Key technologies of China high-speed comprehensive inspection train: CIT450

Abstract

The China comprehensive inspection train (CIT) is designed for evaluating railway infrastructure to ensure safe railway operations. The CIT integrates an array of inspection devices, capable of simultaneously assessing railway health condition parameters. The CIT450, representing the second generation, can reach a top speed of 450 km/h with inspection on the infrastructure. This paper begins by outlining the global evolution of inspection trains. It then focuses on the critical technologies underlying the CIT450, which include: (1) real-time inspection data acquisition with spatial and temporal synchronization; (2) intelligent fusion and centralized management of multi-source inspection data, enabling remote supervision of the inspection process; (3) technologies in inspecting track, train–track interaction, catenary, signalling systems, and train operating environment; and (4) AI-driven analysis and correlation of inspection data. The future developmental directions for comprehensive inspection trains are discussed finally. The CIT450’s approach to real-time railway health monitoring can enrich traditional inspection means, operational, and maintenance methods by enhancing inspection efficiency and automating railway maintenance.

Introduction

Rail transport, known for its large carrying capacity, economic efficiency, and high safety, is an indispensable element of the transportation systems worldwide. As of the end of 2021, China’s railway operating mileage totalled 150,000 kms. The China State Railway Group indicated that by 2035, the national railway network is expected to reach approximately 200,000 kms, with about 70,000 kms of high-speed railways.

With its extensive and increasingly sophisticated high-speed railway network, China faces the significant challenge of maintaining the safety and comfort of train operations. To address this, the development and implementation of advanced track inspection technologies are crucial. To meet these needs, an array of technologies, including inspection robots, trains, and vehicles/trolleys, have been developed [1, 2, 3, 4, 5–6]. Among these, the deployment of inspection trains for on-the-go infrastructure assessment can be a helpful strategy, particularly vital in China due to its vast railway network [7, 8].

The high-speed comprehensive inspection train (CIT) is an effective and efficient technology in railway infrastructure inspection [9]. Its primary functions include track component inspection [10], track geometry assessment [11], catenary geometry analysis [12], pantograph–catenary dynamic interaction examination [13], vehicle dynamic response testing [14], wheel–rail interaction force measurement [15], and communication and signalling quality evaluation. This ensures that each component meets strict safety and performance standards. Globally, there is a diverse range of inspection trains, each designed to inspect one or more specific components within the railway system, as shown in Table 1.

Table 1. Inspection trains designed to inspect one or more specific components within the railway system

Inspection items of diverse trains	Details of the inspected railway infrastructure components
Track component inspection	Granular media issues (ballast, subgrade, and embankment) [16, 17, 18, 19, 20, 21, 22–23] Sleeper–ballast interaction [24, 25] Fastening system flaws [26, 27–28] Rail flaws [6, 29, 30, 31, 32–33]
Track geometry inspection	Track geometry indicators: track gauge, gauge change rate, alignment, longitudinal level, superelevation, cross level, twist, curve radius, and car body lateral and vertical accelerations, along with the corresponding mileage for each indicator [34, 35, 36, 37, 38, 39–40]. The indicator data are used to calculate the track quality index (TQI) [41, 42]
Train–track interaction evaluation system	This system detects the vertical and lateral forces between the wheel and the rail and measures the vibration acceleration of the car body, frame, and axle box [43, 44, 45, 46–47]. It calculates parameters such as derailment coefficient, lateral wheel axle force, and rail corrugation [48]. The system also analyses both local and continuous short-wave irregularities on the tracks [36, 49, 50, 51–52]
Pantograph–catenary inspection system	This system measures the geometric parameters of the overhead contact system [53, 54]. Key inspection items include catenary height/overhead contact line height, catenary stagger, rigid point, pantograph–catenary contact force, electrical arcing, catenary wire spacing, vertical spacing between contact wires, catenary voltage, and catenary mast location [55, 56]
Signal inspection system	This system primarily focuses on detecting electrical parameters of trackside signal equipment, including track circuits, traction return currents, compensation capacitors, and real-time operational data monitoring of train control on-board equipment [57, 58]
Other relevant components and conditions	This category includes geographical and environmental factors, panoramic views, abnormal object detection, and track dynamic inspection using stand-alone accelerometers on the track with vibration energy harvesting [59, 60–61]

Currently, the majority of the inspection trains operate at speeds of 100 km/h or lower while inspecting railway infrastructure, with only a limited number capable of exceeding 200 km/h. To date, there are no inspection trains capable of operating at speeds of 350 km/h or higher. The development of high-speed CIT is therefore of significant importance and demand, particularly for the following reasons:

1. In China, the total mileage of high-speed railways is continuously expanding. Researching and developing higher-speed inspection technology is essential for improving the overall efficiency of railway operation and maintenance.

2. With China’s 400 km/h high-speed railway networks under planning and construction, there is a need for inspection trains that can adapt to these higher-speed operations. Such inspection trains should ideally operate at speeds equivalent to the high-speed trains themselves to avoid disrupting normal operations.

3. The CIT, capable of inspecting multiple infrastructure functions simultaneously, offers high efficiency in terms of data processing and time utilization. They align with the development requirements for low carbon footprint and energy savings [62].

To achieve the aim of inspecting at high speed, CIT has several technology challenges. These include the integrated analysis and application of inspection data, and the modularization and integration of sub-inspection systems with varying functions (such as track geometry and pantograph–catenary inspection systems) within a single vehicle [63, 64–65]. This is especially challenging when operating at speeds exceeding 350 km/h. Therefore, there is an urgent need to upgrade the high-speed CIT. This involves the adoption of cutting-edge measurement equipment and new data processing methods.

Overview of railway inspection trains

Railway infrastructure inspection trains, vehicles, or cars typically operate at varying speeds, with high-speed inspection trains (> 100 km/h). This inspection means is primarily found in Europe, Japan, South Korea, the UK, the USA, and China [66, 67, 68, 69–70].

A prime example is the Diamante 2.0 from the Italian Rete Ferroviaria Italiana (RFI) company. The train has a maximum operational inspection speed of up to 300 km/h and is equipped with over 200 sensors and 98 cameras. It can perform dynamical measurement of more than 500 parameters in the whole railway system, including track geometry, rail wear, wheel–rail interaction, catenary geometry, dynamic catenary–pantograph interaction parameters, and communication signals [71].

Germany’s Deutsche Bahn has developed a specialized inspection train tailored for the German railway system. It utilizes operational passenger cars as platforms and integrates a suite of inspection systems. These systems include a track inspection system, a pantograph–catenary contact force inspection system, and a train operation environment safety monitoring system. This comprehensive integration enables a thorough assessment of the railway infrastructure’s health condition [72].

Japan’s high-speed railway inspection train is the ‘Doctor Yellow’ train. This high-speed CIT is equipped with an array of integrated systems designed for multiple inspection tasks. These systems include track geometry inspection, wheel–rail interaction analysis, catenary status evaluation, and communication signal detection devices. Remarkably, the Doctor Yellow is capable of conducting its inspection operations at speeds of up to 210 km/h.

The China’s current high-speed CIT predominantly utilize ‘Harmony Train’ and ‘Fuxing Train’ high-speed train sets as their platforms. These trains are equipped with integrated multiple inspection systems. Particularly, there is one specialized inspection train designed for the extreme cold regions of Northeast and Northwest China.

While the earlier China’s high-speed CITs can fulfil the tasks of inspecting infrastructure health condition, there are areas that need further development. The operation of existing inspection sub-systems and the analysis of collected data still rely heavily on manual labour. Moreover, the various inspection sub-systems operate in isolation, which hinders the ability to quickly and efficiently correlate data across the different sub-systems. Consequently, the efficiency in analysing inspection data is low, with only a primary focus on predicting specific types of infrastructure defects.

This earlier CIT lacks a systematic methodology for evaluating the overall health condition of the infrastructure. Particularly, there are gaps in the inspection capabilities of on-board measurement equipment, such as the inability to fully inspect ballastless track slabs. Therefore, there is a significant need for further enhancements and expansions of inspection functionalities to meet the evolving demands of high-speed railway infrastructure.

Overall architecture of the China comprehensive inspection train

The second generation of China’s high-speed railway comprehensive inspection train, CIT450, is capable of conducting the following inspections, including track geometry, short-wave track irregularities, pantograph–catenary interaction, catenary geometry, wheel–rail interaction, communication service quality, signal system health status, and safety in train operating environment.

The CIT450 uses a high-precision time and location two-dimensional information synchronization integration architecture. This architecture, coupled with a fully automated inspection operation workflow and data collection management mode, enables the CIT450 to operate with higher efficiency, automation, and intelligence than the first-generation CIT. The train is composed of 8 carriages, the overall appearance is shown in Fig. 1 and the interior layout is shown in Fig. 2.

