010.
[60] B. Champagnon, C. Chemarin, E. Duval, R. Le Parc, Glass structure and light scattering, J. Non Cryst. Solids 274 (1) (2000) 81–86, https://doi.org/10.1016/
S0022-3093(00)00207-6.
[61] P.C. Wait, K. De Souza, T.P. Newson, A theoretical comparison of spontaneous Raman and Brillouin based fibre optic distributed temperature sensors, Opt.
Commun. 144 (1) (1997) 17–23, https://doi.org/10.1016/S0030-4018(97)00482-3.
[62] Y. Horiuchi, M. Suzuki, ‘Data transport system architecture for scientific submarine cable.’ The 3rd International Workshop on Sci-* cited by examiner.
[63] P. Ferdinand, The Evolution of Optical Fiber Sensors Technologies During the 35 Last Years and Their Applications in Structure Health Monitoring, Jul. 2014.
[64] M. Fenta, J. Szanyi, Fibre optic methods of prospecting: a comprehensive and modern branch of geophysics, Surv Geophys (2021), https://doi.org/10.1007/
s10712-021-09634-8.
[65] Y. Muanenda, C. J. Oton, F. Di Pasquale, Application of Raman and Brillouin Scattering Phenomena in Distributed Optical Fiber Sensing, Front. Phys., 7.
Frontiers Media S.A., Oct. 29, 2019. doi: 10.3389/fphy.2019.00155.
[66] M. Tateda, T. Horiguchi, Advances in optical time-domain reflectometry, J. Lightwave Technol. 7 (8) (1989) 1217–1224, https://doi.org/10.1109/50.32386.
[67] Z. Ding et al., Distributed optical fiber sensors based on optical frequency domain reflectometry: A review, Sensors (Switzerland), vol. 18, no. 4. MDPI AG, Apr.
03, 2018. doi: 10.3390/s18041072.
[68] Z. Ding et al., Compensation of laser frequency tuning nonlinearity of a long range OFDR using deskew filter, 2013. [Online]. Available: http://lunainc.com/
wp-.
[69] Institute of Electrical and Electronics Engineers, ICTON 2009 : 11th International Conference on Transparent Optical Networks : Ponta Delgada University, S. Miguel,
Azores, Portugal, 28th June-2nd July 2009.
[70] X. Bao, Y. Wang, Recent advancements in rayleigh scattering-based distributed fiber sensors, Adv. Dev. Instrum. 2021 (2021) 1–17, https://doi.org/10.34133/
2021/8696571.
[71] C. Baker, Y. Lu, J. Song, X. Bao, Incoherent optical frequency domain reflectometry based on a Kerr phase-interrogator, Opt. Express 22 (13) (2014)
15370–15375, https://doi.org/10.1364/OE.22.015370.
[72] M. Inoue, F. Ito, X. Fan, Y. Koshikiya, Very long range quasi-Fourier spectroscopy for narrowband lasers, Opt. Express 20 (26) (2012) B566, https://doi.org/
10.1364/oe.20.00b566.
[73] J.R. Jiang, J. Bin Kao, Y.L. Li, Semi-supervised time series anomaly detection based on statistics and deep learning, Appl. Sci. (Switzerland) 11 (15) (2021),
https://doi.org/10.3390/app11156698.
[74] S. Liehr, N. Nöther, K. Krebber, Incoherent optical frequency domain reflectometry and distributed strain detection in polymer optical fibers, Meas. Sci.
Technol. 21 (1) (2010), https://doi.org/10.1088/0957-0233/21/1/017001.
[75] C. Liang, Q. Bai, M. Yan, Y. Wang, H. Zhang, B. Jin, A comprehensive study of optical frequency domain reflectometry, IEEE Access, vol. 9. Institute of Electrical
and Electronics Engineers Inc., pp. 41647–41668, 2021. doi: 10.1109/ACCESS.2021.3061250.
[76] X. Gui, et al., Distributed optical fiber sensing and applications based on large-scale fiber bragg grating array: review, J. Lightwave Technol. (2023) 1–14,
https://doi.org/10.1109/jlt.2022.3233707.
[77] S. Jamal, H. Wimmer, Performance analysis of machine learning algorithm on cloud platforms: AWS vs Azure vs GCP, in: A. Gibadullin (Ed.), Information
Technologies and Intelligent Decision Making Systems, Springer Nature Switzerland, Cham, 2023, pp. 43–60.