[See PDF for image]

Fig. 1

The overall appearance of CIT450

[See PDF for image]

Fig. 2

The interior layout of CIT450

Inspection functions

Table 2 shows the overview of the inspection functions, corresponding inspection items/indicators/parameters, key highlights, and the associated sensors/methods employed by the CIT450. This table details the items/indicators/parameters that have been developed for CIT450.

Table 2. Overview of key technologies in China high-speed CIT450

Inspection function	Inspection items/indicators/parameters	Highlights	Sensors/methods and reference
Track	Gauge, longitudinal level, alignment, cross level, twist, superelevation, curvature, track stiffness change, TQI	Track stiffness change*	Laser vision; inertial reference method [41]
Train and track dynamics	Car body acceleration (lateral and vertical), axle box acceleration (lateral and vertical), frame acceleration (horizontal and vertical), track impact index (TII), rail corrugation index (RCI)	TII; RCI	Accelerometer [73]
Wheel–rail interaction	Wheel–rail vertical force, wheel–rail lateral force, derailment coefficient, wheel load reduction rate, wheel axle lateral force, frame lateral stability performance, Sperling’s Ride Index (vertical and lateral)	Continuous measurement method	Instrumented wheelset [14]
Catenary	Catenary height/overhead contact line height, catenary stagger, rigid point, pantograph–catenary contact force, electrical arcing, catenary wire spacing, vertical spacing between contact wires, contact catenary voltage, catenary mast location, catenary quality index (CQI), catenary dynamic index (CDI),	CQI, CDI	Visual; Strain sensor; Accelerometer [74]
Communication system	GSM-R network service quality, 5G-R network service quality, public network coverage quality	5G-R network service Quality	Electromagnetic induction [75]
Signalling system	Status of track circuit, compensation capacitor, traction return current, transponder, on-board equipment of CTCS-3		Electromagnetic induction [76]
Train operating environment	Catenary pillar number plate, bird nests on the catenary pole, invasive foreign object, damaged barrier	Intelligent identification	Visual/Radar [77]
Train location	Tag based, GNSS based, visual positioning, electronic map display	Visual positioning	Visual/GPS/RFID [78]

^* The equipment and sensors are installed and integrated into CIT450, and the track stiffness data have been collected. Data analysis and field verification are still under performing

Inspection sub-system integration architecture

The CIT450 adopts a master–slave integrated architecture, centralizing control and coordination, as shown in Fig. 3. At its core, a central control centre functions as the master node, with each inspection sub-system operating as a slave node. These sub-systems are interconnected through a designed network comprising time synchronization, location synchronization, and data transmission channels. Furthermore, a local cloud server is established on the train, tasked with collecting and retrieving data in real time. This enables real-time display of all data sourced from the inspection sub-systems.

[See PDF for image]

Fig. 3

Architecture for integration of inspection sub-systems

To facilitate high-speed train operations exceeding 350 km/h, ultimately reaching up to 450 km/h (= 125 m/s), the inspection sub-systems are designed with separate modules for data collection and data processing. This design meets the strict real-time requirements for data collection and supports the development of multi-procedural data processing technologies for low real-time data. Such an arrangement ensures that the inspection train can deliver reliable inspection results even at the peak speed of 450 km/h.

The architecture’s key improvements are outlined as follows:

• All on-board sensor data are accurately managed, achieving a time synchronization precision of 1 ms for continuous data collection across various sensors.

• The inspection data from all sub-systems are harmonized with the master server’s unified location information, ensuring precise alignment of inspection data from different systems at each specific location (mileage).

• The inspection data are managed via a central cloud service, facilitating rapid access, association, and display on any terminal within the train. This system also incorporates 5G technology, enabling wireless transmission of detected defect data from the inspection train to the ground centre for further processing and review.

Intelligent operation and data management

Conducting track inspections with CITs, such as the first generation of China CIT, typically involve prolonged working hours and repetitive manual tasks. To address this, the CIT450 incorporates an automated inspection workflow, replacing traditional manual procedures and significantly enhancing inspection operation efficiency. The automated inspection operation and the associated data management process are detailed in Fig. 4.

[See PDF for image]

Fig. 4

Flow of intelligent operation and data management

The inspection workflow is comprised of two primary components: the control and operation of inspection equipment and the processing of inspection data. It initiates with the activation of inspection equipment, encompassing the control of equipment, monitoring of equipment state, and diagnosis of equipment faults. The CIT450 features the system control module, enabling the one-button activation of all on-board inspection equipment.

The centralized control system facilitates various functions, such as setting inspection tasks, switching inspection functions and items, and intelligently positioning the train location (mileage). It also monitors the working frequencies of the detection sensors and equipment, the output waveform data from all inspection sub-systems, as well as the status of train operating environment. In case of system faults, the system promptly notifies about the fault, allowing inspection engineers to quickly locate and address the issue in real time.

The heart of the CIT450 operation lies in its inspection data. Given the vast volume of data generated by inspection equipment and sensors spread across different carriages, a specialized system ensures efficient data access for both on-board and remote staffs. The CIT450 is equipped with an on-board transmission network for comprehensive data collection, leading to a unified data storage terminal. Additionally, a cloud service platform on the train enables shared access to inspection data from any terminal. Each sub-system’s data processing terminals convert sensor data into inspection indicators/parameters, aggregating them onto the cloud service platform. Inspection staff utilize data processing software to access and verify infrastructure defects detected by the sub-systems. This includes validating effective data, discarding invalid data, and conducting specialized assessments of structural components like bridges, switches, and transition zones.

Key inspection technology

Track inspection

The track inspection sub-system is a fundamental component of the CIT450, specifically designed for assessing track geometry [79]. This sub-system is capable of measuring a wide range of indicators/parameters, including gauge, gauge change rate, alignment of left and right rails, longitudinal level of left and right rails, superelevation, cross level, twist, curve radius, lateral and vertical accelerations of the car body, train speed, and location determined by mileage.

When comparing the standards of the inspection sub-system indicators to the European standard EN 13848, the inspection sub-system presents several differences:

• Inspection speed range: The sub-system can inspect at speeds ranging from 0 to 450 km/h.

• Detection wavelength range: For both alignment and longitudinal level, the detection wavelength range spans from 1.5 to 200 m. In contrast, EN 13848–2 specifies a wavelength range for level from 3 to 150 m and for alignment from 3 to 200 m.

Table 3 illustrates the inspection indicators and the key parameters associated with these indicators. The table delineates the measurement range of the inspection equipment (range), the baseline length for measuring the indicators (wavelength), and the train operating speed during track inspection (train speed).

Table 3. Inspection indicators and key parameters of these indicators

Inspection indicator	Range (mm)	Maximum permissible error (mm)	Repeatability (mm)	Reproducibility (mm)	Comparability (mm)	Wavelength (m)	Minimum train speed (km/h)	Resolution (mm)
Gauge	-15/ + 50	± 0.8	0.3	0.4	0.6	NA	0	0.1
Alignment	± 50	± 1.0	0.4	0.5	0.7	1.5–25	25	0.1
		± 1.0	0.4	0.5	0.7	1.5–42	25
		± 2.0	0.7	1.0	1.4	1.5–70	50
		± 3.0	1.0	1.5	2.0	1.5–120	120
		± 4.0	1.7	2.5	3.0	1.5–200	150
Longitudinal level	± 50	± 1.0	0.4	0.5	0.7	1.5–25	25	0.1
		± 1.0	0.4	0.5	0.7	1.5–42	25
		± 2.0	0.7	1.0	1.4	1.5–70	50
		± 3.0	1.0	1.5	2.0	1.5–120	120
		± 4.0	1.7	2.5	3.0	1.5–200	150
Superelevation	± 225	± 5	–	–	–	–	0	0.1
Cross level	± 50	± 1.0	0.4	0.5	0.7	–	0	0.1
Twist	± 50	± 1.0	0.4	0.5	0.7	3	0	–
Curvature	–	5 × 10–5 m−1	–	–	–	–	25	–
Gauge change rage	± 10‰	0.2‰	–	–	–	3	0	–
Lateral& vertical accelerations (car body)	± 10 m/s2	± 0.01 m/s2	–	–	–	–	0	0.001 m/s2
Train speed	0–450 km/h	± 1 km/h	–	–	–	–	–	0.1 km/h
Gradient	± 40‰	± 1‰	–	–	–	–	25	0.1‰
Vertical curve curvature	–	5 × 10–5 m−1	–	–	–	–	25	–

The maximum wavelength value of 200 m is selected for several reasons. Numerical simulations indicate that the primary vibration frequency of a high-speed train body is around 1 Hz, and the sensitive wavelength of long-wave irregularities at a speed of 400 km/h is generally within 200 m [80]. When the train operates at speeds between 250 and 400 km/h, the vibration characteristics of the train are sensitive to the following wavelengths of track geometry irregularities: longitudinal level (80–160 m), alignment (40–120 m), cross level (50–160 m), and twist (40–100 m). By achieving dynamic measurement of track irregularities using 200 m as the cut-off wavelength, all types of track geometry irregularity wavelengths can be included. Additionally, this measurement set considers the future condition of increasing train speeds to 450 km/h.

Setting a 200 m cut-off wavelength for track irregularities is insufficient for comprehensive track quality inspection, necessitating additional modifications. Existing track inspection systems typically have a maximum cut-off wavelength of 120 m for track irregularities. Extending the cut-off wavelength alone does not satisfy accuracy requirements. Firstly, traditional segmented inertial measurement units, installed at various positions on the vehicle, can suffer from measurement errors due to relative vibration between the vehicle body and the frame. Secondly, component ageing can lead to accuracy degradation in analog circuit filter units over time, a process that is challenging to monitor. To address these issues, the new system architecture adopts a strapdown and digital inertial measurement unit, improving the accuracy of the core sensor, the fibre optic gyroscope.

In terms of algorithms, the measurement error of long-wave irregularities primarily arises from the displacement projection error caused by the measurement error of the train’s attitude (i.e. the train body relative to the track, including pitch, roll, and yaw). This error increases with integration time. Traditional methods, which rely on conventional complementary recursive filters, use inclinometers to correct for gyroscope integration drift. However, at high speeds of 400 km/h, this approach can result in low-frequency trends.

The new algorithm architecture, illustrated in Fig. 5, incorporates train operation information and the ‘dynamic zero-speed’ assumption. It introduces odometer measurements of train speed and enhances the accuracy of train attitude measurements using an extended Kalman filter framework. This framework dynamically estimates the biases of gyroscopes and accelerometers, thereby improving the accuracy of train inertial measurements. Additionally, a track geometry parameter synthesis algorithm integrates distance measurements from computer vision results with inertial measurement results, enabling the measurement of track irregularities with a cut-off wavelength of 200 m.

[See PDF for image]

Fig. 5

Algorithm architecture of 200 m cut-off wavelength of track irregularity

The track quality index (TQI) is used for assessing track geometry irregularity and is evaluated using the standard deviation method. The calculation of TQI involves considering a 200-m section of the track as a unit and computing the standard deviation for the amplitudes of five key track geometry indicators. These indicators include gauge, alignment (for both left and right rails), longitudinal level (for both left and right rails), superelevation, cross level, and twist. Each of these indicators contributes to a single index, and the TQI is derived by summing these seven individual indices, providing a comprehensive assessment of track irregularity.

The specific calculation equations for TQI, which detail the mathematical process of determining each standard deviation and their aggregation, are

1

$${\overline{x}}_i=\frac{1}{n}\sum _{j=1}^n{x}_{\mathit{ij}},$$

2

$${\sigma }_i=\sqrt{\frac{1}{n}\sum _{j=1}^n{\left(x_{\mathit{ij}}-{\overline{x}}_i\right)}^2},$$

3

$$\text{TQI}=\sum _{i=1}^7{\sigma }_i,$$where $x_{\mathit{ij}}$ is the measured track irregularity value (TQI factors) at the jth sampling point of the ith TQI factor; ${\overline{x}}_i$ is the arithmetic mean of the ith TQI factor; ${\sigma }_i$ is the standard deviation of the ith TQI factor; and n is the number of sampling points. These equations are designed to accurately capture the variations and deviations in track geometry, therefore enabling a precise evaluation of track condition.

The track inspection sub-system employs the inertial reference method for conducting track geometry inspections. This sub-system comprises several key components, including a detection beam, two displacement measurement units, and an inertial measurement unit, designed to work together, as shown in Fig. 6.

[See PDF for image]

Fig. 6

Sensor structure of track geometry inspection sub-system

The detection beam plays a pivotal role in this set-up. It is mechanically connected and securely affixed to the end of the inspection car bogie. To ensure that measurements are as accurate as possible, shock absorbers are installed at the connection points. These pads are crucial as they significantly mitigate the impact of vehicle vibrations on the measurement accuracy. The displacement measurement unit, in conjunction with the inertial measurement unit, works together to capture detailed track geometry data (e.g. alignment, cross level, etc.) used for evaluating the track’s condition [42].

More specific information about this sub-system is explained as follows. The displacement measurement unit utilizes laser technology for obtaining track geometry parameters. For example, the measurement and calculation of gauge employ the principle of laser triangulation. The imaging principle is illustrated in Fig. 7a. Laser lights are actively emitted onto the railhead and rail waist, forming a light band. By processing the light band data (Fig. 7b), the centre of the light band is identified, and the distance between the displacement measurement unit and the rail is calculated. The gauge values are then derived by calculating the distance between points located 16 mm below the maximum points on the top surfaces of the left and right rails.

[See PDF for image]

Fig. 7

Sensor cooperation of track geometry inspection sub-system for the measurement of track geometry parameters: a principle of gauge measurement with displacement measurement unit; b illustration of processing laser light bands on rail; c cooperation of displacement measurement unit and inertial measurement unit; d principle of longitudinal level calculation

The longitudinal level is measured and calculated with the following principle. As shown in Fig. 7d, the distance between the IMU and the rail (denoted as h) is measured using the similar principle as that used for measuring the gauge. Conversely, the variable used for calculating the longitudinal level, denoted as z, represents the movement of the IMU. This variable z is calculated by double integrating the acceleration, denoted as a detected by the accelerometer in the IMU. The longitudinal level, denoted as y, is equal to the difference between the vertical motion of the IMU and the relative distance between the IMU and the rail, h.

For the measurement and calculation of the track geometry parameter—cross level, both the displacement measurement unit and the inertial measurement unit are required. The principle is illustrated in Fig. 7c. In the figure, l_al and l_ar refer to the two points that are used to measure the gauge. The distance between the top surfaces of the two rails, denoted as G, is typically set or measured at 1500 mm. The variable h represents the cross level. The calculation of the cross level is based on the following equations:

4

$$h=Gsin{\varphi }_{\text{t}},$$

5

$${\varphi }_{\text{t}}={\varphi }_{\text{b}}-{\varphi }_{\text{bt}}.$$

The angle ϕ_t is calculated as the difference between the detection beam angle to the ground ϕ_b and the detection beam angle to the rail plane ϕ_bt. The angle ϕ_b is measured using the gyroscope and inclinometer (horizontal acceleration) of the inertial measurement unit. The angle ϕ_bt is obtained through measurements taken by the laser video system, using the following equation:

6

$${\varphi }_{\text{bt}}\approx \frac{\mathit{ab}}{\mathit{oa}}\approx \frac{l_{\text{pr}}-l_{\text{pl}}}{l_{\text{d}}},$$where l_pl is the vertical distance between the left displacement measurement unit and the left rail top surface; l_pr is the vertical distance between the right displacement measurement unit and the right rail top surface; and l_d is the distance between the left and right displacement measurement units. This set-up is also shown in Fig. 7c.

The track geometry parameter, twist, is calculated using cross level values by subtracting a sequence of cross level values over a certain base length.

Short-wave track irregularity detection

Weld defects, rail corrugation, and rail surface issues such as squats represent forms of short-wave track irregularities that can induce abnormal high-frequency vibrations in wheel–rail interaction. These irregularities, if severe, can lead to damage in the components involved in wheel–rail interaction, posing a threat to train operation safety. When considering the wheelset as a rigid body, the axle box acceleration provides a more direct reflection of the abnormal vibrations resulting from these short-wave track irregularities. As the train traverses areas of rail corrugation and poor weld joints, the axle box acceleration signal exhibits high-frequency impact characteristics. Furthermore, changes in train running speed can also result in unsteady signal characteristics.

To supplement track condition inspections, the CIT450 employs accelerometers mounted on the axle box to measure vertical vibration acceleration, focusing on high-frequency components. This approach aids in detecting short-wave defects [47]. The CIT450 axle box is equipped with an accelerometer, which has the resolution 2.5 mm/s², as depicted in Fig. 8.