[78] P. Adhikari, H. G. Rao, Dipl.-I. M. Buderath, Machine Learning based Data Driven Diagnostics & Prognostics Framework for Aircraft Predictive Maintenance.
[79] D.Y. Kononenko, V. Nikonova, M. Seleznev, J. van den Brink, D. Chernyavsky, An in situ crack detection approach in additive manufacturing based on acoustic
emission and machine learning, Additive Manufact. Lett. 5 (2023), 100130, https://doi.org/10.1016/j.addlet.2023.100130.
[80] D. Bala, M.A. Islam, M.I. Hossain, M. Mynuddin, M.A. Hossain, M.S. Hossain, Automated brain tumor classification system using convolutional neural networks
from MRI images, in: 8th International Conference on Engineering and Emerging Technologies, ICEET 2022, Institute of Electrical and Electronics Engineers Inc.,
2022, https://doi.org/10.1109/ICEET56468.2022.10007116.
[81] M. Sands, J. Kim, J. Kim, Output-only structural damage detection via enhanced random vibration analysis using LSTM/GRU model, in: 2022 IEEE/ACIS 20th
International Conference on Software Engineering Research, Management and Applications, SERA 2022, Institute of Electrical and Electronics Engineers Inc., 2022,
pp. 3–9, https://doi.org/10.1109/SERA54885.2022.9806775.
[82] J. Hu, E. Zhu, S. Wang, X. Liu, X. Guo, J. Yin, An efficient and robust unsupervised anomaly detection method using ensemble random projection in
surveillance videos, Sensors (switzerland) 19 (19) (2019), https://doi.org/10.3390/s19194145.
[83] I.H. Sarker, Machine learning: algorithms, real-world applications and research directions, SN Computer Science, vol. 2, no. 3. Springer, May 01, 2021. doi:
10.1007/s42979-021-00592-x.
[84] S. Jamal, M.V. Cruz, S. Chakravarthy, C. Wahl, H. Wimmer, Integration of EEG and eye tracking technology: a systematic review, in: Conference Proceedings -
IEEE SOUTHEASTCON, Institute of Electrical and Electronics Engineers Inc., 2023, pp. 209–216, https://doi.org/10.1109/
SoutheastCon51012.2023.10115167.
[85] J. Lee, F. Wu, W. Zhao, M. Ghaffari, L. Liao, D. Siegel, Prognostics and health management design for rotary machinery systems – reviews, methodology and
applications, Mech. Syst. Signal Process 42 (1–2) (2014) 314–334, https://doi.org/10.1016/j.ymssp.2013.06.004.
[86] Y. Liu, et al., Deep anomaly detection for time-series data in industrial IoT: a communication-efficient on-device federated learning approach, IEEE Internet
Things J 8 (8) (2021) 6348–6358, https://doi.org/10.1109/JIOT.2020.3011726.
[87] U. Islam, et al., A novel anomaly detection system on the internet of railways using extended neural networks, Electronics (switzerland) 11 (18) (2022),
https://doi.org/10.3390/electronics11182813.
[88] S. Omar, A. Ngadi, H. H. Jebur, Machine Learning Techniques for Anomaly Detection: An Overview, 2013.
[89] M. Landauer, S. Onder, F. Skopik, M. Wurzenberger, Deep learning for anomaly detection in log data: A survey, Machine Learn. Appl. 12 (2023), 100470,
https://doi.org/10.1016/j.mlwa.2023.100470.
[90] K. Choi, J. Yi, C. Park, S. Yoon, Deep Learning for Anomaly Detection in Time-Series Data: Review, Analysis, and Guidelines, IEEE Access, vol. 9. Institute of
Electrical and Electronics Engineers Inc., pp. 120043–120065, 2021. doi: 10.1109/ACCESS.2021.3107975.

24
M.A. Rahman et al. Mechanical Systems and Signal Processing 208 (2024) 110983

[91] M.A. Rahman, H. Taheri, J. Kim, Deep learning model for railroad structural health monitoring via distributed acoustic sensing, in: 2023 26th ACIS
International Winter Conference on Software Engineering, Artificial Intelligence, Networking and Parallel/distributed Computing (SNPD-Winter), IEEE, 2023,
pp. 274–281, https://doi.org/10.1109/SNPD-Winter57765.2023.10223971.