[See PDF for image]

Fig. 8

Accelerometer and sensors/cables installed on the axle box [81]

The CIT450 uses the developed track impact index (TII) to evaluate short-wave track irregularity, which relies on a recursive mapping relationship in the axle box acceleration calculations. The process for the determining of TII is explained as follows:

1. Filter the axle box acceleration based on its frequency distribution characteristics related to short-wave track irregularities. The band-pass filter frequency is typically set between 10 and 500 Hz.

2. Calculate the mobile root mean square (RMS) of the filtered axle box acceleration, by

   7

   $${\text{RMS}}_i=\sqrt{\left({\text{RMS}}_{i-1}^2+x_{i+K-1}^2-x_{i-1}^2\right)/K},$$where $i=\text{1,2},\cdots N-K+1$, and $K$ is a moving window length.

3. Normalize the mobile RMS to get the TII by

   8

   $$\text{TII}={\text{RMS}}_i/\overline{\text{RMS}}.$$

4. Divide the TII based on track line sections with a unit length of 50 m, in terms of mileage.

5. Extract the maximum TII in every section, and record the corresponding mileage.

6. Evaluate the short-wave track irregularity according to the maximum TII.

A test case for a particular railway line illustrates the effectiveness of the developed index, TII. Figure 9a, the waveform result of the TII, shows a peak TII value exceeding the threshold of 6 (as defined in [82]), indicating a poor rail joint. A field on-site re-examination, shown in Fig. 9b, reveals severe fish scale defects and squats on the rail surface around the rail joint. After a maintenance, the retest of the TII shows a significantly reduced maximum value of 3.2, which confirms the method's effectiveness in the diagnose of the short-wave track impact irregularities.

[See PDF for image]

Fig. 9

Track impact index— a test case for a certain railway line: a track impact index result demonstration; b verification on the site

To detect rail corrugation, the rail corrugation index (RCI) method has been proposed [47]. The calculation of this index shares similarities with the TII, involving the computation of the mobile RMS and the mean value of the band-pass-filtered axle box acceleration data. However, the RCI method differs in its filtering approach and frequency band. The filter of axle box acceleration is based on the wavelength of the corrugation, with the filtering range tailored to the common wavelengths of rail corrugation found in China’s high-speed railways, typically ranging from 40 to 150 mm and 150 to 300 mm.

For other situations like standard speed lines and heavy haul railway lines with different wave lengths, this method also can still be applied, and the filtering frequency determines through the conversion of the train’s running speed. Additionally, the RCI method introduces an energy factor to quantify the concentration of vibration energy from the axle box acceleration. This factor helps in assessing the severity of the rail corrugation. The energy factor is calculated from the axle box acceleration data in sections where the corrugation index exceeds its threshold, focusing on the energy within a 3 Hz range around the main frequency compared to the total energy of the section. A higher energy factor suggests more pronounced periodicity in the short-wave irregularities, indicating more severe corrugation characteristics. The rail corrugation condition is determined by evaluating both the corrugation index and the energy factor against their respective thresholds. When both exceed the set thresholds, it indicates a poor rail corrugation condition. Further details about the energy factor and rail corrugation index, including the thresholds for these indices, are available in [47].

A specific test case for a railway line demonstrates the method effectiveness. In Fig. 10a, the waveform result of the rail corrugation index shows a peak value surpassing the threshold of 9, with an energy factor exceeding 0.45, indicating rail corrugation at that location. Field verification, as illustrated in Fig. 8b, confirms severe corrugation on the rail surface. After implementing rail maintenance, such as rail grinding, a follow-up test records a significantly reduced maximum value of the rail corrugation index at 1.4, affirming the effectiveness of the RCI method in accurately diagnosing rail corrugation.

[See PDF for image]

Fig. 10

Rail corrugation index— a test case for a certain railway line: a rail corrugation index calculation results; b verification on the site

Catenary inspection

The inspection of the catenary system primarily employs methods such as visual measurement, vibration detection, and video capture. The main aspects of this inspection include evaluating the catenary height/overhead contact line height, catenary stagger, rigid points, among others [56], as detailed in Table 4. These elements are crucial for ensuring the proper functioning and safety of the catenary system.

Table 4. Inspection indicator and parameters for catenary system

Inspection indicator	Measurement range	Resolution	Maximum permissible error
Catenary height (mm)	5000–7000	1	± 10
Catenary stagger (mm)	−625–625	1	± 10
Catenary wire spacing (mm)	0–800	1	± 20
Vertical spacing between contact wires (mm)	0–200	1	± 20
Catenary mast location	–	–	5%
Span (m)	0–80	–	5%
Rigid point (m/s2)	0–980	10	± 10
Pantograph–catenary contact force (N)	0–500	1	± 5
Arc duration (ms)	0–500	1	± 2
Contact catenary voltage	0–31.5 kV	10 V	± 100 V
Trainset grid-side current (A)	0–1000	1	± 10

Table 4 provides a detailed breakdown of the specific inspection indicators and their corresponding parameters. This table serves as a comprehensive guide, outlining the range and precision of each indicator, as well as the maximum permissible errors.

The catenary inspection sub-system is comprehensively equipped with three distinct parts of inspection equipment, focusing on pantograph–catenary dynamic interaction, catenary geometry parameters, and pantograph video capture. Additionally, a catenary dynamic index (CDI) has been developed to evaluate the quality of the catenary system.

Pantograph–Catenary Dynamic Interaction

This aspect of the inspection includes critical parameters such as pantograph–catenary contact pressure, rigid point (vibration acceleration), and arcing (sparking). The pantograph–catenary contact pressure measurement is primarily conducted through pressure sensors situated between the pantograph slide plate and the power supply bracket, as depicted in Fig. 11. This parameter is vital for assessing the efficiency and safety of the electrical power transfer from the catenary to the train.

[See PDF for image]

Fig. 11

Pantograph–catenary contact pressure measurement and corresponding forces/pressures

Arcing is a phenomenon that occurs during the pantograph’s current collection process, typically when the pantograph slider mechanically disengages from the contact wire, as illustrated in Fig. 12. Arcing is not only indicative of the quality of pantograph current collection but also influences the wear and service life of both the pantograph slider and the contact wire.

[See PDF for image]

Fig. 12

Arcing (sparking) phenomenon occurred in-site railway line

The detection of arcing is achieved using ultraviolet arcing sensors equipped with photomultiplier tube technology and an arcing induction sensor (or arc detection sensor), as shown in Fig. 13a. For the pantograph arcing inspection, the system is designed to convert the ultraviolet light signal from arcing into an electrical signal in accordance with the EN50317 standard. This conversion is output as an analog pulse, as demonstrated in Fig. 13b. Concurrently, the system performs high-speed AD sampling of this analog signal, enabling quantitative analysis and processing of the pantograph arcing information.

[See PDF for image]

Fig. 13

Principle of arcing (sparking) detection: a arcing induction sensor; b measurement results demonstration

Catenary geometry inspection

The main geometry parameters of the catenary assessed by the CIT450 include catenary stagger and the height of the catenary contact wire. For accurate measurement of these parameters, the CIT450 employs a multi-camera stereoscopic vision measurement method, as shown in Fig. 14. This system consists of a detection beam fitted with four line-array cameras, which is fixed to the top surface of the inspection train. By synchronously capturing images from these cameras and processing them in real time, the system can pinpoint the location of point P (shown in Fig. 14) in the images captured by each camera. The vertical and lateral displacements of point P in the rooftop coordinate system are then calculated using the disparity principle [55]. In the figure, G₁, G₂, G₃, G₄ represent four cameras. 2D_L and 2D_R are laser beam distance measurement unit, which is also shown in Fig. 7c—marked as ‘1’.

[See PDF for image]

Fig. 14

Catenary geometry inspection equipment—scheme of the four line-array cameras

Pantograph video collection

The pantograph video monitoring equipment is crucial for monitoring the dynamic operating conditions of the pantograph and assisting in the analysis of catenary status. To cater to the imaging requirements of the high-speed CIT450 under varying weather and environmental conditions, such as during daytime, night-time, and inside tunnels, the CIT450 employs improved imaging technologies like high-speed synchronized stroboscopic and wide dynamic range imaging. These technologies facilitate adaptive adjustment of image acquisition parameters, ensuring the clarity and precision of pantograph video imaging (Figs. 15, 16).

[See PDF for image]

Fig. 15

Pantograph video collection equipment and collected images in different environment: a pantograph–catenary video collection components; b images from open track; c images from tunnel

[See PDF for image]

Fig. 16

Diagram illustrating the catenary dynamic index (CDI) calculation method

Catenary dynamic index (CDI)

The CDI serves as a comprehensive evaluation indicator for the catenary system, derived from the integration of inspection equipment for pantograph–catenary interaction. The CDI assesses the quality of catenary sections based on four key inspection parameters: stagger, contact wire height, pantograph contact force, and arcing time. Each of these parameters plays a significant role in the pantograph–catenary interaction and the efficiency of current collection. The CDI values range from 0 to 10, providing a quantifiable measure of catenary quality.