[92] K. Zope, K. Singh, H. Nistala, A. Basak, P. Rathore, V. Runkana, Anomaly Detection and Diagnosis in Manufacturing Systems: A Comparative Study of
Statistical, Machine Learning and Deep Learning Techniques.
[93] A.M.S. Omar, M.K. Osman, M.N. Ibrahim, Z. Hussain, A.F. Abidin, Fault classification on transmission line using LSTM network, Indonesian J. Electr. Eng.
Comp. Sci. 20 (1) (2020) 231–238, https://doi.org/10.11591/ijeecs.v20.i1.pp231-238.
[94] T. Yuan, R. Sheng, Understanding LSTM Networks The strengths of Recurrent Neural Networks(RNN).
[95] J. Kim, S. Kim, H. Wimmer, H. Liu, A Cryptocurrency Prediction Model Using LSTM and GRU Algorithms, in Proceedings - 2021 IEEE/ACIS 6th International
Conference on Big Data, Cloud Computing, and Data Science, BCD 2021, Institute of Electrical and Electronics Engineers Inc., Sep. 2021, pp. 37–44. doi: 10.1109/
BCD51206.2021.9581397.
[96] S. Kim, L. Chen, J. Kim, Intrusion Prediction using LSTM and GRU with UNSW-NB15, in 2021 Computing, Communications and IoT Applications, ComComAp
2021, Institute of Electrical and Electronics Engineers Inc., 2021, pp. 101–106. doi: 10.1109/ComComAp53641.2021.9652926.
[97] N. Michielli, U.R. Acharya, F. Molinari, Cascaded LSTM recurrent neural network for automated sleep stage classification using single-channel EEG signals,
Comput. Biol. Med. 106 (2019) 71–81, https://doi.org/10.1016/j.compbiomed.2019.01.013.
[98] S. Belagoune, N. Bali, A. Bakdi, B. Baadji, K. Atif, Deep learning through LSTM classification and regression for transmission line fault detection, diagnosis and
location in large-scale multi-machine power systems, Measurement (lond) 177 (2021), https://doi.org/10.1016/j.measurement.2021.109330.
[99] L. Alzubaidi, et al., Review of deep learning: concepts, CNN architectures, challenges, applications, future directions, J. Big Data 8 (1) (2021) Dec, https://doi.
org/10.1186/s40537-021-00444-8.
[100] Z. Guo, Y. Hao, H. Shi, Z. Wu, Y. Wu, X. Sun, A fault diagnosis algorithm for the dedicated equipment based on the CNN-LSTM mechanism, Energies (Basel), vol.
16, no. 13, Jul. 2023, doi: 10.3390/en16135230.
[101] G. Petmezas et al., Automated lung sound classification using a hybrid CNN-LSTM network and focal loss function, Sensors, vol. 22, no. 3, Feb. 2022, doi:
10.3390/s22031232.
[102] Y. He, J. Lv, H. Liu, T. Tang, Toward the trajectory predictor for automatic train operation system using CNN–LSTM network, Actuators 11 (9) (2022) Sep,
https://doi.org/10.3390/act11090247.
[103] C. S. Han and K. M. Lee, Hybrid Deep Learning Model for Time Series Anomaly Detection, Association for Computing Machinery (ACM), Aug. 2023, pp. 1–5.
doi: 10.1145/3599957.3606232.
[104] S. Garg, K. Kaur, N. Kumar, G. Kaddoum, A. Y. Zomaya, R. Ranjan, A Hybrid Deep Learning based Model for Anomaly Detection in Cloud Datacentre Networks.
[105] W. Gong et al., A novel deep learning method for intelligent fault diagnosis of rotating machinery based on improved CNN-SVM and multichannel data fusion,
Sensors (Switzerland), vol. 19, no. 7, Apr. 2019, doi: 10.3390/s19071693.
[106] D. Keerthana, V. Venugopal, M.K. Nath, M. Mishra, Hybrid convolutional neural networks with SVM classifier for classification of skin cancer, Biomed. Eng.
Adv. 5 (2023), 100069, https://doi.org/10.1016/j.bea.2022.100069.
[107] M.N. Ab Wahab, A. Nazir, A.T.Z. Ren, M.H.M. Noor, M.F. Akbar, A.S.A. Mohamed, Efficientnet-Lite and Hybrid CNN-KNN implementation for facial expression
recognition on raspberry Pi, IEEE Access 9 (2021) 134065–134080, https://doi.org/10.1109/ACCESS.2021.3113337.
[108] S. Qamar, R. R. Rasmus¨oberg, D. Malyshev, M. Andersson, A hybrid CNN-Random Forest algorithm for bacterial spore segmentation and classification in TEM
images, doi: 10.1101/2023.04.03.535316.
[109] H. Taheri, M. Jones, S. Bueso Quan, M. Gonzalez Bocanegra, M. Taheri, Distributed acoustic sensing (DAS) for intelligent in-motion transportation condition
monitoring, in ASME International Mechanical Engineering Congress and Exposition, American Society of Mechanical Engineers, 2022, p. V02BT02A050.
[110] Z. Wang et al., Novel railway-subgrade vibration monitoring technology using phase-sensitive OTDR, in 25th International Conference on Optical Fiber Sensors,
SPIE, Apr. 2017, p. 103237G. doi: 10.1117/12.2265169.
[111] Y. Shang, C. Wang, Y. Zhang, W. Zhao, J. Ni, G. Peng, Non-intrusive pipeline flow detection based on distributed fiber turbulent vibration sensing, Sensors, 22,
11, 2022, doi: 10.3390/s22114044.
[112] A.M. Biondi, et al., Pipeline structural health monitoring using distributed fiber optic sensing textile, Opt. Fiber Technol. 70 (2022), https://doi.org/10.1016/j.
yofte.2022.102876.
[113] F. Baensch et al., Feasibility study: continuous monitoring of pipes using distributed acoustic and fibre optic sensors. [Online]. Available: http://www.ndt.net/
?id=23521.
[114] K. Singh, A. Godfrey, OptaSense ® Distributed Acoustic Sensing (DAS) System for the Power Network-Integrated SMART-Sensing REAL TIME MONITORING.
[115] X. Li, Y. Zeng, D. Muchiri, W. Yan, Y. Feng, The use of distributed acoustic sensing (DAS) in monitoring the integrity of cement-casing system, J. Pet Sci. Eng.
208 (2021), 109690, https://doi.org/10.1016/j.petrol.2021.109690.
[116] G. Soldi et al., Monitoring of Underwater Critical Infrastructures: the Nord Stream and Other Recent Case Studies, Feb. 2023, [Online]. Available: http://arxiv.
org/abs/2302.01817.
[117] Y. Li, M. Karrenbach, J. Ajo-Franklin, Distributed Acoustic Sensing in Geophysics, Dec. 2021, doi: 10.1002/9781119521808.
[118] X. Liu, M.R. Saat, C.P.L. Barkan, Analysis of causes of major train derailment and their effect on accident rates, Transp Res Rec 2289 (2012) 154–163, https://
doi.org/10.3141/2289-20.
[119] Fiber Optic Availability and Opportunity Analysis for North American Railroads, 2016. [Online]. Available: http://www.fra.dot.gov.
[120] G. Cedilnik, R. Hunt, G. Lees, Advances in Train and Rail Monitoring with DAS, 2018.
[121] S. Gao, H. Dong, B. Ning, Q. Zhang, Cooperative prescribed performance tracking control for multiple high-speed trains in moving block signaling system, IEEE
Trans. Intell. Transp. Syst. 20 (7) (2019) 2740–2749, https://doi.org/10.1109/TITS.2018.2877171.
[122] C. Wiesmeyr et al., Real-time train tracking from distributed acoustic sensing data, Applied Sciences (Switzerland), vol. 10, no. 2, 2020, doi: 10.3390/
app10020448.
[123] S. Kowarik et al., Fiber optic train monitoring with distributed acoustic sensing: Conventional and neural network data analysis, Sensors (Switzerland), vol. 20,
no. 2, Jan. 2020, doi: 10.3390/s20020450.
[124] A. Timofeev, The Rail Traffic Management with Usage of C-OTDR Monitoring Systems. 2015. Accessed: Jun. 04, 2023. [Online]. Available: https://www.
researchgate.net/publication/280566687_The_Rail_Traffic_Management_with_Usage_of_C-OTDR_Monitoring_Systems.