The evaluation of the CDI is structured into units corresponding to catenary structures, such as anchor sections, phase splitting, and track switches. This method allows for a quantitative description of the dynamic performance of each evaluation unit, offering a detailed and comprehensive assessment of the catenary system’s condition and performance.

5G for railway (5G-R) communication network inspection

In August 2020, the China State Railway Group emphasized the importance of enhancing the construction and utilization of new infrastructure like the 5G communication network. Currently, China’s railways are implementing the 5G railway (5G-R) communication system [75]. The country’s existing railway mobile wireless communication system, the GSM-R system, is transitioning directly to the 5G-R system, skipping over the LTE (4G) technology [55]. The CIT450 has been designed to support a composite inspection function that encompasses GSM-R network detection, 5G-R network detection, and public network coverage detection [83].

The 5G-R system architecture, as shown in Fig. 17, is based on the Stand Alone (SA) network architecture defined by 3rd Generation Partnership Project (3GPP) and includes additional application business systems essential for railway operations, such as intelligent network systems and dispatch communication systems.

[See PDF for image]

Fig. 17

5G for railway communication network system architecture

The 5G-R communication network inspection system categorizes the detection objectives into three key areas: hardware inspection, communication network capacity inspection, and application business inspection, with the system architecture shown in Fig. 18.

• Hardware inspection: This focuses on electromagnetic environment interference and network signal strength coverage. It assesses potential interferences in the network operating environment and verifies whether the signal-to-noise ratio fulfils the requirements for wireless signal transmission and reception. The inspection also includes an analysis of wireless signal strength coverage to gauge the range and intensity of electromagnetic wave coverage.

• Communication network capacity inspection: This area concentrates on network access performance, mobility performance, throughput, network latency, packet loss rate, and other indicators. The goal is to evaluate the quality of network service and carry out network optimization based on these insights.

• Application business inspection: This inspection targets services that require low latency, high reliability, and high bandwidth. It involves selecting representative services like voice, train control, and dispatch command services to ascertain whether the 5G-R quality of service can satisfy the demands of railway transport, train operation control, and dispatching commands.

[See PDF for image]

Fig. 18

5G-R communication network inspection system architecture

For testing the signal strength of the 5G-R communication network, TSME6 test instruments developed by Rohde and Schwarz are employed to collect data on signal strength coverage and the electromagnetic environment. Service quality testing is conducted using a 5G-R test phone or a 5G-R module equipped with signalling analysis capabilities. By integrating wireless communication equipment into the CIT450, functionalities such as dispatch command and train number identification are realized.

Abnormality detection in train operating environment

Because the high-speed train travelling speed is very fast, high-speed railway tracks are fully enclosed by the barriers. With the ongoing growth in transportation demands, the safety of railway lines has become a big concern. Various objects, such as falling rocks, wild animals, bird nests on electric poles, and large textile materials (shown in Fig. 19), can potentially cross the barriers from outside the railway line. These objects, if sucked in by the train’s side, can cause significant damage to train and the power supply system, potentially leading to safety incidents. Such occurrences might result in train delays in less severe cases, and in more serious situations, they could even lead to derailments.

[See PDF for image]

Fig. 19

Foreign objects intruding on the high-speed train operating environment

To address this safety issue, the CIT450 has been equipped with high-definition camera components in the train cab, see Fig. 20, designed to capture video data of the train operating environment. This system is further enhanced with an artificial intelligence analysis system, enabling the processing of the collected video data. This set-up facilitates the automatic detection of foreign objects hanging from the catenary, bird nests on the catenary mounting devices, identification numbers on catenary poles, and any damage to barriers and fences along the track.

[See PDF for image]

Fig. 20

Installation of high-definition camera in car cab

For comprehensive coverage and to ensure the detection of foreign objects from both directions of train operation, video capture equipment is installed in both the first and eighth car cabs. This strategic placement of cameras allows for a thorough monitoring of the train’s surroundings, contributing significantly to the enhancement of safety measures on high-speed railways.

In the development of the artificial intelligence model for the CIT450, several innovative approaches were implemented:

1. Metric learning with supervised and unsupervised methods: As shown in Fig. 21, utilizing videos captured from the high-speed train’s operating environment, metric learning is applied through a blend of supervised and unsupervised learning techniques. The deep convolutional neural network has been enhanced in three key areas: feature fusion, object detection, and loss function optimization. This enhancement facilitates the real-time, automatic detection of foreign objects, such as bird nests on catenary poles, without prior knowledge of the object type, number, shape, or size.

2. Deep learning model for abnormality detection: To tackle the challenges posed by the diverse types of enclosed railway facilities, unpredictable abnormalities, and limited abnormal training input, a deep learning model that combines semi-supervised and unsupervised learning has been developed. This model redefines the task of detecting abnormal targets in videos as a similarity measurement problem among objects of interest, effectively identifying abnormalities like damage to sound barriers (shown in Fig. 22) and protective nets.

3. Feature recognition algorithm: As shown in Fig. 23, a novel algorithm focused on region correlation and an improved SVTR (scaled vector transformed rectifiers) network has been formulated for the recognition of features such as catenary pole numbers. The YOLO v4 network is employed to detect pole areas and number plate regions in single-frame images, particularly in overlapping areas and complex structural patterns. This process allows for accurate positioning of the closest pole and its corresponding number plate region. Furthermore, the SVTR tiny network is used to ensure precise identification of catenary pole numbers, adapting to the diversity in scales and variable character lengths. This system ensures accurate recognition of catenary pole numbers across a variety of scenarios.

[See PDF for image]

Fig. 21

Detection of abnormal bird nests

[See PDF for image]

Fig. 22

Intelligent identification of sound barrier damage

[See PDF for image]

Fig. 23

Process of recognizing catenary pole numbers

These innovative approaches in AI model design significantly enhance the CIT450’s capability to monitor and analyse the railway environment, contributing to improved safety and efficiency in railway operations.

Data association and analysis application

The CIT450 integrated various inspection systems, including track, catenary, communication network, and train operating environment. This integration is decisive for the associative analysis of data from these different systems. This analysis is crucial for leveraging the full potential of a comprehensive inspection platform and maximizing the value of the collected data. The CIT450 can gather and use the data from railway line inspections, system data files, and train operation location information. By automatedly recognizing characteristic features of railway infrastructure for mileage verification, it establishes a unified temporal–spatial benchmark coordinate system. This achieves accurate localization and display of railway infrastructure defects.

Precise localization of defects based on infrastructure feature recognition

One of the key technologies of CIT450 is its ability to precisely localize infrastructure defects on the track, and then convenient for guiding railway line maintenance. This precision is achieved by fusing sensor signals that detect features of various railway system components, enabling accurate pinpointing of defects and flaws. Specific instances of this include:

• Feature point recognition: The CIT450 identifies feature points with the means shown in Fig. 24. This process involves three stages:

• Identification of catenary poles and their numbers through analysis of video footage from the train operating environment.

• Detection of transponders located on the railway track, achieved via the transponder sensing antenna.

• Recognition of easement curve starting and ending points based on the waveform characteristics of track geometry parameters.

• Data timeline analysis: The inspection data are methodically arranged along a timeline, with feature points and identified defect information integrated into this sequence. This approach facilitates precise positioning of railway system component defects along the railway line, allowing for the identification of nearby feature points relevant to the detected defects, as shown in Fig. 24.

[See PDF for image]

Fig. 24

Precise defect localization based on feature point recognition in railway system component: a catenary pole number plate; b infrastructure feature points

These advanced methodologies significantly enhance the accuracy and effectiveness of the CIT450 in detecting and localizing railway infrastructure defects, thereby contributing to more effective maintenance strategies and improved overall safety of the railway system.

Application of associative analysis of multiple sub-system inspection data

The CIT450 utilizes associative analysis of data from multiple sub-systems to enhance the efficiency and accuracy of railway inspections. This process involves the real-time collection of data to the on-board central data server, where it is gathered and processed by the data processing system. Through alignment and association based on temporal and spatial benchmarks, the data are dynamically displayed and analysed in unison. Key applications of this associative analysis include:

• Catenary geometry parameter and pantograph video data association: This involves correlating the waveforms from catenary geometry parameter inspections with data from pantograph video recordings. For instance, the blue box in Fig. 25 identifies the position of a catenary anchor. By analysing images of the anchor from the pantograph video alongside the geometry measurement waveforms, a comprehensive assessment of the anchor’s health condition can be achieved.