[125] F. Peng, N. Duan, Y.J. Rao, J. Li, Real-time position and speed monitoring of trains using phase-sensitive OTDR, IEEE Photon. Technol. Lett. 26 (20) (2014)
2055–2057, https://doi.org/10.1109/LPT.2014.2346760.
[126] G. Zhang, Z. Song, A.G. Osotuyi, R. Lin, B. Chi, Railway traffic monitoring with trackside fiber-optic cable by distributed acoustic sensing Technology, Front.
Earth Sci. (lausanne) 10 (2022), https://doi.org/10.3389/feart.2022.990837.
[127] Track Inspector Rail Defect Reference Manual, 2015. Accessed: Jun. 04, 2023. [Online]. Available: https://railroads.dot.gov/elibrary/track-inspector-rail-
defect-reference-manual.
[128] D. Bombarda, G.M. Vitetta, G. Ferrante, Rail diagnostics based on ultrasonic guided waves: an overview, Appl. Sci. (switzerland) 11 (3) (2021) 1–42, https://
doi.org/10.3390/app11031071.
[129] L.N. Wheeler, E. Pannese, N.A. Hoult, W.A. Take, H. Le, Measurement of distributed dynamic rail strains using a Rayleigh backscatter based fiber optic sensor:
Lab and field evaluation, Transp. Geotech. 14 (2018) 70–80, https://doi.org/10.1016/j.trgeo.2017.10.002.
[130] A. Wagner, A. Nash, F. Michelberger, H. Grossberger, G. Lancaster, The effectiveness of distributed acoustic sensing (DAS) for broken rail detection, Energies
(basel) 16 (1) (Jan. 2023) 522, https://doi.org/10.3390/en16010522.
[131] C. Wiesmeyr, A. Papp, H. Garn, G. Neunteufel, W. Zottl, Train tracking and train condition monitoring by Distributed Acoustic Sensing Nomenclature DAS
Distributed Acoustic Sensing OTDR Optical Time Domain Reflectometry SVM Support Vector Machine.

25
M.A. Rahman et al. Mechanical Systems and Signal Processing 208 (2024) 110983

[132] L. Xie, Z. Li, Y. Zhou, W. Xiang, Y. Wu, Y. Rao, Railway track online detection based on optical fiber distributed large-range acoustic sensing, IEEE Internet
Things J. (2023), https://doi.org/10.1109/JIOT.2023.3311173.
[133] Y. Hao, H. Qi, S. Liu, V. Nian, Z. Zhang, Study of noise and vibration impacts to buildings due to urban rail transit and mitigation measures, Sustainability
(Switzerland), vol. 14, no. 5, Mar. 2022, doi: 10.3390/su14053119.
[134] Y. Cao, Q. Xiang, B. Li, Z. Li, Field measurement and energy analysis of ground-borne vibration around high-speed railway viaduct, Shock Vib. 2023 (2023)
1–20, https://doi.org/10.1155/2023/4640726.
[135] Q. Liu, L. Xu, Q. Feng, Intelligent monitoring of vibration and structural-borne noise induced by rail transit, Intelligent Transport. Infrastruct. 2 (2023), https://
doi.org/10.1093/iti/liad013.
[136] J. Morais, P. Morais, C. Santos, A. Paixão, E. Fortunato, Railway track support condition assessment—Initial developments on a vehicle-based monitoring
solution through modal analysis, Proc. Inst. Mech. Eng. F J Rail Rapid Transit. (2022), https://doi.org/10.1177/09544097211064101.
[137] M. Belding, A. Enshaeian, P. Rizzo, Vibration-based approach to measure rail stress: Modeling and first field test, Sensors 22 (19) (2022) Oct, https://doi.org/
10.3390/s22197447.
[138] A.A. Dibazar, A. Yousefi, H.O. Park, B. Lu, S. George, T.W. Berger, Intelligent acoustic and vibration recognition/alert systems for security breaching detection,
close proximity danger identification, and perimeter protection, in: 2010 IEEE International Conference on Technologies for Homeland Security, 2010,
pp. 351–356, https://doi.org/10.1109/THS.2010.5654931.

26