• Real-time catenary pole identification and association with geometric waveforms: The system can identify catenary poles and their corresponding numbers from line environment videos in real time. These data are then associated with track and catenary geometric waveforms to precisely locate defects along the railway line.

• Synchronizing train operating environment video with track geometry inspection data: Observing the train as it enters a curve section, as captured in the train operating environment video, and synchronizing this with the changes in waveform characteristics from the track geometry inspection system (indicated by a green box in Fig. 23) enhance the association of inspection data with the railway system component ledger. This synchronization helps in accurately correlating physical railway components with their respective inspection data (Figs. 26–30).

[See PDF for image]

Fig. 25

Application of associative analysis of multiple sub-system inspection data using CIT450

[See PDF for image]

Fig. 26

Track gauge — the minimum distance measured between lines perpendicular to the running at point P

[See PDF for image]

Fig. 27

Longitudinal level — the deviation of the running table levels on any rail from the smoothed vertical position

[See PDF for image]

Fig. 28

Cross level — the difference in height between adjacent running tables from the angle between the running surface and a horizontal reference plane

[See PDF for image]

Fig. 29

Alignment — the deviation in the y-direction of the position of point P on any rail from the smoothed lateral position

[See PDF for image]

Fig. 30

Analysis method — individual defects are represented by the amplitude from the zero-line to the peak value (V₁)

Conclusion and perspectives

Conclusions

The second generation of China’s high-speed comprehensive inspection train, CIT450 integrates eight sub-systems spanning track maintenance, power supply, and signalling. CIT450 can conduct in-service, in-condition state inspections at an unprecedented speed of 450 km/h, and it can simultaneously test and output over 100 railway infrastructure inspection parameters. The following key technological developments in CIT450 are summarised:

• Addressing the previous CIT inefficiencies in railway system inspections, where each data type required dedicated personnel, the CIT450 features a comprehensive control platform. This platform centralizes control over the train inspection systems, monitors system operational states, manages data collection and sharing, and facilitates train-to-ground wireless transmission interaction. This innovation significantly reduces the need for inspection operators and enhances the efficiency of comprehensive inspections.

• A high-precision time and space synchronization benchmark for the train inspection systems has been established by integrating GPS, PTP (Precision Time Protocol) time synchronization networks, and wheel speedometers on the bogies. Utilizing the IEEE1588v2 protocol, microsecond-level synchronization across all systems’ main clocks is achieved. This precise synchronization allows for dynamic, online, and accurate association of multi-disciplinary inspection data, paving the way for advanced multi-source multimodal data associative analysis.

• The inspection equipment incorporates the following functions, including evaluation of 200-m long-wave and short-wave track geometry irregularities, catenary health status evaluation, and 5G for railway (5G-R) communication testing. Enhanced capabilities for detecting abnormalities in the train operating environment are provided through machine vision technology.

• The CIT450 processes various inspection data with machine learning. This includes automatic filtering of abnormal interference data, precise identification of catenary poles, recognition of foreign objects in the train operating environment, detection of damaged fences, and automatic association and alignment of multi-disciplinary inspection data.

Perspectives

Several challenges remain for the future development and research of high-speed inspection trains.

A comprehensive comparison of the CIT450 with other comprehensive inspection trains has not yet been fully made due to the scarce availability of related information in the literature. Consequently, it is challenging to assess its general applicability across various countries, which may differ in geography, weather, safety regulations, and railway asset management policies. It would be beneficial to compare these inspection trains to better understand the future development needs for both high-speed and heavy haul railways.

Current research on the CIT450 primarily concentrates on prototype design and the verification of key technologies within a laboratory setting. Implementing fast inspection with these technologies necessitates a specialized train for equipment installation. Moreover, given the complexity of field conditions, through sustained and extensive inspection efforts, we aim to continuously gather and analyse engineering experiences from all CIT450 sub-systems to ultimately develop a robust product.

Deepening research in intelligent technologies for inspection is crucial. Intelligent inspection, or monitoring, involves a progressive shift from manual operations to machine automation. In the CIT450, we have made some developments in automating the management of comprehensive inspection workflows. Due to these developments, fewer personnel are now required to oversee the entire detection apparatus; however, inspectors continue to play a crucial role in ensuring that inspection operations run smoothly. Moreover, tasks such as data analysis, defect identification, and the issuance of early warnings for defects or hazards still heavily rely on manual intervention. Consequently, realizing full automation of inspection and evaluation represents a vital direction for future research.

Acknowledgements

This work was supported by the National Natural Science Foundation of China (Grant No. 52272427), the Technology Research and Development Program of China National Railway Group (Grant No. K2021T015), and Development Plan of China Academy of Railway Sciences Corporation Ltd. (Grant No. 2022YJ256). Our colleagues in China Academy of Railway Science are acknowledged for their contributions to the development of CIT450, including Qiang Han, Xinyu Du, Jinghui Meng, Boguang Xia, Chengliang Xia, Jikun Wang, Guoyue Liu, and Haoran Wang. We thank Dr. Yunlong Guo, Delft University of Technology, for his help with this paper.

References

1. Sun Y, Zhao H, Yang H et al (2022) Research and design of a multi-track daily inspection robot for urban rail transit. In: 14th International Conference on Measuring Technology and Mechatronics Automation (ICMTMA). Changsha, pp 528–531

2. Jing, G; Qin, X; Wang, H et al. Developments, challenges, and perspectives of railway inspection robots. Autom Constr; 2022; 138, [DOI: https://dx.doi.org/10.1016/j.autcon.2022.104242] 104242.

3. Li D, Read D, Thompson H et al (2010) Evaluation of ground penetrating radar technologies for assessing track substructure conditions. In: Proceedings of AREMA 2010 Annual Conference. Orlando.

4. Wang, H; Silvast, M; Markine, V et al. Analysis of the dynamic wheel loads in railway transition zones considering the moisture condition of the ballast and subballast. Appl Sci; 2017; 7, 12 1208. [DOI: https://dx.doi.org/10.3390/app7121208]

5. Rahimi M, Liu H, Rahman M et al (2021) Localisation and navigation framework for autonomous railway robotic inspection and repair system. In: Proceedings of the 10th International Conference on Through-Life Engineering Services, online event.

6. Li, Q; Shi, Z; Zhang, H et al. A cyber-enabled visual inspection system for rail corrugation. Future Gener Comput Syst; 2018; 79, pp. 374-382. [DOI: https://dx.doi.org/10.1016/j.future.2017.04.032]

7. Takagi, K. Development of high-speed railways in China. Japan Railw Trans Rev; 2011; 57, pp. 36-41.

8. China Railway (2024) The length of China's high-speed railways in operation has reached 45,000 km. http://www.china-railway.com.cn/xwzx/zhxw/202401/t20240112_132652.html. Assessed Aug 4th 2024

9. Kim, S-S; Park, C; Kim, Y-G et al. Parameter characteristics of rail inspection measurement system of HSR-350x. J Mech Sci Technol; 2009; 23, 4 pp. 1019-1022. [DOI: https://dx.doi.org/10.1007/s12206-009-0332-5]

10. Xiong, L; Jing, G; Wang, J et al. Detection of rail defects using NDT methods. Sensors; 2023; 23, 10 4627. [DOI: https://dx.doi.org/10.3390/s23104627]

11. Andani M (2016) The application of Doppler LIDAR technology for rail inspection and track geometry assessment. Dissertation, Virginia Polytechnic Institute and State University

12. Nezu, K; Matsumura, I; Aboshi, M et al. Contactless measuring method of overhead contact line positions by stereo image measurement and laser distance measurement. Q Rep RTRI; 2015; 56, 3 pp. 181-186. [DOI: https://dx.doi.org/10.2219/rtriqr.56.181]

13. Gao, G; Yan, X; Yang, Z et al. Pantograph–catenary arcing detection based on electromagnetic radiation. IEEE Trans Electromagn Compat; 2019; 61, 4 pp. 983-989. [DOI: https://dx.doi.org/10.1109/TEMC.2018.2841050]

14. Niu, L; Yang, F; Deng, X et al. An assessment method of rail corrugation based on wheel–rail vertical force and its application for rail grinding. J Civ Struct Health Monit; 2023; 13, 4 pp. 1131-1150. [DOI: https://dx.doi.org/10.1007/s13349-023-00700-w]

15. Kaewunruen, S; Ishida, M; Marich, S. Dynamic wheel–rail interaction over rail squat defects. Acoust Aust; 2015; 43, 1 pp. 97-107. [DOI: https://dx.doi.org/10.1007/s40857-014-0001-4]

16. Wang, S; Liu, G; Jing, G et al. State-of-the-art review of Ground Penetrating Radar (GPR) applications for railway ballast inspection. Sensors; 2022; 22, 7 2450. [DOI: https://dx.doi.org/10.3390/s22072450]

17. Silvast, M; Nurmikolu, A; Wiljanen, B et al. An Inspection of Railway Ballast Quality Using Ground Penetrating Radar in Finland. Proceedings of the Institution of Mechanical Engineers, Part F: Journal of Rail and Rapid Transit; 2010; 224, 5 pp. 345-351. [DOI: https://dx.doi.org/10.1243/09544097JRRT367]

18. Yang, Y; Zhao, W. Curvelet transform-based identification of void diseases in ballastless track by ground-penetrating radar. Struct Control Health Monit; 2019; 26, 4 [DOI: https://dx.doi.org/10.1002/stc.2322] e2322.

19. Al-Qadi, IL; Xie, W; Roberts, R. Scattering analysis of ground-penetrating radar data to quantify railroad ballast contamination. NDT E Int; 2008; 41, 6 pp. 441-447. [DOI: https://dx.doi.org/10.1016/j.ndteint.2008.03.004]

20. Hou, Z; Zhao, W; Yang, Y. Identification of railway subgrade defects based on ground penetrating radar. Sci Rep; 2023; 13, 6030. [DOI: https://dx.doi.org/10.1038/s41598-023-33278-w]

21. Zhang, J; Lu, Y; Yang, Z et al. Recognition of void defects in airport runways using ground-penetrating radar and shallow CNN. Autom Constr; 2022; 138, [DOI: https://dx.doi.org/10.1016/j.autcon.2022.104260] 104260.

22. Liu, H; Wang, S; Jing, G et al. Combined CNN and RNN neural networks for GPR detection of railway subgrade diseases. Sensors (Basel); 2023; 23, 12 5383. [DOI: https://dx.doi.org/10.3390/s23125383]

23. De Bold RP (2011) Non-destructive evaluation of railway trackbed ballast. Dissertation, University of Edinburgh

24. Balouchi F, Bevan A, Formston R (2016) Detecting railway under-track voids using multi-train in-service vehicle accelerometer. In: 7th IET Conference on Railway Condition Monitoring 2016 (RCM 2016). Birmingham, pp 1–6

25. Unluoglu, HA; Bryson, LS; Rose, JG. Stress distribution in a railroad track at the crosstie–ballast interface. J Transp Eng Part A Syst; 2023; 149, 8 10. [DOI: https://dx.doi.org/10.1061/JTEPBS.TEENG-7761]

26. Wu, Y; Chen, P; Qin, Y et al. Automatic railroad track components inspection using hybrid deep learning framework. IEEE Trans Instrum Meas; 2023; 72, 5011415.

27. Feng, H; Jiang, Z; Xie, F et al. Automatic fastener classification and defect detection in vision-based railway inspection systems. IEEE Trans Instrum Meas; 2014; 63, 4 pp. 877-888. [DOI: https://dx.doi.org/10.1109/TIM.2013.2283741]

28. Liu, J; Huang, Y; Zou, Q et al. Learning visual similarity for inspecting defective railway fasteners. IEEE Sens J; 2019; 19, 16 pp. 6844-6857. [DOI: https://dx.doi.org/10.1109/JSEN.2019.2911015]

29. Datta, D; Scalea, FL. High-Speed inspection of rails by passive ultrasonic monitoring. Journal of Nondestructive Evaluation, Diagnostics and Prognostics of Engineering Systems; 2022; 5, 4 [DOI: https://dx.doi.org/10.1115/1.4055382] 041007.

30. Zheng, J; Wang, L; Liu, J et al. An inspection method of rail head surface defect via bimodal structured light sensors. Int J Mach Learn Cybern; 2023; 14, 5 pp. 1903-1920. [DOI: https://dx.doi.org/10.1007/s13042-022-01736-y]

31. Li, Q; Ren, S. A real-time visual inspection system for discrete surface defects of rail heads. IEEE Trans Instrum Meas; 2012; 61, 8 pp. 2189-2199. [DOI: https://dx.doi.org/10.1109/TIM.2012.2184959]

32. Li HR, Shi YH, Gao B et al (2021) Dynamic electromagnetic thermography system for rail inspection. In: 2021 IEEE Far East NDT New Technology & Application Forum (FENDT). Kunming, pp 99–103

33. Thomas, HM; Heckel, T; Hanspach, G. Advantage of a combined ultrasonic and eddy current examination for railway inspection trains. Insight Non Destr Test Cond Monit; 2007; 49, 6 pp. 341-344.

34. Ren, S; Gu, S; Xu, G et al. A new track inspection car based on a laser camera system. Chin Opt Lett; 2011; 9, 3 pp. 31202-31205. [DOI: https://dx.doi.org/10.3788/COL201109.031202]

35. Wrobel, S.A (2013) Multi-function LIDAR sensors for non-contact speed and track geometry measurement in rail vehicles. Dissertation, Virginia Tech

36. Hao, X; Yang, J; Yang, F et al. Track geometry estimation from vehicle–body acceleration for high-speed railway using deep learning technique. Veh Syst Dyn; 2023; 61, 1 pp. 239-259. [DOI: https://dx.doi.org/10.1080/00423114.2022.2037669]

37. Hsieh, CC; Lin, YW; Tsai, LH et al. Offline deep-learning-based defective track fastener detection and inspection system. Sens Mater; 2020; 32, 10 3429.

38. Hsieh, CC; Hsu, TY; Huang, WH. An online rail track fastener classification system based on YOLO models. Sensors; 2022; 22, 24 9970. [DOI: https://dx.doi.org/10.3390/s22249970]

39. Ge, H; Huat, DCK; Koh, GC et al. Guided wave–based rail flaw detection technologies: state-of-the-art review. Struct Health Monit; 2021; 21, 3 pp. 1287-1308. [DOI: https://dx.doi.org/10.1177/14759217211013110]

40. Fischer, S; Liegner, N; Bocz, P et al. Investigation of track gauge and alignment parameters of ballasted railway tracks based on real measurements using signal processing techniques. Infrastructures; 2023; 8, 2 26. [DOI: https://dx.doi.org/10.3390/infrastructures8020026]

41. Zhao, W; Qiang, W; Yang, F et al. Data-driven ballast layer degradation identification and maintenance decision based on track geometry irregularities. Int J Rail Transp; 2024; 12, 4 pp. 581-603. [DOI: https://dx.doi.org/10.1080/23248378.2023.2228802]

42. Dai P, Wang S, Huang Y et al (2017) Visual indexing of large scale train-borne video for rail condition perceiving. IEICE Trans Inf & Syst E100.D(9):2017–2026

43. Chang, C; Ling, L; Chen, S et al. Dynamic performance evaluation of an inspection wagon for urban railway tracks. Measurement; 2021; 170, [DOI: https://dx.doi.org/10.1016/j.measurement.2020.108704] 108704.

44. Hou, Y; Wang, X; Wei, J et al. Measured dynamic load distribution within the in situ axlebox bearing of high-speed trains under polygonal wheel–rail excitation. Railway Engineering Science; 2024; [DOI: https://dx.doi.org/10.1007/s40534-024-00344-6]

45. Martínez-Llop, PG; de Dios Sanz Bobi J, Solano Jiménez Á, et al. Condition-based maintenance for normal behaviour characterisation of railway car-body acceleration applying neural networks. Sustainability; 2021; 13, 21 12265. [DOI: https://dx.doi.org/10.3390/su132112265]

46. Lederman, G; Chen, S; Garrett, JH et al. A data fusion approach for track monitoring from multiple in-service trains. Mech Syst Signal Process; 2017; 95, pp. 363-379. [DOI: https://dx.doi.org/10.1016/j.ymssp.2017.03.023]

47. Xu, X; Liu, J; Sun, S et al. Dynamic diagnosis method and quantitative characterization of rail corrugation. Proceedings of the Institution of Mechanical Engineers, Part F: Journal of Rail and Rapid Transit; 2022; 237, 3 pp. 297-308. [DOI: https://dx.doi.org/10.1177/09544097221107985]

48. Xu X, Sun S, Yang F, et al (2022) A study on dynamic response character of high-speed railway joint. In: 7th International Conference on Intelligent Transportation Engineering (ICITE). Beijing, pp 122–127

49. Huang, W; Zhang, W; Du, Y et al. Detection of rail corrugation based on fiber laser accelerometers. Meas Sci Technol; 2013; 24, 9 [DOI: https://dx.doi.org/10.1088/0957-0233/24/9/094014] 094014.

50. Xiao B, Mao X, Liu J et al (2021) An improved marginal index method to diagnose poor welded joints of heavy-haul railway. In: 2021 Global Reliability and Prognostics and Health Management (PHM-Nanjing). Nanjing, pp 1–7

51. Choi, S. Identifying parametric models used to estimate track irregularities of a high-speed railway. Machines; 2022; 11, 1 6. [DOI: https://dx.doi.org/10.3390/machines11010006]

52. Yan, T-H; Costa, MDA; Corman, F. Developing and extending status prediction models for railway tracks based on on-board monitoring data. Transportation Research Record: Journal of the Transportation Research Board; 2023; 2677, 6 pp. 708-719. [DOI: https://dx.doi.org/10.1177/03611981221150245]

53. Zhong, J; Liu, Z; Han, Z et al. A CNN-based defect inspection method for catenary split pins in high-speed railway. IEEE Trans Instrum Meas; 2019; 68, 8 pp. 2849-2860. [DOI: https://dx.doi.org/10.1109/TIM.2018.2871353]

54. Zhang W, Mu M, Wang J et al (2021 ) Research on Contact Wire Uplift of Typical High-speed Railway at 300km/h and 350km/h. In: 2nd China International Youth Conference on Electrical Engineering (CIYCEE). Chengdu, pp 1–6

55. Wang, J. Research on wireless communication scheme of train control system compatible with 5G-R and GSM-R interoperability. Railway Signalling & Communication Engineering; 2021; 18, 5 42.

56. Yao L, Qiu J, Gao S et al (2021) Defect detection in high-speed railway overhead contact system: importance, challenges, and methods. In: International Conference on Security, Pattern Analysis, and Cybernetics(SPAC). Chengdu, pp 76–80

57. Du, X; Wu, D. Visual inspection system for trackside communication and signal infrastructure. Proceedings of the Institution of Mechanical Engineers, Part F: Journal of Rail and Rapid Transit; 2020; 235, 1 pp. 121-130. [DOI: https://dx.doi.org/10.1177/0954409719899201]

58. Xu Q, Meng J, Luo Y et al (2021) Uplink transmission performance evaluation and prediction of railway balise based on AHP-WNN. In: 21st International Conference on Software Quality, Reliability and Security Companion (QRS-C). Hainan, pp 169–176

59. Liu, J; Chen, S; Lederman, G et al. Dynamic responses, GPS positions and environmental conditions of two light rail vehicles in Pittsburgh. Sci Data; 2019; 6, 1 146. [DOI: https://dx.doi.org/10.1038/s41597-019-0148-9]

60. Wang, Y; Li, S; Wang, P et al. A multifunctional electromagnetic device for vibration energy harvesting and rail corrugation sensing. Smart Mater Struct; 2021; 30, 12 [DOI: https://dx.doi.org/10.1088/1361-665X/ac31c5] 125012.

61. Jiang, X; Wang, S. Railway panorama: a fast inspection method for high-speed railway infrastructure monitoring. IEEE Access; 2021; 9, pp. 150889-150902. [DOI: https://dx.doi.org/10.1109/ACCESS.2021.3125645]

62. Li R, Bai Z, Chen B et al (2020) High-speed railway track integrated inspecting by GNSS-INS multisensor. In: 2020 IEEE/ION Position, Location and Navigation Symposium (PLANS). Portland, pp 798–809

63. Hou W (2010) Comprehensive inspection technologies of high-speed railways. In: 2010 Joint Rail Conference. Urbana, pp 323–328

64. Yang, J; Bai, X; Zhang, Z et al. Research on the application of BDS/GIS/RS technology in the high-speed railway infrastructure maintenance. IOP Conf Ser: Earth Environ Sci; 2021; 783, 1 [DOI: https://dx.doi.org/10.1088/1755-1315/783/1/012168] 012168.

65. Liu, C; Wang, D; Lin, P et al. Research on combined location method of dual rail inspection vehicle based on adaptive Kalman filter. J Phys: Conf Ser; 2020; 1626, 1 012098.

66. Liu J, Wei Y, Bergés M et al (2019) Detecting anomalies in longitudinal elevation of track geometry using train dynamic responses via a variational autoencoder. In: Sensors and Smart Structures Technologies for Civil, Mechanical, and Aerospace Systems 2019. Denver

67. Hisa, T; Kanaya, M; Sakai, M et al. Rail and contact line inspection technology for safe and reliable railway traffic. Hitachi Review; 2012; 61, 7 pp. 325-330.

68. Matsuda, H; Takikawa, M; Nanmoku, T et al. Track test monitoring system using a multipurpose experimental train. WIT Transactions on The Built Environment; 2010; 114, pp. 701-708. [DOI: https://dx.doi.org/10.2495/CR100641]

69. Weston, P; Roberts, C; Yeo, G et al. Perspectives on railway track geometry condition monitoring from in-service railway vehicles. Veh Syst Dyn; 2015; 53, 7 pp. 1063-1091. [DOI: https://dx.doi.org/10.1080/00423114.2015.1034730]

70. Roberts, C; Goodall, RM. Strategies and techniques for safety and performance monitoring on railways. IFAC Proceedings Volumes; 2009; 42, 8 pp. 746-755. [DOI: https://dx.doi.org/10.3182/20090630-4-ES-2003.00123]

71. Marijolovic, K (2022) A diamond on the tracks: FS reveals new predictive diagnostics train. https://decode39.com/3045/diamond-fs-diagnostics-train-luigi-ferraris/. Assessed Aug 4th 2024

72. Egelkraut, K. On-Site ultrasonic rail testing. Non-Destr Test; 1968; 1, 5 pp. 297-305. [DOI: https://dx.doi.org/10.1016/0029-1021(68)90035-2]

73. Wei, Z; Sun, X; Yang, F et al. Carriage interior noise-based inspection for rail corrugation on high-speed railway track. Appl Acoust; 2022; 196, [DOI: https://dx.doi.org/10.1016/j.apacoust.2022.108881] 108881.

74. Han, Y; Liu, Z; Lyu, Y et al. Deep learning-based visual ensemble method for high-speed railway catenary clevis fracture detection. Neurocomputing; 2020; 396, pp. 556-568. [DOI: https://dx.doi.org/10.1016/j.neucom.2018.10.107]

75. He, R; Ai, B; Zhong, Z et al. 5G for railways: next generation railway dedicated communications. IEEE Commun Mag; 2022; 60, 12 pp. 130-136. [DOI: https://dx.doi.org/10.1109/MCOM.005.2200328]

76. Xu, W; Dai, X; Zhao, C et al. Parallel testing for centralized traffic control systems of intelligent railways. IEEE Trans Intell Veh; 2023; 8, 9 pp. 4249-4262. [DOI: https://dx.doi.org/10.1109/TIV.2023.3305543]

77. Zhao H, Li Y, Wu S, et al (2022) An MPI programming model for fast bird nest detection on the railway catenary. In: 13th International Symposium on Parallel Architectures, Algorithms and Programming (PAAP). Beijing, pp 1–6

78. Gao, L; Jiu, Y; Wei, X et al. Anomaly detection of trackside equipment based on GPS and image matching. IEEE Access; 2020; 8, pp. 17346-17355. [DOI: https://dx.doi.org/10.1109/ACCESS.2020.2966783]

79. Ma, S; Liu, X; Zhang, B et al. Differential deformation identification of high-speed railway substructures based on dynamic inspection of longitudinal level. Sensors (Basel); 2022; 23, 1 219. [DOI: https://dx.doi.org/10.3390/s23010219]

80. Xiao, X; Xu, H; Yang, Y. Analysis of the influence of track irregularity on high-speed train ride comfort. Veh Syst Dyn; 2023; 62, 7 pp. 1658-1685. [DOI: https://dx.doi.org/10.1080/00423114.2023.2250888]

81. Xu, X; Sun, S; Niu, L et al. An approach for the estimation of vertical wheel/rail force using dynamic signals. Veh Syst Dyn; 2024; 62, 4 pp. 1022-1036. [DOI: https://dx.doi.org/10.1080/00423114.2023.2214256]

82. Xu X, Liu J, Sun S et al (2018) Track short wave irregularity of high-speed railway insulated joint evaluation based on track impact index method. In: Proceedings of the Asia-Pacific Conference on Intelligent Medical 2018 & International Conference on Transportation and Traffic Engineering 2018. Beijing, pp 6–10

83. Liu T, He D, Guan K et al (2022) Channel characterization for 5G-R indoor communication at 2.1 GHz. In: 16th European Conference on Antennas and Propagation (EuCAP). Madrid, pp 1–5